Print 1

How Credit Card Payments Increase Unhealthy Food Purchases: Visceral Regulation of Vices
MANOJ THOMAS KALPESH KAUSHIK DESAI SATHEESHKUMAR SEENIVASAN

Some food items that are commonly considered unhealthy also tend to elicit impulsive responses. The pain of paying in cash can curb impulsive urges to purchase such unhealthy food products. Credit card payments, in contrast, are relatively painless and weaken impulse control. Consequently, consumers are more likely to buy unhealthy food products when they pay by credit card than when they pay in cash. Results from four studies support these hypotheses. Analysis of actual shopping behavior of 1,000 households over a period of 6 months revealed that shopping baskets have a larger proportion of food items rated as impulsive and unhealthy when shoppers use credit or debit cards to pay for the purchases (study 1). Follow-up experiments (studies 24) show that the vice-regulation effect of cash payments is mediated by pain of payment and moderated by chronic sensitivity to pain of payment. Implications for consumer welfare and theories of impulsive consumption are discussed.
he past two decades have witnessed a rapid increase in obesity among U.S. consumers. According to the Center for Disease Control, 34% of U.S. adults are obese, up from 23% in 1988. An additional 33% are overweight (Ogden et al. 2006). These results suggest that the consumption of unhealthy food is increasing and have prompted
Manoj Thomas is assistant professor of marketing at Cornell University, 353 Sage Hall, Ithaca, NY 14850 (mkt27@cornell.edu). Kalpesh Kaushik Desai is associate professor of marketing at State University of New York, Binghamton (kdesai@binghamton.edu). Satheeshkumar Seenivasan is a doctoral candidate at State University of New York, Buffalo (ss383 @buffalo.edu). This article has beneted from stimulating discussions of related research papers in the behavioral marketing journal club at Cornell University. The authors gratefully acknowledge Robert Frank, Sachin Gupta, and the review team at JCR for their helpful comments, and the Center of Relationship Marketing, State University of New York at Buffalo, for making the scanner panel data available for the eld study. The authors thank Napatsorn Jiraporn and Bora Park for assisting with data collection and Barbara Drake for proofreading the manuscript. Baba Shiv served as editor and Joel Huber served as associate editor for this article. Electronically published October 6, 2010
several researchers to examine the factors that inuence consumers decisions to buy unhealthy food. Intriguingly, this period has also witnessed an increase in relatively painless forms of payment such as credit and debit cards (Humphrey 2004; Nilson Report 2007). The share of cash in consumer payments has fallen by a third, from 31% in 1974 to 20% in 2000. Cards are replacing cash as the preferred mode of payment; about 40% of purchases in 2006 were made using credit and debit cards. The average American carries 4.4 cards in his/her wallet. These trends raise important, but hitherto unaddressed, questions: Does the mode of payment inuence consumers ability to control their impulsive urges? Are consumers more likely to buy unhealthy food products when they pay by credit or debit cards than when they pay in cash? We address these questions from a psychological perspective in this research. Our conceptualization and hypotheses draw on two distinct streams of research: the literature on impulsive consumption (Baumeister 2002; Hoch and Loewenstein 1991; Khan and Dhar 2006; Kivetz and Keinan 2006; Loewenstein 1996; Metcalfe and Mischel 1999; Raghunathan, WalkerNaylor, and Hoyer 2006; Ramanathan and Menon 2006; Rook 1987; Shiv and Fedorikhin 1999; Vohs and Heatherton
126
2010 by JOURNAL OF CONSUMER RESEARCH, Inc. Vol. 38 June 2011 All rights reserved. 0093-5301/2011/3801-0001$10.00. DOI: 10.1086/657331
This content downloaded from 59.92.103.7 on Wed, 27 Mar 2013 16:38:50 PM All use subject to JSTOR Terms and Conditions
VISCERAL REGULATION OF VICES
127
2000; Wertenbroch 1998; Ubel 2009) and the literature on the psychological effects of different modes of payment (Feinberg 1986; Mishra, Mishra, and Nayakankuppam 2006; Prelec and Loewenstein 1998; Raghubir and Srivastava 2008, 2009; Soman 2001; Soman and Cheema 2002). Based on the rst stream of literature, we suggest that some unhealthy food products trigger impulsive purchase urges because of the desire activated by emotive imagery and associated sensations. Consumers impulsively buy such products even though they consider the products to be unhealthy and experience regret after the purchase. Following Wertenbroch (1998; also see Kivetz and Keinan 2006), we characterize such food items as vice products. Based on the second stream of literature, we propose that mode of payment can inuence decisions to purchase these vice products. Specically, paying in cash feels more painful than paying by credit or debit card. This pain of paying in cash can curb impulsive responses and thus reduce the purchase of such vice products. The notion that mode of payment can curb impulsive purchase of unhealthy food products is substantively important. The epidemic increase in obesity suggests that regulating impulsive purchases and consumption of unhealthy food products is a steep challenge for many consumers. Several factors contribute to this: the automatic nature of visceral responses to vice products (Shiv and Fedorikhin 1999), the depletability of cognitive resources that override these visceral responses (Vohs and Heatherton 2000), the chronicity of impulsive goals (Puri 1996; Ramanathan and Menon 2006), the belief in the unhealthy p tasty intuition (Raghunathan et al. 2006), and biases induced by contextual factors (Cheema and Soman 2008; Wansink 2006; Wansink and Chandon 2007). Given that many consumers struggle to regulate their impulsive responses, the nding that mode of payment could serve as a self-regulation tool has substantive relevance for consumer welfare. Our conceptualization and empirical results augment the extant literature in several ways. First, we show that the effects of mode of payment are contingent on the type of product. Prior research has demonstrated that pain of payment affects the willingness to spend money (see Prelec and Loewenstein 1998); our results qualify this nding by demonstrating that relative to deliberative purchase decisions, impulsive purchase decisions are more likely to be inuenced by pain of payment. This result calls for a more nuanced conceptualization of the effects of mode of payment on consumer behavior. Second, our results contribute to the debate on whether impulsive decisions to purchase unhealthy products can be characterized as rational choices. Some economists have argued that the decision to buy unhealthy food is a rational choice (see Ubel [2009] for a discussion of this debate). Per the rational choice model, consumers buy unhealthy food items because their utility from immediate consumption exceeds the disutility from long-term unhealthiness. The nding that purchases of impulsive products (e.g., cookies) are inuenced by pain of payment while those of planned products (e.g., oatmeal) are
not implies that these two types of purchase decisions cannot be characterized by the same rational choice model. This result supports Loewensteins (1996) suggestion that descriptive choice models should incorporate the effects of visceral states. Finally, our results suggest that self-control is not entirely volitional; it can be facilitated or impeded by seemingly unrelated contextual factors that inuence visceral responses.
CONCEPTUAL FRAMEWORK Impulsive Purchase of Vice Products

Although consumption of unhealthy food products can be caused by several factors, such as faulty beliefs and lack of knowledge, impulsivity seems to be one of the, if not the most, inuential antecedents of unhealthy food consumption. Most scholars who have attempted to conceptualize impulsivity concur that impulsive purchase decisions are based on spontaneous desires elicited by emotive imagery and associated sensations and that such decisions could be inconsistent with ones long-term plans and goals (Baumeister 2002; Hoch and Loewenstein 1991; Kivetz and Keinan 2006; Loewenstein 1996; Metcalfe and Mischel 1999; Rook 1987; Scott et al. 2008; Wertenbroch 1998). For example, Baumeister (2002, 670) dened impulsive purchasing as behavior that is not regulated and that results from an unplanned spontaneous impulse. In particular, impulsive purchasing involves getting a sudden urge to buy something without advance intention or plan and then acting on that impulse without carefully or thoroughly considering whether the purchase is consistent with ones long range goals, ideals, resolves, and plans. Adopting a similar perspective, in this paper we examine consumers impulsive purchases of unhealthy food products. It is reasonable to assume that most people cherish long and healthy lives. So, when in a reective frame of mind, most people would want to minimize their consumption of food items that they consider to be unhealthy. However, their purchase decisions are not always based on such deliberative thinking. When consumers encounter vice productssuch as cookies, cakes, and piesthe emotive imagery and associated desire trigger impulsive purchase decisions (Loewenstein 1996; Shiv and Fedorikhin 1999; Wertenbroch 1998). These visceral factors can prod them to include such vice products in their shopping baskets even though they consider such products to be unhealthy. Since the desire that triggers impulsive behavior is caused by visceral factors, it can be weakened by other aversive visceral factors. Aversive visceral responses, such as feelings of pain, can extinguish consumptive desires. With the extinction of desire, vice products no longer seem so appealing. Stated differently, the desire to consume a vice product is based on a visceral state, and it recedes with aversive visceral responses. This notion is consistent with Metcalfe and Mischels (1999, 11) assertion that internal activation of irrelevant hot nodes allows the diversion of considerable cognitive-affective energy and hence serves as
128
JOURNAL OF CONSUMER RESEARCH
an effective distracter. This line of reasoning implies that environmental factors that trigger feelings of pain can dissipate impulsive urges and thus curb impulsive purchases. We propose that mode of payment is one such environmental factor: pain of payment can reduce the pleasure of anticipated indulgence and thus curb impulsive purchases.
liberative evaluations. This implies that the proportion of vice products relative to virtue products in the basket will change depending on the mode of payment. Based on this reasoning, we hypothesize: H1: The number of unhealthy and impulsive food items (vice products) in the shopping basket will be lower when consumers pay in cash than when they pay by credit cards. Mode of payment will not inuence the number of healthy and deliberative food items (virtue products) in the shopping basket. Further, our conceptualization posits that pain of payment is caused by the form of payment and not by other mechanisms such as payment decoupling or time discounting of delayed payments. This implies that all types of cards credit as well as debitwill reduce pain of payment and thereby increase the purchase of vice products. Like credit card payments, debit card payments are also less vivid and emotionally more inert than cash payments; therefore, they feel less painful. Hence, we expect that debit cards will also weaken impulse control. Formally, H2: Both credit cards and debit cards will reduce pain of payment and thus increase purchases of vice products.
Pain of Paying in Cash

Several researchers have suggested that the mode of payment can inuence pain of payment. Based on their model of hedonic mental accounting, Prelec and Loewenstein (1998) posited that paying in cash elicits greater pain than paying by other modes of payment even when the modes are normatively equivalent. Raghubir and Srivastava (2008) tested the effect of pain of payment on willingness to spend in several experiments. In one of their experiments (study 3), some participants were given a $50 bill while others were given a $50 scrip certicatea certicate whose value is recognized by the payer and payee. Participants then responded to a simulated shopping study. The authors predicted that since paying by the scrip will feel less painful than paying in cash, participants will spend more with the scrip certicate. Consistent with their prediction, participants spent more when they were given scrip than when they were given an equivalent amount in cash. Several other studies offer converging empirical evidence for the proposition that cash payments feel different from other less vivid and emotionally more inert modes of payments (Mishra et al. 2006; Raghubir and Srivastava 2008, 2009; Soman 2001).
Individual Differences in Sensitivity to Pain of Payment

Previous research suggests that pain of payment is not only a situational factor, it is also an individual difference variable. Individuals differ in their sensitivity to pain of payment; some are chronically more sensitive to pain than others. Based on this premise, Rick, Cryder, and Loewenstein (2008) suggest that consumers can be identied as tightwads or spendthrifts. They label as tightwads those consumers whose affective reaction to spending may lead them to spend less than their more deliberative selves would prefer. In contrast, they label as spendthrifts those consumers who experience minimal pain of payment and, therefore, end up spending more than what they themselves would consider as normatively appropriate. Our conceptualization predicts that the vice-regulation effect of cash payments will vary across these two types of consumers. Relative to spendthrifts, tightwads would be more sensitive to pain of payment, and therefore pain of paying in cash will have a larger effect on their consumption of vice products. Formally, H3: The vice-regulation effect of cash payments will be stronger for consumers who are more sensitive to pain of payment. In the following sections, we describe the empirical studies conducted to test these hypotheses.
Deliberative Purchase of Virtue Products

Not all purchase decisions are based on spontaneous impulses. While purchase decisions of vice productssuch as cookies, cakes, and piesare inuenced by spontaneous impulsive responses, purchase decisions of virtue products such as fat-free yogurt and whole wheat breadtend to be more deliberative. Previous research (e.g., Kross, Ayduk, and Mischel 2005) suggests that deliberative thinking can reduce the intensity of negative emotions. Further, since the purchase of virtue products is based on justiable considerations (I am buying something I need), consumers might be able to rationalize or explain away the pain of payment. Based on this reasoning, we predict that while purchasing a virtue product such as fat-free yogurt, a consumer can explain away the pain of paying in cash. However, a consumer might not be able to explain away the pain of payment for a vice product since the purchase decision is based on a visceral response. Consequently, the pain of paying in cash is likely to have a larger effect on purchase decisions of vice than of virtue products. To summarize our discussion thus far, the pain of paying in cash can weaken desires and thus curb impulsive purchases of vice products. However, the pain of paying in cash is less likely to inuence purchase decisions of virtue products because such decisions tend to be based on more de-
129
TABLE 1 DESCRIPTIVE STATISTICS (STUDY 1) Variable Average number of transactions at the focal retailer Proportion of transactions paid with credit cards (%) Proportion of transactions paid with debit cards (%) Proportion of households paying with credit or debit cards on more than 90% of transactions (%) Proportion of households not using credit or debit card at all (%) Average basket size, cash purchases ($) Average basket size, credit card purchases ($) Average basket size, debit card purchases ($) Average basket impulsiveness (19 scale) Average basket unhealthiness (19 scale) Mean values 37.9 40.5 8.9 15.9 13.9 37.9 67.6 60.1 4.1 4.4
STUDY 1: MODE OF PAYMENT AND HEALTHINESS OF SHOPPING BASKET A FIELD STUDY

To seek preliminary evidence for the hypothesized effect of painless payment on consumption of unhealthy food items, we analyzed data from a large retailer that operates several grocery stores in the northeastern region of the United States. The data included information on what products each household in the panel purchased during their trips to the store and the prices of the products. More importantly, this data set is distinct from other typically used scanner panel data sets in that it provides information on whether each purchase was made using a credit card, debit card, or cash. Taking advantage of this unique aspect of the data set, we test whether consumers buy more impulsive and unhealthy food products when they pay by credit cards than when they pay in cash (hypothesis 1). Further, this study also tests whether the proposed effect will manifest for both debit and credit cards or only for credit cards. Our conceptualization predicts that both debit and credit cards will increase the proportion of unhealthy and impulsive food products in the shopping basket, because the card payment format makes payments less vivid and thus reduces the pain of payment (hypothesis 2).
NOTE.Basket size refers to the amount of money spent on the 100 food items included in this study.
Data
Shopping Data. The data cover a period of 6 months (January to June) in the year 2003. We use a random sample of 1,000 loyal single-member households who primarily purchase from the chain stores for our analysis. We restrict our analyses to single-member households because in multimember households, it is not clear whether the observed effects are due to mode of payment or due to differences in shopping behaviors of individual members of the household. For example, a younger member of the household might always use credit cards while an older member might always use cash. In such a situation, it will be difcult to delineate the effect of mode of payment on purchases. By restricting our analyses to single-member households, we can avoid this confound. For these 1,000 households, we obtained data on what products they purchased during the 6-month period on each visit to the store and how they paid during the trip. Table 1 presents a summary of the relevant statistics for the sample. Note that about 41% of the transactions were paid by credit cards and 9% were paid by debit cards. Further, there is a lot of within-household variability in card usage; only 16% of the households used credit or debit cards on more than 90% of their shopping trips whereas 14% always paid in cash. A majority of the households switched between card and cash payments. Since we restrict our analysis to single-member households, these data offer us the opportunity to test our hypothesis on how mode of payment affects consumer behavior. Our primary interest was in assessing whether the use of credit and debit cards increases the proportion of impulsive
vice products in the shopping basket. Since impulsiveness of a product is easier to assess in the context of food categories, we selected 100 major food categories (baby food, tea, gelatin, meat, etc.) based on the sales of the products. These 100 categories were selected only on the basis of their sales and accounted for 73% of the food sales in our sample. Thus, these 100 categories represent the types of food products consumers typically buy. We used the same category labels that the retail store management uses to organize and display the products on the retail shelves.
Survey for Impulsiveness and Healthiness Ratings of Categories. Our analyses required the ratings of these 100
categories on the impulsiveness-planned and healthy-unhealthy scales. Since these ratings were not available in the data set, we obtained them through a survey. Note that our interest is in perceptions of unhealthiness rather than actual unhealthiness of food items. If a consumer does not perceive a product as unhealthy, then she is unlikely to consider it as a vice product and experience postpurchase regret. Seventy-eight undergraduate students at a northeastern U.S. university were asked to rate each of these 100 food categories. To ensure that the results were not biased by a repeatedmeasure design, we asked about one-half of the pretest survey participants to rate the categories on impulsiveness while the other half were asked to rate each category on healthiness. One group of participants (n p 37) were asked to rate each category on planned-impulsive purchase continuum. Specically, they were informed that an impulsive product is one that is usually purchased without prior planning. Even when they do not plan to buy the product, consumers often buy such products spontaneously when they are in a store. In contrast, planned products are based on prior deliberation. Consumers come into the store with the intention of buying such products. Participants evaluated each of the 100 categories on a 9-point scale anchored at 1 p Planned and 9 p Impulsive. Participants responses were
130
averaged to compute the average impulsiveness rating for each category. Beans, barley, rice, baby food, vegetables, milk, and meat were some of the less impulsive categories, and ice cream, candies, cookies, gum, donuts, potato chips, and pudding were some of the more impulsive product categories in the list of 100 categories. The second group of participants evaluated each of the 100 categories on a 9-point scale anchored at 1 p Healthy and 9 p Unhealthy. Their responses were also averaged to compute the average unhealthiness rating for each category. As expected, impulsiveness and unhealthiness ratings collected from separate groups of respondents were highly correlated (r p .59, p ! .01); categories that were rated as impulsive were also rated as unhealthy.
Index ht p b 0 b1 Creditcard_Paymentht b 2 Debitcard_Paymentht b 3 Basket_Size ht b 4 Last_Basket_Sizeht b 5Weekend ht yht . Our focal variables here are the credit card and debit card payment modes that are operationalized as binary dummy variables. Specically, the credit card variable is set to 1 for the trips where the consumer paid using a credit card and 0 for the trips where they used other modes of payment. The debit card variable was operationalized likewise. Since mode of payment is likely to be correlated with other variables that inuence purchase decisions, we control for the available trip-specic variables. We control for basket size on the current trip as well as the previous trip, both measured by the value of the shopping basket. We also control for whether the shopping day is a weekend or weekday using a dummy variable which is set to 1 for weekends. The model was run using clustered regression (for repeated measurements) in SAS separately for each of the indices.
Analysis
Dependent Variables. To test the robustness of our results, we created two separate indices for measuring the unhealthiness and impulsiveness of a shopping basketa simple-average index and a dollar share weighted-average index. The rst index, the average unhealthiness, is based on a simple average of the category level unhealthiness scores for all the categories in the consumers basket during a shopping trip. For instance, if a shopper has 16 categories in the basket, then the average unhealthiness measure is constructed by summing up the unhealthiness ratings of all these 16 categories and dividing the sum by 16. The second index is constructed using share-weighted sum-of-categorylevel scores where the weight is dollar share of that category in the basket. Under this measure, if a category accounts for 60% of the total basket value, then the unhealthiness rating of that category is given a weight of 0.6 in the computation of total basket unhealthiness. While the rst index is inuenced by only the number of unhealthy products, the second one captures the relative amount of money spent on unhealthy products. We used these two indices to test whether mode of payment affects the number of unhealthy categories purchased, the money spent on unhealthy products, or both. For example, if card payments induce consumers to buy more expensive brands without affecting the number of categories, then we expect the mode of payment to inuence only the weighted-average index. These two indices were highly correlated (0.94). Since we are concerned with the unhealthiness as well as the impulsiveness of the basket, these two indices were created for both ratingsthe impulsiveness ratings and the unhealthiness ratings. Thus, we created four different indices to test the effect of mode of payment on purchases: average impulsiveness, weighted-average impulsiveness, average unhealthiness, and weighted-average unhealthiness. Independent Variables. Next, we regressed each of the four indices separately on the mode of payment controlling for some trip-level variables. Our model specication for a household h in shopping trip t is as follows:
Results
The results of the empirical analyses are presented in table 2. Our results are consistent with hypotheses 1 and 2 and indicate that both credit and debit card payments have a signicant positive impact on the impulsiveness and unhealthiness of the basket. These results suggest that consumers are more likely to purchase impulsive and unhealthy food items when they pay with credit or debit cards on the shopping trip. Further, the results are consistent across both the number-based and value-based measures of basket impulsiveness and unhealthiness. The parameters for other predictor variables are generally consistent with our expectations and thus conrm the validity of the estimated parameters. The basket sizes of the current and the last shopping trips have a positive effect, implying that larger-basket shoppers are likely to be more susceptible to impulsive purchase of unhealthy products. This nding is consistent with previous research (Ainslie and Rossi 1998; Bell and Lattin 1998), which suggests that basket size could be a proxy measure of individual differences in pain of payment; consumers who experience more (less) pain of paying tend to have fewer (more) items in their shopping baskets. The day of shopping has a signicant negative effect indicating that consumers shopping on weekends are less likely to be impulsive. This could be because of the shopping list effect: weekend shopping trips tend to be based on shopping lists, and therefore purchases on such trips are less susceptible to impulsive urges. To examine the role of impulsiveness, we tested whether the effect of credit cards on the unhealthiness of the basket

TABLE 2 RESULTS OF ACTUAL SHOPPING BEHAVIOR ANALYSES (STUDY 1) Dependent variable Parameter estimates (SE) Intercept Credit card payment Debit card payment Basket size Last basket size Weekend Pseudo R 2
**p ! .05. ***p ! .01.
131
Average impulsiveness 3.901*** (.015) .129*** (.016) .154*** (.026) .001*** (.0001) .003** (.001) .110*** (.014) .205
Weighted-average impulsiveness 3.888*** (.015) .150*** (.017) .164*** (.028) .002*** (.0001) .004*** (.001) .118*** (.015) .293
Average unhealthiness 4.285*** (.017) .107*** (.018) .156*** (.030) .002*** (.0002) .004*** (.002) .097*** (.016) .273
Weighted-average unhealthiness 4.337*** (.017) .155*** (.019) .176*** (.031) .004*** (.0001) .004*** (.002) .097*** (.017) .492
is stronger for very impulsive products than for planned products. We split the products into two groups based on their ratings on the planned-impulsive scale. Categories with ratings greater than or equal to 6 on the 9-point scale were considered as very impulsive, and categories with ratings less than or equal to 4 were considered planned. This resulted in 19 very-impulsive categories and 46 planned categories (out of 100 food categories in our sample). We ran the weighted-average unhealthiness models for the two sets of categories separately. Consistent with our prediction, there was a signicant difference between the effects of mode of payment on unhealthy purchases for very impulsive and planned categories. The coefcient of credit cards was higher for very impulsive product categories (b p .329) than for planned categories (b p .138), and this difference was signicant (t p 143.9, p ! .01). The same was true for debit cards: b p .326 for very impulsive categories versus b p .179 for planned categories (t p 67.1, p ! .01). These results show that the effect of card payment on the purchase of unhealthy products is stronger when the unhealthy products tend to be more impulsive.
Discussion
Several important conclusions emerge from this study. First, this study offers preliminary support for our hypothesis that painless card payments can weaken impulse control. We observe that consumers tend to have a larger proportion of food items pretested as impulsive and unhealthy in their shopping baskets when they pay by credit card than when they pay in cash. This result supports hypothesis 1. Second, consistent with hypothesis 2, this effect manifested for debit cards also; shoppers who used debit cards had a larger proportion of vice products in their baskets. The nding that debit and credit cards have the same effect, even though debit card payments are immediately charged to the consumers bank account, suggests that it is not discounting of
delayed payments but, rather, it is the abstract and emotionally inert nature of card payments that reduces the pain of payment. Finally, our hypothesis that card payments can increase unhealthy purchases hinges on the assumption that some food items that are commonly considered unhealthy also tend to elicit impulsive purchase decisions. The pretest reported in this study offers a robust test of this assumption. Across 100 food product categories that account for about two-thirds of the value of the food items included in a typical shopping basket, we nd that perceived impulsivity and unhealthiness are highly correlated. Although these results are consistent with our hypotheses, given the nature of the data in this study, we are unable to establish the direction of causality. We do not know whether the consumers make the decision about the mode of payment before adding the products to the basket or whether they decide about the mode of payment after adding the products to the basket. Further, it is likely that mode of payment is correlated with other unobserved variables that might be driving these effects. To test the causal role of pain of payment, we conducted three experiments. These experiments also test the underlying psychological mechanism and rule out some other alternative accounts.
STUDY 2: MANIPULATING MODE OF PAYMENT

In this study, we manipulated the mode of payment in a laboratory experiment holding all other elements constant. If we observe that paying by credit card increases the purchase of unhealthy-impulsive food items in such a setting, then we can uniquely attribute the observed effect to mode of payment. Participants in this experiment responded to a simulated shopping task and were presented with 10 virtue products that were pretested as less impulsive and healthy and 10 vice products that were pretested as more impulsive and unhealthy. Thus, this study used a 2 (payment mode:
132
credit card vs. cash) # 2 (product type: vice vs. virtue) mixed factorial design with 10 replicates each of vice and virtue products. Mode of payment was a between-subjects factor while product type was a within-subjects factor. Additionally, since this was a controlled laboratory experiment, we also measured and analyzed the response time for purchase decisions to get insights into the underlying process.
to the credit card condition saw the logos of four credit card companiesMasterCard, Visa, Discover, and American Expressaccompanied by the statement The store accepts all major credit cards. Such signs are usually displayed at the entrance of most retail stores in the United States. Participants assigned to the cash only condition were informed that the new store accepts only cash payments; credit cards or checks were not accepted at this store.
Method
Participants. One hundred and fty-one undergraduate
and graduate students (47% women) from Cornell University participated in this study for a small sum of money.
Results
Vice Products in Shopping Basket. For each participant, we computed the number of virtue and the number of vice products in the shopping basket. These measures were submitted to a 2 # 2 mixed factorial ANOVA with mode of payment (credit card vs. cash) as a between-subjects factor and type of product (vice vs. virtue) as a withinsubjects factor. The main effect of the type of product (F(1, 149) p 43.92, p ! .01) was qualied by a signicant mode-of-payment by type-of-product interaction (F(1, 149) p 4.52, p p .035). There were more vice products in the basket when the mode of payment was credit card (M p 2.85) than when the mode of payment was cash (M p 2.02; F(1, 149) p 6.15, p p .01). Further, the mode of payment did not affect the number of virtue products in the basket (MCreditCard p 3.85 vs. MCash p 3.97, F ! 1). We also ran a repeated-measures choice model and found that controlling for price differences of vice and virtue products and order of presentation did not mitigate the interaction effect of mode of payment and type of product. Thus, hypothesis 1 is supported. We did a similar analysis for the values of vice and virtue products in the shopping basket. The pattern of results was similar to that reported above. Participants spent more on impulsive vice products when the mode of payment was credit card than when it was cash (MCreditCard p $14.07 vs. MCash p $9.89; F(1, 149) p 6.35, p ! .01). However, the mode of payment did not affect the amount spent on virtue products in the basket (MCreditCard p 17.50 vs. MCash p 17.43; F ! 1). The pattern of means is depicted in gure 1. Response Time for Purchase Decisions. To test our assumption that purchase decisions for vice products are spontaneously elicited, we analyzed the response time for participants purchase decisions. We computed the logarithmic transforms of response times in milliseconds and removed outliers that were more than 3 standard deviations away from the mean. Since the order of presentation was a signicant predictor of response time and accounted for a large proportion of the variance, the analysis was done using an ANCOVA to control for the effects of order. The logtransform of response time was submitted to a mixed factorial ANCOVA with mode of payment (credit card vs. cash) as a between-subjects factor, type of product (virtue vs. vice) as a within-subjects factor, and order of presentation as a covariate. Only the main effects of the order of presentation (F(1, 2,849) p 472.63, p ! .01) and the type of product were signicant (F(1, 149) p 55.27, p ! .01). Participants
Stimuli and Pretest. In this study, participants considered 20 food products, half of which were virtue products while the other half were vice products (see the appendix for the full list). To conrm that our manipulation of product types served the intended purpose, a pretest was conducted with 27 students who did not participate in the main experiment. Twelve subjects rated the impulsiveness (1 p impulsive purchase and 7 p planned purchase), and 15 participants rated the healthiness (1 p unhealthy and 7 p healthy) of 20 food products. The 10 products that we considered as vice products were perceived to be more impulsive (M p 2.68) than the products that we considered virtue products (M p 4.61; t(11) p 7.01, p ! .01). The vice products were also perceived to be less healthy (M p 2.11) than the virtue products (M p 5.44; t(14) p 17.17, p ! .01). Further, the impulsivity ratings and the unhealthiness ratings of the products collected from two separate sets of participants were highly correlated (r p .91, p ! .01). This result corroborates our pretest results from study 1 and suggests that unhealthy products also tend to be impulsive. Procedure. Participants were invited to participate in a Food Shopping Study, which was ostensibly conducted by a large retail chain that was planning to open a store in the town, to understand what types of food consumers buy on a typical shopping trip. They were told to imagine that they were in this new grocery store. The study was administered on computers. All participants saw the 10 vice and 10 virtue food products, one at a time, with the impulsive and virtue products interspersed. The 20 products were ordered randomly using a random number generator and presented sequentially. The products were presented in the same random order to all the participants. On each screen, they saw the name of the product (e.g., Oreo Cookies), a picture of the product (approximately 2 # 2 centimeters), and its price. There were two response buttons at the bottom of the screen. One button had the picture of a shopping basket and was labeled Add to Shopping Cart while the other one was to continue shopping. Participants proceeded to the next screen at their own pace by clicking on one of the two buttons. The computer unobtrusively recorded the response time for each product. The mode of payment was manipulated immediately before the participants saw the products. Participants assigned

FIGURE 1 THE EFFECT OF CREDIT CARDS ON IMPULSIVE PURCHASE DECISIONS (STUDY 2)
133
by the interplay of the positive and negative emotional responses elicited by impulsive products and the pain of paying in cash, respectively. Cash payments are painful, and card payments are relatively painless. The negative emotion elicited by cash payments counters the positive emotion elicited by impulsive products and thus facilitates impulse control. In the following experiment, we directly test this account. We also test whether the reduced pain of payment in the credit card condition is due to inattention to prices.
STUDY 3: MEDIATION BY PAIN OF PAYMENT

This study replicates and extends the results from the previous experiment. This study was designed with three objectives in mind. First, to test the robustness of our results, instead of using students as participants, we gathered responses from a more representative sample of consumers from an online panel. The present experiment was administered as an online study by an independent market research rm to respondents in their online panel. Second, we included questions to measure the pain elicited by the mode of payment. We tested whether this pain of payment mediates the effect of mode of payment on the purchase of impulsive vice food items. If pain of payment mediates the effect of mode of payment on vice purchases but not on virtue products, then this moderated mediation would be direct evidence for the proposed account. Third, we tested whether inattention to prices plays a role in reducing pain of payment for credit card users. Do credit card users experience less pain because they do not pay attention to prices? It is likely that credit card users pay less attention to prices, and this inattention to prices might reduce their pain of payment. If this is the case, then credit card users should be less accurate in recalling the amount of money they spent on the shopping trip in a surprise recall task. Alternatively, is credit card a less vivid and more emotionally inert medium that does not elicit the same pain of payment as cash does? This account posits that even when they carefully process the prices, credit card users do not experience the same pain of paying as consumers who use cash. To test these competing accounts, we asked participants to estimate their expenses in a surprise recall task.
NOTE.Vice products are unhealthy food items purchased impulsively, while virtue products are healthy food items purchased after deliberative evaluations.
took less time to decide about vice products (M p 574 milliseconds per product) than about virtue products (M p 619 milliseconds per product). Neither the effect of mode of payment (F ! 1) nor its interaction with type of product was signicant (p 1 .23). Although response time patterns are not conclusive evidence for the underlying process, these results are consistent with our predictions. These response time results suggest that purchase decisions for vice products were more spontaneous than those for virtue products.
Discussion
These results from the laboratory study support our hypothesis (hypothesis 1) that paying by credit card increases the proportion of impulsive vice products in the shopping basket. Participants assigned to the credit card condition were more likely to buy food products that were rated as impulsive even though these food items also had low healthiness ratings. However, the mode of payment did not affect purchase decisions of virtue products that were based on deliberative decisions. Together, studies 1 and 2 offer convergent evidence that mode of payment can weaken impulse control and thus increase purchases of unhealthy food items. However, this study does not clarify why mode of payment inuences the purchase of unhealthy food items. Our conceptualization assumes that the observed effect is caused
Method
Participants. We used the paid service of an online market research rm to collect data for this experiment. Participants were consumers who were enrolled in the panel of an online market research rm (Qualtrics) for monetary benets. One hundred and twenty-ve participants (48% women) completed the online questionnaires. One participants responses were inconsistent and therefore excluded from the analyses. The average age of participants was 42, and 78% of them had a household income of $30,000 or more per annum.
134
Procedure. The experiment used the same mixed factorial design as that in the previous study: 2 (payment mode: credit card vs. cash) # 2 (product type: vice vs. virtue) design with 10 replicates of each product type. The questionnaire was created using an online software package. This software randomly assigned the participants to one of the two between-subjects conditions, that is, cash or credit card. The cover story, the stimuli, and the simulated shopping format used in this experiment were identical to those in the previous laboratory experiment. However, unlike the previous study, in this study, each participant saw a different random order of presentation of the 20 products. Further, several additional variables designed to measure the process were added to the questionnaire after the shopping task. Attention to Price. Immediately after the shopping task, participants were asked to recall from memory how many items they included in the shopping basket and how much money they spent on the items put in the shopping basket. Participants were prevented from accessing the previous screens on which they made the shopping decisions; therefore, they had to make these estimates based on the prices of the products in their memory. Pain of Payment. Then participants were asked to indicate how they felt about spending money during the shopping trip using two distinct scales. The rst scale measured pain using a nonverbal faces scale. Specically, they responded to the following question: A stores payment policy can inuence how consumers feel while spending money. How did you feel about spending money on this shopping trip? Participants indicated their responses on a 5-point nonverbal faces pain scale with a sad face ( ) at the lower end of the scale and a happy face ( ) at the higher end of the scale. The responses in this scale were reverse coded such that a higher number on this scale indicated more pain of payment. Subsequently, to interpret their responses to the nonverbal scale, they were shown a list of negative feelings and were asked to check all the words that described their feelings. Participants saw eight words: Irritated, Restricted, Annoyed, Powerless, Controlled, Suffocated, Inhibited, and None of the above. For each participant, we computed the number of negative feeling words that were checked. Regret Spending Money. Subsequently, participants were asked to indicate to what extent they would regret spending money on food items if they were to purchase the food items. This measure was collected for all 20 products, one at a time. For each of the 20 products, they predicted their regret on a 5-point scale anchored at 1 p Denitely not regret spending money on this item and 5 p Denitely regret spending money on this item. Healthiness of Food. Participants were asked to rate each of the 20 food items on healthiness, one at a time. Participants rated each product on a 5-point scale anchored at 1p Unhealthy and 5 p Healthy. Finally, participants were asked to submit demographic and lifestyle information
such as gender, age, household income, diet control behavior, and whether they had a medical condition that restricted their diet.
Results
Vice Products in Shopping Basket. As in the previous study, for each participant we computed the total number of virtue and vice products in the shopping basket. This measure was submitted to a 2 # 2 mixed factorial ANCOVA with mode of payment (credit card vs. cash) as a betweensubjects factor and type of product (virtue vs. vice) as a within-subjects factor. Since the data were collected from a heterogeneous sample of consumers, some of the demographic variables (diet control habits, medical condition, and household income) and their interactions with type of product were entered as covariates in this mixed model. The main effect of the mode of payment (F(1, 119) p 4.04, p p .046) was qualied by a signicant mode-of-payment by type-of-product interaction (F(1, 119) p 4.46, p p .036). There were more unhealthy impulsive items in the basket when the mode of payment was a credit card (M p 2.90) than when the mode of payment was cash (M p 1.73; F(1, 119) p 7.60, p ! .01). Further, the mode of payment did not affect the number of virtue items in the basket (MCreditCard p 3.31 vs. MCash p 2.96, F ! 1). These results show that the ndings in the previous study hold even with a more representative sample and thus support hypothesis 1. We did a similar analysis for the values of virtue and vice products in the shopping basket. The pattern of results was similar to that reported above. Participants spent more on vice products when the mode of payment was a credit card than when it was cash (MCreditCard p $14.2 vs. MCash p $8.2; F(1, 119) p 8.06, p ! .01). However, the mode of payment did not affect the amount spent on virtue products in the basket (MCreditCard p 14.3 vs. MCash p 12.5; F ! 1). Pain of Payment. Participants responses to the facespain question and the negative feelings word-list question were submitted to a one-way ANOVA. Participants in the cash condition reported greater feelings of pain of payment (MCreditCard p 2.67 vs. MCash p 3.36; F(1, 122) p 13.35, p ! .01) and used more negative feeling words to describe their feeling (MCreditCard p 0.77 vs. MCash p 1.30; F(1, 122) p 6.42, p p .012) than those in the credit card condition. Moderated Mediation: Pain Mediates the Effect of Cash. Our conceptualization predicts that pain of payment
will mediate the effect of mode of payment on impulsive products, but not virtue products. To test this moderated mediation hypothesis, we ran a series of regressions recommended by Muller, Judd, and Yzerbyt (2005). A schematic summary of the mediation model is presented in gure 2. In the regressions, the cash payment condition was coded as 1 and the credit card condition as 0. Mode of payment was a signicant predictor of the number of impulsive products in the shopping basket (b p 1.1, p ! .01). Mode of payment also predicted pain of payment (b p .69, p ! .01).

FIGURE 2 PAIN OF PAYMENT MEDIATES THE EFFECT OF CASH PAYMENTS ON IMPULSE CONTROL (STUDY 3)
135
Further, pain of payment was a signicant predictor of the number of impulsive products in the shopping basket (b p .77, p ! .01). However, when both pain of payment and cash payment were entered as predictors in the same model, the effect of cash payment was no longer signicant (b p .60, p p .14), but the effect of pain of payment continued to be signicant (b p .68, p ! .01). A Sobel test conrmed the mediation (z p 2.61, p ! .01). However, these mediation results did not hold for virtue products. Thus, these moderated mediation results support our conceptualization that pain of paying in cash selectively reduces the proportion of vice products in the shopping basket.
vs. vice) as a within-subjects factor. Only the main effect of type of product was signicant. As expected, participants perceived the 10 vice products to be less healthy than the 10 virtue products (MVicep 2.01 vs. MVirtue p 4.07; F(1, 122) p 528.1, p ! .01). This result clearly shows that participants randomly assigned to the credit card payment condition bought more impulsive products despite being aware that these products were unhealthy.
Attention to Price. To test whether participants paid attention to the price of the products that they were putting into the shopping basket, they were asked to recall the total amount of money they spent on the shopping trip. Participants in the credit card condition reported spending more money than those in the cash condition (MCreditCard p $30.6 vs. MCash p $22.1), and these amounts were quite close to the actual amount that they spent (MCreditCard p $28.5 vs. MCash p $20.7). (This is the total amount, that is, the sum of the amounts spent on vice and virtue products.) To test whether the attention to price varied across the two conditions, we computed the difference between the reported amount and the actual value of the products in the basket for each participant and submitted this measure to a oneway ANOVA. The values of this recall accuracy measure did not change across the two modes of payment, F ! 1. These results suggest that the painlessness of paying by credit card is not due to price neglect. Even when participants pay attention to price, paying by credit card reduces the pain of payment. Perceived Healthiness. For each participant, we computed the average healthiness rating of the 10 virtue products and the average healthiness ratings of the 10 impulsive products. These two measures were submitted to a 2 # 2 mixed factorial ANOVA with mode of payment (credit card vs. cash) as a between-subjects factor and type of product (virtue
Regret Spending Money. The anticipated regret on spending money on the two types of products was submitted to a similar analysis. We computed the average regret ratings for the 10 virtue products and the average regret ratings for the 10 impulsive products. These two measures were submitted to a 2 # 2 mixed factorial ANOVA with mode of payment (credit card vs. cash) as a between-subjects factor and type of product (virtue vs. vice) as a within-subjects factor. The main effect of type of product was signicant; as expected, participants reported that they will experience more regret spending money on vice products than on the virtue products (MVice p 3.45 vs. MVirtue p 2.81; F(1, 122) p 41.75, p ! .01). Further, the main effect of mode of payment was also signicant. Participants reported that they will experience greater regret in spending money when they pay in cash than when they pay by credit cards (MCash p 3.34 vs. MCreditCard p 2.92; F(1, 122) p 7.39, p ! .01). The two-way interaction was not signicant, F ! 1. These results are important because they show that participants regret impulsively spending money on unhealthy food products. Impulsive purchases of unhealthy food products seem to be made on momentary feelings rather than on deliberative consideration of consequences of the consumption.
Discussion
These results are useful in several ways. First, we replicate our basic effectcredit card payments can increase consumption of unhealthy productsusing a more representative sample of consumers. This shows that the results observed in study 2 cannot be attributed to idiosyncratic
136
behavior of the student population. These results also lend more credibility to the results in (eld) study 1 with a similar prole of consumers. Second, and more important, these results offer insights into the underlying process. We use the pain scale to measure the pain of payment and demonstrate that this pain mediates the effect of mode of payment. Thus, this study tests and nds support for the central thesis proposed in this paper that pain of payment plays a benecial functional role in impulse control. Consumers can control their impulsive purchases by deciding to use cash instead of credit cards. We nd that paying by credit cards reduced pain of payment even when these participants were paying attention to prices. Note that we do not rule out the possibility that credit cards might induce price neglect. Whether or not consumers will accurately recall prices is contingent on several factors, including the design of the experiment. Indeed, it is likely that in some situations consumers might simply neglect prices when they pay by credit cards. The present results should be interpreted to imply that even in a situation where participants are carefully paying attention to prices, credit cards can reduce pain of payment, and this, in turn, can weaken impulse control. Furthermore, the alleviation of pain of payment caused by using less vivid and emotionally more inert modes of payment cannot be compensated by paying closer attention to prices. Finally, consistent with Ubels (2009) suggestion, these results suggest that the propensity to buy more impulsive products might not be based on justiable reasoning. Participants unambiguously acknowledged that the impulsive products were unhealthy and that they would experience regret after purchasing these products. Yet, when they were paying by credit cards they were more inclined to buy such products. These results, thus, suggest that participants purchase decisions in this experiment were inuenced by the interplay of the momentary experiences elicited by the pain of payment and impulsive urges. In the following study, we seek further support for this conceptualization.
periment. After the simulated shopping task, participants were asked to complete the four-item Spendthrift-Tightwad (ST-TW) scale developed by Rick et al. (2008). This scale comprises four items that measure respondents spending habits on their usual shopping trips. A higher score on this scale indicates that the respondent experiences less pain of payment and is a spendthrift. Before completing this scale, participants were explicitly instructed that the questions on this scale refer to their usual shopping behaviors.
Results
Vice Products in Shopping Basket. Participants responses to the four-item scale were combined to form a composite ST-TW index. The scores on the ST-TW scale did not change across the two mode-of-payment conditions (p 1 .20). As in the previous two studies, for each participant we computed the total number of virtue and vice products in the basket. This measure was submitted to a mixed factorial ANCOVA with three factors: mode of payment (credit card vs. cash) as a between-subjects factor, type of product (virtue vs. vice) as a within-subjects factor, and the ST-TW scale value as a continuous variable. As in the previous study, some of the demographic variables (diet control habits, medical condition, and household income) and their interactions with type of products were entered as covariates in this mixed model. The main effects of type of product (F(1, 118) p 7.15, p ! .01), and the ST-TW scale value (F(1, 118) p 7.42, p ! .01) were signicant. Further, as in the previous studies, the mode-of-payment # type-of-product interaction was signicant (F(1, 118) p 4.29, p p .04). More important, the three-way interaction was also significant (F(1, 118) p 4.11, p p .045). A similar analysis on the amount of money spent on vice and virtue products yielded convergent results; the main effects (ps ! .05), the two-way interaction (F(1, 118) p 5.00, p p .027), and the three-way interaction were signicant (F(1, 118) p 4.33, p p .039). To make sense of this three-way interaction, following Rick et al.s (2008) approach, we split the participants into three almost equal groups based on the scores on the STTW scale. The group with the lower scores on this scale was labeled as Tightwads, the one with the higher scores as Spendthrifts, and those with intermediate scores as Unconicted. We then computed simple contrasts to test the effect of credit cards on the purchase of vice and virtue products for each of these three groups. The pattern of means for the two groups with extreme scoresSpendthrifts and Tightwadsis shown in gure 3. Mode of payment had a signicant effect on Tightwads purchase decisions; they were more likely to buy impulsive products when paying by credit card (M p $12.8) than when they had to pay in cash (M p $5.6; F(1, 116) p 6.67, p p .01). However, as predicted, mode of payment did not inuence impulsive purchase decisions of Spendthrifts and Unconicted participants (Fs ! 1). Further, for all three groups of participants, mode of payment did not inuence purchase decisions of
STUDY 4: TIGHTWADS AND SPENDTHRIFTS

Rick et al. (2008) have argued that individuals differ in their chronic propensity to experience pain of payment. Tightwads experience more pain than spendthrifts. Our conceptualization suggests that the effect of cash on impulse control will be stronger for tightwads than for spendthrifts (hypothesis 3) because they will experience greater pain. The present study tests this hypothesis.
Method
One hundred and twenty-ve undergraduate students (37% women) from the State University of New York, Binghamton, participated in this online experiment for partial course credit. The experiment was administered using the same software as in the previous study. The cover story, stimuli, and procedure were identical to the previous ex-

FIGURE 3 THE EFFECT OF PAIN OF PAYMENT ON TIGHTWADS AND SPENDTHRIFTS (STUDY 4)
137
NOTE.Vice products are unhealthy food items purchased impulsively, while virtue products are healthy food items purchased after deliberative evaluations.
virtue products (Fs ! 1). A spotlight analysis at one standard deviation below mean of the ST-TW scale showed a signicant difference such that Tightwads were more likely to spend on unhealthy products with credit cards than with cash (b p 5.41; t p 2.32, p p .02). But the effects of mode of payment were not signicant at the mean level of the ST-TW scale (p 1 .36) and one standard deviation above the mean (p 1 .28). These results support hypothesis 3 and our conceptualization that the painless nature of credit cards weakens impulse control.
GENERAL DISCUSSION Summary and Implications

Four empirical studies were conducted to test the inuence of mode of payment on the proportion of unhealthy food products in consumers shopping baskets. The results from all these studies offer convergent support for our hypothesis that card payments increase the purchase and, presumably, the consumption of unhealthy food products. Our conceptualization is based on the premise that when consumers encounter vice productssuch as cookies, cakes, and piesthe emotive imagery and associated desire trigger impulsive purchase decisions. These visceral factors entice them to include such vice products in their shopping baskets, even though they consider such products to be unhealthy.
Pain of payment can curb the impulsive responses and thus reduce the purchase of such vice products. Since paying in cash feels more painful than paying by credit or debit cards, paying in cash can reduce the purchase of unhealthy food items. To test this hypothesis, in study 1 we analyzed the shopping baskets of 1,000 households over a period of 6 months. We found that shopping baskets had a larger proportion of impulsive and unhealthy food items when shoppers paid by credit or debit cards than when they paid in cash. Studies 24 replicated this effect using samples from different populations and offered insights into the underlying process. Study 2 showed that although participants in the cash payment condition purchased fewer vice products, they did not take more time to make purchase decisions. This result suggests that the effect of cash payments is caused by spontaneous pain of payment rather than by deliberative reasoning. The moderated mediation analyses in study 3 offered direct evidence for the mediating effect of pain of payment. In study 4, using the spendthrift-tightwad scale, we found that individual differences in sensitivity to pain of payment can moderate the effect of mode of payment. Together, these studies persuasively implicate pain of payment as the mechanism that underlies the vice-regulation effect of cash payments. These results have implications for consumer welfare. In the popular press as well as in academic publications, the growing obesity problem and its economic consequences
138
have been attributed to consumers failures to control impulsive urges (Ubel 2009). Further, researchers have identied several factors that make impulse control a challenging goal for consumers. Given this background, the nding that at least some consumers might be able to curb their impulsive urges by paying in cash is of substantive importance. The results also augment our knowledge of how mode of payment inuences buyers behavior. Although several studies have shown that credit cards inuence buyer behavior, how it does this has been a matter of debate. It has been suggested that people are conditioned to spend more with credit cards (Feinberg 1986; McCall and Belmont 1996). The conditioning account for mode-of-payment effects does not grant any role to affective processes. In contrast to this view, Prelec and Loewenstein (1998) proposed that credit cards reduce the pain of payment. The present research offers direct evidence for Prelec and Loewensteins visceralfactors account for mode of payment effects. We nd that experienced pain of payment mediates the effect of cash payments on impulse control (see g. 2) and that chronic sensitivity to pain of payment moderates the effect (see g. 3). Further, we show that the effect of pain of payment is contingent on whether the purchase decision is made impulsively or deliberately. Our results also suggest that the painlessness of paying by credit card is not due to price neglect. Participants in study 3 paying by credit cards did not experience pain even though they were paying attention to prices, as conrmed by the surprise recall task. This result is important because several consumers might believe that they can curb their impulsive behavior by paying attention to prices. Our results suggest that such beliefs might be misplaced. Consumers experience less pain when they use less vivid and emotionally more inert modes of payment and succumb to their impulsive urges, and this effect may not be mitigated by paying closer attention to prices.
with the same amount of money. An empirical test of this hypothesis might reveal new insights about the role of abstract and concrete construals in mode of payment effects. Second, it might be useful to compare and contrast visceral regulators such as cash against willpower. Although in this paper we do not directly pit visceral regulation against willpower, previous research suggests that visceral regulators are likely to be more effective in self-control because willpower gets depleted (Vohs and Heatherton 2000) and requires cognitive capacity (Shiv and Fedorikhin 1999). However, despite the obvious advantages of visceral regulators, consumers might tend to rely on their willpower based on the potentially misplaced belief that willpower is as effective as visceral regulators. Preliminary evidence for this proposition comes from the anecdotal observation that although over 60% of American adults are overweight or obese, only 14% of American consumers use cash payments on their shopping trips. Examining when visceral regulators are likely to be more effective than willpower and investigating whether people who are more effective at impulsecontrol tend to delegate the control function to visceral regulators could be fruitful avenues for future research.
APPENDIX
Vice products (Impulsive-Unhealthy): Chips Ahoy Chocolate Chip Cookies $3.91, Coca Cola Classic $4.99, Ghirardelli Hot Cocoa $6.39, Little Debbie MufnsBanana Nut $3.85, Mrs. Smiths Apple Pie $5.85, Mrs. Smiths Pumpkin Pie $5.85, Oreo CookiesChocolate Sandwich $3.71, Oreo CookiesGolden Sandwich $3.71, Sara Lee Cheesecake $8.99, Drakes Coffee Cakes $5.15 Virtue products (Deliberative-Healthy): Aquana Pure Watersix pack $6.99, Arnold/Brownberry 100% Whole Wheat Bread $3.85, Bushs Baked Bean $2.19, Cheerios CerealHoney Nut $7.95, Del Monte Diced Peaches $3.95, Health Valley Granola $4.59, Kashi Go Lean Crunch Cereal $3.85, Quaker Oatmeal $5.39, Special K Cereal $5.15, Yoplait 99% Fat Free Yogurt $1.45
Unaddressed Questions
The results from our studies also raise several unaddressed questions that could be fruitful avenues for future research. We identify two such questions here. First, this research adds to a burgeoning body of literature that offers evidence that emotional sensitivity to modes of payment can vary (Mishra et al. 2006; Prelec and Loewenstein 1998; Raghubir and Srivastava 2008, 2009). However, it is not clear why this happens. Why are card payments emotionally more inert than cash payments? One possibility is that the pain of payment is a function of the abstractness of the transaction. Card payments might be construed abstractly, whereas cash might be construed concretely in consumers minds. Parting with a hundred dollar bill is a very vivid and concrete action. However, charging $100 to ones credit or debit card is an abstract and less vivid action. Just as a concrete and vivid mental representation of a can of Coke is more likely to elicit a stronger visceral response than the abstract construal of a beverage, abstract construal of parting with money is likely to be less painful than concrete construal of parting
REFERENCES
Ainslie, Andrew and Peter E. Rossi (1998), Similarities in Choice Behavior across Product Categories, Marketing Science, 17 (2), 91106. Baumeister, Roy F. (2002) Yielding to Temptation: Self-Control Failure, Impulsive Purchasing, and Consumer Behavior, Journal of Consumer Research, 28 (4), 67076. Bell, David R. and James M. Lattin (1998), Shopping Behavior and Consumer Preference for Store Price Format: Why Large Basket Shoppers Prefer EDLP, Marketing Science, 17 (1), 6688. Cheema, Amar and Dilip Soman (2008), The Effect of Partitions on Controlling Consumption, Journal of Marketing Research, 45 (December), 66575. Feinberg, Richard A. (1986), Credit Cards as Spending Facilitating Stimuli: A Conditioning Interpretation, Journal of Consumer Research, 13 (3), 34856.
PS YC HOLOGICA L SC IENCE
Short Report
Buying Behavior as a Function of Parametric Variation of Number of Choices

Avni M. Shah and George Wolford Dartmouth College
There is accumulating evidence that having many options from which to choose may be counterproductive (Iyengar & Lepper, 2000; Iyengar, Well, & Schwartz, 2006; Schwartz, 2004). In the consumer world, Procter & Gamble noticed a 10% increase in sales of its Head & Shoulders brand after they reduced the number of varieties (Goldstein, 2001). Iyengar and Lepper (2000) carried out three experiments in which subjects had to choose from a small set of 6 options or from a larger set of 24 or 30 options. Subjects rated the situation with more options as more pleasant than the situation with fewer options, but purchased more when there were fewer options and were more satised with their choices. Most previous research on this topic compared only two set sizes for number of choices: a medium value, such as 6, and a large value, such as 24 (but see Iyengar, Jiang, & Huberman, 2004). We were interested in exploring the inuence of number of choices in a more parametric fashion. Several processes might inuence buying behavior as the number of choices increases. Presumably, as the number of choices increases, consumers are more likely to nd an item that meets their needs. If this is the only process involved, then increasing choice should lead to increased buying. However, as choice increases, consumers are more likely to have two or more items that meet their criteria and are close to one another in subjective value, making the choice difcult. Also, as the number of options increases, the cognitive effort in evaluating those options increases (Keller & Staelin, 1987). We predicted that the interaction of these proposed processes would lead to an inverted-U-shaped function; that is, we predicted that as the number of choices increased, buying would initially increase and then decrease. For the product in our experiment, we chose pens, a commodity that college students are interested in and often purchase. We varied the number of options from 2 to 20 in increments of 2.
Address correspondence to George Wolford, 6207 Moore Hall, Dartmouth College, Hanover, NH 03755, e-mail: wolford@dartmouth. edu.
METHOD
Subjects We received permission from our institutional review board to collect data from subjects without prior consent. We set up a table in a high-trafc corridor connecting the two major sections of Dartmouths library. As people passed, we encouraged them, one at a time, to rate a set of pens. One hundred people participated in the main experiment. The vast majority were Dartmouth undergraduates. An additional 20 Dartmouth undergraduates participated in a preliminary rating study.
Materials We purchased two dozen of each of 20 different pens. The pens all had roller-ball delivery systems and came from several different manufacturers. We selected only pens that had black ink and that ranged in cost from $1.89 to $2.39. Despite these restrictions, the pens differed markedly in appearance, feel, and mechanism. Before the main experiment, 20 subjects rated each of the 20 pens on a scale from 1, highly undesirable, to 10, highly desirable. Each subject rated each pen twice in random order. The ratings were averaged to produce a single rating for each pen. We used those ratings to choose the pens that would be included in each set of options.
Procedure For the main experiment, we set up a table in the library corridor and invited passersby to help us choose pens. They were told that the Department of Psychological and Brain Sciences was planning to buy several hundred pens for its supply closet and wanted to buy the best possible pens. The number of pens on the table varied from 2 to 20 in increments of 2. For the set of 2 pens, we included the top-rated pen from the preliminary ratings and the 10th-rated pen. For each successively larger set, we included the options in the previous set (as in Iyengar & Lepper,
Volume 18Number 5
Copyright r 2007 Association for Psychological Science
369
Number of Choices and Buying Behavior
jects who purchased a pen across these three bins. The effect of number of options was significant, w2(2, N 5 100) 5 9.722, p 5 .008, j 5 .312.
DISCUSSION
Fig. 1. Proportion of subjects who bought any pens as a function of the number of choices available.
As we hypothesized, buying behavior was a curvilinear function of number of choices, peaking at a value of 10 pens. Our data are consistent with the ndings of Iyengar and Lepper (2000), as an additional 20% of subjects purchased pens when there were 6 choices, compared with when there were 20; but examination of the more complete function yields a slightly different conclusion, namely, that more choice does not always lead to less buying. Rather, it is only after the optimal point has been exceeded that more choice results in less buying. Quite possibly, the peak and shape of the function would vary depending on the product, although we predict that many functions relating purchasing to number of choices would be curvilinear. AcknowledgmentsWe wish to thank Devin Riley for his help.
2000) and added 2 more pens, a relatively high-rated pen and a relatively low-rated pen. Each subject was asked to select the pen that he or she liked best. That choice was recorded. The subject was then informed that the pens were all in the $2.00 range, but that as a token of our appreciation, he or she could buy any pens on the table for $1 each. Whether or not a pen was purchased was the primary dependent measure.
RESULTS
REFERENCES Goldstein, N. (2001). Inside inuence report: Is your company offering too much? When more is less. In C. Cibbarelli & B. Gordon (Eds.), Inuence at work. Retrieved July 18, 2005, from www.insideinuence.com/year03/08/MoreisLess/ Iyengar, S.S., Jiang, W., & Huberman, G. (2004). How much choice is too much?: Contributions to 401(k) retirement plans. In O.S. Mitchell & S. Utkus (Eds.), Pension design and structure: New lessons from behavioral nance (pp. 8395). Oxford, England: Oxford University Press. Iyengar, S.S., & Lepper, M.R. (2000). When choice is demotivating: Can one desire too much of a good thing? Journal of Personality and Social Psychology, 79, 9951006. Iyengar, S.S., Wells, R.E., & Schwartz, B. (2006). Doing better but feeling worse: Looking for the best job undermines satisfaction. Psychological Science, 17, 143150. Keller, K.L., & Staelin, R. (1987). Effects of quality and quantity of information on decision effectiveness. Journal of Consumer Research, 14, 200213. Schwartz, B. (2004). The paradox of choice: Why more is less. New York: Harper Collins. (RECEIVED 4/26/06; REVISION ACCEPTED 8/28/06; FINAL MATERIALS RECEIVED 10/2/06)
Figure 1 presents the proportion of subjects who purchased a pen as a function of set size. We used two statistical analyses to test our prediction that there would be a curvilinear relation between number of choices and buying behavior. First, we carried out a logistic regression analysis with number of choices as a rst-order and as a second-order polynomial. The overall model was significant, w2(2) 5 9.17, p 5 .01, corrected r 5 .342. The linear component was significant, Wald w2(1) 5 5.35, p 5 .019, as was the quadratic component, Wald w2(1) 5 7.051, p 5 .008. Second, we grouped the number of choices into three bins (the lowest three, the middle four, and the highest three) and carried out a chi-square test comparing the proportion of sub-
370
Volume 18Number 5
Opportunity Cost Neglect

SHANE FREDERICK NATHAN NOVEMSKY JING WANG RAVI DHAR STEPHEN NOWLIS*
To properly consider the opportunity costs of a purchase, consumers must actively generate the alternatives that it would displace. The current research suggests that consumers often fail to do so. Even under conditions promoting cognitive effort, various cues to consider opportunity costs reduce purchase rates and increase the choice share of more affordable options. Sensitivity to such cues varies with chronic dispositional differences in spending attitudes. We discuss the implications of these results for the marketing strategies of economy and premium brands.
tudents of economics are taught that decisions require the consideration of opportunity coststhe unrealized ow of utility from the alternatives a choice displaces (Alchian 1968; Buchanan 1969; Nozick 1977). The assumption that consumers consider the opportunity costs of a decision is not only upheld as a law of consumer behavior applied to idealized consumers in economic textbooks but also appears to be widely assumed about actual consumers. For example, Becker, Ronen, and Sorter (1974, 327) contend: Decision makers confronted with a showcase of beluga caviar consider how much hamburger they could buy with the money [that] a pound of caviar costs. . . . People intuitively take opportunity costs into account. Okada and Hoch (2004, 313) similarly conclude: The opportunity cost
*Shane Frederick (Shane.Frederick@yale.edu) and Nathan Novemsky (Nathan.Novemsky@yale.edu) are both associate professors of marketing at the Yale School of Management, Yale University, PO Box 208200, New Haven, CT 06520-8200. Jing Wang (janewang@smu.edu.sg) is assistant professor of marketing at Singapore Management University, Lee Kong Chian School of Business, 50 Stamford Road, Singapore, 178899. Ravi Dhar (Ravi.Dhar@yale.edu) is George Rogers Clark Professor of Marketing at the Yale School of Management (School of Management, Yale University, PO Box 208200, New Haven, CT 06520-8200. Stephen Nowlis (Stephen.nowlis@asu.edu) is the AT&T Distinguished Research Professor of Marketing at the Carey School of Business, Arizona State University, Main Campus PO Box 874106, Tempe, AZ 85287-4106. The authors thank Daylian Cain, Zoe Chance, Eric Gold, Ryan Hamilton, Daniel Kahneman, Barak Libai, John Little, George Loewenstein, John Lynch, Daniel Mochon, Draz en Prelec, Daniel Read, Eldar Shar, Paige Skiba, Catherine Tucker, Ray Weaver, Juan Juan Zhang, and Ezra Zuckerman for valuable comments. The authors thank Rebecca Ratner for proposing study 4. Correspondence concerning this article should be addressed to Shane Frederick. John Deighton served as editor and Ann McGill served as associate editor for this article. Electronically published April 22, 2009
of money is easy to assess. Money has a readily exchangeable market, is highly liquid and fungible, and can be saved. A dollar is a dollar . . . and so what comes to mind as the next best use for money remains fairly constant across situations. Evaluating opportunity costs requires consumers to consider outside options that are not explicit components of a purchase decision. The assumption that they do so is inconsistent with much psychological research showing that judgments and preferences are based primarily on information that is explicitly presented (Kahneman and Frederick 2002; Slovic 1972). The assumption is also inconsistent with an experience one of us had while shopping for stereos. This customer was frozen in indecision between a $1,000 Pioneer and a $700 Sony, and the salesman intervened, framing the choice as follows: Well, think of it this waywould you rather have the Pioneer or the Sony and $300 worth of CDs? Remarkably, the decision that seemed so difcult just moments before was no longer even closethe Sony was at the cash register moments after the word CDs escaped the salesmans mouth. A big pile of new CDs seemed far too steep a price to pay for the Pioneers slightly more attractive speakers. The consumer in the forgoing story could subtract $700 from $1,000 and was capable, in principle, of recognizing that $300 can be used to purchase $300 worth of CDs. Nevertheless, that particular perspective was overlooked despite nearly an hour of contemplating the choice. While this anecdote may not be representative of all consumer decisions, we believe that opportunity costs are often neglected and that effort alone will not typically overcome this neglect. We propose that consumers may not spontaneously consider opportunity costs as economics textbooks hypothesize and as consumer behavior researchers often assume. As a result,
553
2009 by JOURNAL OF CONSUMER RESEARCH, Inc. Vol. 36 December 2009 All rights reserved. 0093-5301/2009/3604-0006$10.00. DOI: 10.1086/599764
554
consumer preferences can be inuenced by various manipulations that bring to mind opportunity costs, as we demonstrate in several studies described below.
OPPORTUNITY COST NEGLECT

In accounting parlance, incurred expenses and other negative cash ows are termed out-of-pocket costs, in contrast with opportunity costs, which refer to the absence of potential positive cash ows (e.g., salary that is not earned while one is in school). In this article, we propose that opportunity costs are not merely underweighted relative to some corresponding out-of-pocket costs (as argued by Thaler [1980]) but may often be neglected entirely; that is, consumers may not explicitly consider the outside goods that an expenditure displaces. This contention draws support from the nding that decision makers restrict their thoughts to salient situational elements and neglect relevant information that remains implicit. For example, when participants were permitted to ask questions about some opportunity (such as going to see a lm in a foreign city) before deciding whether to do it, their inquiries pertained almost exclusively to the focal event rather than to possible alternatives (such as visiting a museum or attending a sporting event; Legrenzi, Girotto, and Johnson-Laird 1993). In studies of probability judgments, outcomes not explicitly represented are often ignored or underweighted (Fischhoff, Slovic, and Lichtenstein 1978; Tversky and Koehler 1994). In research on affective forecasting, judgments about ones current or future well-being are excessively sensitive to current mood or the domain that the research instrument happens to make momentarily accessible, to the neglect of other relevant factors (e.g., Loewenstein and Frederick 1997; Schkade and Kahneman 1998; Wilson et al. 2000). In research on intertemporal choice, Loewenstein and Prelec (1993) found that consumers chose differently when the implicit alternative to dining out was explicitly described as eating at home. The repeated nding that people focus only on explicitly presented details and fail to spontaneously ll in the logical consequences of a choice or judgment suggests that opportunity costs are likely to be neglected, since computing the opportunity cost of a decision requires the decision maker to actively generate alternatives that are not explicitly provided. Thus, frames that enhance the salience of opportunity costs, such as that used by the stereo salesman, should evoke thoughts about alternative uses of money that consumers would not have generated themselves. This, in turn, should deter any focal purchase and make cheaper options more attractive than expensive options. The stereo salesmans suggestion to use the $300 price difference on CDs (rather than on the more expensive stereo) may have been particularly effective because CDs complement the stereo. Even if our narrator had spontaneously conjured alternate uses for the $300, he may not have considered anything as attractive as the CDs. However, our experiments will show that cues to consider opportunity costs will inuence choice, even the
placebic reminder that money preserved by forgoing some purchase will be available for other purchases. In studies 1a1c, we show that purchase rates are reduced when the not buy option is described as keeping money for other purchases and that preferences shift toward cheaper options when the price difference is made explicit, even without mentioning other purchases. Study 2 replicates these ndings and argues against another account: that drawing attention to the price difference segregates it from the total price, thereby increasing its impact. Study 3 provides direct evidence that specifying the price difference as residual cash brings to mind thoughts of purchasing outside goods and suggests that the efcacy of such frames remains even under conditions promoting deliberation. Study 4 shows that opportunity costs can be brought to mind by priming alternative uses of money outside the purchase context. The nal study shows that sensitivity to manipulations that bring opportunity costs to mind is moderated by individual differences in the pain of paying. We conclude by discussing the relevance of our results for managers and suggesting questions for future research.
STUDY 1A: THE EFFECT OF OPPORTUNITY COST SALIENCE ON WILLINGNESS TO PURCHASE

In the rst study, we tested whether a buy/no buy purchase decision can be inuenced by a minimal reminder to consider opportunity costsspecically, by framing not buying as keeping money for other purchases (without suggesting any particular use of the money). If opportunity costs are spontaneously considered, this frame should have no effect. However, we hypothesize that opportunity costs are routinely neglected and that this frame will diminish interest in a focal purchase.
Method
One hundred and fty students at Arizona State University were asked to imagine they could purchase a DVD for $14.99. As shown in gure 1, each was randomly assigned to one of two conditions in which the decision against purchasing was worded either as Not buy or Keep the $14.99 for other purchases. The two descriptions are, of course, normatively equivalent, since not buying implies keeping the money for other purchases.
Results and Discussion

Although the principal reason against purchasing an entertaining video (or any other good) is presumably to save the money for something else, describing the Not buy option as Keeping money for other purchases caused willingness to purchase to fall from 75% to 55% (x 2 (1) p 6.57, p ! .05). We found similar results for choices involving a dinner event or a pair of sunglasses. Such manipulations should have no effect if participants spontaneously

FIGURE 1 STUDY 1A
555
considered outside goods before rendering their decisions, and thus this provides some evidence that they do not.
97), the price difference between the two iPods was left implicit; for the other half (n p 99), we added the phrase leaving you $100 in cash, as shown in gure 2.
STUDY 1B: OPPORTUNITY COST SALIENCE ENHANCES PREFERENCE FOR THE CHEAPER OPTION
The prior study showed that opportunity cost cues increased the likelihood of declining a single option, when only one was offered, suggesting that choosers might have failed to consider forgone opportunities without prompting. Single-option choices may indeed be special in that the choice context fails to evoke comparisons (Jones et al. 1998). When one chooses among several options, however, it is clear that selecting one necessitates forgoing others. This could remind decision makers of opportunity costs more broadlythat is, if I am giving this up, what else might I be giving up?and hence render additional opportunity cost cues redundant. We test this in the next study.

In each condition, roughly half of the respondents (n p 51 in both conditions) declined to purchase either iPod. However, among those who chose to purchase one, we found that merely describing the cost difference as a residual cash amount increased the choice share of the cheaper iPod from 37% to 73% (x 2 (1) p 12.3, p ! .001). Just mentioning the cost savings seems to encourage participants to consider alternate uses of that money that they overlook when the difference is left implicit, thereby making an additional $100 seem too steep a price to pay for the enhanced features of the more expensive option. These results might be understood as a demand effect if participants regarded the mention of residual cash as a suggestion that frugality is important or that they should avoid unnecessary expenditure. To test this account, we ran a similar study in which respondents rated the importance of being frugal, the amount of guilt they would experience purchasing the more expensive alternative, and how frivolous such a purchase would be. Although our opportunity cost reminder replicated the effect discussed
Method
As part of a short Web-based questionnaire, 196 participants were asked to choose between purchasing one of two iPods or declining both. For half of the participants (n p
FIGURE 2 STUDY 1B
556
above, these additional measures showed no differences between conditions (ts ! 1).
STUDY 1C: OPPORTUNITY COST SALIENCE IN CONSEQUENTIAL CHOICE

Our rst two studies involved hypothetical choices, and thus they may not have commanded as much consideration as participants would devote to a consequential choice, which could foster consideration of opportunity costs. Furthermore, consequential choices are more likely to reect participants preferences than their guesses about the choice that would most please the experimenter. Thus, in the current study, we apply our manipulation from the previous study to a real choice where actual money is exchanged for one of two actual mugs.
Method
Eighty-eight students at Arizona State University were each paid $10 cash for their participation in an unrelated study. Participants were then informed that they would use the money they had just received to make an actual choice between an insulated metallic portable coffee mug for $10 and a simpler ceramic mug for $3.99. Half of the participants saw a description of the cheaper mug that included the phrase leaving you with an extra $6.01 in cash to spend on something else.
this alternate account, we introduce here a third condition that also explicitly mentions the price difference but that does so in a way that less readily brings to mind alternative uses for money. Specically, in a choice between two stereo systems, we added a third condition in which the phrase spend $300 more was added to the description of the more expensive option, to frame the price difference as the price premium needed to purchase the superior stereo. This also segregates the price difference from the total price, but it less readily brings to mind thoughts about alternative uses since it identies how the segregated sum will be spent (on the more expensive stereo). If segregation explains the prior results, we would expect the choice share of the cheaper stereo to increase whether the additional cost of the expensive stereo is segregated in the form of a premium (spend $300 more) or a discount (leaving you $300). By contrast, our opportunity cost account predicts that only the discount manipulation will inuence preferences.
Method
We randomly assigned 110 MIT students in an MBA marketing class to receive one of three versions of a choice between two stereos (see below). The preamble in each case was the same: Suppose you have just won $1,000 playing a scratch-off lottery ticket and are shopping for a new stereo system. Check the option you would choose. The only difference was the way in which the options were described, as shown in gure 3.

The effects observed in the rst two studies persisted when real money and real mugs were at stake. When opportunity costs were brought to mind, the choice share of the cheaper mug increased from 40% to 60% (x 2 (1) p 3.63, p p .05). This result suggests that our opportunity cost manipulation is not operating through experimental demand, and it casts doubt on the idea that motivation to consider the choice is a sufcient antidote for opportunity cost neglect.

The cheaper stereo was chosen more often in our standard opportunity cost condition than in either the control condition (82% vs. 59%; x 2 (1) p 4.47, p ! .05) or the price premium condition (82% vs. 51%; x 2 (1) p 8.31, p ! .01). Importantly, the price premium condition did not differ signicantly from the control condition (51% vs. 59%; x 2 (1) p 0.47, p p .5), which argues against a segregation interpretation.
STUDY 2: EXPLORING A SEGREGATION ACCOUNT OF THIS EFFECT

Study 2 focuses on an alternative account of our ndings based on prospect theory. Thaler (1985) invoked the prospect theory value function (Kahneman and Tversky 1979) to propose that small amounts of money have greater impact when segregated from a larger amount in which they are embedded. He found, for example, that a pair of outcomes (losing $200 and getting $25) was judged to be more attractive than the combined outcome (losing $175). In other words, $25 has a greater impact as a separate gain than as a reduction of a larger loss. Our results could be interpreted similarly: explicitly mentioning a residual cash amount might segregate this amount from the total cost, thereby increasing the impact of the price difference and, in turn, the attractiveness of the cheaper option (e.g., v($100) + v($299) 1 v($399)). To test
STUDY 3: A DIRECT MEASURE OF OPPORTUNITY COST SALIENCE

Thus far, our studies demonstrate that purchase decisions are inuenced by subtle frames that bring to mind opportunity costs. These ndings are consistent with the notion that individuals neglect information that remains implicit while they retain the ability to recognize its relevance when the choice is framed in those terms. Since participants found opportunity costs relevant when external cues brought them to mind, they might bring them to mind spontaneously if they simply thought harder about the choice. While study 1c suggests that real consequences are not sufcient to cause participants to bring opportunity costs to mind, in this study we manipulate cognitive effort more directly by having participants record advantages and disadvantages of each option

FIGURE 3 STUDY 2
557
before making their decision. This procedure also provided a manipulation check of whether our opportunity cost frame does in fact increase consideration of outside goods.
Method
Two hundred and thirty-ve Yale University students were randomly assigned to one of four conditions of a 2 # 2 between-subjects design that crossed our opportunity cost manipulation (whether the price difference was left implicit or was emphasized by adding the phrase leaving you $300 in cash) with a cognitive involvement manipulation. Those assigned to the low-involvement conditions simply made a choice. Those in the high-involvement conditions rst listed all the advantages and disadvantages they could generate for each option.
ers as either mentioning outside goods or not. There were very few coding disagreements, and these were resolved through discussion. Participants generated essentially the same number of thoughts in the control condition (4.04) and the opportunity cost condition (4.12, t (71) ! 1). However, the content of those thoughts differed as expected. In the opportunity cost condition, 30% of participants mention outside goods (e.g., Ill have leftover money to buy CDs; Ill have $300 for shopping for clothes) as a consideration affecting their stereo decision, as compared to just 13% in the control condition (x 2 (1) p 6.35, p ! .05). As further evidence that our manipulation is inuencing choice by bringing to mind thoughts about outside goods, a Sobel test conrmed that the mention of outside goods was a significant mediator of stereo choice ( z p 1.6, p p .05, onetailed).

The cheaper stereo was chosen more often in the opportunity cost condition than in the control condition whether involvement was low (87% vs. 66%; x 2 (1) p 4.30, p ! .05) or high (86% vs. 70%; x 2 (1) p 6.77, p ! .05). A logistic regression revealed a signicant main effect of the opportunity cost manipulation (b p 1.21, p ! .05) but no effect of involvement (b p 0.17, p 1 .6) and no interaction (b p 0.18, p 1 .8). We should note that, in a separate study, we also tested whether the imposition of cognitive load (holding a long number in memory prior to choice) increased the effect of the opportunity cost manipulation. It did not. Participants thought listings allow us to examine the prevalence of thoughts regarding outside goods across conditions. These thoughts were coded by two independent rat-
STUDY 4: PRIMING OPPORTUNITY COSTS

Thus far, we have shown that explicit mention of the price difference increases the share of the cheaper option, and we have proposed that it does so by encouraging consideration of outside goods. This account suggests that choices should be inuenced by any procedure that brings outside options to mind. In study 4, we test another such procedure: having participants list several things they would like to buy in an ostensibly unrelated prior study.
Method
As part of a large packet of questionnaires that participants were paid $5 to complete, 150 Yale University undergraduate students chose between a pair of cell phones, with the
558
higher-quality option costing $20 more than the other option. Participants were randomly assigned to either a control condition, in which they simply chose a cell phone, or to a priming condition, in which the cell phone choice was preceded by a study in which they listed several items costing around $20 that they would like to buy. This listing study was presented in a different font as part of a separate study with a different title to help mask the connection to the subsequent decision between cell phones.

Using the denition provided by Rick and colleagues, we classied participants as tightwads or spendthrifts based on their composite score on the four-item measure (a 1 .7). We then conducted a binary logistic regression to assess the effect of our manipulation of opportunity cost salience and participants score on the tightwad/spendthrift scale on participants choice between the cheaper or more expensive stereo. This analysis revealed a signicant main effect of opportunity cost salience (b p 1.51, p ! .001) and a marginally signicant interaction (b p .75, p p .08). Among spendthrifts, explicit references to residual cash increased the choice share of the cheaper stereo from 41% to 87% (x 2 (1) p 14.7, p ! .001). By contrast, our manipulation had a smaller and statistically unreliable effect among tightwads, increasing the choice share of the cheaper stereo from 65% to 80% (x 2 (1) p 1.84, p 1 .15). The nding that external cues to consider opportunity costs affect tightwads less than spendthrifts suggests that tightwads may be more inclined to spontaneously construe purchase decisions in terms of opportunity costs.
Results
All participants in the priming condition listed at least one other product they would like to buy for $20 (e.g., CDs, books, clothing). Participants in this condition chose the cheaper cell phone signicantly more often (47% vs. 30% in the control condition, x 2 (1) p 4.11, p ! .05). This manipulation casts doubt on explanations involving experimental demand as the priming manipulation (the listing task) was ostensibly unrelated to the cell phone choice.
STUDY 5: INDIVIDUAL DIFFERENCES IN SENSITIVITY TO OPPORTUNITY COST CUES

We have posited that many consumers fail to spontaneously bring to mind opportunity costs when making purchase decisions. Of course, this tendency may vary across individuals. Recent studies suggest chronic differences in the extent to which consumers experience pain when paying for a good (Rick, Cryder, and Loewenstein 2008). We propose that this pain is determined in part by the degree to which a consumer focuses on the opportunities given up. Correspondingly, we predict that tightwads who experience greater pain of paying will be less affected by manipulations that increase the salience of opportunity costs because they are more prone to think about purchases this way without the intervention of external cues. Study 5 tests this idea.
GENERAL DISCUSSION
Our research challenges the widespread presumption that consumers spontaneously consider opportunity costs when making a purchase decision. Frames that evoke opportunity costs reduced the likelihood of purchasing a given product (study 1a) and increased the choice share of the cheaper option (studies 1b5). Such manipulations remained effective when the choices were consequential (study 1c) and when participants were forced to deliberate about the decision (study 3). Such effects are not readily attributable to enhanced price salience or to the segregation of the price difference from the total price (study 2) but appear instead to reect the degree to which a particular description of the options brings to mind outside goods (study 3). Study 4 shows that priming opportunity costs affects subsequent purchase decisions even when the priming task was seemingly unrelated to the focal choice. Study 5 suggests that the efcacy of these manipulations depends on chronic spending attitudes as consumers who self-identify as tightwads are less inuenced by frames intended to highlight opportunity costs. Unlike the salesmans comment in the motivating anecdote, several of our manipulations left the alternative uses of the potential savings unspecied. Although mention of any specic outside good may promote greater consideration of alternatives, estimates of opportunity cost may be affected by the particular opportunity that is mentioned. In a study not reported above, we found that the manipulation that referenced a $300 cash savings was much less effective when accompanied by an unattractive example of how that money could be spent (specically, a weekend trip to Des Moines, Iowa). Future research could examine how specic opportunities mentioned in a decision context differentially inuence choice.
Method
Three hundred individuals were recruited via the Internet to participate in a Web-based questionnaire. They encountered the stereo choice described earlier and received either the control condition (in which the price difference was left implicit) or the residual cash condition (in which opportunity costs were brought to mind by adding the phrase leaving you $300 in cash to the description of the cheaper option). After making their choice, they completed a four-item Spendthrift-Tightwad scale (Rick et al. 2008) intended to characterize people according to their self-reported ease or difculty spending money. Our hypothesis was that tightwads would be more likely to spontaneously consider opportunity costs and thus be less susceptible to our manipulation.
559
Since the consideration of opportunity costs has been shown to inuence relative preferences, the effect of decision context on the accessibility of particular outside opportunities may be an important source of preference instability. In study 4, we found that the request to generate just a few uses of a specic dollar amount affected subsequent purchase decisions. In another study, we found that this task actually affected consumers attitudes toward money, increasing their predicted reaction to nding (or losing) a $20 bill and decreasing their willingness to take a pay cut in exchange for a more satisfying job. Our contention that consumers often fail to consider opportunity costs might seem to contradict abundant evidence of price sensitivity. However, price sensitivity may result from many considerations other than computation of opportunity costs. The price of a product could affect consumers evaluation of the deal relative to an adopted reference price (Thaler 1985; Weaver and Frederick 2009), their assessment of whether the ratio or difference in the prices is commensurate with the perceived differences in quality, or the amount of pain that accompanies expenditure. Recent neural imaging research (Knutson et al. 2007) nds that monetary losses directly activate primary rewards centers, which supports the idea that payment may have a deterrent force that requires no cognitive elaboration of the specic opportunities foreclosed by an expenditure (Prelec and Loewenstein 1998; Rick et al. 2008). Although sensitivity to price may reect some vague recognition of sacrice, that need not mean that, when consumers decide whether a product is worth it, they consider specic alternative uses for that money or attempt to simulate and compare the utility those other goods would deliver. Indeed, explicit reference to outside goods is notably rare in written protocols about purchases. Brown (2005) found that fewer than 10% of his participants mentioned outside goods when instructed to describe how they arrived at their maximum willingness to pay, we found that only 13% of participants mentioned outside goods in the control condition of study 3, and Rick et al. (2008) found only two comments resembling an explicit contemplation of opportunity costs in over 2,200 protocols regarding past purchase regrets. While our studies may suggest that opportunity cost neglect occurs in many consumer settings, there are important boundaries worth noting. Typical purchase situations may fail to evoke considerations of outside goods because the amount of money consumers have at their disposal is quite vague and thus trade-offs are rarely explicit. In the language used by Zauberman and Lynch (2005), monetary budgets have considerable slackany particular expenditure (e.g., having wine with dinner) does not unambiguously jeopardize the satisfaction of any other particular purchase goal (e.g., upgrading ones stereo system). This view of opportunity cost neglect suggests several important classes of situations where one would expect no such neglect. First, very poor individuals or those on xed incomes may be keenly aware of opportunity costs in many decisions because their
binding budget constraints may frequently necessitate a careful comparison of mutually exclusive options. Similarly, those who budget narrowly and impose rigid constraints on the pertinent category of expenditure (see Heath and Soll 1996) may feel that they are giving something up each time they allot money to a specic purchase. The relationship between wealth, mental budgeting, and spontaneous consideration of opportunity costs is an important unexamined topic. Some research suggests that time budgets contain more perceived slack than monetary budgets (Okada and Hoch 2004; Zauberman and Lynch 2005). Therefore, opportunity costs of time might be even less likely to be considered than opportunity costs of moneyexcept, perhaps, by those who charge by the hour (e.g., lawyers, consultants), who may grow accustomed to converting time into equivalent revenue. Future research could examine when opportunity costs of time are spontaneously considered and how they might be externally cued. It seems natural to assume that frames that foster the recognition of opportunity costs lead to better choices since the generation of alternatives is universally regarded as an essential component of good decision making (Hammond, Keeney, and Raiffa 2002). However, we have shown that enhancing the salience of opportunity costs will tend to dissuade consumers from selecting the higher-quality, more expensive alternativea decision they rarely regret (Frederick 1998; Kivetz and Keinan 2006). This phenomenon is hinted at by the tagline on a long-running commercial for Acura automobiles: Paying for quality can be a difcult decision at rst, but over time it gets a lot easier to live with. We agree with Acuras descriptive claim but not its normative tone. The fact that consumers do not regret spending a lot of money does not mean that they should not regret it. As noted by Gilovich and Medvec (1995) in their landmark article on regret, the passage of time impedes our ability to track the cost of a previous action, and the opportunity costs of past expenditures may be neglected even more than the opportunity cost of present expenditures. Future research could explore how consideration of opportunity costs inuences short-term and long-term regret and satisfaction.
Marketing Implications
Given that consumers who bring to mind opportunity costs become more price sensitive, manufacturers of less expensive brands interested in increasing price sensitivity may promote their products more effectively by reminding consumers to consider the opportunity costs of the price premium of more expensive competitors. Rather than advertising a brands low prices in some general way (as when Pontiac advertised its vehicles with the tagline: Your money hasnt gone this far since you lived with your parents) or emphasizing proportional or cumulative cost savings vis-a `vis some specic competitor (e.g., 20% cheaper than our leading competitor; Consumers have saved billions by switching to MCI), rms may better promote low-price prod-
560
ucts by cueing consumers to think about the leftover cash and possible attractive uses for it. For example, Volkswagen could emphasize the economy of purchasing their vehicle in terms of the new wardrobe of clothes one would then be able to afford, perhaps illustrated by a smartly dressed driver ferrying her unstylish friends in her new Beetle. Consistent with this strategy, an advertisement for IKEA furniture depicted, on the left panel, an unhappy woman standing next to a cabinet containing a single pair of shoes. The caption beneath reads Customized cabinet ($1,670) + 1 pair of shoes ($30) p $1,700. By contrast, the right panel depicted a woman beaming at her daughter in front of a more modest IKEA cabinet that was overowing with shoes. The caption beneath showed the price of the cabinet ($245) plus the price of 48 pairs of shoes ($1,440) p $1,685. Similarly, an ad for Sun America investments frames the cost of a diamond ring ($13,000) as the reduction of $60,296 from the beloveds retirement account and urges its clients to nd less expensive ways of saying I love you. Our research suggests that another effective promotional tactic for less expensive brands would be to bundle their product with another good that could be purchased for the difference in price between their product and their competitors more expensive product. In a study not reported above, we found that bundling a cheaper stereo with a $300 VCR made that option more attractive than a description that neither mentionednor constrained the use ofthe $300 price difference. Although this is a worse deal for consumers (because it effectively forces any consumer considering the cheaper option to spend $300 on the VCR rather than on other things), it serves the function of the CDs in the motivating anecdote by dramatizing the opportunity costs of choosing the more expensive option. Manufacturers of expensive products or premium brands need to consider how to guard against (or exploit) opportunity costs in consumer choice. In such cases, the strategy would not be to emphasize the opportunity costs of purchasing a competitors product but to downplay the opportunity costs of purchasing ones own. For example, a DeBeers ad depicts two large diamond earrings and the tagline: Redo the kitchen next year to (misleadingly) imply that the cost of the diamonds is merely a delay in the kitchens renovationsomething a homeowner may be inclined to want anyway. Since consumers may readily accept the offered characterization of opportunity costs, proposing small ones may be another strategy for encouraging spending. For example, relief agencies often frame requests for donations to impoverished children in terms of trivial opportunity costs (For the price of a cup of coffee, you could . . .). Amusingly, on one ostensibly antiwar Web site, the cost of the war in Iraq (then estimated at $300 billion) was illustrated as the loss of nine Twinkies per American per day for a yeara rather unimpressive opportunity cost that could, perversely, increase support for that war.
Conclusion
One can dispute the normative issue of how much consumers ought to dwell on opportunity costs, but the descriptive phenomenon is clear: bringing to mind opportunity costs can markedly affect preferences. Our experimental results implying that opportunity costs are commonly neglected seem less surprising than the prevalence of the assumption that opportunity costs are routinely computedthat consumers routinely generate an exhaustive set of alternatives and successively simulate the utility of various combinations against the utility of the focal good whose purchase is being considered. To illustrate the implausibility of this assumption, consider the following example. Shortly after having had the stereo experience, our narrator purchased a $3 cognac trufe, which he quickly consumed. Afterward, his friend asked him, Was it worth the money? Before responding, he rst considered what else he could have purchased with $3six Snickers bars, a copy of The Sporting News, or a ner glass of wine with dinner. Or he could save the moneyit is not much, but along with other sacrices, maybe he could get a bigger apartment next year. He also recalled that satellite TV costs $49 a month and that he had hardly been watching any TV lately. With the $49 he would save, he could have all the trufes he wants. Bested by his friends question, recognizing that such thoughts could go on endlessly, he nally admitted, I dont know. The role of opportunity costs in decision making has relevance beyond the domain of consumer products. Excerpts from two political speeches warrant comparison. First consider the State of the Union address delivered by George W. Bush on January 29, 2002, just prior to the onset of war with Iraq: My budget includes the largest increase in defense spending in two decadesbecause while the price of freedom and security is high, it is never too high. Whatever it costs to defend our country, we will pay. Contrast this with a passage from Eisenhowers Chance for Peace Speech, which he delivered on April 16, 1953, as he was leaving ofce: The cost of one modern heavy bomber is this: a modern brick school in more than 30 cities. It is two electric power plants each serving a town of 60,000 people. It is two ne, fully equipped hospitals. It is some 50 miles of concrete highway. We pay for a single ghter with a half million bushels of wheat. We pay for a single destroyer with new homes that could have housed more than 8,000 people. Note that Bush, who was making a case for an expensive war, carefully avoided references to its opportunity costs, while Eisenhower, who was selling peace, liberally invoked them. It is now clear that Eisenhowers speech failed to curb our nations military spending, but our research suggests that his rhetorical strategy has psychological force. If it cannot stop war, it might, at least, be used to increase the market share of affordable stereos.
REFERENCES
Alchian, Armen A. (1968), Cost, in International Encyclopedia of the Social Sciences, Vol. 3, New York: Macmillan, 40415.
REPORTS
The Psychological Consequences of Money

Kathleen D. Vohs,1* Nicole L. Mead,2 Miranda R. Goode3 Money has been said to change peoples motivation (mainly for the better) and their behavior toward others (mainly for the worse). The results of nine experiments suggest that money brings about a self-sufficient orientation in which people prefer to be free of dependency and dependents. Reminders of money, relative to nonmoney reminders, led to reduced requests for help and reduced helpfulness toward others. Relative to participants primed with neutral concepts, participants primed with money preferred to play alone, work alone, and put more physical distance between themselves and a new acquaintance . eople long have debated the effects of money on human behavior. Some scholars have pointed to its role as an incentive, insofar as people want money in order to trade it for prized goods or services (1, 2). Others, however, have deplored money for undermining interpersonal harmony (3). We propose that both outcomes emerge from the same underlying process: Money makes people feel self-sufficient and behave accordingly. In this Report, money refers to a distinct entity, a particular economic concept. Consistent with other scholarly uses of the term (1), we use the term money to represent the idea of money, not property or possessions. Our research activates the concept of money through the use of mental priming techniques, which heighten the accessibility of the idea of money but at a level below participants conscious awareness. Thus, priming acts as a nonconscious reminder of the concept of money. We tested whether activating the concept of money leads people to behave self-sufficiently, which we define as an insulated state wherein people put forth effort to attain personal goals and prefer to be separate from others. The term as we define it does not imply a value judgment and encompasses a mixture of desirable and undesirable qualities, which may help explain the positive and negative consequences of money (4). The self-sufficiency hypothesis encapsulates findings from extant research on money. If money brings about a state of self-sufficiency, then a lack of money should make people feel ineffectual. Previous research indicates that physical and mental illness after financial strain due to job loss is statistically mediated by reduced feelings of personal control (5). A recent theory by Lea and Webley (1), which characterizes money as both a tool and a drug, emphasizes that people value money for its instrumentality: Money enables people to achieve goals without aid from others. Therefore, we predicted that reminders of money would lead to
1 Department of Marketing, Carlson School of Management, University of Minnesota, 3-150 321 19th Avenue South, Minneapolis, MN 55455, USA. 2Department of Psychology, Florida State University, Tallahasse, FL 323064301, USA. 3 Marketing Division, Sauder School of Business, University of British Columbia, Vancouver, BC V6T 1Z2, Canada.
*To whom correspondence should be addressed. E-mail: kvohs@csom.umn.edu
Fig. 1. Percentage of participants who asked for help as a function of money prime and length of time that had elapsed while working on (A) a difficult task (from Experiment 1) or (B) an unsolvable task (from Experiment 2).
1154
17 NOVEMBER 2006
VOL 314
SCIENCE
www.sciencemag.org
Downloaded from www.sciencemag.org on March 27, 2013
changes in behavior that suggest a feeling of selfsufficiency. When reminded of money, people would want to be free from dependency and would also prefer that others not depend on them. In Experiment 1, participants were randomly assigned to three conditions. In two conditions (play money and money prime), participants were reminded of money; control participants were not reminded of money (6). All participants first completed a descrambling task (7), which activated neutral concepts (control and play money) or money (money prime). The descrambling task consisted of 30 sets of five jumbled words. Participants created sensible phrases using four of the five words. In the control and play-money conditions, the phrases primed neutral concepts (e.g., cold it desk outside is became it is cold outside). In the money-prime condition, 15 of the phrases primed the concept of money (e.g., high a salary desk paying became a high-paying salary), whereas the remaining 15 were neutral phrases (6). Participants in the play-money condition were primed with money by a stack of Monopoly money in their visual periphery while completing the neutral descrambling task. Next, participants were given a difficult but solvable problem that involved arranging 12 disks into a square with five disks per side. As the experimenter exited the room, he offered that he was available to help if the participant wanted assistance. Persistence on the problem before asking for help was the dependent measure (8). As predicted, participants who were reminded of money (play money and money prime) worked longer than control participants before requesting help [F(2,49) = 3.73, P < 0.04; mean (M) money prime = 314.06 s, SD = 172.79; M play money = 305.22 s, SD = 162.47; M control = 186.12 s, SD = 118.09]. The two money conditions did not differ from each other [t(49) < 1], but each was significantly different from the control group [money prime versus control: t(49) = 2.44, P < 0.02; Cohens d = 0.86; play money versus control: t(49) = 2.30, P < 0.03; Cohens d = 0.84]. Percentages of participants who requested help are shown in Fig. 1A. In Experiment 2, we made two key changes to increase the generalizability of the findings of Experiment 1. First, we equated status differences between the would-be helper and the participant to ensure that differences in requests for help in Ex-
periment 1 were not due to differential sensitivity to the experimenters higher status. The second change was to the manipulation of the money prime. We hypothesized that money primes are unlikely to activate the idea of meager finances rather, monetary wealth is probably what is activated. This reasoning suggests that directly reminding people of meager finances will not lead to the same effects as reminders of financial affluence, which we tested systematically in Experiment 2. Participants were randomly assigned between two manipulations; one condition activated the idea of an abundance of money (high money) and the other activated the idea of restricted amount of money (low money). Participants first read aloud an essay in front of a video camera. Participants in the high-money condition read about growing up having abundant financial resources, whereas lowmoney participants read about growing up having meager resources. Afterward, all participants were given the opportunity to ask for help. The indicator of self-sufficiency was persistence on an impossible task before asking for help. The participants job was to outline all segments of a geometric figure once and only once without lifting the pencil or retracing any segments. Unbeknownst to participants, the figure was unsolvable. After 2 min of working alone, the experimenter and a confederate (who was blind to the participants condition) entered the room. The experimenter said
REPORTS
that the confederate was another participant who had just completed this experiment and therefore could be asked for help, if needed. Results indicated that participants in the high-money condition worked significantly longer than participants in the low-money condition before asking for help [t(35) = 2.03, P = 0.05; Cohens d = 0.65; M high money = 1058.48 s, SD = 210.12; M low money = 876.63 s, SD = 334.42]. Percentages of participants asking for help are shown in Fig. 1B. Thus, the effects of money did not depend on relative status differences between the participant and the helper. In Experiment 3, we predicted that people who value self-sufficiency would be less helpful than others because they expect that each person will take care of him- or herself. Hence, we expected that participants primed with money would volunteer less time relative to control participants. Participants were randomly assigned to one of two conditions, one that primed money and one with neutral concepts. The priming manipulations were the money and neutral (control condition) descramble tasks from Experiment 1. After the priming task, the experimenter explained that she was an undergraduate who was looking for help coding data and asked whether the participant would be able to help (9). She explained that each data sheet takes approximately 5 min to code. Participants were left alone to indicate how many data sheets, if any, they would be willing to code and also to provide their contact information. Participants in the money condition volunteered to help code fewer data sheets than did participants in the control condition [t(37) = 2.06, P < 0.05; Cohens d = 0.66] (Table 1). Translated into time, control condition participants volunteered an average of 42.5 min of their time, whereas participants in the money condition volunteered only slightly more than half that much (~25 min). Experiment 3 showed that participants primed with money offered less help to the experimenter than did participants primed with neutral concepts. Yet, it may be that by asking for help for sometime in the future, the experimenter suggested that she was not in dire straits (in which case, she likely would have asked for immediate aid); thus, money condition participants may have failed to realize that help was truly needed. Accordingly, it was important to move beyond promises of help to measuring real helping behavior. In Experiment 4, two between-subject conditions were used to prime money or neutral concepts. Each participant completed the descramble tasks (from Experiment 1). Next, the participant was left alone to complete irrelevant questionnaires. Meanwhile, the experimenter reentered with a confederate (who was blind to the participants priming condition) and introduced her as another participant. The experimenter explained that there was no space in the laboratory and therefore the confederate must share a room with the participant. After pretending to work for one minute, the confederate asked the participant to explain the directions for the task she was given because she did not understand what to do. Time spent helping the confederate was the measure of helping. Participants who were primed with money were less helpful than participants not primed with money [t(42) = 2.13, P < 0.04; Cohens d = 0.63]. The data showed that participants primed with money spent half as much time helping the confused confederate as did participants in the control condition (Table 1). Apparently, participants who were primed with money believed that the confederate should figure out on her own how to perform the task, as a self-sufficient person would do. In Experiment 5, we wanted to give moneyprimed participants a helping opportunity that required no skill or expertise, given that the help that was needed in the two previous experiments may have been perceived as requiring knowledge or special skill to enact. The opportunity to help in the current experiment was quite easy and obvious, in that it involved helping a person who spilled a box of pencils. Participants were randomly assigned to one of three conditions that were manipulated in two steps. Each participant first played the board game Monopoly with a confederate (who was blind to the participants condition) posing as another participant. After 7 min, the game was cleared except for differing amounts of play money. Participants in the high-money condition were left with $4000, which is a large amount of Monopoly money. Participants in the low-money condition were left with $200. Control condition participants were left with no money. For highand low-money participants, the play money remained in view for the second part of the manipulation. At this step, participants were asked to imagine a future with abundant finances (high money), with strained finances (low money), or their plans for tomorrow (control). Next, a staged accident provided the opportunity to help. A new confederate (who was blind to the participants priming condition) walked across the laboratory holding a folder of papers and a box of pencils, and spilled the pencils in front of the participant. The number of pencils picked up (out of 27 total) was the measure of helpfulness. As predicted, the money prime influenced helpfulness [F(2, 32) = 4.34, P < 0.03]. Participants in the high-money condition gathered fewer pencils than did participants in the low-money condition [t(32) = 2.75, P < 0.02; Cohens d = 0.81] or those in the control condition [t(32) = 2.13, P < 0.05; Cohens d = 1.23] (Table 1). Helpfulness did not differ between the low-money group and the control group [t < 1, not significant). Even though gathering pencils was an action that all participants could perform, participants reminded of financial wealth were unhelpful. Experiment 6 tested for the psychological effects of money by operationalizing helpfulness as monetary donations. Upon arrival to the laboratory, participants were given $2 in quarters in exchange for their participation. The quarters were said to have been used in an experiment that was now complete; in actuality, giving participants quarters ensured that they had money to donate (9). Participants were randomly assigned to one of two conditions, in which they descrambled phrases (as in Experiment 1) that primed money or neutral concepts. Then participants completed some filler questionnaires, after which the experimenter told them that the experiment was finished and gave them a false debriefing. This step was done so that participants would not connect the donation opportunity to the experiment. As the experimenter exited the room, she mentioned that the lab was taking donations for the University Student Fund and that there was a box by the door if the participant wished to donate. Amount of money donated was the measure of helping. We found that Participants primed with money donated significantly less money to the student fund than participants not primed with money [t(38) = 2.13, P < 0.05; Cohens d = 0.64] (Table 1). To convincingly demonstrate that money makes people self-sufficient, we tested the hypothesis in new contexts. The final experiments tested the effects of money on social intimacy, desire to engage in leisure activities alone, and preference to work alone. In Experiment 7,
Table 1. Helpfulness as a function of experimental condition in Experiments (Exp.) 3 to 6. The data are means SD; higher numbers indicate greater helpfulness. Within each experiment, means from the money and no-money conditions are different from each other at P < 0.05.
Exp. no. 3 Money condition 5.10 3.99 No-money condition 8.47 5.99 Dependent variable Number of data sheets participants volunteered to code Time spent helping a peer (seconds) Number of pencils gathered
4 5
67.35 84.65 18.00 1.96
147.81 158.15 20.30 1.77 (control) 19.72 2.28 (low money) 1.34 1.02
0.77 0.74
Monetary donations (in $)
www.sciencemag.org
SCIENCE
VOL 314
17 NOVEMBER 2006
1155
REPORTS
participants were randomly assigned to one of three priming conditions. Participants sat in front of a computer while completing questionnaires. After 6 min, one of three screensavers appeared. Participants in the money condition saw a screensaver depicting various denominations of currency floating underwater (fig. S1). Participants in the fish condition saw a screensaver with fish swimming underwater (fig. S2). Participants in the no-screensaver condition saw a blank screen. Afterwards, participants were told they would be having a get-acquainted conversation with another participant. Participants were asked to move two chairs together while the experimenter left to retrieve the other participant. The dependent measure was distance between the two chairs (10). Participants primed with money placed the two chairs farther apart than did participants in the fish condition [t(33) = 2.37, P < 0.05; Cohens d = 1.07] and the no-screensaver condition [t(33) = 2.30, P < 0.05; Cohens d = 0.85] (Table 2). Chair distance did not differ between fish and blank screensaver conditions [t(33) < 1, not significant]. Hence, participants primed with money put more physical distance between themselves and a new acquaintance than participants not primed with money. In Experiment 8, we tested whether moneyprimed participants would place a premium on being alone even when choosing leisure activities that could be enjoyed with friends and family. Participants were randomly assigned to one of three priming conditions. Participants first sat at a desk, which faced one of three posters, to complete filler questionnaires. In the money condition, the desk faced a poster showing a photograph of various denominations of currency (fig. S3). In two control conditions, the desk faced a poster showing either a seascape or a flower garden (figs. S4 and S5). Subsequently, participants were presented with a nine-item questionnaire that asked them to choose between two activities. Within each item, one option was an experience that only one person could enjoy and the other option was for two people or more (e.g., an in-home catered dinner for four versus four personal cooking lessons). Participants primed with money chose more individually focused leisure experiences than participants primed with either of the two neutral primes [F(2, 58) = 4.04, P < 0.05; money versus seascape: t(58) = 2.75, P < 0.05; Cohens d = 0.59; money versus flowers: t(58) = 2.10, P < 0.05; Cohens d = 1.06] (Table 2). The choice of activities did not differ between neutral conditions [t(58) < 1, not significant]. Thus, money primes lead people to be less social relative to those in nonmoney prime conditions. In Experiment 9, a more rigorous test of the self-sufficiency hypothesis was tested: We asked whether people reminded of money would choose to work alone. Working on a task with a co-worker presumably means less work for each person, but the co-worker may prefer to rely on the participant, which would be an affront to self-sufficiency. Participants were given the option of working on a project with a peer or alone. Participants were randomly assigned to three priming conditions. As in Experiment 7, screensavers showing money, fish, or no screensaver primed money or nonmoney concepts. Participants were then told that their next task was an advertisement development task on which they could work alone or with a peer. Participants were left alone to indicate their choice. Participants desire to work with a peer was significantly affected by priming condition [X 2(2, n = 37) = 10.10, P < 0.01] (Table 2). Choosing to perform the task with a co-worker was reduced among money condition participants relative to participants in both the fish [X2(1) = 7.00, P < 0.05; odds ratio = 11.25] and no-screensaver conditions [X 2(1) = 8.22, P < 0.05; odds ratio = 15.00]. There was no difference in choice between the fish and no-screensaver conditions [t(34) < 1, P > 0.05, not significant]. Nine experiments provided support for the hypothesis that money brings about a state of selfsufficiency. Relative to people not reminded of money, people reminded of money reliably performed independent but socially insensitive actions. The magnitude (11) of these effects is notable and somewhat surprising, given that our participants were highly familiar with money (12) and that our manipulations were minor environmental changes or small tasks for participants to complete. Research on the repercussions of studying economics dovetails nicely with our results. Frank, Gilovich, and Regan (13) reported that university students majoring in economics made self-interested moves in social dilemma games more often than students of other disciplines. Economics students also were more convinced than noneconomists that their competitors would make self-interested moves, a result that echoes the present thesis that money evokes a view that everyone fends for him- or herself. The self-sufficient pattern helps explain why people view money as both the greatest good and evil. As countries and cultures developed, money may have allowed people to acquire goods and services that enabled the pursuit of cherished goals, which in turn diminished reliance on friends and family. In this way, money enhanced individualism but diminished communal motivations, an effect that is still apparent in peoples responses to money today.
References and Notes
1. S. E. G. Lea, P. Webley, Behav. Brain Sci. 29, 161 (2006). 2. A. Furnham, M. Argyle, The Psychology of Money (Routledge, London, 1998). 3. P. R. Amato, S. J. Rogers, J. Marriage Fam. 59, 612 (1997). 4. The term self-sufficiency has been used in the psychological literature in two ways. One use (typically in research on recovery after injury) connotes a positive meaning of being free from needing others in order to effectively perform a task. The second use (typically in psychotherapy writings) takes on a discernibly negative meaning. Self-sufficiency in this case is considered a barrier to intimacy and is often seen in narcissistic personality disorders. Our use of the term incorporates both interpretations. We use self-sufficiency in part to suggest the autonomous agent who competently works toward personal goals, as well as the socially insensitive narcissist. We use the term not to suggest a stable trait (as in previous writings) but rather to signify a transitory psychological state brought on by reminders of money. 5. R. H. Price, J. N. Choi, A. D. Vinokur, J. Occup. Health Psychol. 7, 302 (2002). 6. Materials and methods are available as supporting material on Science Online. 7. T. K. Srull, R. S. Wyer, Jr., J. Pers. Soc. Psychol. 37, 1660 (1979). 8. R. S. Schwab, in Fatigue, W. F. Floyd, A. T. Welford, Eds. (Lewis, London, 1953), pp. 14348. 9. J. M. Twenge, R. F. Baumeister, C. N. DeWall, N. J. Ciarocco, J. M. Bartels, J. Pers. Soc. Psychol., in press. 10. C. N. Macrae, G. V. Bodenhausen, A. B. Milne, J. Jetten, J. Pers. Soc. Psychol. 67, 808 (1994). 11. J. Cohen, Psychol. Bull. 112, 155 (1992). 12. The majority of the participants in our experiments were raised in Canada, the United States, China, and Hong Kong (in decreasing order of prevalence). 13. R. H. Frank, T. Gilovich, D. T. Regan, Ethol. Sociol. 14, 247 (1993). 14. This work benefited from financial support from the Social Sciences and Humanities Research Council and the Canada Research Chair Council, both to K.V. We thank research assistants A. Boyce, R. Chan, L. Chen, A. Connolly, S. Curtis, V. Ding, S. Gonzalez, A. Kaikati, S. Sartain, J. Suydam, A. Talbot, and N. Van Den Berg.
Table 2. Social distance preferences as a function of experimental condition in Experiments (Exp.) 7 to 9. The data are means SD; higher numbers indicate preferences for greater social distance. In Experiments 7 and 9, the neutral 1 condition represents the fish screensaver condition, whereas the neutral 2 condition represents the no-screensaver condition. In Experiment 8, the neutral 1 condition represents the flower poster, whereas the neutral 2 condition represents the seascape poster. Within each experiment, means for the money condition differ from means in both neutral conditions at P < 0.05.
Exp. no. 7 Money condition 118.44 41.63 Neutral 1 condition 79.48 30.43 Neutral 2 condition 80.54 47.06 Dependent variable Physical distance between participant and partner (centimeters) Number of solitary activity selections Proportion of participants who opted to work alone
Supporting Online Material

www.sciencemag.org/cgi/content/full/314/5802/1154/DC1 Materials and Methods Figs. S1 to S5 References 14 July 2006; accepted 18 September 2006 10.1126/science.1132491
8 9
4.00 1.20 0.83 0.39
2.82 1.00 0.31 0.48
3.10 1.80 0.25 0.45
1156
17 NOVEMBER 2006
VOL 314
SCIENCE
www.sciencemag.org
PS YC HOLOGICA L SC IENCE
Research Article
Contagion and Differentiation in Unethical Behavior

The Effect of One Bad Apple on the Barrel
Francesca Gino,1 Shahar Ayal,2 and Dan Ariely2
1
University of North Carolina at Chapel Hill and 2Duke University
ABSTRACTIn
a world where encounters with dishonesty are frequent, it is important to know if exposure to other peoples unethical behavior can increase or decrease an individuals dishonesty. In Experiment 1, our confederate cheated ostentatiously by nishing a task impossibly quickly and leaving the room with the maximum reward. In line with social-norms theory, participants level of unethical behavior increased when the confederate was an in-group member, but decreased when the confederate was an out-group member. In Experiment 2, our confederate instead asked a question about cheating, which merely strengthened the saliency of this possibility. This manipulation decreased the level of unethical behavior among the other group members. These results suggest that individuals unethicality does not depend on the simple calculations of cost-benet analysis, but rather depends on the social norms implied by the dishonesty of others and also on the saliency of dishonesty. It is almost impossible to open a newspaper or turn on a television without being exposed to a report of dishonest behavior of one type or another. Names such as Enron, Tyco, and Arthur Andersen provide extreme examples; other examples include cheating on taxes, insurance fraud, employee theft, academic dishonesty, athletes use of illegal drugs, and of course illegal downloading of software and digital content. Given so many rst- and second-hand encounters with unethical behavior, one important question that comes to mind concerns the effect of such exposure on otherwise honest individuals. Do they tend to start engaging in unethical behavior? In the current work, we explored this very question by examining the conditions under which exposure to the unethical
behavior of another person increases or decreases individuals dishonesty.

OTHER INDIVIDUALS UNETHICAL BEHAVIOR: THREE SOURCES OF INFLUENCE
Address correspondence to Francesca Gino, Kenan-Flagler Business School, University of North Carolina, McColl Building, Room 4725, Chapel Hill, NC 27599-3490, e-mail: fgino@unc.edu.
The unethical behavior of other individuals can inuence observers behavior in (at least) three possible ways. First, when exposed to the dishonesty of others, individuals may change their estimate of the likelihood of being caught cheating (e.g., a student who sees a peer cheating on an exam and getting away with it may change his or her estimation of the probability of being caught in the act). Together with the amount to be gained from cheating and the expected punishment, the likelihood of being caught cheating is a central input in the rational crime theory (Allingham & Sandmo, 1972; Becker, 1968). In this rational framework, the individual engages in a cost-benet calculation that leads to the ultimate decision about dishonesty (support for this perspective is evident in work by Hill & Kochendorfer, 1969; Leming, 1980; Michaels & Miethe, 1989; Steininger, Johnson, & Kirts, 1964; Tittle & Rowe, 1973; and Vitro & Schoer, 1972). As a consequence of such cost-benet analysis, any change in the estimation of the likelihood of being caught cheating can inuence the magnitude of dishonesty an individual chooses to engage in (e.g., the student who sees a peer cheating on an exam and getting away with it changes his or her estimation of the probability of being caught in the act and is thus more likely to cheat). A second way in which observing others behavior may change ones own dishonesty concerns the saliency of ethicality at the moment one is considering a particular behavior. Previous research has shown that when the categorization of a particular behavior is not clear-cut, people can, and in fact often do, categorize their own actions in positive terms, avoiding negative updating to their moral self-image (Baumeister, 1998; Schweitzer & Hsee, 2002). However, Mazar, Amir, and Ariely
Volume 20Number 3
Copyright r 2009 Association for Psychological Science
393
Contagious Unethicality
(2008) found that drawing peoples attention to moral standards could reduce dishonest behaviors. For example, after being asked to recall the Ten Commandments, participants who were given the opportunity to cheat, and gain nancially from this action, did not cheat at all; in contrast, participants who had the same opportunity to cheat but had not been given the moral reminder cheated substantially. These results suggest that when unethical behavior is made salient, people may pay greater attention to their own moral standards and categorize the ethicality of their own behavior more rigidly. Such momentary uctuations in moral standards are also evident in a study by Vohs and Schooler (2008), who found that priming participants to believe in determinism (e.g., by making them read statements endorsing determinism) led to higher levels of dishonesty than inducing participants to believe in free will. The saliency hypothesis suggests that when people observe someone behaving dishonestly (e.g., when they read about a new corruption scheme), the saliency of this act increases, making them pay attention to honesty and to their own standards of honesty, and, as a consequence, decreasing their tendency to engage in dishonest acts. A third possible inuence of observing the unethicality of another person is that it simply changes ones understanding of the social norms related to dishonesty (Cialdini & Trost, 1998). Cialdini, Reno, and Kallgren (1990) dened two types of social norms: descriptive norms, which specify what most people do in a particular situation, and injunctive norms, which specify the particular behaviors that most people approve or disapprove of. According to norm-focus theory (Cialdini et al., 1990; Reno, Cialdini, & Kallgren, 1993), the social context determines which of these two types of norms people attend to at a particular time and how these norms will impinge on an individuals immediate behavior. For example, Cialdini et al. had a confederate litter in the environment in front of some participants or simply walk through the environment in front of others. Participants who saw the confederate litter subsequently littered more than those who did not see the confederate litter if the environment was clean, but this effect was reversed when the environment was dirty. This kind of social learning by observing other peoples behavior was also demonstrated in Banduras classic studies (Bandura, 1965; Bandura, Ross, & Ross, 1961, 1963), in which children exposed to an aggressive model reproduced considerably more aggressive behaviors toward a Bobo doll than did children who were not exposed to the aggressive model. Moreover, children reproduced more aggressive behaviors when an adult did not comment on the aggressive models actions (or when an adult was not present in the room) than when the adult disapproved of those actions using negative comments (Hicks, 1968; Siegel & Kolin, 1959). Children might have interpreted the lack of evaluative comments on the models aggressive behavior and the absence of an adult in the room as signs of permission, through social norms.
The social-norms account implies another important factor that might inuence the degree to which people are affected by the unethical behavior of others around them: the degree to which they identify with those others. The idea is that when the identication is strong, the behaviors of others will have a larger inuence on observers social norms. Field evidence for this idea was obtained in a large-scale survey of Australian citizens (Wenzel, 2004), which found that the presence of social norms elicited consistent behavior (i.e., tax compliance), but only when respondents identied with the group to which the norms were attributed. These ndings can be explained by social-identity theory (Tajfel, 1982; Tajfel & Turner, 1979, 1986), according to which group members tend to use their own group to maintain or enhance a positive social identity and self-esteem, and as a consequence are motivated to conform with norms that provide them with an in-group identity, rather than an out-group one (see also Rubin & Hewstone, 1998). On the basis of social-identity theory, we hypothesized that the inuence of the social norm triggered by observed unethical behavior will depend on whether the actor is an in-group or outgroup member. When an in-group member is observed engaging in unethical behavior, other group members may make him or her the standard for the descriptive norm and, as a result, engage in increased unethical behavior themselves. In contrast, when an out-group member engages in unethical behavior, the injunctive norm is likely to be salient, and non-group members may want to distance themselves from this bad apple in order to maintain a distinctive and positive social identity (Brewer, 1993; Tajfel & Turner, 1986). In summary, the three mechanisms we have described (i.e., a change in the estimated likelihood of being caught, saliency, and social norms) lead to different predictions about the inuence of being exposed to other peoples dishonesty. The estimated-likelihood-of-being-caught mechanism suggests that observing another person behaving dishonestly will increase ones propensity to act dishonestly. The saliency mechanism suggests that observing another person behaving dishonestly will decrease ones propensity to act dishonestly. Finally, the social-norms mechanism suggests that when the observed other is an out-group member, in-group members will show a reduced likelihood of engaging in dishonest behaviors, but when the observed other is an in-group member, the social norms will change and the other members of the group will be more likely to engage in dishonest behavior. We conducted two laboratory experiments to investigate how cost-benet analysis, saliency, and social norms interact to promote or inhibit unethical behavior. In these experiments, participants were asked to solve simple math problems in the presence of others. In some of the conditions, participants were given the opportunity to cheat by misreporting their performance and earning undeserved money; this manipulation allowed us to measure their level of dishonesty. Most important, in some of the conditions, participants were exposed to a confed-
394
Volume 20Number 3
Francesca Gino, Shahar Ayal, and Dan Ariely
erate who either exhibited dishonesty or merely asked a question about the possibility of cheating; this manipulation allowed us to test the three proposed mechanisms by which observing the behavior of others can inuence levels of dishonesty.
EXPERIMENT 1: EFFECTS OF THE CONFEDERATES IDENTITY
The rst experiment consisted of four conditions. The control condition, which gave participants no opportunity to cheat, served as a baseline for performance on the task. The second conditionthe shredder conditionserved as a baseline for the magnitude of cheating when participants had the opportunity to cheat but were not exposed to someones dishonesty. For the nal two conditions (in-group-identity and out-group-identity), we hired a professional actor who served as a confederate and made it clear to the participants at the onset of the experiment that he was cheating to the greatest extent possible. By comparing these four conditions, we were able to test the inuence of the three mechanisms for dishonesty: According to the cost-benet (rational) perspective, the confederates behavior, whether in the in-group-identity or the out-group-identity condition, should have made it clear that participants could get away with cheating, and hence should have increased the tendency to cheat; thus, this perspective predicted that cheating would be lowest in the control condition, at an intermediate level in the shredder condition, and highest in the in-group-identity and out-group identity conditions. According to the saliency mechanism, cheating after exposure to the confederates unethical behavior would be expected to be lower than the cheating baseline (i.e., in the shredder condition), regardless of the confederates identity; thus, this perspective predicted that cheating would be lowest in the control condition, intermediate in the in-group-identity and out-group-identity conditions, and highest in the shredder condition. Finally, according to the social-norms account, cheating would be highest when the confederate was a member of the in-group, and much lower when he was a member of the out-group and participants wanted to distance themselves from him; thus, this perspective predicted that cheating would be lowest in the control condition, higher in the out-group-identity condition, still higher in the shredder condition, and highest in the in-group-identity condition.
Method Participants One hundred forty-one Carnegie Mellon University students (79 male, 62 female) participated in the study for a maximum payment of $10. Participants were randomly assigned to one of four conditions: control, shredder, shredder with in-group confederate, and shredder with out-group confederate. The average age of participants was 22 years (SD 5 4.57).
Design and Procedure Twelve sessions, each lasting about 15 min, were conducted. We randomly assigned conditions to sessions. Between 8 and 14 participants were included in each session (average 5 12). Each participant received a brown envelope that contained $10 (eight $1 bills and four half-dollar coins) and an empty white envelope. Participants also received two sheets of paper: The rst was a worksheet with 20 matrices, each containing 12 numbers consisting of an integer and two decimals (e.g., 6.39), and the second was a collection slip on which participants were supposed to report their performance and answer questions about their gender and age. Once the experiment started, participants had 5 min to nd two numbers per matrix that added up to 10. The allotted time was not sufcient for anyone to solve all 20 matrices. For each pair of numbers correctly identied, participants were allowed to keep $0.50 from their supply of money, and at the end of the allotted time, they were asked to transfer the unearned amount to the white envelope. There were two boxes in the room: a blue recycling box for the questionnaires and a cardboard box for the white envelopes containing unearned money. In the control condition, the two boxes were located on the experimenters desk. After the 5 min had passed, participants were asked to line up near the desk and hand the test to the experimenter. The experimenter checked how many matrices each participant had solved correctly, wrote down that score on the collection slip, and deposited both sheets in the blue recycling box. Next, the experimenter made sure the participant left the correct amount of unearned money in the white envelope (based on the participants performance) and deposited the white envelope into the cardboard box. In the shredder condition, the boxes were located in two different corners of the classroom, with the recycling box standing next to an electronic paper shredder. After the 5 min had passed, participants were asked to count the number of matrices they had solved correctly, write this number down on the collection slip, walk to the electronic shredder, and shred their worksheet. Next, participants transferred their unearned money from the brown envelope into the white envelope, and placed the white envelope and the collection slip in the cardboard box. During this procedure, the experimenter remained at her desk and did not engage in any process to check that the participants followed her instructions. Finally, in the two identity conditions (the in-group-identity and the out-group-identity conditions), we hired a professional actor to be our confederate. These conditions followed the same general procedure used in the shredder condition, with one difference: About 60 s after the experiment started (such a short time that it would have been clear to the participants that the person was lying or cheating), the confederate stood up and said loudly: Ive solved everything. What should I do? The experimenter reminded him about the procedure (shred, transfer money, deposit the white envelope and collection sheet). When the actor nished with the shredding, he said: I solved every-
Volume 20Number 3
395
thing. My envelope for the unearned money is empty. What should I do with it? The experimenter replied: If you dont have money to return, you have nished and are free to go. The actor then said, I want to thank you. I wish all of you a wonderful day. He left the room smiling, without going to the returnedmoney box. The only difference between the in-group-identity and the out-group-identity conditions was the T-shirt that the confederate was wearing. Because the study was conducted at Carnegie Mellon University, the confederate wore a plain T-shirt in the in-group-identity condition and a University of Pittsburgh T-shirt in the out-group-identity condition (see Levine, Prosser, Evans, & Reicher, 2005, for a similar manipulation of in-group and out-group membership in a laboratory setting). Results and Discussion A hierarchical linear model with session as a factor nested under condition revealed signicant differences in the number of matrices that were reported to have been solved correctly across the four conditions, F(3, 129) 5 26.01, p < .001. The effect of session was not signicant. As Figure 1 shows, participants in the control condition reported the lowest number of correctly solved matrices, and the level of cheating in the shredder condition was elevated by more than 50%. Most important, the level of cheating was dramatically inuenced by the confederate, such that cheating increased further in the in-group-identity condition and decreased in the out-group-identity condition. Simple planned contrasts demonstrated that the number of matrices reported to be solved correctly increased from the control to the shredder condition (p < .001). In addition, this number increased signicantly from the shredder condition to the in-group-identity condition (p < .001) and decreased signicantly from the shredder condition to the out-group-identity condition (p < .01). Finally, the difference between the outgroup-identity condition and the control condition was also signicant (p < .05), suggesting that although the out-group manipulation reduced the level of cheating, it did not eliminate
cheating. Overall, the pattern of results supports the socialnorms hypothesis, as cheating was lowest in the control condition, higher in the out-group-identity condition, still higher in the shredder condition, and highest in the in-group-identity condition. The results do not support the cost-benet perspective (according to which the level of cheating would be highest and approximately the same in the two identity conditions) or the saliency hypothesis (according to which the level of cheating would be intermediate and approximately the same in the two identity conditions). The social-norms hypothesis was also supported by the percentage of participants who imitated the confederates behavior and reported solving all 20 matrices (none of them claimed to have solved all the problems before the end of the 5-min period). This percentage was signicantly higher in the in-group-identity condition (24.3%, 9 out of 37) than in the out-group-identity condition (3.6%, 1 out of 28), w2(1, N 5 65) 5 5.27, p < .05.
EXPERIMENT 2: EFFECTS OF SALIENCY
Number of Questions (out of 20)
16 14 12 10 8 6 4 2 0 Control Shredder In-Group Out-Group
The results of Experiment 1 made it clear that social norms and, in particular, exposure to a dishonest act by an in-group versus an out-group memberhave a large effect on dishonesty. In addition, given that the probability of being caught was the same in the two identity conditions, Experiment 1 also made it clear that cost-benet trade-offs cannot be a main cause for dishonesty (at least in our setting). However, because the identity manipulation used in Experiment 1 involved both identity and salience, it was premature to dismiss the saliency account, as saliency and social norms may have worked in concert. In Experiment 2, we tested the independent effect of saliency. This experiment consisted of three conditions: the control condition, which gave participants no opportunity to cheat; the shredder condition, which provided participants an opportunity to cheat; and the saliency condition, in which participants had an opportunity to cheat and a confederate strengthened the saliency of cheating behavior without changing the groups norms. If the saliency hypothesis is correct, participants will cheat when they have the opportunity to do so, but will cheat to a smaller extent when the saliency of honesty increases.
Method Participants Ninety-two students (49 male, 43 female) participated in this experiment for extra class credit in their introductory business courses. In addition, participants received a monetary payment based on their self-reported performance on the task. Participants average age was 20 years (SD 5 1.16). The participants were randomly assigned to one of the three conditions.
Experimental Condition
Fig. 1. Number of problems reported to have been solved correctly in Experiment 1 as a function of experimental condition. Error bars represent standard errors.
396
Volume 20Number 3
Francesca Gino, Shahar Ayal, and Dan Ariely
Design and Procedure Six sessions, each lasting about 15 min, were conducted. We randomly assigned conditions to sessions. Between 15 and 17 participants were included in each session (average 5 15). The control and the shredder conditions matched the corresponding conditions in Experiment 1 exactly, but the saliency condition was novel. The saliency condition followed the same general procedure used in the shredder condition, with one difference: When the experimenter gave instructions explaining the procedure for the study, the actor, posing as a participant, asked aloud, So, is it OK to cheat? The experimenter answered, You can do whatever you want. Aside from this planned interruption, the confederate behaved in the same way as all the participants throughout the session. The logic for this manipulation was that it increased the saliency of cheating, but did not create any social norm because the confederate did not leave the room before any of the other participants, so that his actions were not observed. Note that from the rational cost-benet perspective, the response of the experimenter (You can do whatever you want) should have increased cheating because this statement made it clear that there were no negative consequences to such behavior. Results and Discussion A hierarchical linear model with session as a factor nested under condition revealed signicant differences in the number of matrices that were reported to have been solved correctly across the four conditions, F(2, 86) 5 8.47, p < .001. The effect of session was not signicant. Participants in the control condition, who did not have the opportunity to cheat, reported the lowest number of matrices solved; participants in the shredder condition reported the highest number of matrices solved; and participants in the saliency condition reported an intermediate number of matrices solved (see Fig. 2). Simple planned contrasts demonstrated that differences across the three conditions were all signicant at the .05 level.
These results conrm previous ndings by Mazar et al. (2008) showing that study participants cheated when given the opportunity to do so, although not to a large degree. More important, the addition of the confederates question to the shredder manipulation (the saliency condition) decreased the level of cheating. This effect of the saliency condition casts further doubt on the applicability of the cost-benet perspective on dishonesty, because, if anything, the noncommittal answer of the experimenter should have decreased the perceived cost of acting dishonestly and increased cheating. At the same time, this condition supports the role of saliency in reducing dishonesty, suggesting that both social norms and saliency contribute to the exhibited level of (dis)honesty in society. Finally, it is possible that the results observed in Experiment 1 were due to the combined inuence of social norms and saliency. If so, and if saliency reduced the extent of cheating in Experiment 1 in the same way that it operated in Experiment 2, one could hypothesize that the inuence of social norms (as seen in our in-group-identity condition) could result in even higher levels of dishonesty if the saliency of these acts were reduced. This is indeed a worrisome speculation.
GENERAL DISCUSSION
16 14 12 10 8 6 4 2 0 Control Shredder Saliency
Experimental Condition
Fig. 2. Number of problems reported to have been solved correctly in Experiment 2 as a function of experimental condition. Error bars represent standard errors.
The results of the two experiments show that people react to the unethical behavior of others, and that their reaction depends on the social norms implied by the observed dishonesty and also on the saliency of dishonesty. Our results also show that reaction to the dishonesty of others does not seem to depend on changes in the calculations of cost-benet analysis. In Experiment 1, observing an in-group peer engaging in unethical behavior increased participants likelihood of acting unethically themselves. However, observing an out-group peer engaging in unethical behavior reduced participants likelihood of acting unethically themselves. In Experiment 2, we tested the independent effect of saliency and found support for the idea that when the saliency of dishonesty increases (at least when social norms are not implied), cheating decreases. Taken together, the ndings suggest that peer inuence is an important factor in unethical behavior. Prior research has shown that ethical climate and ethical culture are important predictors of the frequency of unethical acts within groups and organizational settings (for a review, see Loe, Ferrell, & Manseld, 2000, or Ford & Richardson, 1994). Although we recognize the importance of such macrocomponents, we believe that microelements, such as the behavior of one particular individual in a group, can also have large consequences. The question of whether unethical behaviors, such as cheating, stealing, and dishonesty, are contagious is fundamental to both organizations and society. Healthy work and social environments depend on the ability of individuals (e.g., leaders and other role models) to spread ethical norms and values, while reducing the attractiveness of unethical misconduct, either
Number of Questions (out of 20)
Volume 20Number 3
397
through appropriate sanctioning rules or through an ethical culture. Our ndings suggest that relatively minor acts of dishonesty by in-group members can have a large inuence on the extent of dishonesty, and that techniques that help to stigmatize the bad apples as out-group members and strengthen the saliency of their behavior could be useful tools to ght dishonesty. As Anthony Eden stated, Corruption never has been compulsory (quoted by wikiquote.org, 2007). But, as our results show, under certain conditions, dishonest behavior can be contagious. AcknowledgmentsThe authors gratefully acknowledge comments from David Harrison and Maurice Schweitzer and also thank two anonymous reviewers for their useful suggestions and guidance during the revision process. The authors greatly appreciate the support and facilities of the Center for Behavioral Decision Research at Carnegie Mellon University, where the experiments were conducted.
REFERENCES Allingham, M.G., & Sandmo, A. (1972). Income tax evasion: A theoretical analysis. Journal of Public Economics, 1, 323338. Bandura, A. (1965). Inuence of models reinforcement contingencies on the acquisition of imitative responses. Journal of Personality and Social Psychology, 1, 589595. Bandura, A., Ross, D., & Ross, S.A. (1961). Transmission of aggression through imitation of aggressive models. Journal of Abnormal and Social Psychology, 63, 575582. Bandura, A., Ross, D., & Ross, S.A. (1963). Imitation of lm-mediated aggressive models. Journal of Abnormal and Social Psychology, 66, 311. Baumeister, R.F. (1998). The self. In D.T. Gilbert, S.T. Fiske, & G. Lindzey (Eds.), Handbook of social psychology (Vol. 1, pp. 680 740). New York: McGraw-Hill. Becker, G.S. (1968). Crime and punishment: An economic approach. Journal of Political Economy, 76, 169217. Brewer, M. (1993). The role of distinctiveness in social identity and group behavior. In M.A. Hogg & D. Abrams (Eds.), Group motivation: Social psychological perspectives (pp. 116). New York: Harvester Wheatsheaf. Cialdini, R.B., Reno, R.R., & Kallgren, C.A. (1990). A focus theory of normative conduct: Recycling the concept of norms to reduce littering in public places. Journal of Personality and Social Psychology, 58, 10151026. Cialdini, R.B., & Trost, M.R. (1998). Social inuence: Social norm, conformity, and compliance. In D.T. Gilbert, S.T. Fiske, & G. Lindzey (Eds.), Handbook of social psychology (Vol. 2, pp. 151 192). New York: McGraw-Hill. Ford, R.C., & Richardson, W.D. (1994). Ethical decision making: An overview of the empirical literature. Journal of Business Ethics, 13, 205221. Hicks, D.J. (1968). Effects of co-observers sanctions and adult presence on imitative aggression. Child Development, 39, 303309.
Hill, J., & Kochendorfer, R.A. (1969). Knowledge of peer success and risk of detection as determinants of cheating. Developmental Psychology, 1, 231238. Leming, J.S. (1980). Cheating behavior, subject variables, and components of the internal-external scale under high and low risk conditions. Journal of Educational Research, 74, 8387. Levine, M., Prosser, A., Evans, D., & Reicher, S. (2005). Identity and emergency intervention: How social group membership and inclusiveness of group boundaries shape helping behavior. Personality and Social Psychology Bulletin, 31, 443453. Loe, T.W., Ferrell, L., & Manseld, P. (2000). A review of empirical studies assessing ethical decision making in business. Journal of Business Ethics, 25, 185204. Mazar, N., Amir, O., & Ariely, D. (2008). The dishonesty of honest people: A theory of self-concept maintenance. Journal of Marketing Research, 45, 633644. Michaels, J.W., & Miethe, T.D. (1989). Applying theories of deviance to academic cheating. Social Science Quarterly, 70, 870885. Reno, R.R., Cialdini, R.B., & Kallgren, C.A. (1993). The transsituational inuence of social norms. Journal of Personality and Social Psychology, 64, 104112. Rubin, M., & Hewstone, M. (1998). Social identity theorys self-esteem hypothesis: A review and some suggestions for clarication. Personality and Social Psychology Review, 2, 4062. Schweitzer, M.E., & Hsee, C.K. (2002). Stretching the truth: Elastic justication and motivated communication of uncertain information. Journal of Risk and Uncertainty, 25, 185201. Siegel, A.E., & Kohn, L.G. (1959). Permissiveness, permission, and aggression: The effect of adult presence or absence on aggression in childrens play. Child Development, 30, 131141. Steininger, M., Johnson, R.E., & Kirts, D.K. (1964). Cheating on college examinations as a function of situationally aroused anxiety and hostility. Journal of Educational Psychology, 55, 317 324. Tajfel, H. (1982). Social identity and intergroup relations. Cambridge, England: Cambridge University Press. Tajfel, H., & Turner, J.C. (1979). An integrative theory of intergroup conict. In W.G. Austin & S. Worchel (Eds.), The social psychology of intergroup relations (pp. 3347). Pacic Grove, CA: Brooks/Cole. Tajfel, H., & Turner, J.C. (1986). The social identity theory of intergroup behavior. In S. Worchel & W.G. Austin (Eds.), Psychology of intergroup relations (pp. 724). Chicago: Nelson-Hall. Tittle, C.R., & Rowe, A.R. (1973). Moral appeal, sanction threat and deviance: An experimental test. Social Problems, 20, 488498. Vitro, F.T., & Schoer, L.A. (1972). The effects of probability of test success, test importance, and risk of detection on the incidence of cheating. Journal of School Psychology, 10, 269277. Vohs, K.D., & Schooler, J.W. (2008). The value of believing in free will: Encouraging a belief in determinism increases cheating. Psychological Science, 19, 4954. Wenzel, M. (2004). An analysis of norm processes in tax compliance. Journal of Economic Psychology, 25, 213228. wikiquote.org. (2007). Quotes about corruption. Retrieved January 9, 2008, from http://en.wikiquote.org/wiki/Corruption (RECEIVED 7/14/08; REVISION ACCEPTED 9/12/08)
398
Volume 20Number 3
NINA MAZAR, ON AMIR, and DAN ARIELY*

People like to think of themselves as honest. However, dishonesty paysand it often pays well. How do people resolve this tension? This research shows that people behave dishonestly enough to profit but honestly enough to delude themselves of their own integrity. A little bit of dishonesty gives a taste of profit without spoiling a positive self-view. Two mechanisms allow for such self-concept maintenance: inattention to moral standards and categorization malleability. Six experiments support the authors theory of self-concept maintenance and offer practical applications for curbing dishonesty in everyday life.
Keywords: honesty, decision making, policy, self
The Dishonesty of Honest People: A Theory of Self-Concept Maintenance

It is almost impossible to open a newspaper or turn on a television without being exposed to a report of dishonest behavior of one type or another. To give a few examples, wardrobingthe purchase, use, and then return of the used clothingcosts the U.S. retail industry an estimated $16 billion annually (Speights and Hilinski 2005); the overall magnitude of fraud in the U.S. property and casualty insurance industry is estimated to be 10% of total claims payments, or $24 billion annually (Accenture 2003); and the tax gap, or the difference between what the Internal Revenue Service estimates taxpayers should pay and what they actually pay, exceeds $300 billion annually (more than 15% noncompliance rate; Herman 2005). If this evidence is not disturbing enough, perhaps the largest contribution to dishonesty comes from employee theft and fraud, which has been estimated at $600 billion a year in the United States alonean amount almost twice the market capitalization of General Electric (Association of Certified Fraud Examiners 2006). WHY ARE PEOPLE (DIS)HONEST? Rooted in the philosophies of Thomas Hobbes, Adam Smith, and the standard economic model of rational and
*Nina Mazar is Assistant Professor of Marketing, Joseph L. Rotman School of Management, University of Toronto (e-mail: nina.mazar@ utoronto.ca). On Amir is Assistant Professor of Marketing, Rady School of Management, University of California, San Diego (e-mail: oamir@ucsd. edu). Dan Ariely is Visiting Professor of Marketing, Fuqua School of Business, Duke University (e-mail: dandan@duke.edu). The authors thank Daniel Berger, Anat Bracha, Aimee Drolee, and Tiffany Kosolcharoen for their help in conducting the experiments, as well as Ricardo E. Paxson for his help in creating the matrices. Pierre Chandon served as associate editor and Ziv Carmon served as guest editor for this article.
selfish human behavior (i.e., homo economicus) is the belief that people carry out dishonest acts consciously and deliberatively by trading off the expected external benefits and costs of the dishonest act (Allingham and Sandmo 1972; Becker 1968). According to this perspective, people would consider three aspects as they pass a gas station: the expected amount of cash they stand to gain from robbing the place, the probability of being caught in the act, and the magnitude of punishment if caught. On the basis of these inputs, people reach a decision that maximizes their interests. Thus, according to this perspective, people are honest or dishonest only to the extent that the planned trade-off favors a particular action (Hechter 1990; Lewicki 1984). In addition to being central to economic theory, this external costbenefit view plays an important role in the theory of crime and punishment, which forms the basis for most policy measures aimed at preventing dishonesty and guides punishments against those who exhibit dishonest behavior. In summary, this standard external costbenefit perspective generates three hypotheses as to the forces that are expected to increase the frequency and magnitude of dishonesty: higher magnitude of external rewards (Ext-H1), lower probability of being caught (Ext-H2), and lower magnitude of punishment (Ext-H3). From a psychological perspective, and in addition to financial considerations, another set of important inputs to the decision whether to be honest is based on internal rewards. Psychologists show that as part of socialization, people internalize the norms and values of their society (Campbell 1964; Henrich et al. 2001), which serve as an internal benchmark against which a person compares his of her behavior. Compliance with the internal values system provides positive rewards, whereas noncompliance leads to negative rewards (i.e., punishments). The most direct evi633 Journal of Marketing Research Vol. XLV (December 2008), 633644
2008, American Marketing Association ISSN: 0022-2437 (print), 1547-7193 (electronic)
634
JOURNAL OF MARKETING RESEARCH, DECEMBER 2008 they derive some financial benefit from behaving dishonestly but still maintain their positive self-concept in terms of being honest. To be more precise, we posit a magnitude range of dishonesty within which people can cheat, but their behaviors, which they would usually consider dishonest, do not bear negatively on their self-concept (i.e., they are not forced to update their self-concept).1 Although many mechanisms may allow people to find such a compromise, we focus on two particular means: categorization and attention devoted to ones own moral standards. Using these mechanisms, people can record their actions (e.g., I am claiming $x in tax exemptions) without confronting the moral meaning of their actions (e.g., I am dishonest). We focus on these two mechanisms because they support the role of the self-concept in decisions about honesty and because we believe that they have a wide set of important applications in the marketplace. Although not always mutually exclusive, we elaborate on each separately. Categorization We hypothesize that for certain types of actions and magnitudes of dishonesty, people can categorize their actions into more compatible terms and find rationalizations for their actions. As a consequence, people can cheat while avoiding any negative self-signals that might affect their self-concept and thus avoid negatively updating their selfconcept altogether (Gur and Sackeim 1979). Two important aspects of categorization are its relative malleability and its limit. First, behaviors with malleable categorization are those that allow people to reinterpret them in a self-serving manner, and the degree of malleability is likely to be determined by their context. For example, intuition suggests that it is easier to steal a $.10 pencil from a friend than to steal $.10 out of the friends wallet to buy a pencil because the former scenario offers more possibilities to categorize the action in terms that are compatible with friendship (e.g., my friend took a pencil from me once; this is what friends do). This thought experiment suggests not only that a higher degree of categorization malleability facilitates dishonesty (stealing) but also that some actions are inherently less malleable and therefore cannot be categorized successfully in compatible terms (Dana, Weber, and Kuang 2005; for a discussion of the idea that a medium, such as a pen, can disguise the final outcome of an action, such as stealing, see Hsee et al. 2003). In other words, as the categorization malleability increases, so does the magnitude of dishonesty to which a person can commit without influencing his or her self-concept (Baumeister 1998; Pina e Cunha and Cabral-Cardoso 2006; Schweitzer and Hsee 2002). The second important aspect of the categorization process pertains to its inherent limit. The ability to categorize behaviors in ways other than as dishonest or immoral can be incredibly useful for the self, but it is difficult to imagine that this mechanism is without limits. Instead, it may be possible to stretch the truth and the bounds of mental representations only up to a certain point (what
dence of the existence of such internal reward mechanisms comes from brain imaging studies that reveal that acts based on social norms, such as altruistic punishment or social cooperation (De Quervain et al. 2004; Rilling et al. 2002), activate the same primary reward centers in the brain (i.e., nucleus accumbens and caudate nucleus) as external benefits, such as preferred food, drink, and monetary gains (Knutson et al. 2001; ODoherty et al. 2002). Applied to the context of (dis)honesty, we propose that one major way the internal reward system exerts control over behavior is by influencing peoples self-conceptthat is, the way people view and perceive themselves (Aronson 1969; Baumeister 1998; Bem 1972). Indeed, it has been shown that people typically value honesty (i.e., honesty is part of their internal reward system), that they have strong beliefs in their own morality, and that they want to maintain this aspect of their self-concept (Greenwald 1980; Griffin and Ross 1991; Josephson Institute of Ethics 2006; Sanitioso, Kunda, and Fong 1990). This means that if a person fails to comply with his or her internal standards for honesty, he or she will need to negatively update his or her selfconcept, which is aversive. Conversely, if a person complies with his or her internal standards, he or she avoids such negative updating and maintains his or her positive selfview in terms of being an honest person. Notably, this perspective suggests that to maintain their positive selfconcepts, people will comply with their internal standards even when doing so involves investments of effort or sacrificing financial gains (e.g., Aronson and Carlsmith 1962; Harris, Mussen, and Rutherford 1976; Sullivan 1953). In our gas station example, this perspective suggests that people who pass by a gas station will be influenced not only by the expected amount of cash they stand to gain from robbing the place, the probability of being caught, and the magnitude of punishment if caught but also by the way the act of robbing the store might make them perceive themselves. The utility derived from behaving in line with the selfconcept could conceivably be just another part of the cost benefit analysis (i.e., adding another variable to account for this utility). However, even if we consider this utility just another input, it probably cannot be manifested as a simple constant, because the influence of dishonest behavior on the self-concept will most likely depend on the particular action, its symbolic value, its context, and its plasticity. In the following sections, we characterize these elements in a theory of self-concept maintenance and test the implications of this theory in a set of six experiments. THE THEORY OF SELF-CONCEPT MAINTENANCE People are often torn between two competing motivations: gaining from cheating versus maintaining a positive self-concept as honest (Aronson 1969; Harris, Mussen, and Rutherford 1976). For example, if people cheat, they could gain financially but at the expense of an honest selfconcept. In contrast, if they take the high road, they might forgo financial benefits but maintain their honest selfconcept. This seems to be a winlose situation, such that choosing one path involves sacrificing the other. In this work, we suggest that people typically solve this motivational dilemma adaptively by finding a balance or equilibrium between the two motivating forces, such that
1Our self-concept maintenance theory is based on how people define honesty and dishonesty for themselves, regardless of whether their definition matches the objective definition.
The Dishonesty of Honest People Piaget [1950] calls assimilation and accommodation). If we assume that the categorization process has such built-in limits, we should conceptualize categorization as effective only up to a threshold, beyond which people can no longer avoid the obvious moral valence of their behavior. Attention to Standards The other mechanism that we address in the current work is the attention people pay to their own standards of conduct. This idea is related to Duval and Wicklunds (1972) theory of objective self-awareness and Langers (1989) concept of mindlessness. We hypothesize that when people attend to their own moral standards (are mindful of them), any dishonest action is more likely to be reflected in their self-concept (they will update their self-concept as a consequence of their actions), which in turn will cause them to adhere to a stricter delineation of honest and dishonest behavior. However, when people are inattentive to their own moral standards (are mindless of them), their actions are not evaluated relative to their standards, their self-concept is less likely to be updated, and, therefore, their behavior is likely to diverge from their standards. Thus, the attentionto-standards mechanism predicts that when moral standards are more accessible, people will need to confront the meaning of their actions more readily and therefore be more honest (for ways to increase accessibility, see Bateson, Nettle, and Roberts 2006; Bering, McLeod, and Shackelford 2005; Diener and Wallbom 1976; Haley and Fessler 2005). In this sense, greater attention to standards may be modeled as a tighter range for the magnitude of dishonest actions that does not trigger updating of the self-concept or as a lower threshold up to which people can be dishonest without influencing their self-concept. Categorization and Attention to Standards Whereas the categorization mechanism depends heavily on stimuli and actions (i.e., degree of malleability and magnitude of dishonesty), the attention-to-standards mechanism relies on internal awareness or salience. From this perspective, these two mechanisms are distinct; the former focuses on the outside world, and the latter focuses on the inside world. However, they are related in that they both involve attention, are sensitive to manipulations, and are related to the dynamics of acceptable boundaries of behavior. Thus, although the dishonesty that both self-concept maintenance mechanisms allow stems from different sources, they both tap the same basic concept. Moreover, in many real-world cases, these mechanisms may be so interrelated that it would be difficult to distinguish whether the source of this type of dishonesty comes from the environment (categorization) or the individual (attention to standards). In summary, the theory of self-concept maintenance that considers both external and internal reward systems suggests the following hypotheses:
Ext&Int-H1: Dishonesty increases as attention to standards for honesty decreases. Ext&Int-H2: Dishonesty increases as categorization malleability increases. Ext&Int-H3: Given the opportunity to be dishonest, people are dishonest up to a certain level that does not force them to update their self-concept.
635 EXPERIMENT 1: INCREASING ATTENTION TO STANDARDS FOR HONESTY THROUGH RELIGIOUS REMINDERS The general setup of all our experiments involves a multiple-question task, in which participants are paid according to their performance. We compare the performance of respondents in the control conditions, in which they have no opportunity to be dishonest, with that of respondents in the cheating conditions, in which they have such an opportunity. In Experiment 1, we test the prediction that increasing peoples attention to their standards for honesty will make them more honest by contrasting the magnitude of dishonesty in a condition in which they are reminded of their own standards for honesty with a condition in which they are not. On the face of it, the idea that any reminder can decrease dishonesty seems strange; after all, people should know that it is wrong to be dishonest, even without such reminders. However, from the self-concept maintenance perspective, the question is not whether people know that it is wrong to behave dishonestly but whether they think of these standards and compare their behavior with them in the moment of temptation. In other words, if a mere reminder of honesty standards has an effect, we can assert that people do not naturally attend to these standards. In Experiment 1, we implement this reminder through a simple recall task. Method Two hundred twenty-nine students participated in this experiment, which consisted of a two-task paradigm as part of a broader experimental session with multiple, unrelated paper-and-pencil tasks that appeared together in a booklet. In the first task, we asked respondents to write down either the names of ten books they had read in high school (no moral reminder) or the Ten Commandments (moral reminder). They had two minutes to complete this task. The idea of the Ten Commandments recall task was that independent of peoples religion, of whether people believed in God, or of whether they knew any of the commandments, knowing that the Ten Commandments are about moral rules would be enough to increase attention to their own moral standards and thus increase the likelihood of behavior consistent with these standards (for a discussion of reminders of God in the context of generosity, see Shariff and Norenzayan 2007). The second, ostensibly separate task consisted of two sheets of paper: a test sheet and an answer sheet. The test sheet consisted of 20 matrices, each based on a set of 12 three-digit numbers. Participants had four minutes to find two numbers per matrix that added up to 10 (see Figure 1). We selected this type of task because it is a search task, and though it can take some time to find the right Figure 1
A SAMPLE MATRIX OF THE ADDING-TO-10 TASK
1.69 4.67 5.82 6.36
1.82 4.81 5.06 5.19
2.91 3.05 4.28 4.57
636
JOURNAL OF MARKETING RESEARCH, DECEMBER 2008 izenry overclaims on taxes, the very act of overclaiming is generally accepted and therefore not necessarily considered immoral. However, if this interpretation accounted for our findings, increasing peoples attention to morality (Ten Commandments/recycle condition) would not have decreased the magnitude of dishonesty. Therefore, we interpreted these findings as providing initial support for the self-concept maintenance theory. Note also that, on average, participants remembered only 4.3 of the Ten Commandments, and we found no significant correlation between the number of commandments recalled and the number of matrices the participants claimed to have solved correctly (r = .14, p = .29). If we use the number of commandments remembered as a proxy for religiosity, the lack of relationship between religiosity and the magnitude of dishonesty suggests that the efficacy of the Ten Commandments is based on increased attention to internal honesty standards, leading to a lower tolerance for dishonesty (i.e., decreased self-concept maintenance threshold). Finally, it is worth contrasting these results with peoples lay theories about such situations. A separate set of students (n = 75) correctly anticipated that participants would cheat when given the opportunity to do so, but they anticipated that the level of cheating would be higher than what it really was (Mpred_Books/recycle = 9.5), and they anticipated that reminding participants of the Ten Commandments would not significantly decrease cheating (Mpred_Ten Commandments/recycle = 7.8; t(73) = 1.61, p = .11). The contrast of the predicted results with the actual behavior we found suggests that participants understand the economic motivation for overclaiming, but they overestimate its influence on behavior and underestimate the effect of the self-concept in regulating honesty. EXPERIMENT 2: INCREASING ATTENTION TO STANDARDS FOR HONESTY THROUGH COMMITMENT REMINDERS Another type of reminder, an honor code, refers to a procedure that asks participants to sign a statement in which they declare their commitment to honesty before taking part in a task (Dickerson et al. 1992; McCabe and Trevino 1993, 1997). Although many explanations have been proposed for the effectiveness of honor codes used by many academic institutions (McCabe, Trevino, and Butterfield 2002; see http://www.academicintegrity.org), the self-concept maintenance idea may shed light on the internal process underlying its success. In addition to manipulating the awareness of honesty standards through commitment reminders at the point of temptation, Experiment 2 represents an extension of Experiment 1 by manipulating the financial incentives for performance (i.e., external benefits); in doing so, it also tests the external costbenefit hypothesis that dishonesty increases as the expected magnitude of reward from the dishonest act increases (Ext-H1). Method Two hundred seven students participated in Experiment 2. Using the same matrix task, we manipulated two factors between participants: the amount earned per correctly solved matrix ($.50 and $2, paid to each participant) and the attention to standards (control, recycle, and recycle + honor code).
answer, when it is found, the respondents could unambiguously evaluate whether they had solved the question correctly (assuming that they could add two numbers to 10 without error), without the need for a solution sheet and the possibility of a hindsight bias (Fischhoff and Beyth 1975). Moreover, we used this task on the basis of a pretest that showed that participants did not view this task as one that reflected their math ability or intelligence. The answer sheet was used to report the total number of correctly solved matrices. We promised that at the end of the session, two randomly selected participants would earn $10 for each correctly solved matrix. In the two control conditions (after the ten books and Ten Commandments recall task, respectively), at the end of the four-minute matrix task, participants continued to the next task in the booklet. At the end of the entire experimental session, the experimenter verified their answers on the matrix task and wrote down the number of correctly solved matrices on the answer sheet in the booklet. In the two recycle conditions (after the ten books and Ten Commandments recall task, respectively), at the end of the fourminute matrix task, participants indicated the total number of correctly solved matrices on the answer sheet and then tore out the original test sheet from the booklet and placed it in their belongings (to recycle later), thus providing them with an opportunity to cheat. The entire experiment represented a 2 (type of reminder) 2 (ability to cheat) betweensubjects design. Results and Discussion The results of Experiment 1 confirmed our predictions. The type of reminder had no effect on participants performance in the two control conditions (MBooks/control = 3.1 versus MTen Commandments/control = 3.1; F(1, 225) = .012, p = .91), which suggests that the type of reminder did not influence ability or motivation. Following the book recall task, however, respondents cheated when they were given the opportunity to do so (MBooks/recycle = 4.2), but they did not cheat after the Ten Commandments recall task (MTen Commandments/recycle = 2.8; F(1, 225) = 5.24, p = .023), creating a significant interaction between type of reminder and ability to cheat (F(3, 225) = 4.52, p = .036). Notably, the level of cheating remained far below the maximum. On average, participants cheated only 6.7% of the possible magnitude. Most important, and in line with our notion of self-concept maintenance, reminding participants of standards for morality eliminated cheating completely: In the Ten Commandments/recycle condition, participants performance was undistinguishable from those in the control conditions (F(1, 225) = .49, p = .48). We designed Experiment 1 to focus on the attention-tostandards mechanism (Ext&Int-H1), but one aspect of the resultsthe finding that the magnitude of dishonesty was limited and well below the maximum possible level in the two recycle conditionssuggested that the categorization mechanism (Ext&Int-H2) could have been at work as well. A possible alternative interpretation of the books/recycle condition is that over their lifetime, participants developed standards for moral behavior according to which overclaiming by a few questions on a test or in an experimental setting was not considered dishonest. If so, these participants could have been completely honest from their point of view. Similarly, in a country in which a substantial part of the cit-
The Dishonesty of Honest People In the two control conditions, at the end of five minutes, participants handed both the test and the answer sheets to the experimenter, who verified their answers and wrote down the number of correctly solved matrices on the answer sheet. In the two recycle conditions, participants indicated the total number of correctly solved matrices on the answer sheet, folded the original test sheet, and placed it in their belongings (to recycle later), thus providing them an opportunity to cheat. Only after that did they hand the answer sheet to the experimenter. The recycle + honor code condition was similar to the recycle condition except that at the top of the test sheet, there was an additional statement that read, I understand that this short survey falls under MITs [Yales] honor system. Participants printed and signed their names below the statement. Thus, the honor code statement appeared on the same sheet as the matrices, and this sheet was recycled before participants submitted their answer sheets. In addition, to provide a test for ExtH1, we manipulated the payment per correctly solved matrix ($.50 and $2) and contrasted performance levels between these two incentive levels. Results and Discussion Figure 2 depicts the results. An overall analysis of variance (ANOVA) revealed a highly significant effect of the attention-to-standards manipulation (F(2, 201) = 11.94, p < .001), no significant effect of the level of incentive manipulation (F(1, 201) = .99, p = .32), and no significant interaction (F(2, 201) = .58, p = .56). When given the opportunity, respondents in the two recycle conditions ($.50 and $2) cheated (Mrecycle = 5.5) relative to those in the two control conditions ($.50 and $2: Mcontrol = 3.3; F(1, 201) = 15.99, p < .001), but again, the level of cheating fell far below the maximum (i.e., 20); participants cheated only 13.5% of the possible average magnitude. In line with our findings in Experiment 1, this latter result supports the idea that we were also observing the workings of the categorization mechanism. Between the two levels of incentives ($.50 and $2 conditions), we did not find a particularly large difference in the Figure 2
EXPERIMENT 2: NUMBER OF MATRICES REPORTED SOLVED
637 magnitude of cheating; cheating was slightly more common (by approximately 1.16 questions), though not significantly so, in the $.50 condition (F(1, 201) = 2.1, p = .15). Thus, we did not find support for Ext-H1. A possible interpretation of this decrease in dishonesty with increased incentives is that the magnitude of dishonesty and its effect on the categorization mechanism depended on both the number of questions answered dishonestly (which increased by 2.8 in the $.50 condition and 1.7 in the $2 condition) and the amount of money inaccurately claimed (which increased by $1.4 in the $.50 condition and $3.5 in the $2 condition). If categorization malleability was affected by a mix of these two factors, we would have expected the number of questions that participants reported as correctly solved to decrease with greater incentives (at least as long as the external incentives were not too high). Most important for Experiment 2, we found that the two recycle + honor code conditions ($.50 and $2: Mrecycle + honor code = 3.0) eliminated cheating insofar as the performance in these conditions was undistinguishable from the two control conditions ($.50 and $2: Mcontrol = 3.3; F(1, 201) = .19, p = .66) but significantly different from the two recycle conditions ($.50 and $2: Mrecycle = 5.5; F(1, 201) = 19.69, p < .001). The latter result is notable given that the two recycle + honor code conditions were procedurally similar to the two recycle conditions. Moreover, the two institutions in which we conducted this experiment did not have an honor code system at the time, and therefore, objectively, the honor code had no implications of external punishment. When we replicated the experiment in an institution that had a strict honor code, the results were identical, suggesting that it is not the honor code per se and its implied external punishment but rather the reminder of morality that was at play. Again, we asked a separate set of students (n = 82) at the institutions without an honor code system to predict the results, and though they predicted that the increased payment would marginally increase dishonesty (Mpred_$2 = 6.8 versus Mpred_$.50 = 6.4; F(1, 80) = 3.3, p = .07), in essence predicting Ext-H1, they did not anticipate that the honor code would significantly decrease dishonesty (Mpred_recylce + honor code = 6.2 versus Mpred_recycle = 6.9; F(1, 80) = .74, p = .39). The contrast of the predicted results with the actual behavior suggests that people understand the economic motivation for overclaiming, that they overestimate its influence on behavior, and that they underestimate the effect of the self-concept in regulating honesty. In addition, the finding that predictors did not expect the honor code to decrease dishonesty suggests that they did not perceive the honor code manipulation as having implications of external punishment. EXPERIMENT 3: INCREASING CATEGORIZATION MALLEABILITY Making people mindful by increasing their attention to their honesty standards can curb dishonesty, but the theory of self-concept maintenance also implies that increasing the malleability to interpret ones actions should increase the magnitude of dishonesty (Schweitzer and Hsee 2002). To test this hypothesis, in Experiment 3, we manipulate whether the opportunity for dishonest behavior occurs in terms of money or in terms of an intermediary medium (tokens). We posit that introducing a medium (Hsee et al.
10 8 6 4 2 0
Notes: Mean number of solved matrices in the control condition (no ability to cheat) and the recycle and recycle + honor code (HC) conditions (ability to cheat). The payment scheme was either $.50 or $2 per correct answer. Error bars are based on standard errors of the means.
638
JOURNAL OF MARKETING RESEARCH, DECEMBER 2008 EXPERIMENT 4: RECOGNIZING ACTIONS BUT NOT UPDATING THE SELF-CONCEPT Our account of self-concept maintenance suggests that by engaging only in a relatively low level of cheating, participants stayed within the threshold of acceptable magnitudes of dishonesty and thus benefited from being dishonest without receiving a negative self-signal (i.e., their selfconcept remained unaffected). To achieve this balance, we posit that participants recorded their actions correctly (i.e., they knew that they were overclaiming), but the categorization and/or attention-to-standards mechanisms prevented this factual knowledge from being morally evaluated. Thus, people did not necessarily confront the true meaning or implications of their actions (e.g., I am dishonest). We test this prediction (Ext&Int-H3) in Experiment 4. To test the hypothesis that people are aware of their actions but do not update their self-concepts, we manipulated participants ability to cheat on the matrix task and measured their predictions about their performance on a second matrix task that did not allow cheating. If participants in a recycling condition did not recognize that they overclaimed, they would base their predictions on their exaggerated (i.e., dishonest) performance in the first matrix task. Therefore, their predictions would be higher than the predictions of those who could not cheat on the first task. However, if participants who overclaimed were cognizant of their exaggerated claims, their predictions for a situation that does not allow cheating would be attenuated and, theoretically, would not differ from their counterparts in the control condition. In addition, to test whether dishonest behavior influenced peoples self-concept, we asked participants about their honesty after they completed the first matrix task. If participants in the recycling condition (who were cheating) had lower opinions about themselves in terms of honesty than those in the control condition (who were not cheating), this would mean that they had updated their self-concept. However, if cheating did not influence their opinions about themselves, this would suggest that they had not fully accounted for their dishonest behaviors and, consequently, that they had not paid a price for their dishonesty in terms of their self-concept. Method Forty-four students participated in this experiment, which consisted of a four-task paradigm, administered in the following order: a matrix task, a personality test, a prediction task, and a second matrix task. In the first matrix task, we repeated the same control and recycle conditions from Experiment 2. Participants randomly assigned to either of these two conditions had five minutes to complete the task and received $.50 per correctly solved matrix. The only difference from Experiment 2 was that we asked all participants (not just those in the recycle condition) to report on the answer sheet the total number of matrices they had correctly solved. (Participants in the control condition then submitted both the test and the answer sheets to the experimenter, who verified each of their answers on the test sheets to determine payments.) In the second, ostensibly separate task, we handed out a ten-item test with questions ranging from political ambitions to preferences for classical music to general abilities. Embedded in this survey were two questions about partici-
2003) will offer participants more room for interpretation of their actions, making the moral implications of dishonesty less accessible and thus making it easier for participants to cheat at higher magnitudes. Method Four hundred fifty students participated in Experiment 3. Participants had five minutes to complete the matrix task and were promised $.50 for each correctly solved matrix. We used three between-subjects conditions: the same control and recycle conditions as in Experiment 2 and a recycle + token condition. The latter condition was similar to the recycle condition, except participants knew that each correctly solved matrix would earn them one token, which they would exchange for $.50 a few seconds later. When the five minutes elapsed, participants in the recycle + token condition recycled their test sheet and submitted only their answer sheet to an experimenter, who gave them the corresponding amount of tokens. Participants then went to a second experimenter, who exchanged the tokens for money (this experimenter also paid the participants in the other conditions). We counterbalanced the roles of the two experimenters. Results and Discussion Similar to our previous findings, participants in the recycle condition solved significantly more questions than participants in the control condition (Mrecycle = 6.2 versus Mcontrol = 3.5; F(1, 447) = 34.26, p < .001), which suggests that they cheated. In addition, participants magnitude of cheating was well below the maximumonly 16.5% of the possible average magnitude. Most important, and in line with Ext&Int-H2, introducing tokens as the medium of immediate exchange further increased the magnitude of dishonesty (Mrecyle + token = 9.4) such that it was significantly larger than it was in the recycle condition (F(1, 447) = 47.62, p < .001)presumably without any changes in the probability of being caught or the severity of the punishment. Our findings support the idea that claiming more tokens instead of claiming more money offered more categorization malleability such that people could interpret their dishonesty in a more self-serving manner, thus reducing the negative self-signal they otherwise would have received. In terms of our current account, the recycle + token condition increased the threshold for the acceptable magnitude of dishonesty. The finding that a medium could be such an impressive facilitator of dishonesty may explain the incomparably excessive contribution of employee theft and fraud (e.g., stealing office supplies and merchandise, putting inappropriate expenses on expense accounts) to dishonesty in the marketplace, as we reported previously. Finally, it is worth pointing out that our results differ from what a separate set of students (n = 59) predicted we would find. The predictors correctly anticipated that participants would cheat when given the opportunity to do so (Mpred_recycle = 6.6; t(29) = 5.189, p < .001), but they anticipated that being able to cheat in terms of tokens would not be any different than being able to cheat in terms of money (Mpred_recycle + token = 7; t(57) = 4.5, p = .65). Again, this suggests that people underestimate the effect of the selfconcept in regulating honesty.
The Dishonesty of Honest People pants self-concept as it relates to honesty. The first question asked how honest the participants considered themselves (absolute honesty) on a scale from 0 (not at all) to 100 (very). The second question asked participants to rate their perception of themselves in terms of being a moral person (relative morality) on a scale from 5 (much worse) to 5 (much better) at the time of the survey in contrast to the day before. In the third task, we surprised participants by announcing that they would next participate in a second five-minute matrix task, but before taking part in it, their task was to predict how many matrices they would be able to solve and to indicate how confident they were with their predictions on a scale from 0 (not at all) to 100 (very). Before making these predictions, we made it clear that this second matrix task left no room to overclaim because the experimenter would check the answers given on the test sheet (as was done in the control condition). Furthermore, we informed participants that this second test would consist of a different set of matrices, and the payment would depend on both the accuracy of their prediction and their performance. If their prediction was 100% accurate, they would earn $.50 per correctly solved matrix, but for each matrix they solved more or less than what they predicted, their payment per matrix would be reduced by $.02. We emphasized that this payment scheme meant that it was in their best interest to predict as accurately as possible and to solve as many matrices as they could (i.e., they would make less money if they gave up solving some matrices, just to be accurate in their predictions). Finally, the fourth task was the matrix task with different number sets and without the ability to overclaim (i.e., only control condition). Thus, the entire experiment represented a two-condition between-subjects design, differing only in the first matrix task (possibility to cheat). The three remaining tasks (personality test, prediction task, and second matrix task) were the same. Results and Discussion The mean number of matrices solved in the first and second matrix tasks appears in Table 1. Similar to our previous experiments, on the first task, participants who had the ability to cheat (recycle condition) solved significantly more questions than those in the control condition (t(42) = 2.21, p = .033). However, this difference disappeared in the
639 second matrix task, for which neither of the two groups had an opportunity to cheat (t(42) = .43, p = .67), and the average performance on the second task (M2ndMatrixTask = 4.5) did not differ from the control conditions performance on the first task (M1stMatrixTask/control = 4.2; t(43) = .65, p = .519). These findings imply that, as in the previous experiments, participants cheated when they had the chance to do so. Furthermore, the level of cheating was relatively low (on average, two to three matrices); participants cheated only 14.8% of the possible average magnitude. In terms of the predictions of performance on the second matrix task, we found no significant difference (t(42) ~ 0, n.s.) between participants who were able to cheat and those who were not in the first matrix task (Mcontrol = 6.3, and Mrecycle = 6.3). Moreover, participants in the control and recycle conditions were equally confident about their predictions (Mforecast_control = 72.5 versus Mforecast_recycle = 68.8; t(42) = .56, p = .57). Together with the difference in performance in the first matrix task, these findings suggest that those who cheated in the first task knew that they had overclaimed. As for the ten-personality-questions survey, after the first task, participants in both conditions had equally high opinions of their honesty in general (t(42) = .97, p = .34) and their morality compared with the previous day (t(42) = .55, p = .58), which suggests that cheating in the experiment did not affect their reported self-concepts in terms of these characteristics. Together, these results support our selfconcept maintenance theory and indicate that peoples limited magnitude of dishonesty flies under the radar; that is, they do not update their self-concept in terms of honesty even though they recognize their actions (i.e., that they overclaim). In addition, we asked a different group of 39 students to predict the responses to the self-concept questions (absolute honesty and relative morality). In the control condition, we asked them to imagine how an average student who solved four matrices would answer these two questions. In the recycle condition, we asked them to imagine how an average student who solved four matrices but claimed to have solved six would answer these two questions. As Table 1 shows, they predicted that cheating would decrease both a persons general view of him- or herself as an honest person (t(37) = 3.77, p < .001) and his or her morality compared with the day before the test (t(37) = 3.88, p < .001).2 This finding provides further support for the idea that people do not accurately anticipate the self-concept maintenance mechanism. EXPERIMENT 5: NOT CHEATING BECAUSE OF OTHERS Thus far, we have accumulated evidence for a magnitude of cheating, which seems to depend on the attention a person pays to his or her own standards for honesty as well as categorization malleability. Moreover, the results of Experiment 4 provide some evidence that cheating can take place without an associated change in self-concept. Overall, these
2We replicated these findings in two other prediction tasks (within and between subjects). Students anticipated a significant deterioration in their own self-concept if they (not another hypothetical student) were to overclaim by two matrices.
Table 1
EXPERIMENT 4: PERFORMANCE ON THE MATRIX AND PERSONALITY TESTS Matrix Task First Matrix Task Condition Control Recycle Matrices Solved (0 to 20) First Task 4.2 6.7 Personality Test Absolute Honesty (0 to 100) Actual 85.2 79.3 67.6 32.4 Relative Morality (5 to +5) Predicted .4 1.4 Actual .4 .6
Second Task Predicted 4.6 4.3
Notes: Number of matrices reported as correctly solved in the first and second matrix task, as well as predicted and actual self-reported measures of absolute honesty and relative morality in the personality test after the control and recycle conditions, respectively, of the first matrix task.
640
JOURNAL OF MARKETING RESEARCH, DECEMBER 2008 allotted (four matrices, which is the accurate number, or eight matrices, which is an exaggeration). Again, the dependent variable was the number of matrices reported as being solved correctly. The experiment represented a 2 2 between-subjects design. Results and Discussion On average, participants in the two control conditions solved 3.3 and 3.4 matrices, and those in the corresponding recycle conditions solved 4.5 and 4.8 matrices (in the 4 and 8 believed standard performance conditions, respectively). A two-factorial ANOVA of the number of matrices solved as a function of the ability to cheat and the belief about others performances showed a main effect of the ability to cheat (F(1, 104) = 6.89, p = .01), but there was no main effect of the beliefs about average performance levels (F(1, 104) = .15, p = .7) and no interaction (F(1, 104) = .09, p = .76). That is, when participants had a chance to cheat, they cheated, but the level of cheating was independent of information about the average reported performance of others. This finding argues against drive toward achievement, threshold due to external costs, or norm compliance as alternative explanations for our findings. EXPERIMENT 6: SENSITIVITY TO EXTERNAL REWARDS Because the external costs of dishonest acts are central to the standard economic costbenefit view of dishonesty, we wanted to test its influence more directly. In particular, following Nagin and Pogarskys (2003) suggestion that increasing the probability of getting caught is much more effective than increasing the severity of the punishment, we aimed to manipulate the former type of external costthat is, the likelihood of getting caught on three levelsand to measure the amount of dishonesty across these three cheating conditions. If only external costbenefit trade-offs are at work in our setup, we should find that the level of dishonesty increases as the probability of being caught decreases (Ext-H2). Conversely, if self-concept maintenance limits the magnitude of dishonesty, we should find some cheating, but the level of dishonesty should be roughly of the same magnitude, regardless of the probabilities of getting caught. Method This experiment entailed multiple sessions with each participant sitting in a private booth (N = 326). At the start of each session, the experimenter explained the instructions for the entire experiment. The first part of the experimental procedure remained the same for all conditions, but the second part varied across conditions. All participants received a test with 50 multiple-choice, general-knowledge questions (e.g., How deep is a fathom? How many degrees does every triangle contain? What does 3! equal?), had 15 minutes to answer the questions, and were promised $.10 for each question they solved correctly. After the 15 minutes, participants received a bubble sheet onto which they transferred their answers. Similar to Scantron sheets used with multiple-choice tests, for each question, the bubble sheet provided the question number with three circles labeled a, b, and c, and participants were asked to mark the corresponding circle. The manipulation of our conditions pertained to the bubble sheet and to what participants did with it after transferring their answers.
findings are in line with our theory of self-concept maintenance: When people are torn between the temptation to benefit from cheating and the benefits of maintaining a positive view of themselves, they solve the dilemma by finding a balance between these two motivating forces such that they can engage to some level in dishonest behavior without updating their self-concept. Although these findings are consistent with our theory of self-concept maintenance, there are a few other alternative accounts for these results. In the final two experiments, we try to address these. One possible alternative account that comes to mind posits that participants were driven by self-esteem only (e.g., John and Robins 1994; Tesser, Millar, and Moore 1988; Trivers 2000). From this perspective, a person might have cheated on a few matrices so that he or she did not appear stupid compared with everybody else. (We used the matrix task partially because it is not a task that our participants related to IQ, but this account might still be possible.) A second alternative for our findings argues that participants were driven only by external, not internal, rewards and cheated up to the level at which they believed their dishonest behavior could not be detected. From this perspective, participants cheated just by a few questions, not because some internal force stopped them but because they estimated that the probability of being caught and/or the severity of punishment would be negligible (or zero) if they cheat by only a few questions. As a consequence, they cheated up to this particular thresholdin essence, estimating what they could get away with and cheating up to that level. A third alternative explanation is that the different manipulations (e.g., moral reminders) influenced the type of social norms that participants apply to the experimental setting (see Reno, Cialdini, and Kallgren 1993; for focusing effects, see Kallgren, Cialdini, and Reno 2000). According to this norm compliance argument, a person who solves three matrices but knows that, on average, people report having solved six should simply go ahead and do what others are doing, namely, report six solved matrices (i.e., cheat by three matrices). What these three accounts have in common is that all of them are sensitive to the (expected) behavior of others. In contrast, our self-concept maintenance theory implies that the level of dishonesty is set without reference to the level of dishonesty exhibited by others (at least in the short run). This contrast suggests a simple test in which we manipulate participants beliefs about others performance levels. If the level of cheating is driven by the desire for achievement, external costs, or norm compliance, the number of matrices that participants claim to have solved should increase when they believe that the average performance of others is higher. However, if the level of cheating is driven by selfconcept maintenance considerations, the belief that others solve many more matrices should have no effect on the level of dishonesty. Method One hundred eight students participated in a matrix task experiment, in which we manipulated two factors between participants: the ability to cheat (control and recycle, as in Experiments 2) and beliefs about the number of matrices the average student solves in the given condition in the time
The Dishonesty of Honest People In the control condition, participants received a standard bubble sheet. When they finished transferring their answers, they handed both the test and the bubble sheet to the experimenter, who checked their answers, summed up the number of correct answers, and paid the participants $.10 for each correct answer. In the no-recycle condition (first cheating condition), the bubble sheet had the correct answers premarked, such that the circles representing the correct answers were shaded in gray. This design prompted a dilemma for participants when they faced a question they had answered incorrectly on their test sheet; they could be honest and fill in the corresponding incorrect bubble or be dishonest and fill in the correct bubble. After participants finished transferring their answers, they summed up the number of their correct answers, wrote that number at the top of the bubble sheet, and handed both the test and the bubble sheet to the experimenter, who paid them according to their self-summed score. In this condition, participants could cheat with some risk that the experimenter might discover it, if the experimenter compared the answers on the bubble sheet with the answers on the test sheet. The recycle condition (second cheating condition) was similar to the norecycle condition, with the difference that participants were instructed to transfer their answers to the premarked bubble sheet and then walk to a shredder, shred their original test sheet, and take only the bubble sheet to the experimenter, at which point they would be paid accordingly. Because of the shredding, this condition offered a lower probability of being caught cheating than the no-recycle condition. Finally, the recycle+ condition (third cheating condition) further decreased the probability of being caught by instructing participants to shred both their test sheet and the bubble sheet, walk over to a large jar with money at the corner of the room, and take the amount they earned. In addition, by making the payment self-service, the recycle+ condition eliminated any interactions with the experimenter, thus decreasing social concerns with cheating.3 At the start of each experimental session of the recycle+ condition, the jar was filled with different denominations that totaled $100. After each session (out of the sight of students), we collected the jar and measured the amount of money in it.4 Results and Discussion On average, participants in the control condition solved 32.6 questions, and those in the no-recycle, recycle, and recycle+ conditions solved 36.2, 35.9, and 36.1 questions, respectively. An overall ANOVA of the number of questions reported as solved revealed a highly significant effect of the conditions (F(3, 322) = 19.99, p < .001). The average reported performance in the three cheating conditions was significantly higher than in the control condition (F(1, 322) = 56.19, p < .001), but there was no difference in
3In a separate study, we asked participants to estimate the probability of being caught across the different conditions and found that these conditions were indeed perceived in the appropriate order of the likelihood of being caught (i.e., no recycle > recycle > recycle+). 4The goal of the recycle+ condition was to guarantee participants that their individual actions of taking money from the jar would not be observable. Therefore, it was impossible to measure how much money each respondent took in this condition. We could record only the sum of money missing at the end of each session. For the purpose of statistical analysis, we assigned the average amount taken per recycle+ session to each participant in that session.
641 dishonesty across the three cheating conditions (F(2, 209) = .11, p = .9), and the average magnitude of dishonesty was approximately 20% of the possible average magnitude, which was far from the maximal possible dishonesty in these conditions (similar to findings by Goldstone and Chin 1993). These latter results suggest that participants in all three cheating conditions seemed to have used the same threshold to reconcile the motivations to benefit financially from cheating and maintain their positive self-concept. Experiment 6 is also useful in testing another possible alternative explanation, which is that the increased level of cheating we observed in the three cheating conditions was due to a few bad apples (a few people who cheated a lot) rather than to a general shift in the number of answers reported as correctly solved (many people cheating just by a little bit). As Figure 3 shows, however, the dishonesty seemed to be due to a general increase in the number of correct responses, which resulted in a rightward shift of the response distribution.5 To test this stochastic dominance assumption, we subjected the distributions to a series of quantile regressions and found that the cheating distributions dominated the control distribution at every possible point (e.g., at the 10th, 20th, 30th, 40th, 50th, 60th, 70th, 80th, and 90th percentiles, the number of questions solved was significantly higher in the cheating conditions than in the control condition: t(210) = 3.65, 3.88, 4.48, 4.10, 2.92, 3.08, 2.11, 2.65, and 3.63, ps < .05), but the distributions across the cheating conditions did not differ from one another (no ps < .35). Although Experiment 6 was particularly useful for this analysis (because it included multiple cheating conditions),
5This analysis did not include the recycle+ condition, because we were not able to measure individual-level performance; instead, we were limited to measuring performance per session.
Figure 3
EXPERIMENT 6: NUMBER OF MULTIPLE-CHOICE QUESTIONS REPORTED SOLVED
Notes: Frequency distribution of number of solved questions in the control condition (no ability to cheat) and two cheating conditions: norecycle and recycle. The values on the y-axis represent the percentage of participants having solved a particular number of questions; the values on the x-axis represent 1 ranges around the displayed number (e.g., 21 = participants having solved 20, 21, or 22 questions).
642
JOURNAL OF MARKETING RESEARCH, DECEMBER 2008 trade-off and are unappreciative of the regulative effectiveness of the self-concept.6 In principle, the theory we propose can be incorporated into economic models. Some formalizations related to it appear in recent economic theories of utility maximization based on models of self-signaling (Bodner and Prelec 2001) and identity (Bnabou and Tirole 2004, 2006). These models can be adopted to account for self-concept maintenance by incorporating attention to personal standards for honesty (meta-utility function and salience parameter s1, respectively) and categorization malleability (interpretation function and probability 1 , respectively). These approaches convey a slowly spreading conviction among economists that to study moral and social norms, altruism, reciprocity, or antisocial behavior, the underlying psychological motivations that vary endogenously with the environment must be understood (see also Gneezy 2005). The data presented herein offer further guidance on the development of such models. In our minds, the interplay between these formal models and the empirical evidence we provide represents a fruitful and promising research direction. Some insights regarding the functional from in which the external and internal rewards work together emerge from the data, and these findings could also provide worthwhile paths for further investigations in both economics and psychology. For example, the results of Experiment 2 show that increasing external rewards in the form of increasing benefits (monetary incentive) decreased the level of dishonesty (though insignificantly). This observation matches findings from another matrix experiment in which we manipulated two factors between 234 participants: the ability to cheat (control and recycle) and the amount of payment to each participant per correctly solved matrix ($.10, $.50, $2.50, and $5). In this 2 4 design, we found limited dishonesty in the $.10 and $.50 conditions but no dishonesty in the $2.50 and $5 conditions. Furthermore, the magnitude of dishonesty was approximately the same for $.10 and $.50. Together, these observations raise the possibility of a step functionlike relationshipconstant, limited amount of dishonesty up to a certain level of positive external rewards, beyond which increasing the external rewards could limit categorization malleability, leaving no room for under-the-radar dishonesty. In this way, dishonesty may actually decrease with external rewards. Finally, it is worthwhile noting some of the limitations of our results. The first limitation is related directly to the relationship between external and internal rewards. Arguably, at some point at which the external rewards become very high, they should tempt the person sufficiently to prevail (because the external reward of being dishonest is much larger than the internal reward of maintaining a positive self-concept). From that point on, we predict that behavior would be largely influenced by external rewards, as the standard economic perspective predicts (i.e., ultimately, the magnitude of dishonesty will increase with increasing, high external rewards).
6Note that our manipulations in their general form may be viewed as priming. In this sense, our results may generalize to a much larger class of manipulations that would curtail cheating behavior and may be useful when, for example, the Ten Commandments or honor codes are not a feasible solution, such as purchasing environments.
a stronger test would be to determine whether this conclusion also holds across all six experiments. To do so, we converted the performance across all the experiments to be proportional, that is, the number of questions reported solved relative to the maximum possible. Analyzing all conditions across our experiments (n = 1408), we again find strict stochastic dominance of the performance distributions in conditions that allowed cheating over conditions that did not ( = .15, t(1406) = 2.98, p = .003). We obtain similarly reliable differences for each quantile of the distributions, suggesting that the overall mean difference ( = .134, t(1406) = 9.72, p < .001) was indeed caused by a general shift in the distribution rather than a large shift of a small portion of the distribution. GENERAL DISCUSSION People in almost every society value honesty and maintain high beliefs about their own morality; yet examples of significant dishonesty can be found everywhere in the marketplace. The standard costbenefit model, which is central to legal theory surrounding crime and punishment, assumes that dishonest actions are performed by purely selfish, calculating people, who only care about external rewards. In contrast, the psychological perspective assumes that people largely care about internal rewards because they want, for example, to maintain their self-concept. On the basis of these two extreme starting points, we proposed and tested a theory of self-concept maintenance that considers the motivation from both external and internal rewards. According to this theory, people who think highly of themselves in terms of honesty make use of various mechanisms that allow them to engage in a limited amount of dishonesty while retaining positive views of themselves. In other words, there is a band of acceptable dishonesty that is limited by internal reward considerations. In particular, we focus on two related but psychologically distinct mechanisms that influence the size of this bandcategorization and attention to standardswhich we argue have a wide set of important applications in the marketplace. Across a set of six experiments we found support for our theory by demonstrating that when people had the ability to cheat, they cheated, but the magnitude of dishonesty per person was relatively low (relative to the possible maximum amount). We also found that, in general, people were insensitive to the expected external costs and benefits associated with the dishonest acts, but they were sensitive to contextual manipulations related to the self-concept. In particular, the level of dishonesty dropped when people paid more attention to honesty standards and climbed with increased categorization malleability (Dana, Weber, and Kuang 2005). Some of the results provide more direct evidence for the self-concept maintenance mechanism (Experiment 4) by showing that even though participants knew that they were overclaiming, their actions did not affect their self-concept in terms of honesty. Note also that, in contrast, predictors expected dishonest actions to have a negative effect on the self-concept. This misunderstanding of the workings of the self-concept also manifested in respondents inability to predict the effects of moral reminders (Ten Commandments and honor code) and mediums (tokens), suggesting that, in general, people expect others to behave in line with the standard economic perspective of an external costbenefit
The Dishonesty of Honest People Another limitation is that our results did not support a sensitivity to others reported behaviors, implying that, for example, self-esteem or norm compliance considerations do not influence peoples decisions about being dishonest. We do not imply that such effects are not prevalent or perhaps even powerful in the marketplace. For example, it could be that the sensitivity to others operates slowly toward changing a persons global internal standards for honesty, rather than having a large influence on the local instances of dishonesty, such as those that took place in our experiments. From a practical perspective, one of the two main questions about under-the-radar dishonesty pertains to its magnitude in the economy. By its very nature, the level of dishonesty in the marketplace is difficult to measure, but if our studies are any indication, it may far exceed the magnitude of dishonesty committed by standard, run-of-the-mill criminals, who consider only the external rewards in their decision. Across the six experiments (excluding the recycle + token condition), among the 791 participants who could cheat, we encountered only 5 (.6%) who cheated by the maximal amount (and, thus, presumably engaged in external costbenefit trade-off analysis, leading to standard rational dishonesty), whereas most cheated only slightly (and, thus, presumably engaged in a trade-off of external and internal rewards, leading them to engage in limited dishonesty that flies under the self-concept radar). Furthermore, the total costs incurred as a result of limited dishonesty were much greater than those associated with the maximal dishonesty. Taken at face value, these results suggest that the effort that society at large applies to deterring dishonestyespecially standard rational dishonesty might be misplaced. Another important applied speculation involves the medium experiment. As society moves away from cash and electronic exchanges become more prevalent, mediums are rapidly more available in the economy. Again, if we take our results at face value, particular attention should be paid to dishonesty in these new mediums (e.g., backdating stocks) because they provide more opportunities for underthe-radar dishonesty. In addition, we observed that the medium experiment not only allowed people to cheat more but also increased the level of maximal cheating. In the medium experiment, we observed 24 participants who cheated maximally, which indicated that the tokens not only allowed them to elevate their acceptable magnitude of dishonesty but also liberated them from the shackles of their morality altogether. When we consider the applied implications of these results, we must emphasize that our findings stem from experiments not with criminals but with students at elite universities, people who are likely to play important roles in the advancement of the United States and who are a lot more similar to the general public. The prevalence of dishonesty among these people and the finding that, on an individual level, they were mostly honest rather than completely dishonest suggest the generalizability of our results. As Goldstone and Chin (1993) conclude, people seem to be moral relativists in their everyday lives. From a practical perspective, the next question is thus related to approaches for curbing under-the-radar dishonesty. The results of the honor code, Ten Commandments, and token manipulations are promising because they sug-
643 gest that increasing peoples attention to their own standards for honesty and decreasing the categorization malleability could be effective remedies. However, the means by which to incorporate such manipulations into everyday scenarios in which people might be tempted to be dishonest (e.g., returning used clothes, filling out tax returns or insurance claims), to determine how abstract or concrete these manipulations must be to be effective (see Hayes and Dunning 1997), and to discover methods for fighting adaptation to these manipulations remain open questions. REFERENCES
Accenture (2003), One-Fourth of Americans Say Its Acceptable to Defraud Insurance Companies, (February 12), (accessed December 1, 2006), [available at http://www.accenture.com/xd/ xd.asp?it=enweb&xd=_dyn%5Cdynamicpressrelease_577. xml]. Allingham, Michael G. and Agnar Sandmo (1972), Income Tax Evasion: A Theoretical Analysis, Journal of Public Economics, 1 (November), 32338. Aronson, Elliot (1969), A Theory of Cognitive Dissonance: A Current Perspective, in Advances in Experimental Social Psychology, Vol. 4, Leonard Berkowitz, ed. New York: Academic Press, 134. and J. Merrill Carlsmith (1962), Performance Expectancy as a Determinant of Actual Performance, Journal of Abnormal and Social Psychology, 65 (3), 17882. Association of Certified Fraud Examiners (2006), 2006 ACFE Report to the Nation on Occupational Fraud & Abuse, (accessed August 1, 2008), [available at http://www.acfe.com/ documents/2006-rttn.pdf]. Bateson, Melissa, Daniel Nettle, and Gilbert Roberts (2006), Cues of Being Watched Enhance Cooperation in a Real-World Setting, Biology Letters, 2 (June), 41214. Baumeister, Roy F. (1998), The Self, in Handbook of Social Psychology, Daniel T. Gilbert, Susan T. Fiske, and Gardner Lindzey, eds. New York: McGraw-Hill, 680740. Becker, Gary S. (1968), Crime and Punishment: An Economic Approach, Journal of Political Economy, 76 (2), 169217. Bem, Daryl J. (1972), Self-Perception Theory, in Advances in Experimental Social Psychology, Vol. 6, Leonard Berkowitz, ed. New York: Academic Press, 162. Bnabou, Roland and Jean Tirole (2004), Willpower and Personal Rules, Journal of Political Economy, 112 (4), 84886. and (2006), Identity, Dignity and Taboos, working paper, Department of Economics and Woodrow Wilson School, Princeton University. Bering, Jesse M., Katrina McLeod, and Todd K. Shackelford (2005), Reasoning About Dead Agents Reveals Possible Adaptive Trends, Human Nature, 16 (4), 36081. Bodner, Ronit and Drazen Prelec (2001), Self-Signaling and Diagnostic Utility in Everyday Decision Making, working paper, Sloan School of Management, Massachusetts Institute of Technology. Campbell, Ernest Q. (1964), The Internalization of Moral Norms, Sociometry, 27 (4), 391412. Dana, Jason, Roberto A. Weber, and Jason Xi Kuang (2005), Exploiting Moral Wiggle Room: Experiments Demonstrating an Illusory Preference for Fairness, working paper, Department of Psychology, University of Illinois at UrbanaChampaign. De Quervain, Dominique J.-F., Urs Fischbacher, Valerie Treyer, Melanie Schelthammer, Ulrich Schnyder, Alfred Buck, and Ernst Fehr (2004), The Neural Basis of Altruistic Punishment, Science, 305 (August 27), 125458. Dickerson, Chris A., Ruth Thibodeau, Elliot Aronson, and Dayna Miller (1992), Using Cognitive Dissonance to Encourage
Justified Ethicality 2
Justified Ethicality:
Observing Desired Counterfactuals Modifies Ethical Perceptions and Behavior Forthcoming in Organizational Behavior and Human Decision Processes
Shaul Shalvi1, Jason Dana2, Michel J. J. Handgraaf1, and Carsten K.W. DeDreu Department of Psychology, University of Amsterdam, 2Department of Psychology, University of Pennsylvania
Abstract Employing a die-under-cup paradigm, we study the extent to which people lie when it is transparently clear they cannot be caught. We asked participants to report the outcome of a private die roll and gain money according to their reports. Results suggest that the degree of lying depends on the extent to which self-justifications are available. Specifically, when people are allowed to roll the die three times to ensure its legitimacy, but only the first roll is supposed to count, we find evidence that the highest outcome of the three rolls is reported. Eliminating the ability to observe more than one roll reduces lying. Additional results suggest that observing desired counterfactuals, in the form of additional rolls not meant to determine pay, attenuates the degree to which people perceive lies as unethical. People seem to derive value from self-justifications allowing them to lie for money while feeling honest.
Justified Ethicality 3 Justified Ethicality: Observing Desired Counterfactuals Modifies Ethical Perceptions and Behavior Morality, like art, means drawing a line somewhere. Oscar Wilde (1854-1900) Both within and outside organizations, daily life provides individuals ample opportunities to gain financially by bending the rules and lying. To some extent, the decision whether to lie is an economic calculation driven by whether the potential gain from lying is greater than the likelihood of getting caught times the magnitude of subsequent punishment (Becker, 1968; Alingham & Sandmo, 1972). To some extent, however, lying or not may depend on purely intrinsic moral standards and ethical considerations. For example, the more aversive the consequences of ones dishonesty on others outcomes, the less likely it is that one will lie (Gneezy, 2005). Furthermore, people avoid major lies even when the chances of getting caught are essentially zero (Fischbacher & Heusi, 2008; Shalvi, Handgraaf & De Dreu, in press. Mazar, Amir and Ariely (2008) explain the tendency to lie a little bit but not as much as one possibly could by proposing that people lie to some degree to increase their profit, but not so much as to threaten their positive self-concept as honest individuals. To paraphrase Oscar Wilde, it seems that when considering lying or not, people draw the line somewhere. But the exact location of the line remains elusive, and it is yet unknown what leads people to decide that for a certain amount they would lie, but for a larger amount they would not. We study lying by employing a die-under-cup paradigm (Shalvi et al., in press; based on Fischbacher and Heusi, 2008) that allows participants to report the outcome of a die roll only they can see and gain money according to their reports. Consistent with previous studies (Fischbacher & Heusi, 2008; Shalvi, et al. in press; Mazar, et al. 2008; Gino, Ayal & Ariely, 2009; Gino, Norton & Ariely, 2010; Lundquist, Ellingson & Johannesson, 2009) we find evidence of incomplete dishonesty. Reports differ significantly from the distribution of an honest die roll, but participants still lie far less than they could have. Novel to the current work, we find that the extent to which people allow themselves to lie depends critically on the availability of self-justifications that no one else knows about. Participants who were allowed to roll the die additional times to verify that it was legitimate lied to a greater extent and appear to report the largest number they saw on any roll. Participants reported the largest number they saw even though they knew that additional rolls were not supposed to count for determining pay. Observing a desired higher number on a roll that was not meant to count seems to justify lying using this one specific high number. Importantly, our paradigm insures that these justifications are completely private; no one else knows what numbers the participant has seen. Additional studies corroborate this pattern of behavioral results by showing that of all possible profit boosting possibilities, participants report that they are most likely to use one of the additional rolls as means to increase their pay. Addressing the psychological driver for the behavioral pattern we found, we obtained evidence that participants judge dishonest reports to be less dishonest when they are equal to one of the additional rolls. The idea that people need self-justifications for lying is grounded in the notion that people are likely to arrive to conclusions that they want to arrive at, but their ability to do so is constrained by their ability to construct seemingly reasonable justifications for these conclusions (Kunda, 1990, p. 480).
Justified Ethicality 4 Self Justifying Unethical Behavior Initial evidence that people derive value from having justifications to dishonestly benefit themselves comes from work by Batson and colleagues (Batson, Kobrynowicz, Dinnerstein, Kampf, & Wilson, 1997; Batson, Thompson, Seuferling, Whitney, & Strongman, 1999). Participants in their studies had to determine whether they or another participant would have to perform an undesirable task and were provided with a coin they could toss to assist them in making this decision, so that they could either make the decision themselves, or use a coin-toss to make the decision for them. Supporting their moral hypocrisy prediction, people who claimed to use the coin to make the decision won significantly more often (between 80% and 90%) than predicted by an honest toss. Participants were thus able to appear fair by flipping the coin, yet still serve self-interest by ignoring the coin (Batson & Thompson, 2001, p. 55). This line of research supports the idea that people find value in appearing fair and moral. They seek a fair procedure (coin toss) to justify their self benefiting outcome to others. A second line of evidence supporting the notion that people seek to appear fair while serving their self interest comes from work on the ultimatum bargaining game, in which a proposer offers a division of a commodity (e.g., chips to be converted to money), and a responder can accept or reject the proposed division. If the responder accepts, the commodity is divided as proposed; if the responder rejects, neither party receives anything (Gth, Schmittberger & Schwarze, 1982). Pillutla and Murnighan (1995; 2003; see also Kagel, Kim, and Moser, 1996) used a modified version of the game providing only proposers with the value of the chips for themselves as well as for responders. By manipulating the value of the chips for the responders to be lower vs. equal to the value of the proposers chips, Pillutla and Murninghan were able to disentangle between proposers desire to act in a fair way (i.e., propose offers that are fair in monetary terms) or merely appear fair (i.e., propose offers that seem fair in terms of chips offered but are actually self-serving in monetary terms). Results indeed indicated that proposers made offers that seemed fair while they actually were not. Further empirical evidence supporting the notion that justifications lead people to act on their self interest while appearing moral comes from early work by Snyder, Kleck, Strenta and Mentzer (1979). In this study, people had to choose in which of two rooms they would watch a movie and fill out a questionnaire one with a handicapped person or with a non- handicapped person. When the same movie clip was presented in both rooms, people were more likely to sit with the handicapped individual. But when different movies were presented, providing a justification for selecting one room over the other, a majority avoided the handicapped person. This finding further suggests that when people have an apparent justification for not sitting with the handicapped person by way of their preference for one of the two clips, their preferred clip matches the room which allows more morally questionable behavior (i.e., avoiding a handicapped person). These lines of work clearly indicate that people find value in that they can appear fair and moral in the eyes of others and that if questioned about their decision they will be able to justify it by the honest procedure they employed (i.e., a coin toss), the fair offer they place on the table, or their preferred means of entertainment (i.e., a specific video clip). However, in many daily situations we act in solitude without having to justify our (un)ethical behavior to anyone but ourselves. Mazar, Amir and Ariely (2008) indeed demonstrated that people value maintaining their honest self-concept; that is, they value feeling and not just appearing honest. Even though participants in their studies could (and did) lie anonymously, these participants refrained from lying to the fullest extent that they could. Mazar and collegues explained this tendency to refrain from lying to large extents by suggesting that people have a desire to hold a positive self-view which includes seeing themselves as ethical and honest people. Lying modestly thus allows one to serve two desires simultaneously to benefit financially as well as to maintain an honest self-concept.
Justified Ethicality 5 We propose that justifications - here, in the form of observed desired counterfactuals - have an economic value in this equation. Justifications are valuable as they allow one to lie for money while maintaining an honest self-concept. Most importantly, our approach does not require a restriction of the magnitude of the lie as means to allow maintaining an honest self-concept. We propose that people would be able to maintain feeling honest when lying to different degrees (even a lot) conditional on having a justification for their unethical acts. Consider, for example, a person using an online tax calculator to determine if she is entitled to a tax refund. If, after entering all her information, the initial results are disappointing, she may explore modified scenarios that would lead her closer to the desired refund. During this process of exploration, she may eventually discover that if one item had been different e.g., if her partner had moved in with her on the 30th of June instead of the actual 1st of July, making them a shared household for one day more than six months instead of one day less than 6 months, her refund would be better. Experiencing the feeling of a desired outcome (a better tax return) resulting from this observed counterfactual information (moving in sooner) makes lying on the return in this one specific way seem more acceptable. We argue that observing desired counterfactual information justifies lying, as it modifies the extent to which a dishonest act is perceived as unethical, increasing ones likelihood to lie. Importantly, we further maintain that this psychological mechanism occurs even when the justification is only for oneself. That is, that people find value in feeling honest even when they know that they lie. Desired Counterfactuals as Justifications The proposition that self-justifications are valuable in allowing people to feel honest while lying implies that honesty is not perceived as a sharp contrast between lying versus being honest, but rather as a continuum that stretches between these two ends. Deontologically, a lie is a lie and is wrong. We however propose that when a desired state is observed peoples perception of what is ethical and what is unethical is modified. Accordingly, a given dishonest act can be perceived as more or less ethical depending on the availability of a self-justification for doing it, and whether such justification is used. Initial support for the idea that ethical evaluations (and subsequent behavior) are not a right vs. wrong dichotomy but rather a continuum ranging between the two ends, comes from work on the elastic justifications (Hsee, 1995; 1996). For example, Schweitzer and Hsee (2002) asked sellers of a car to provide a buyer with a mileage estimate from a range of possible values. Sellers lied more when the provided range was wide rather than narrow, as they could justify the lie by their increased uncertainty about the true mileage. It seems that sellers processed the information about the cars mileage in a self-serving manner (Dunning, 2005; Dunning, Heath & Suls, 2004; Ehrlinger & Dunning, 2003), allowing them to gain financially. Conversely, lying may decrease when available information provides no means to justify an unethical act. The question remains, how did observing a desirable piece of information influence lying? Car sellers in Schweitzer and Hsees study (2002) who received a wide range estimate may have been influenced by the (desired) high end of the range, which potentially served as a cognitive anchor to which they adjusted their mileage estimates (Kahneman, 1992; Mussweiler, 2003; Ritov, 1996; Tversky & Kahneman, 1974). Alternatively, merely observing a desired outcome that could have been true may have liberated some peoples standards enough so that reporting the highest number seemed ethically acceptable. Similarly, in the tax example above, learning that reporting moving in with a partner one day before the move actually took place may make this one specific lie more legitimate than other lies which are based on information that was not observed. Past work indeed reveals that considering upward counterfactuals that are evaluatively better than actuality (Roese, 1997; p. 134), influences peoples (emotional) interpretation of factual reality (Mandel & Dhami, 2005; Zeelenberg et al. 1998) and their subsequent decisions
Justified Ethicality 6 (Ritov & Baron, 1995; Roese, 1994; Reichert & Slate, 2000; Morris & Moore, 2000; Markman, McMullen & Elizaga, 2008; for a review see, Epstude & Roese, 2008). But how does observing a desired piece of information justify lying? And why would lying using this specific piece of observed information feel more legitimate compared to lying using other pieces of information that were not observed? Kahneman and Millers Norm Theory (1986) postulates that norms are computed after the event has occurred rather than in advance. Specifically, they proposed that each stimulus selectively recruits its own alternatives (Garner, 1962; 1970) and is interpreted in a rich context of remembered and constructed representations of what it could have been, might have been, or should have been (Kahneman & Miller, 1986; p. 136). A key aspect in Norm Theory is the level of mutability of the stimulus (e.g. information, object, circumstances) which is the extent to which this stimulus is modifiable. Low mutability suggests that it is very difficult to imagine the situation had been different, while high mutability means that this is easy. When it is possible to think of alternative scenarios leading to an unfavorable outcome (i.e., high mutability) people tend to focus on upward counterfactual comparisons (Roese & Olson, 2007; Markman, Gavanski, Sherman & McMullen, 1993). The reason for this is that considering such desirable (upward) counterfactuals leads to reflective as if thinking: People reflect on the counterfactual as if it was real, thereby reducing the contrast between this piece of counterfactual information and factual reality (Markman & McMullen, 2003; Kahneman & Varey, 1990). Simply put, when reality is disappointing and desirable information becomes available, it is much easier to imagine the specific alternative scenario which would have happened if this piece of counterfactual information was true. Here we propose that people observing a desired piece of information may feel that it is more legitimate to use this piece of information since it nearly occurred, serving their self- interest as a result. When a desired piece of information is observed, it is this specific information, more than other desired alternatives that were not observed, that becomes readily available for consideration as an alternative fact. Put differently, when reality is disappointing and a desired piece of information is observed, people more easily construe reality by using the observed information that is available to them than any other information that was not observed. We thus suggest that when an observed counterfactual is desirable compared to factual reality, dishonestly using this counterfactual to boost profit may feel less unethical than using other counterfactuals that were not observed. Overview and Predictions We placed people in a situation in which they could lie to gain financially and manipulated whether they were able to observe desired counterfactuals. Specifically, we adapted a simple paradigm introduced by Fischbacher and Heusi (2008) in which people privately roll a die and are paid based on what outcome they report. Across their different experimental conditions including manipulations of the profit generated by the lie, the influence of the report on another persons outcomes, and the level of privacy in receiving pay for the task (i.e., by the experimenter vs. paying oneself from an envelope), a robust pattern of monotonically increasing reports emerged higher outcomes were reported more often compared to lower outcomes. The deviation from the expected distribution of a fair roll reflected that people lied to some degree for financial gain. Based on our reasoning that people might use observed desired counterfactuals to boost profit, we focused on one aspect of Fischbacher and Heusis paradigm that was not investigated before, namely the irrelevant rolls used to verify the dies legitimacy and not to determine pay. If reporting a desired counterfactual indeed seems like a more legitimate lie, people may report the highest outcome they observed on all rolls instead of only the one that should actually determine pay. People may have been reporting the highest of the multiple rolls they saw that is, the most desirable counterfactual they observed. To address this possibility, we adapted Fischbacher and Heusis (2008) paradigm to
Justified Ethicality 7 eliminate these extra rolls. Holding all other aspects of the paradigm constant, we introduced a condition we call single-roll that allows participants to verify that the die are legitimate while only rolling the die once during the task. We contrast this condition to a multiple-rolls condition in which participants verify the dies legitimacy by rolling it additional times after the roll that is to determine pay. Our main prediction was thus: Hypothesis 1: Reducing peoples ability to observe desired counterfactuals reduces lying. Put differently, seeing only one die roll will reduce lying compared to seeing additional rolls because seeing only one roll reduces ones ability to justify the lie to him or herself by reporting a desired counterfactual that almost happened. Because reports using the die rolling paradigm were private, we could only statistically infer what strategies participants used. In Experiment 2, we explicitly asked participants who were explained this task to indicate what they would report when faced with various combinations of first and additional rolls. Doing so assessed if people indeed over-report the highest of these rolls more than any other profit boosting option available to them. Experiments 3 and 4 were designed to test the proposed psychological mechanism accounting for this effect. Specifically, in these experiments we tested the prediction that the reason people report observed desired counterfactuals in order to profit financially is because reporting these justified outcomes seems less unethical than dishonestly reporting a desired outcome that was never observed. Simply put, we tested if seeing a justification makes lying for profit feel more honest compared to a similar lie without justification. The influence of having a justification on lying should thus be mediated by the modification in ethical perceptions. Therefore, a given lie feels less unethical with a justification and thus is more likely to be used. Specifically, we tested the following prediction: Hypothesis 2: Dishonestly reporting a desired counterfactual that was observed is considered less unethical than other desired dishonest reports that were not observed. Experiment 1 Participants rolled a die and earned money according to what they reported rolling (Fischbacher & Heusi, 2008; Shalvi, et al., in press). We placed the die under a paper cup with small hole in the top and had participants shake the cup to roll the die and look through the hole to see the result. This procedure assured participants that only they could see the die (see Figure 1). The first roll was to determine pay equivalent in US Dollars to the number they reported. After rolling for pay, participants were also instructed to roll at least two more times to ensure that the die was legitimate. This procedure also insured that their first roll could not be observed by anyone, even after they left the experiment. In another condition, holding all other situational parameters constant, we eliminated participants ability to observe more than one roll before reporting their outcomes. We tested if eliminating peoples ability to observe a desired counterfactual in the form of a higher number on one of the subsequent rolls (meant only to verify the dies legitimacy and not to determine pay), reduces lying. Participants and procedure. One hundred and twenty night students at an eastern US university were randomly assigned to one of two die roll conditions (multiple-rolls vs. single- roll). Participants arrived at the lab in groups of 8 to 19, were seated in individual cubicles and performed the die-under-cup task, by rolling a six-faced die under a cup, reporting the outcome, and earning in dollars the number they reported. Participants in the multiple-rolls condition were instructed to shake the cup once, check the outcome that determined their pay, and then roll at least two more times to make sure that the die was legitimate. They were then asked to write down the first roll outcome. In the single-roll condition, participants first saw a box
Justified Ethicality 8 containing all the dice that were used in their experimental session and were allowed to test that the dice were legitimate in any way they liked, including rolling them. After testing the dice, participants were then instructed on and received a die-under-cup. Participants were asked to roll and check the outcome, but not instructed to roll again. To assure participants that their outcomes were private in this condition, the experimenters passed around a box into which each participant swept his cup and die. This procedure was demonstrated by the experimenter before participants began the actual task. In summary, both the multiple rolls and the single roll conditions allowed the participants to privately roll a die, verify that it was legitimate, and hide the result from being viewed by anyone in the future, but the two conditions differed critically in that the multiple roll condition allowed participants to roll the die multiple times after they knew what the task was. Results If people use observed desired counterfactuals to unethically increase their profit, we should find less lying in the single-roll compared to the multiple-rolls condition, and the distribution of multiple-rolls reports should be consistent with reporting the best out of 3 outcomes. Indeed the multiple-rolls distribution differed from the uniform distribution expected from a fair die (Kolmogorov-Smirnov Z = 2.03, p < .01). This was due to underreporting of 1s, 2s, and 3s (95%CIs < chance [16.67%] for 1s and 2s and 90%CI < chance for 3s) and over-reporting 6s (95%CI > chance). In the single-roll condition, lying was reduced and the distribution did not significantly differ from chance (Kolmogorov- Smirnov Z = 1.16, ns). Importantly, as predicted by Hypothesis 1, people lied significantly less in the single-roll condition (M = $3.97, SD = 1.56) than in the multiple-rolls condition (M = $4.45, SD = 1.59), Mann-Whitney Z = 1.93, p = .05, see Figure 2. We compared the distributions in each condition to the theoretical distribution of choosing the highest value of three die rolls, a mechanism for lying that we propose. In the theoretical distribution of choosing the highest of 3 rolls (see Figure 2, left side), the likelihood of seeing 1 as the highest value of the three rolls is very low (it happens only if one rolls 1 three times, which is 1 of 63 = 216 combinations), whereas seeing 6 as the highest value of three rolls is a more likely event (it happens in 91 of 216 three die roll combinations). If people report desired counterfactuals (the highest value they observed) the distribution of the multiple rolls condition should resemble the distribution resulting from choosing the highest of several rolls. Indeed, supporting Hypothesis 1, the multiple-rolls distribution did not differ from the theoretical distribution of choosing the highest of three rolls (Kolmogorov- Smirnov Z = .97, ns), whereas the single-roll distribution did (Kolmogorov-Smirnov Z = 2.34, p < .01). Hence, we could not reject the hypothesis that people who rolled more than once chose the highest of three rolls they have observed. The behavioral pattern collected in this study allows us to assess the potential range of honest people as well as income maximizers that would lie in all cases (see Fischbacher & Heusi, 2008). There is no reason to assume that anyone reporting a 1 would be lying. In our sample, 8% in both single and multiple rolls conditions reported rolling 1 suggesting that at least 8% of people would not lie whether or not they have a justification to do so. On the other extreme, 34% of participants in the multiple rolls condition and 19% in the single roll condition reported rolling a 6. Because 6 would be reported an expected 1/6th of the time if participants were reporting honestly, we can calculate the proportion of people who lie to the maximum by subtracting 1/6th from the observed proportions and multiplying the outcome by 6/5. As clarified by Fischbacher and Heusi, this multiplication is required to take into account those participants that actually rolled a 6 but would have lied if they have rolled a lower number. In the multiple rolls condition, the
Justified Ethicality 9 proportion of income-maximizers is thus (34%- 17%) * 6/5 = 20% which is very similar to Fischbacher and Heusis 22% estimate. However, as demonstrated here, the proportion of incomemaximizers might be much lower when a justification in the form of extra rolls is removed. Calculating the proportion of liars in the single rolls condition reveals that only (19%-17%) * 6/5 = 2.5% of people may be considered as people who would lie to the maximum extent regardless of the outcome they observed. Discussion and Introduction to Experiment 2 Supporting Hypothesis 1, limiting ones ability to observe desired counterfactuals reduced lying. People who saw only one roll lied less compared to those who saw multiple rolls. Furthermore, providing initial support to Hypothesis 1, we could not reject the hypothesis that people who rolled more than once reported rolling the highest of 3 rolls they observed (which was the minimal number of times they were instructed to roll). The method employed in experiment 1 made it absolutely transparent that ones actions can never be observed. As such, it rooted out fears of detection and allowed testing internal standards. Justifications in the form of the extra rolls were in the eyes of the person rolling only; no one else could know they had these justifications. Statistical analyses further demonstrated that the distribution of reported outcomes for participants, who could roll more than one time, resembled the theoretical distribution of choosing the highest of three rolls. We could not, however, directly observe what people saw before reporting the outcome. In experiment 2, we asked people directly what they would have reported in such ethically challenging situations. While this procedure does not motivate participants reports financially, it does allow a more direct test of the prediction that people report the desired counterfactual they observe more than any other profit boosting options that are available to them. Participants and procedures. Sixty eight people (51% females, Mage = 26.71) were recruited online and paid for participation. We employed a 6 (1st roll: 1-6) X 6 (2nd roll: 1-6) within subject design. Participants read the instructions of the multiple-rolls condition and indicated what they would have reported in all randomly presented two die roll combinations. For example, a participant was shown rolls of 2 and 5 and asked what she would report. Comprehension checks. Participants were asked three questions to verify that they understood the task: whether the first roll counts for pay, whether second roll was used to verify that the die was legitimate, and whether when one sees a 3 on the first roll that was also the number one should have reported. Because incorrect comprehension of the task could lead to a false appearance of using counterfactuals e.g., participants might think that the task demands reporting the highest roll - seventeen participants who did not answer these questions correctly were excluded from all analyses. Including these participants did not modify any of the reported results. Results Most people stated that they would have reported honestly. Specifically, 69.6% would have honestly reported rolling a 1, 73.9% when rolling a 2, 81.7% when rolling a 3, 88.6% when rolling a 4, 89.2% when rolling a 5, and 94.4% when rolling a 6. An ANOVA predicting the stated likelihood to lie from rolled outcome revealed a linear effect, F (1, 50) = 21.64, p < .0001, 2 = .30. The higher the actual roll was, the more likely the participants were to give an honest report. These greater reports of honesty than were seen in Experiment 1 are not surprising in some sense since participants did not gain financially from reporting higher numbers. Providing further support to Hypothesis 1, among those reporting that they would have lied, the outcome most commonly used was the desired counterfactual, see Table 1. For all dishonest reports, we calculated the proportion of people who reported the second roll outcome conditional on
Justified Ethicality 10 each first roll outcome. We then tested whether this proportion differed from that specified by a null hypothesis in which dishonest reports were chosen randomly from a uniform distribution of higher outcomes. For example, of the 83 dishonest reports after seeing 1 on the first roll, 40 (48.2%) were equal to the value that appeared on the second roll. A binominal test revealed that this proportion was higher than the 20% predicted by our null hypothesis (e.g., 2, 3, 4, 5, or 6), p <.0001. Similarly, 29 of the 61 dishonest reports after seeing a 2, 18 of the 34 dishonest reports after seeing a 3, and 10 of the 13 dishonest reports after seeing a 4 were 2nd roll values, all proportions larger than expected by chance (p <.001, p =.03, and p =.09 respectively). Taken together, 51% (97 of 191) dishonest reports were justified lies (i.e., equal to the observed desired counterfactual). This proportion was almost twice as high as the proportion of choosing a higher number at random (26%, p< .01).1 Discussion and Introduction to Experiment 3 Results of Experiment 2 provided further support for our main prediction that people use the observed desired counterfactual to unethically boost profit more than any other option available to them. In Experiment 1, support was gained by eliminating the ability to observe more than one roll, which reduced lying. Furthermore, in the multiple rolls condition the distribution of reported outcomes did not differ from the distribution of reporting the highest of three rolls, while in the single roll condition it did. Taken together we concluded that, as predicted, people seem to be reporting the highest of the rolls they observed. That said, as peoples rolls were completely private, these behavioral results provided strong but indirect evidence for our predicted mechanism. The results of Experiment 2 provided more direct evidence that among people who estimated that they would report dishonestly, the most common reported outcome, across all die roll combinations, was the highest observed value. In Experiments 3 and 4, we moved forward to test our predictions about the underlying psychological mechanism leading people to report the highest of the rolls they saw. We predicted that reporting an observed value that is not supposed to count would feel less unethical compared to reporting any other inflated value that was not observed. In turn, this reduced level of perceived unethicality will increase ones likelihood to lie. Put differently, seeing a high value on a roll not meant to count makes it more legitimate to lie as long as the observed value is reported. Note that this is all the more interesting, since if one decides to lie, one might as well always report the highest possible outcome, rather than the observed desirable counterfactual. To gauge peoples perceptions of what is ethical and what is not, Experiment 3 asked people to evaluate to what extent different dishonest reports were considered to be unethical. Specifically, we presented people with different combinations of die rolls and reports, and asked them to what extent they considered these reports to be lies. We tested whether observing desired counterfactuals makes a dishonest report seem less unethical compared to a dishonest report occurring without observing the counterfactual. That is, whether observing desired counterfactuals justifies lying. Participants and procedures. One hundred fifteen people (44% females, M age= 29.98) were recruited online and paid for participation. We employed a 6 (1st roll: 1-6) X 6 (2nd roll: 1-6) X 6 (report: 1-6) within subjects design, except that we omitted combinations in which the report was lower than the first roll (lying downward), resulting in 126 combinations. Each participant was randomly assigned to evaluate 48 of these combinations in one of three different randomly predetermined blocks (the order of presented combinations within blocks was also randomized). Blocking did not moderate any of the described results and is thus not discussed any further. Participants were asked to rate for each roll/report combination the extent to which they considered each report to be a lie (1 = not at all to 7 = very much). Comprehension checks. As in Experiment 2, participants were asked three questions to verify that they understood the task: whether the first roll counts for pay, whether second roll was used to verify that the die was legitimate, and whether when one sees a 3 on the first roll that was
Justified Ethicality 11 also the number one should have reported. The 32 participants who did not understand these rules were excluded from all analyses. Analyzing the data without those participants was a conservative test to our hypothesis. This is because participants who fail to understand the task were likely to be lenient about reporting the highest of the observed rolls merely because they failed to understand that such behavior violates the tasks rules. Including these participants did not modify any of the reported results. Results Die roll/report combinations were recoded to reflect whether the person rolling was honest (e.g., rolled 2 then 3 and reported 2) or lied to increase profit (e.g., rolled 2 then 3 and reported 4). Not surprisingly, an ANOVA with honest vs. dishonest reports as within subject factor, revealed that reporting honestly was consider less of a lie (M =1.87, SD =1.28) than reporting dishonestly (M =5.72, SD =1.47), F(1, 82) =189.54, p <.001, 2 =.70. Honest reports received ranking higher than 1 due to some participants tendency not to use the full Likert scale available to them. Notwithstanding this tendency, across all truthful combinations the mode was always 1 assuring that most participants evaluated honest reports by choosing the most honest end of the scale. We further recoded justified lies as untruthful reports equal to the desired counterfactual (e.g., rolled 2 then 5 and reported 5) and unjustified lies as untruthful reports that were different from the desired counterfactual (e.g., rolled 2 then 3 and reported 4). Supporting Hypothesis 2, an ANOVA with justified vs. unjustified lies as within subject factor revealed that an untruthful report that was justified (i.e., when the report was equal to the 2nd roll value) was considered less of a lie (M =5.57, SD =1.59) compared to an unjustified (i.e., when the report was not equal to the 2nd roll value) untruthful report (M =5.87, SD=1.47), F(1,82) =10.52, p =.001, 2 =.11. This effect was primarily driven by unjustified lies which were higher than the second roll outcome (p < .01), which is the key comparison for our reasoning as such combinations provide absolutely no justification for the dishonest report. Discussion and Introduction to Experiment 4 Results from Experiment 3 supported hypothesis 2. When considering whether a report was ethical, people took into consideration the outcome obtained on a roll not meant to determine pay. People considered dishonest reports to be less unethical with (rather than without) a possible justification for a lie. In Experiment 4, we manipulated the number of observed desired counterfactuals by providing information about the first roll that counted for pay, and two additional rolls assessing the dies legitimacy. Participants had to evaluate the extent to which they consider another persons lying as unethical after this person saw no (vs. one vs. two) desired counterfactuals. They were also asked to estimate the extent to which they were likely to act the same (i.e., lie). Participants. Sixty nine undergraduate students at an eastern US university completed a die roll survey after participating in another experiment. Design and Procedures. Participants filled out a survey with 25 three die roll combinations. Participants learned that a person reported rolling a 6 after rolling a lower number than 6 on the first roll and rolling no 6 (vs. one 6 vs. two 6s) in two subsequent rolls determining the dies legitimacy. We employed a 5 (1st roll: 1-5) X 3 (combined number of 6s appearing in the second and third rolls: 0 vs. 1 vs. 2) within subject design. Participants indicated for each combination the extent to which they considered the report a lie and the likelihood that they would have done the same (both on a 1not at all to 7 very much scale). All participants reported that they understood the rules.
Justified Ethicality 12 Results Is this a lie? We recoded all combinations to indicate the number of 6s they include in the two subsequent rolls (0 vs. 1 vs. 2)2. Supporting Hypothesis 2, an ANOVA with number of 6s (0 vs. 1 vs. 2) as within subject factor predicted the extent to which reporting a 6 was considered a lie: reporting a 6 without seeing a 6 was considered a greater lie (M=5.76, SD =.56) than doing so after seeing one 6 (M =5.29, SD =1.08), which in turn was considered a greater lie than doing so after seeing two 6s (M =4.91, SD =1.59), F (1,68) =23.98, p <.001, 2 =.26 (simple effects, Fs >18.78, ps <.001). Would you do the same? Supporting Hypothesis 1, an ANOVA with number of 6s (0 vs. 1 vs. 2) as a within subject factor predicted the extent to which participants reported that they would have done the same (i.e. lie). Participants indicated that they were less likely to have reported a 6 without seeing a 6 (M =1.92, SD =1.17) than doing so after seeing one 6 (M =2.51, SD =1.47) and after seeing two 6s (M = 2.82, SD = 1.72), F(1,68) = 32.58, p<.001, 2 =.32 (simple effects, Fs >14.79, ps <.001)3, see Figure 3. Further analyses revealed that the modification in ethical perceptions partially mediated the effect of observing a desired counterfactual (a 6) on participants reported likelihood that they would lie as well.4 Experiment 4 Discussion Experiment 4 provided further support for hypotheses H1 and H2: Participants evaluated an untruthful report as less unethical as a function of whether a desired counterfactual was observed. They further indicated that they were more likely to lie when observing such desired counterfactuals. Additional analyses provided support for the idea that the relationship between seeing a desired counterfactual and ones likelihood to lie was partially mediated by ones modification in the perceived unethicality of the act. One limitation to the results of Experiment 4 relates to the fact that since it was a paper and pencil survey in which order of presented die roll combinations was not randomized, we could not rule out order effects. The possibility that order effects were driving any of the reported results of Experiment 4 seems unlikely however, as these results replicate the results obtained in Experiment 3, in which randomized combinations were presented. General Discussion The current work contributes to our understanding of peoples perceptions of ethicality and how these perceptions translate into unethical behavior. Employing a simple die-under- cup paradigm that allowed participants to lie anonymously and gain financially, we found that observing desired counterfactuals leads people to (1) lie more than people who do not observe desired counterfactuals, (2) evaluate unethical acts as more ethical, and (3) assess their own likelihood to commit an unethical act to be higher due to the modification in ethical perceptions. In Experiment 1, we found that allowing people the opportunity to lie without getting caught led to dishonesty. However, while evidence of lying was clear, the amount of lying was modest. For example, only 34% of participants in the multiple rolls condition reported a 6, when in fact all of them could have gotten away with reporting a 6. This finding confirms other recent work showing that people limit the magnitude of their lying even when they cannot be caught (Fischbacher & Heusi, 2008; Shalvi, et al. in press; Mazar, et al. 2008; Gino, et al., 2009, 2010). However, we also show that beyond the magnitude of lying, people rely critically on self-justifications to enable themselves to lie, even when these justifications will never be available to other people. This conclusion stems from three pieces of evidence. First, in Experiment 1, the distribution of reported
Justified Ethicality 13 outcomes did not differ from the theoretical distribution of choosing the highest of three rolls, the mechanism for lying we proposed. Second,eliminating peoples ability to observe more than one roll reduced lying. This solid piece of evidence speaks to the fact that the extra rolls were worth money to participants although these rolls were clearly for their eyes only. In the single roll condition, people made less money because they could not avail themselves of these justifications. Third, in Experiment 2, participants indicated what they would have reported in all possible die roll combinations. Over all possible combinations, people reported the outcome of the second roll value more than any other profit boosting value available to them. Together, these findings suggest that private justifications translate to larger lies. We further proposed that self justifications allowed people to lie more for money while feeling honest. That is, we suggested that justifications make a given dishonest act feel less unethical compared to the exact same dishonest act that occurs without a justification. Results of Experiments 3 and 4 supported this idea. We found that people viewed a lie more severely when it did not match the counterfactual support of an additional die roll. In Experiment 4 we gained support for the notion that the modification in ones perception of ethicality partially mediates the relation between observing a desired counterfactual and lying. Our results do not imply that people do not care about their ability to appear honest in the eyes of others. Obviously, as pointed our in the introduction, there is ample evidence that such considerations do play an important role in determining ones likelihood to lie. The contribution of the current work is that over and above such a desire to appear honest in the eyes of others (which was constant across our experimental conditions), people value feeling honest in their own eyes. Importantly, we found that having a justification to behave unethically, in the form of observing desired counterfactuals, allow people to maintain a feeling of honesty even when lying quite a bit. An interesting finding in Experiment 4 was the different level of variance observed for both ratings of how unethical is it to lie without seeing a 6, when seeing one 6 and when observing two 6s as well as how likely people are to lie in these conditions. The variance increased when more justifications were observed. While this issue did not pose a statistical problem in interpreting the data, from a theoretical point of view this finding may be of interest as it seems to suggest that people are more polarized in their opinions when evaluating justified lies compared to when evaluating unjustified lies. Put differently, when justifications are not available people seem to be rather unanimous about the fact that lying is unethical while when justifications are available people are relatively more diverse in their interpretation of the lies. This potentially points out to some individual difference on this issue. Future research may be needed to further explore this possibility in greater detail. Limitations The die-under-cup paradigm prevented us from assessing who actually lied. Some may consider the fact that we cannot trace individual reports back to observed die rolls as the downside of the paradigm. Indeed, in Experiment 1, we were unable to collect self-report data to assess the mediating process for dishonesty. We believe, however, that the indispensable feature of our approach is that it makes absolutely transparent to participants that their actions could never be observed. In this way, we can be confident in our data interpretation that our participants were not concerned that their actions would be discovered and that our experiment simply tests their propensity to lie as governed by their internal standards (for similar approach see e.g., Mazar, et al., 2008; Greene & Paxton, 2009; Lammers, Stapel & Galinsky, 2010). As such, using justifications in the form of the additional rolls that were observed were absolutely for the self. The fact that we could not trace back actual reports in Experiment 1 forced us to assess lying indirectly by comparing the distribution of outcomes to different theoretical distributions (i.e., honest, best of three). The significant difference between an observed distribution in the multiple rolls condition and a theoretical honest distribution provided solid indication for lying. However, the
Justified Ethicality 14 lack of difference between the observed distribution and the theoretical distribution of choosing the highest of three provided indirect support as it did not reject the null (no difference) hypothesis. We used a hypothetical experiment to fill in this gap. Results of Experiment 2 helped in that they clarified that indeed a significant portion of people indicating that they would lie chose the value appearing on the second roll as a way to boost their profit. Together, these pieces of evidence converge to support our proposition that people use private justifications when deciding the extent to which they can lie. Theoretical contribution and boundary conditions Gneezy (2005) noted that people have non-consequential preferences in which they treat the same monetary outcome differently, depending on the process that leads up to it (p.392). Corroborating this notion, we found that peoples likelihood to lie for a given outcome was influenced by whether the outcome was observed while assessing the dies legitimacy. It seems that the utility of a given monetary outcome varies depending on whether it is attained by a (non) justified lie. Recent work (Shalvi, et al., in press) reveals that the likelihood that people will lie for a specific amount depends on the additional profit boosting alternatives that are available. People are less likely to lie when the available options are narrowed to include only major lies (lying to the maximum extent) or minor ones (increasing profit only marginally above an exit-option) compared with situations that also include middle-of-the- road lies (increasing profit substantially but not to the maximum). Narrowing the range of potential lies may have reduced peoples ability to justify a potential unethical act, in turn decreasing lying. The idea that justifications modify perceptions of ethicality further communicates with work on bounded ethicality (Bazerman & Banaji, 2004; Chugh, Bazerman & Banaji, 2005) proposing that implicit biases in peoples awareness systematically prevent them from identifying others wrongdoing. Attempting to detect cheaters, people were less tolerant when they observed an abrupt change in performance rather than multiple, gradual changes (Gino & Bazerman, 2009). Our findings may suggest that peoples lack of awareness of others misconduct, leading to an ethical slippery slope, may be due to the extent to which the others unethical behavior may be justified. Non-obvious lies may be more likely to be perceived in ethically lenient ways. Future research is needed to conclusively address this possibility and determine to what extent our ethicality is bounded or justified. Our reasoning linking justifications and counterfactual thinking relied on Kahneman and Millers Norm Theory, which suggests that norms are established after the fact and based on available information and ones ability to reinterpret this information. A key factor in this theory is the extent to which the available information is mutable (i.e., modifiable). This is one key aspect that may provide more insight into the boundary conditions of the phenomenon described in our studies. There are several leading candidates for such boundary conditions. First, a mutable event is an event that people may influence and modify, or at least feel that they may. A likely boundary condition to our results is thus reducing peoples control over generating the justification, which should reduce their likelihood to use the justification. Manipulating whether the participant or another person rolls the die may yield interesting data pertaining to this issue. Second, it is possible that merely imagining the alternative scenarios (rather than actually observing them) may be sufficient to justify lying. Attempting to clarify this issue would help determine whether it is the mental simulation or the actual evidence that is used as a justification for lying. Third, past work found that recent events are more mutable than earlier events (Miller & Gunasegaram, 1990). It is thus possible that manipulating the roll that counts for pay to be the first vs. a later roll would influence peoples likelihood to use the non relevant for pay rolls as justifications. This line of reasoning suggests that lying about the outcome of roll X (e.g., the 5th roll) using the values that were observed before would be relatively difficult as all the
Justified Ethicality 15 counterfactual evidence that preceded the roll that count is less mutable at the time of reporting. Studying these potential boundary conditions to our results will not only shed light about what type of information is more likely to be used as justification, but will also contribute to the more general discussion about the relation between mutability and counterfactual thinking. Concluding thoughts It seems that Oscar Wilde was right in suggesting that morality, like art, means drawing a line somewhere. The current results suggest that this line is drawn exactly where justifications end. The current work contributes to our understanding of how justifications that are available only for the self shape ones ethical perceptions and behavior. Using a simple die-under-cup paradigm we provided evidence that people report rolling the highest of the rolls they see in order to gain money, while they know that the rules clearly indicate that they should report only the first roll outcome. As such, people seem to drive value from self- justifications and lie for money as long as they can feel honest.
B R I T I S H J O U R N A L O F P S YC H I AT RY ( 2 0 0 5 ) , 1 8 7, 3 2 0 ^ 3 2 5
Prefrontal white matter in pathological liarsy

YALING YANG, ADRIAN R AINE, TODD LENCZ, SUSAN BIHRLE, LORI LACASSE and PATRICK COLLET TI
scanning revealed major atrophy of the right superior temporal gyrus (Raine et al, 2000). Full demographic, cognitive and physical characteristics of the three groups of participants are shown in Table 1. Full informed, written consent was obtained from all participants in accordance with institutional review board procedures at the University of Southern California.
Assessment of lying Background Studies have shown increased bilateral activation in the prefrontal cortex when normal individuals lie, butthere have been no structural imaging studies of deceitful individuals. Aims To assess whether deceitful individuals show structural abnormalities in prefrontal grey and white matter volume. Method Prefrontal grey and white matter volumes were assessed using structural magnetic resonance imaging in 12 individuals who pathologically lie, cheat and deceive (liars),16 antisocial controls and 21normal controls. Results Liars showed a 22^26% increase in prefrontal white matter and a 36^42% reduction in prefrontal grey/ white ratios compared with both antisocial controls and normal controls. Conclusions These findings provide the first evidence of a structural brain deficitinliars, theyimplicate theyimplicatethe the prefrontal cortex as an important (but not sole) component in the neural circuitry underlying lying and provide an initial neurobiological correlate of a deceitful personality. Declaration of interest None. Funding detailed in Acknowledgements.
Despite many clinical and psychological studies on lying and deception (Rogers, 1997; McCann, 1998), and although it has been hypothesised that there is a neurobiological basis to lying, cheating and manipulative behaviour (Ford et al, 1988), this hypothesis has not been tested. Several functional magnetic resonance imaging (fMRI) studies on normal individuals who lie or feign memory impairments have found increased bilateral activation in the prefrontal cortex during lying (Spence et al, 2001; Lee et al, 2002; Ganis et al, 2003). To provide initial empirical data on the structural brain imaging correlates of lying and deception, we assessed the volume of prefrontal grey and white matter in individuals who lie, cheat or deceive to test the hypothesis that such individuals have an abnormality within the prefrontal cortex. We used a symptom-based approach (Halligan & David, 2001) to define a group of liars and investigated the neurobiological correlates of lying that are not shared by either an antisocial control group or a normal control group. Participants were defined as liars if they fulfilled: (a) criteria for pathological lying on the Psychopathy Checklist Revised (PCL R; Hare, 1991); or (b) criteria for conning/manipulative behaviour on the PCLR; or (c) the deceitfulness criterion for DSMIV (American Psychiatric Association, 1994) antisocial personality disorder (lifelong repeated lying, use of aliases or conning others for personal profit or pleasure); or (d) criteria for malingering as indicated by admitting to telling lies to obtain sickness benefits in a self-report crime interview (see below). The term liars is intended as a short-hand specifically to denote the above four symptoms. A symptom-based orientation was employed because it has a number of significant advantages over a more traditional syndromal approach (Bentall et al, 1988; Costello, 1992; Halligan & David, 2001), especially in this particular field, which lacks diagnostic boundaries. Normal controls (n21) were selected from the remaining pool on the basis that they fulfilled none of the four criteria for lying. They also failed to meet criteria for either DSMIV antisocial personality disorder or DSMIV conduct disorder, and were matched as closely as possible to the 12 liars with respect to gender and ethnicity. Because the liar group was significantly antisocial, any structural brain differences could be an artefact of antisocial personality, which has been associated with an 11% reduction in prefrontal grey matter in this group (Raine et al, 2000). Consequently, an antisocial control group (n16) was formed by matching liars with individuals who did not fulfil criteria for lying, but who scored as highly as liars on DSMIV measures of antisocial personality disorder and conduct disorder.
METHOD Participants
All participants were taken from a total sample of 108 community volunteers drawn from five temporary employment agencies in Los Angeles (Raine et al, 2000). Groups consisted of 12 participants (11 male, 1 female) with a history of lying (liars), 21 normal controls (15 male, 6 female) who had neither antisocial personality disorder nor a history of pathological lying and 16 antisocial controls (15 male, 1 female) with antisocial personality disorder but no history of pathological lying. Exclusion criteria were: age under 21 or over 45 years, non-fluency in English, a history of epilepsy, claustrophobia, a pacemaker and metal implants. One individual was excluded a priori because brain
See invited commentary, pp. 326^327, this issue.
32 0
P R E F RON R ON TA L W H I T E M AT T E R IN IN L I A R S
T Table able 1 Demographic, cognitive and physical, and diagnostic characteristics of the study groups1
Characteristic
Normal controls Antisocial controls (n=21) (n=16)
Liars (n12)
Statistics
Group comparisons
Demographic Age, years: mean (s.d.) Socio-economic status: mean (s.d.) Gender (male/female) Ethnicity, % White Cognitive and physical Handedness: mean (s.d.) Total IQ: mean (s.d.) Verbal IQ minus performance IQ: mean (s.d.) Head circumference, inches: mean (s.d.) Period of unconsciousness, min: mean (s.d.) Hospitalised head trauma, % present Diagnostic Total psychopathy score: mean (s.d.) Total APD score: mean (s.d.) APD diagnosis, % Conduct disorder, % Alcohol dependence/misuse, % present Drug dependence/misuse, % present Alcohol/drug dependence/misuse, % present 10.7 (5.3) 1.4 (2.2) 0 0 38.1 40.0 47.6 17.8 (4.0) 5.6 (2.3) 25 37.5 56.3 43.8 62.5 21.1 (7.7) 6.2 (3.7) 25 33.3 58.3 58.3 66.7 F (2,46)14.8, P0.0001 F (2,46)17.0, P0.0001 w26.1, d.f.2, P0.047 w29.5, d.f.2, P0.009 w 1.8, d.f.2, P0.46
2
31.4 (6.9) 38.8 (10.2) 15/6 66.7 33.3 (10.8) 106.6 (14.3) 75.9 (15.8) 56.4 (2.15) 363.44 (1439.1) 33.3
29.5 (5.5) 34.3 (9.2) 15/1 31.3 33.9 (10.2) 94.2 (11.3) 72.8 (15.3) 57.0 (1.95) 68.2 (257.5) 56.3
36.5 (5.3) 35.7 (9.1) 11/1 33.3 31.8 (13.0) 101.0 (20.1) 11.2 (22.8) 57.8 (1.26) 9.18 (29.8) 50.0
F (2,46)4.6, P0.01 F (2,45)1.0, P0.36 w24.0, d.f.2, P0.13 w25.7, d.f.2, P0.056 F (2,46)0.12, P0.88 F (2,45)3.0, P0.056 F (2,45)3.6, P0.036 F (2,46)2.1, P0.12 F (2,46)0.63, P0.54 w22.0, d.f.2, P0.08
NC, AC5L
AC5NC
AC5NC NC, AC5L
NC5AC, L NC5AC, L NC5AC, L NC5AC, L
w21.1, d.f.2, P0.59 w21.4, d.f.2, P0.49
APD, antisocial personality disorder; NC, normal controls; AC, antisocial controls; L, liars. 1. All group comparisons are two-tailed, P50.05.
All clinical ratings and diagnoses were performed by clinical PhD graduate research assistants who had both been trained and supervised by A.R. and also had undergone a standardised training and quality assurance programme for diagnostic assessment (Ventura et al, 1998). Pathological lying and conning/manipulative characteristics were assessed using the PCLR, which was supplemented by five sources of collateral data (Raine et al, 2000). These were the Interpersonal Measure of Psychopathy (IMP; Kosson et al, 1997), which provides an interviewers ratings of the participants interpersonal behaviours and which has been validated for use with incarcerated and non-incarcerated samples; self-reported crime as assessed by an adult extension (Raine et al, 2000) of the National Youth Survey self-report delinquency measure (Elliott et al, 1983); official criminal records; data derived from, and behavioural observations made during, the Structured Clinical Interview for DSMIV Mental Disorders (SCIDI; First et al, 1995a) and the Structured Clinical Interview for DSMIV Axis II Personality Disorders (SCIDII; First et al, 1995b). The deceitfulness trait of antisocial personality disorder was ascertained using the SCIDII, whereas
malingering (telling lies to obtain sickness benefits) was self-reported self-reported on the adult extension of the National Youth Survey self-report delinquency measure. Comparisons of the study groups are given in Table 1. The two antisocial groups did not differ with respect to rates of antisocial personality disorder and conduct disorder, but rates for both were significantly higher than for normal controls. The same pattern was observed for total psychopathy scores and total antisocial personality scores (the latter created by summing SCID scores on the seven features of antisocial personality disorder). All three groups did not differ significantly with respect to social class, ethnicity, IQ, handedness, history of head injury, height, head circumference and DSMIV diagnoses of alcohol/drug misuse/dependence. However, groups differed significantly with respect to age, with a higher mean age in the liar group than both control groups. Liars also had significantly higher verbal relative to performance IQ compared with both control groups. There were also trends for group differences in ethnicity (P0.056) and total IQ (P0.056), with antisocial controls tending to have lower total IQ and a greater representation of individuals from Black and
minority controls.
ethnic
groups
than
normal
Demographic, cognitive and physical measures

Estimated IQ was based on five sub-tests (vocabulary, arithmetic, digit span, digit symbol, block design) of the Wechsler Adult Intelligence Scale Revised (WAISR; Wechsler, 1981), with verbalperformance discrepancy scores computed by subtracting performance IQ from verbal IQ. Right v. left hand preference was assessed using the abbreviated Oldfield Inventory (Bryden, 1977), with high scores indicating a stronger preference for right-handedness. History of head injury was defined as head trauma resulting in hospitalisation and the amount of time (in minutes) the subject was rendered unconscious from any head injury. Social class was measured using the Hollingshead classification system (Hollingshead, 1975). A physical examination was conducted to derive measures of height and head circumference.
Magnetic resonance imaging

Structural MRI was conducted on a Philips S15/ACS scanner (Selton, Connecticut, USA)
3 21
YA N G E T A L
with a magnet of 1.5 Tesla field strength. Following an initial alignment sequence of one midsagittal and four parasagittal scans (spin-echo T1-weighted image acquisition, time to repetition600 ms, echo time 20 ms) to identify the anterior commissure/ posterior commissure (AC/PC) plane, 128 three-dimensional T1-weighted gradientecho coronal images (time to repetition 34 ms, echo time12.4 ms, flip angle 358, thickness1.7 mm, 2566256 matrix, field of view23 cm) were taken in the plane directly orthogonal to the AC/PC line. Brain images were reconstructed in three dimensions using a SPARC workstation and semi-automated CAMRA S200 ALLEGRO software (Sun Microsystems Inc., Santa Clara, California, USA) was used for grey/white cerebrospinal fluid segmentation. The prefrontal region was defined as all cortex anterior to the genu of the corpus callosum, and divided into left and right hemispheres along the longitudinal fissure (Raine et al, 2000). Segmentation of grey and white matter was performed using a thresholding algorithm, with the operator unaware of group membership, and applying a cut-off value to the signal intensity histogram to optimally differentiate white from grey matter, areas of which were defined using an automated seeding algorithm on each slice. Whole brain volume was defined as all cerebral cortex, excluding the ventricles, pons and cerebellum. The pons was excluded by drawing a straight line between the two innermost points that form the superior border.
Colliculi were excluded when no longer attached to the cerebral hemispheres. For volume measures, areas on each slice (mm2) were multiplied by slice thickness (1.7 mm) and added to provide volumes in cubic centimetres. centimetres. Interrater reliability (intraclass correlation coefficient) based on 23 scans (raters unaware of each others ratings and group membership) were as follows: total brain volume (0.99), left prefrontal grey (0.99), right prefrontal grey (0.99), left prefrontal white (0.93), right prefrontal white (0.94) and total brain volume (0.99). Volumes of grey and white matter were calculated separately for each hemisphere and a grey/white ratio was calculated for each hemisphere, with lower scores indicating increased white matter compared with grey.
RESULTS Magnetic resonance imaging prefrontal volumes

Liars showed a significant increase in prefrontal white matter and slightly reduced grey matter. A 3 (groups)62 (left/right hemisphere)62 (grey/white) repeatedmeasures analysis of variance (ANOVA) using the multivariate procedure showed no main effect for group (F(2,46)0.729, P0.488) but a significant group6 grey/white interaction (F(2,46)9.049, P0.0001, Z20.282). To break down this interaction, separate analyses were run for grey and white matter using a Bonferroni correction (a0.017) for pairwise comparisons. Liars had significantly greater
prefrontal white matter volume than both antisocial controls (t3.1, d.f.26, P0.004) and normal controls (t2.7, d.f.31, P0.01). Liars had a 25.7% increase (13.3 cm3) in prefrontal white matter compared with antisocial controls and a 22.2% increase (11.8 cm3) compared with normal controls (Fig. 1). For grey matter, liars had non-significantly reduced volumes compared with normal controls (t2.1, d.f.31, P0.04) but not compared with antisocial controls (t0.79, d.f.26, P0.43; 0.43; Fig. 1). Liars had a 14.2% decrease (10.5 cm3) in prefrontal grey matter compared with normal controls. No group6hemisphere interaction for grey/white volumes was found (F(2,46) 0.848, P40.43). Antisocial control and normal control groups did not differ from each other in either grey (t0.39, d.f.35, P0.23) or white matter volumes (t0.39, d.f.35, P0.69).
Prefrontal grey/white ratio

Liars had relatively more prefrontal white than grey matter. A multiple analysis of variance (MANOVA) on grey/white ratios showed a significant main effect for group (F(2,46)10.25, P0.0001, Z20.308). Liars had lower prefrontal grey/white ratios (mean1.15, s.d.0.21) than antisocial controls (mean1.56, s.d.0.38, t3.6, P0.001) or normal controls (mean1.63, s.d.0.27, t5.3, P0.0001). Liars had a 35.7% decrease (0.41) in prefrontal grey/ white ratio compared with antisocial controls and a 41.7% decrease (0.48) compared with normal controls (Fig. 2).
Correction for whole brain volumes

It could be argued that group differences in prefrontal volume were an artefact of group differences in whole brain volume. Consequently, the above analyses on prefrontal grey and white matter were repeated using whole-brain corrected volumes. The same results were found. A repeated-measures ANOVA showed no main effect for group (F(2,46)0.971, P0.386) and no group6 hemisphere interaction (F(2,46)0.966, P0.388) but did show a significant group6grey/white interaction (F(2,46) 9.333, P0.0001, Z20.289). A one-way ANOVA on whole-brain corrected grey/ white ratios again showed a significant group effect (F(2,46)10.34, P0.0001). A one-way ANOVA on whole-brain corrected grey matter was non-significant
Fig. 1 Prefrontal grey and white matter volumes in liars (&), normal controls (&) and antisocial controls ( ).
32 2
P R E F RON R ON TA L W H I T E M AT T E R IN IN L I A R S
Fig. 2 ( ).
Prefrontal grey/white matter ratio in liars
(&), normal controls (&) and antisocial controls
(F(2,46)1.73, P0.18) but a one-tailed test on the previously significant reduction in liars compared with normal controls was marginally significant (P0.031). When prefrontal white matter was expressed as a function of whole brain volume, groups again differed significantly (F(2,46)8.031, P0.001). Liars had significantly higher prefrontal white/whole brain ratios (mean0.069, s.d.0.011) compared with both antisocial controls (mean0.054, s.d.0.011, t3.4, P0.002) and normal controls (mean0.054, s.d.0.010, t3.7, P0.001).
Potential demographic, cognitive and antisocial confounding variables

Groups differed significantly with respect to age, verbalperformance IQ discrepancy scores, psychopathy, antisocial personality disorder and conduct disorder, and also showed trends for differences with respect to ethnicity and full-scale IQ. To rule out the effect of age, psychopathy and antisocial personality disorder, these measures were included as covariates in repeatedmeasures ANOVA. The grey/white matter 6group interaction remained significant significant after correcting for age (F(2,45)5.76, P0.006), ethnicity (F(2,45)8.046, P 0.001), verbalperformance IQ discrepancy scores (F(2,45)6.605, P0.003), full-scale IQ (F(2,45)9.503, P0.0001), psychopathy (F(2,45)4.826, P0.01), antisocial personality disorder (F(2,45)7.421, P 0.002) and conduct disorder (F(2,45) 7.372, P0.002).
lie, cheat and manipulate others. Liars had increased prefrontal white matter volumes and reduced grey/white ratios compared with normal controls. The effect size was substantial, with group membership explaining 28.2% of the variance in prefrontal volume. Furthermore, liars were found to have these same differences compared with the antisocial control group. The inclusion of an antisocial control group is viewed as a significant strength since this is rarely included in imaging studies. In addition, the use of a symptom-based approach is felt to be an initial first step in delineating a neurobiological basis of deception (Halligan & David, 2001). Because lying has been argued to be associated not just with antisocial personality but also with several other personality disorders (Ford et al, 1988), the results of the present study may also have wider psychiatric applicability. The result could not be attributed to group differences in age, ethnicity, IQ, head injury or substance misuse/dependence. Furthermore, group differences remained after a strict control for antisocial personality disorder, psychopathy and conduct disorder, again indicating specificity to lying in particular rather than antisocial behaviour in general. Consistent with prior research on pathological liars (Ford et al, 1988), liars had significantly higher verbal relative to performance IQ scores than both control groups, but higher verbal scores could not account for group differences in prefrontal white matter. The results further implicate the prefrontal cortex as an important (but not sole) component in the neural circuitry underlying lying, and provide an initial neurobiological correlate of a deceitful personality.
Neurodevelopmental theory of pathological lying

The most significant finding of this study is the increase of prefrontal white matter and decrease in grey/white ratio in the liar group. Compared with normal controls, the liar group had a 22.2% increase in prefrontal white matter and a 41.7% decrease in grey/white ratio, and compared with antisocial controls they showed a 25.7% white matter increase and a 35.7% decrease in prefrontal grey/white ratio. Children with autism are less capable of lying than normal children (Sodian & Firth, 1992) and, intriguingly, brain neurodevelopmental studies of autism show the
converse pattern of grey/white ratios to that shown by the liar group. When 2- to 3-yearold children with autism reach 9.511 years of age, their white matter increases only 13% compared with 45% in normal children (Carper et al, 2002). Similarly, Courchesne et al (2001) found only a 10% white matter increase in children with autism compared with a 59% increase in normal children from 23 years of age to 1216 years, and an increased cortical grey/white ratio in children with autism compared with normal controls (i.e. the reverse of liars). Although autism is a complex condition, these results on children with autism, combined with the previous fMRI findings on lying in normal controls and our current findings on adult liars, suggest that the prefrontal cortex is centrally involved in the capacity to lie. Why should increased white matter predispose to a deceitful personality? Although a complete explanation inevitably requires more extensive investigation, an initial working hypothesis is that increased prefrontal white matter developmentally provides the individual with the cognitive capacity to lie. From an evolutionary perspective, it is known that deception in primates is correlated with degree of neocortical expansion (Byrne & Corp, 2004). From a neurodevelopmental perspective, brain weight reaches adult values between the ages of 10 and 12 years, with a very significant increase in the absolute volume of white matter (Paus et al, 2001) that exceeds the developmental reduction in grey matter (Sowell et al, 2002). Psychosocial behavioural research also indicates that while young children are poor liars, by 10 years of age they become much more adept at lying (McCann, 1998). Consequently, the neurodevelopmental increase in white matter parallels developmental changes in the ability to lie. It is conceivable therefore that the increased prefrontal white matter found in adult liars predisposes to lying. The relative reduction in prefrontal grey matter relative to white may also predispose to a general antisocial disinhibited tendency which, coupled with increased white matter, results in excessive lying.
DISCUSSION Prefrontal component of lying circuitry

To our knowledge, this study is the first to show a brain abnormality in people who
Clinical conceptualisation of malingering

The results may have implications for research on the clinical concept of malingering (i.e feigning illness to obtain benefits). While biomedical models of malingering
323
YA N G E T A L
have been put forward and debated (Halligan et al, 2003), there appear to be no studies of the biological characteristics (Raine, 2003). 2003). Of the 12 liars in this study, 6 would be classified as malingerers in that they admitted to telling lies to obtain sickness benefits. Comparison of these malingerers with others in the liar group confirms that they too are characterised by both relatively increased prefrontal white matter (66.0 cm3 v. 64.3 cm3 in malingering and nonmalingering liars, respectively) and a reduced prefrontal grey/white ratio (1.09 v. 1.21). Malingering is not currently viewed as a clinical disorder but is included in DSMIV (American Psychiatric Association, 1994) as a V code to mark it as a condition requiring further attention. If the current findings can be replicated and extended to other populations of malingerers, this could have implications for a more clinical conceptualisation of malingering.
CLINICAL IMPLICATIONS
&
Pathological lying is associated with changes in the prefrontal cortex.This also has implications for psychopathy and antisocial personality disorder. The possibility of a clinical conceptualisation of malingering is raised.
& &
We propose a neurodevelopmental theory of pathological lying that also helps explain the onset of proficient lying in children.
LIMITATIONS
& & &
The sample size was modest. Few females were assessed.
We may have underestimated the extent of prefrontal abnormalities in pathological liars because we did not screen the normal control group for moderate levels of lying.
Symptom-based, neurobiological approach to lying

Several neuroscience paradigms are beginning to converge on an initial answer to the elusive question of what is the neurobiological basis to lying. Prior research on normal controls who lie has attempted to identify psychophysiological correlates of lying (Patrick & Iacono, 1991). More recent fMRI research has identified prefrontal activation as a correlate of lying in normal controls. We have reversed the usual research paradigm by using a symptom-based approach to address the question of what characterises individuals who pathologically lie and to provide a provisional answer of excessive prefrontal white matter. Nevertheless, we caution that the neurobiological basis of lying is likely to be complex, involving brain circuits extending well beyond the prefrontal cortex. Future studies are required to examine changes in brain anatomy during the critical neurodevelopmental time period in childhood alongside changes in lying ability to test further our preliminary hypothesis on the link between prefrontal white matter and lying.
YALING YANG, BS, ADRIAN RAINE, DPhil, Department of Psychology, University of Southern California; TODD LENCZ, PhD, Department of Research, Hillside Hospital (North Shore ^ Long Island Jewish Health System); SUSAN BIHRLE, PhD, LORI LACASSE, BA, Department of Psychology, University of Southern California; PATRICKCOLLETTI, MD, Department of Radiology, University of Southern California School of Medicine, USA Correspondence: DrYalingYang, DrYaling Yang, Department of Psychology,University of Southern California, Los Angeles,CA 90 90089^1061,USA.Tel: 089^1061,USA.Tel: +1 213 720 2220; fax: +1 213 74 0 0897; e-mail: yalingy @usc.edu (First received 5 January 2004, final revision 9 November 2004, accepted 17 November 2004)
grants to Y.Y. from the National Institute of Mental Health (Research Scientist Development Award K02 MH01114^01, Independent Scientist Award K02 MH01114^01 and 5 RO3 MH50940^02) and from the Wacker Foundation.
allocating more research time to the study of symptoms. British Journal of Psychiatry, 160, 304^308.
Courchesne, E., Karns, C. M., Davis, H. R., et al (2001) Unusual brain growth patterns in early life in
patients with autistic disorder. An MRI study. Neurology, 57, 245^254.

Elliott, D. S., Ageton, S., Huizinga, D., et al (1983)
REFERENCES
American Psychiatric Association (1994) Diagnostic and Statistical Manual of Mental Disorders (4th edn) (DSM ^ IV).Washington, DC: APA. Bentall, R. P., Jackson, H. F. & Pilgrim, D. (1988)
The Prevalence and Incidence of Delinquent Behavior: 1976^1980 (National Y outh Survey, Report no. 26). Boulder, CO: Behavior Research Institute.
First, M. B., Spitzer, R. L., Gibbon, M., et al (1995a)
Abandoning the concept of schizophrenia: some implications of validity arguments for psychological research into psychotic phenomena. British Journal of Clinical Psychology, 27, 303^324.
Bryden, M. P. (1977) Measuring handedness with
Structured Clinical Interview for DSM ^ IV Axis I Disorders ^ Patient Edition (SCID ^ I/P, Version 2.0). New York: Y ork: Biometrics Research Department, New Y York ork State Psychiatric Institute.
First, M. B., Spitzer, R. L., Gibbon, M., et al (1995b)
questionnaires. Neuropsychologia, 15, 617^624.

Byrne, R. W. & Corp, N. (2004) Neocortex size predicts deception rate in primates. Proceedings of the Royal Society of London. Series B: Biological Sciences, 271, 1693^1699. Carper, R. A., Moses, P., Tigue, Z. D., et al (2002)
Structured Clinical Interview for DSM ^ IV Axis II Personality Disorders (SCID ^ II, Version 2.0). New Y York: ork: Biometrics Research Department, New York Y ork State Psychiatric Institute.
Ford, C.V., King, B. H. & Hollender, M. H. (1988) Lies and liars: psychiatric aspects of prevarication. American Journal of Psychiatry, 145, 554^562. Ganis, G., Kosslyn, S. M., Stose, S., et al (2003)
ACKNOWLEDGEMENTS
We thank Jennifer Bobier, Nicole Diamond, Kevin Ho, Blane Horvath, Shari Mills, KristenTaylor, Jennice Vilhauer and Pauline Yaralian for assistance in data collection and scoring. This study was supported by
Cerebral lobes in autism: early hyperplasia and abnormal age effects. Neuroimage, 16, 1038^1051.
Costello, C. G. (1992) Research on symptoms versus
Neural correlates of different types of deception: an fMRI investigation. Cerebral Cortex, 13, 830^836.
Halligan, P. W. & David, A. S. (2001) Cognitive neuropsychiatry: towards a scientific psychopathology. Nature Reviews Neuroscience, 2, 209^215.
research on syndromes ^ arguments in favour of
324
REPORTS
Washing Away Your Sins: Threatened Morality and Physical Cleansing

Chen-Bo Zhong1* and Katie Liljenquist2 Physical cleansing has been a focal element in religious ceremonies for thousands of years. The prevalence of this practice suggests a psychological association between bodily purity and moral purity. In three studies, we explored what we call the Macbeth effectthat is, a threat to ones moral purity induces the need to cleanse oneself. This effect revealed itself through an increased mental accessibility of cleansingrelated concepts, a greater desire for cleansing products, and a greater likelihood of taking antiseptic wipes. Furthermore, we showed that physical cleansing alleviates the upsetting consequences of unethical behavior and reduces threats to ones moral self-image. Daily hygiene routines such as washing hands, as simple and benign as they might seem, can deliver a powerful antidote to threatened morality, enabling people to truly wash away their sins.
Table 1. Summary of Results. Study 1 measured the effect of recalling ethical versus unethical behavior on the mental accessibility of cleansing-related words. Study 3 explored the effect of recalling ethical versus unethical behavior on the likelihood of choosing antiseptic wipes (over pencils). Study 4 assessed the effect of hand cleansing on the likelihood of engaging in moral compensatory behaviors (i.e., offering help).
Study 1: Average number of cleansingrelated words completed (SEM) Ethical recall (n 0 30) .90 (1.88) Unethical recall (n 0 30) 1.43 (1.77) Study 3: Percentage who chose antiseptic wipes Ethical recall (n 0 16) 33.3% Unethical recall (n 0 16) 66.7% Study 4: Percentage who volunteered to help Cleansed (n 0 22) 40.9% Not cleansed (n 0 23) 73.9%
Department of Organizational Behavior and HR Management, Joseph L. Rotman School of Management, University of Toronto, Toronto, Ontario M5S 3E6, Canada. 2Department of Management and Organizations, Kellogg Graduate School of Management, Northwestern University, Chicago, IL 60208, USA. *To whom correspondence should be addressed. E-mail: chenbo.zhong@rotman.utoronto.ca
www.sciencemag.org
SCIENCE
VOL 313
8 SEPTEMBER 2006
1451
hen we find ourselves in morally compromising situations, how do we deal with the consequences of unethical behavior, given that most if not all of us desire a moral self-image? This paper investigates a basic coping mechanism that has been used by religions for centuries: washing away one_s sins. Physical cleansing, such as bathing or washing hands, is at the core of many religious rituals. Baptism, for instance, is a water purification ritual practiced by Christians, Mandaeanists, and Sikhs. Christians follow the admonition, BArise and be baptized, and wash away your sins[ (1), with faith that through the symbolic cleansing of their bodies they might also achieve a cleansing of conscience. Physical cleansing is also central to Islam; wudu (often translated as Bablution[) is the Muslim act of washing parts of the body in clean water to prepare for worship. Likewise, Hinduism requires great attention to bodily purity (2). Thus, many major religions discipline bodily purity, suggesting that physical cleansing ceremonies can purify the soul. Research on the correspondence between physical and moral purity (3) has speculated that people are predisposed to use categories that are based on bodily experience (such as clean versus dirty) to construct complex social categories (such as moral versus immoral) (4). For example, in English, words such as Bclean[ and Bpure[ describe both physical and moral states (e.g., he has a clean record). Likewise, the Mandarin phrase Ba pair of dirty hands[ refers to a person who steals. The association between bodily and moral purity may be based not only in cognition, but in emotion as well. As an example,
Bdisgust[ represents an emotion that is experienced in both physical and moral domains. Pure disgust was originally a gustatory emotion rooted in evolution to avoid the intake of potentially hazardous food. Over time, it has taken on social and cultural meanings and has expanded to encompass broader categories of aversions including social or moral violations (5, 6). Although the experience of pure disgust devoid of moral connotations can be subjectively and behaviorally differentiated from the experience of disgust with moral connotations (7), they coincide considerably. Specifically, previous research suggests that pure disgust and moral disgust not only lead to similar facial expressions and physiological activation (6) but also recruit partially overlapping brain regions, mainly in the frontal and temporal lobes (7). Given the psychological, physiological, and neurological overlap between physical and moral disgust, physical cleansing acts that mitigate physical disgust might also reduce social or moral disgust, thereby alleviating moral condemnation. Thus, Lady Macbeth_s hope that a little bit of water would clear her of the treacherous murder of King Duncan might not have been a product of literary creativity, but of Shakespeare_s acute understanding of the
human psyche. If physical and moral purity are so psychologically intertwined, Lady Macbeth_s desperate obsession with trying to wash away her bloodied conscience while crying, BOut, damned spot! Out, I say![ (8) may not have been entirely in vain. Given that physical cleansing might function as a surrogate for moral purification, we set out to investigate (i) whether a threat to moral purity activates a need for physical cleansing (i.e., the Macbeth effect) and (ii) whether physical cleansing is actually efficacious in helping people cope with moral threats. We first determined whether a threat to moral purity increases the mental accessibility of cleansing-related words. We asked participants to recall in detail either an ethical or unethical deed from their past and to describe any feelings or emotions they experienced. Then they engaged in a word completion task in which they converted word fragments into meaningful words (9). Of the six word fragments, three (W _ _ H, SH _ _ ER, and S _ _ P) could be completed as cleansing-related words (wash, shower, and soap) or as unrelated words (e.g., wish, shaker, and step). Participants who recalled an unethical deed generated more cleansingrelated words than those who recalled an ethical deed EF(1,58) 0 4.26, P 0 0.04^, suggesting that unethical behavior enhances the accessibility of cleansing-related concepts (Table 1). Was this accessibility the result of an urge to cleanse one_s body when moral integrity was threatened? Study 2 investigated whether an implicit threat to moral purity produces a psychological desire for cleansing, through expressed preferences for cleansing products. Participants were told that we were investigating the relationship between handwriting and personality and were asked to hand-copy a short story written in the first person. The story described either an ethical, selfless deed (helping a co-worker) or an unethical act (sabotaging a co-worker) (9). Participants then rated the desirability of various products from 1 (completely undesirable) to 7 (com-
REPORTS
pletely desirable). Cleansing products included Dove shower soap, Crest toothpaste, Windex cleaner, Lysol disinfectant, and Tide detergent; other products included Post-it Notes, Nantucket Nectars juice, Energizer batteries, Sony CD cases, and Snickers bars. As expected, copying the unethical story increased the desirability of cleansing products as compared to copying the ethical story EF(1,25) 0 6.99, P 0 0.01^, with no differences between conditions for the noncleansing products EF(1,25) 0 0.02, P 0 0.89^ (Fig. 1). We sought to replicate the results of Study 2 using behavioral measures, so our next study examined the likelihood of taking an antiseptic cleansing wipe after recalling an ethical or unethical deed. Participants engaged in the same recall task as in Study 1 and were then offered a free gift and given a choice between an antiseptic wipe and a pencil (verified in a control condition to be equally attractive offerings). Those who recalled an unethical deed were more likely to take the antiseptic wipe (67%) than were those who recalled an ethical deed (33%) ( c2 0 4.57, P 0 0.03) (Table 1). These three studies provided evidence for the Macbeth effect: Exposure to one_s own and even to others_ moral indiscretions poses a moral threat and stimulates a need for physical cleansing. Our final study investigated the efficacy of physical cleansing can it actually wash away moral sins? Physical cleansing may wash away moral sins through symbolic self-completion (10); that is, people are motivated to complete their selfdefinitions (e.g., musicians) when indicators or symbols of this definition are lacking (e.g., skills) by engaging in activities that complete the symbols (e.g., training). Thus, when moral self-definition is at stake, such as when one has indulged in morally questionable activities, one should naturally be motivated to engage in activities that will restore moral integrity. For instance, Tetlock and colleagues (11) have shown that the mere contemplation of violating one_s core values spurs intent to take actions that will restore and protect those values. The restoration or completion of the moral self can be achieved through direct restitution, but it may also be achieved through substitutable symbols or activities that are not directly related (10, 11). Given the demonstrated association between physical cleansing and moral purity, cleansing activities that improve physical cleanliness may also compensate for moral impurity. Thus, we expected that a threat to the moral self would motivate the restoration of moral purity through direct compensatory behaviors (e.g., volunteering to help). If, however, physical cleansing restores the moral self, then individuals should have less need to engage in direct compensatory behaviors after physically cleansing themselves. This is indeed what we found. In Study 4, participants described an unethical deed from their past (the same recall task as in Study 1). Afterwards, they either cleansed their hands with an antiseptic wipe or not. Then they completed a survey regarding their current emotional state (9). After completing the survey, participants were asked if they would volunteer without pay for another research study to help out a desperate graduate student. Presumably, participants who had cleansed their hands before being solicited for help would be less motivated to volunteer because the sanitation wipes had already washed away their moral stains and restored a suitable moral self. As predicted, physical cleansing significantly reduced volunteerism: 74% of those in the not-cleansed condition offered help, whereas only 41% of participants who had a chance to cleanse their hands offered help (c2 0 5.02, P 0 0.025). Thus, the direct compensatory behavior (i.e., volunteering) dropped by almost 50% when participants had a chance to physically cleanse after recalling an unethical behavior (Table 1). Physical cleansing also influenced participants_ emotional state. Based on an exploratory factor analysis (9), the assessed emotions clustered into two categories: moral emotions (i.e., disgust, regret, guilt, shame, embarrassment, and anger; Cronbach Alpha 0 0.90) and nonmoral emotions (i.e., confidence, calm, excitement, and distress; Cronbach Alpha 0 0.65). As expected, participants who cleansed their hands after the unethical recall reported reduced moral emotions (M 0 1.75, SEM 0 0.19) compared with those who did not (M 0 2.23, SEM 0 0.26), F(1,41) 0 2.94, P 0 0.047. Hand washing, however, did not influence nonmoral emotions, F(1,41) 0 0.25, P 0 0.31 (12). These four studies document a psychological association between physical and ethical cleanliness: Threats to moral purity activate a VOL 313 SCIENCE need for physical cleansing, which can assuage moral emotions and reduce direct compensatory behaviors. Although there are surely limits to the absolution afforded by a bar of soap, our findings shed light on Lady Macbeth_s feverish attempts to physically cleanse herself after the murder of King Duncan. If even an implicit threat to one_s moral image can produce a psychological need to engage in cleansing behaviors, it is only natural that those who suffer genuine guilt would be all the more relentless in their attempts to restore a pure conscience. The implications of this research may be substantial. Future studies that specifically address the psychological and behavioral consequences of physical cleanliness will provide valuable insight into regulatory mechanisms that drive ethical decisions. Given the boost to one_s moral self afforded by physical cleansing, how might it influence subsequent behavior? Would adherence to a rigorous hygiene regimen facilitate ethical behavior? Or, would cleansing ironically license unethical behavior? It remains to be seen whether clean hands really do make a pure heart, but our studies indicate that they at least provide a clean conscience after moral trespasses.
References and Notes
1. The Holy Bible (King James Version), Acts 22:16. 2. C. J. Fuller, The Camphor Flame: Popular Hinduism and Society in India (Princeton Univ. Press, Princeton, NJ, 1992). 3. J. Haidt, S. Algoe, in Handbook of Experimental Existential Psychology, J. Greenberg, S. L. Koole, T. Pyszczynski, Eds. (Guilford, New York, 2004), pp. 322335. 4. G. Lakoff, Women, Fire, and Dangerous Things (Univ. of Chicago Press, Chicago, 1987). 5. J. Haidt, P. Rozin, C. McCauley, S. Imada, Psychol. Dev. Soc. 9, 107 (1997). 6. P. Rozin, L. Lowery, R. Ebert, J. Pers. Soc. Psychol. 66, 870 (1994). 7. J. Moll et al., Cogn. Behav. Neurol. 18, 68 (2005). 8. W. Shakespeare, Macbeth, act 5, scene 1, line 38, in Signet Classic Edition, S. Barnet, Ed. (Penguin, London, 1998). 9. Materials and methods are available as supporting material on Science Online. 10. R. A. Wicklund, P. M. Gollwitzer, Basic Appl. Soc. Psychol. 2, 89 (1981). 11. P. E. Tetlock, O. V. Kristel, S. B. Elson, M. C. Green, J. S. Lerner, J. Pers. Soc. Psychol. 78, 853 (2000). 12. We included participants sex as a covariate. Sex itself had no impact on emotional state or offers of help. We used one-tailed tests for the effect of cleansing on emotional state because we had predicted that hand washing would reduce moral emotions but not affect the other emotions. 13. This research was supported by a grant from Dispute Resolution Research Center at the Kellogg School of Management at Northwestern University and an NSF Graduate Fellowship to K.L. We thank A. Galinsky for his insights and support and G. Ku and K. Murnighan for their thoughtful comments on earlier drafts of the paper.
Supporting Online Material

www.sciencemag.org/cgi/content/full/313/5792/1451/DC1 Materials and Methods Tables S1 and S2 References and Notes 1 June 2006; accepted 12 July 2006 10.1126/science.1130726
Fig. 1. Effect of hand-copying an ethical (n 0 16) vs. unethical story (n 0 11) on the desirability of cleansing and noncleansing products on a scale of 1 (low) to 7 (high). Error bars represent standard error.
1452
8 SEPTEMBER 2006
www.sciencemag.org
BEHAVIORAL AND BRAIN SCIENCES (2011) 34, 1 56

doi:10.1017/S0140525X10001354
The evolution and psychology of self-deception

William von Hippel
School of Psychology, University of Queensland, St Lucia, QLD 4072, Australia billvh@psy.uq.edu.au http://www.psy.uq.edu.au/directory/index.html?id 1159
Robert Trivers
Department of Anthropology, Rutgers University, New Brunswick, NJ 08901 trivers@rci.rutgers.edu http://anthro.rutgers.edu/ index.php?option com_content&task view&id 102&Itemid 136
Abstract: In this article we argue that self-deception evolved to facilitate interpersonal deception by allowing people to avoid the cues to conscious deception that might reveal deceptive intent. Self-deception has two additional advantages: It eliminates the costly cognitive load that is typically associated with deceiving, and it can minimize retribution if the deception is discovered. Beyond its role in specic acts of deception, self-deceptive self-enhancement also allows people to display more condence than is warranted, which has a host of social advantages. The question then arises of how the self can be both deceiver and deceived. We propose that this is achieved through dissociations of mental processes, including conscious versus unconscious memories, conscious versus unconscious attitudes, and automatic versus controlled processes. Given the variety of methods for deceiving others, it should come as no surprise that self-deception manifests itself in a number of different psychological processes, and we discuss various types of self-deception. We then discuss the interpersonal versus intrapersonal nature of self-deception before considering the levels of consciousness at which the self can be deceived. Finally, we contrast our evolutionary approach to self-deception with current theories and debates in psychology and consider some of the costs associated with self-deception. Keywords: deception; evolutionary psychology; motivated cognition; self-deception; social psychology
Why would people deceive themselves? What is the mental architecture that enables the same person to be both deceiver and deceived? How does self-deception manifest itself psychologically? And how do these three questions interrelate? In this article we address these issues with an evolutionary account of self-deception, according to which self-deception evolved to facilitate deception of others. First, we dene what we mean by self-deception and describe how self-deception serves the goal of interpersonal deception. We then discuss the non-unitary nature of the mind and how different types of psychological dualism enable the same person to be both deceiver and deceived. Next we describe different varieties of self-deception and the evidence for these different varieties. We then discuss the interpersonal versus intrapersonal nature of self-deception before considering the levels of consciousness at which the self can be deceived. Finally, we contrast our evolutionary approach to self-deception with current theories and debates in psychology. 1. Deception and self-deception There are many ways to deceive other people. An obvious choice is to tell an outright lie, but it is also possible to deceive others by avoiding the truth, obfuscating the truth, exaggerating the truth, or casting doubt on the truth. Just as these processes are useful in deceiving
# Cambridge University Press 2011 0140-525X/11 $40.00
others, they can also be useful in deceiving the self. For example, if I can deceive you by avoiding a critical piece of information, then it stands to reason that I can deceive myself in the same manner. Thus, we consider various types of self-deception, including biased information search strategies, biased interpretive processes, and biased memory processes. What marks all of these varieties of self-deception is that people favor welcome over unwelcome information in a manner that reects their goals or motivations (in this sense, our approach to selfdeception is consistent with Kunda 1990; Mele 1997; Pyszczynski & Greenberg 1987). We also consider classic cases of self-deception such as rationalization and convincing the self that a lie is true.
WILLIAM VON HIPPEL , professor of psychology at the University of Queensland, Australia, conducts research in social cognition and evolutionary psychology. ROBERT TRIVERS , professor of anthropology at Rutgers University, is best known for his theories of reciprocal altruism, parental investment and sexual selection, parental control of offspring sex ratio, and parent offspring conict. Trivers is recipient of the Crafoord Prize for his fundamental analysis of social evolution, conict and cooperation.
von Hippel & Trivers: The evolution and psychology of self-deception Our approach consists of treating self-deception as a variety of different processes that are directly comparable to those involved in interpersonal deception. This approach has the advantage of tying self-deception to processes that have been studied in interpersonal deception. But it has the disadvantage that some behaviors are ambiguous concerning whether they should be classied as selfdeception. This sort of ambiguity arises to the same degree, however, in the study of interpersonal deception. For example, consider a man who has returned home late from work because he stopped to talk to a female colleague and who is confronted by his wife, who wants to know why he is late. If he responds by claiming his boss gave him an extra assignment, then he is clearly being deceptive. If he responds by saying that he stopped to talk to this female colleague, then he is clearly being truthful. But if he changes the subject perhaps by saying that dinner smells great and thereby distracts his wife from her line of inquiry, then it is not clear whether he was deceptive or simply failed to answer the question. There is no way to know if he is deceiving his wife (by avoiding the truth) without knowing his intent in changing the subject or without knowing more about the relationships among him, his wife, and his colleague. In the same manner, when people avoid unpleasant truths in their own lives, it is often impossible to know if self-deception or some other process is at work without knowing more about their motives or situations. Nevertheless, just as interpersonal deception can be studied in the absence of complete condence in all classications, self-deception can also be studied despite the inevitability of such ambiguous cases. Our approach of treating self-deception as informationprocessing biases that give priority to welcome over unwelcome information also differs from classic accounts that hold that the self-deceiving individual must have two separate representations of reality, with truth preferentially stored in the unconscious mind and falsehood in the conscious mind (see Gur & Sackeim 1979). As is clear in our review of information-processing biases in section 5, people can deceive themselves by preventing unwanted information from being encoded in the rst place. This act can be demonstrably motivational, for example, if people stop gathering information when they like the early returns but keep gathering more information when they do not (Ditto & Lopez 1992). The exible nature of this information-gathering bias also reveals that people have some awareness that upcoming information may be inconsistent with what they have already discovered. Thus, biases of this sort are consistent with classic denitions of self-deception that emphasize simultaneous knowing and not-knowing, in the sense that the individual consciously knows the welcome information that has been gathered but also has some awareness that unwelcome information could be around the next corner (we refer to this as potential awareness). In this case, however, true knowing of unwelcome information is precluded because the individual ends the information search before anything unwelcome is ever encountered. Thus, the individual need not have two representations of reality to self-deceive. Rather, people can self-deceive in the same way that they deceive others, by avoiding critical information and thereby not telling (themselves) the whole truth. It is important to note, however, that not all biases in information processing are self-deceptive. For example, 2
BEHAVIORAL AND BRAIN SCIENCES (2011) 34:1
biases can reect cognitive shortcuts, errors, and differential weighting of prior and new information that have nothing to do with motivational concerns (e.g., Chambers & Windschitl 2004). From our perspective, biases in information processing can be considered self-deceptive only when they favor welcome over unwelcome information in a manner that reects the individuals goals. For example, if bolstering a persons self-image makes that individual more willing to search out negative information about himself (Trope & Neter 1994), we have some evidence that prior avoidance of negative information about the self was motivated and self-deceptive. Similarly, if people spend more time learning about the negative than the positive features of someone else only when they expect that person will reject them (Wilson et al. 2004), we again have some evidence that focusing on the negative aspects of another was motivated and self-deceptive. But if biases are impervious to such manipulations or irrelevant to such concerns, then the evidence suggests that they are not self-deceptive. For example, people often rely on peripheral cues such as source expertise when processing arguments about issues that do not concern them (Petty & Cacioppo 1986). Such biases do not reect self-deception, but rather are simply evidence that these individuals are insufciently motivated to study the arguments carefully and are satised to rely on a heuristic that experts tend to be correct. We return to this issue of differentiating self-deceptive from non-selfdeceptive biases in section 5. 2. Self-deception in the co-evolutionary struggle between deceiver and deceived
If (as Dawkins argues) deceit is fundamental in animal communication, then there must be strong selection to spot deception and this ought, in turn, select for a degree of selfdeception, rendering some facts and motives unconscious so as to not betray by the subtle signs of self-knowledge the deception being practiced. Thus, the conventional view that natural selection favors nervous systems which produce ever ve view more accurate images of the world must be a very na of mental evolution. (Trivers 1976/2006, p. 20)
In the struggle to accrue resources, a strategy that has emerged over evolutionary time is deception. For example, people frequently lie to those on whom they are dependent to receive resources that might not otherwise be provided (DePaulo & Kashy 1998; Steinel & De Dreu 2004). Indeed, approximately half of peoples daily deceptions are intended to gain a resource for the self (DePaulo & Kashy 1998). Such deceptive practices instigate a co-evolutionary struggle, because selection favors the deceived evolving new means of detection and the deceiver evolving new means of deception. Self-deception may be an important tool in this co-evolutionary struggle, by allowing deceivers to circumvent detection efforts (Trivers 1976/2006; 1985; 2000; 2009). In the case of deception among humans, there are at least four general categories of cues (beyond fact-nding itself) that people can use to detect deception in others: Signs of nervousness, suppression, cognitive load, and idiosyncratic sources. Nervousness typically results from consideration of the potential costs of detected deception, and thus cues to nervousness can reveal deception (DePaulo et al. 2003). Nevertheless, nervousness is an
von Hippel & Trivers: The evolution and psychology of self-deception imprecise indicator of deception, in part because many situations induce nervousness independent of deception including interrogations themselves as people are often nervous that they will be falsely accused of deception (Bond & Fahey 1987). For this reason, reliance on nervousness to detect deception can lead to a high rate of false positives. Deception can also be detected via physical indicators of the act of suppression. For example, in an effort to control nonverbal signs of nervousness that might reveal deceptive intent, people try to control their face, trunk, and limbs. This act of muscle control leaves telltale cues, such as increased vocal pitch (DePaulo et al. 2003), which then demand countervailing efforts at muscle relaxation. Deception can be detected when control efforts are disjointed or when efforts at relaxation fail to eliminate signs of suppression. Cognitive load results when people must maintain two types of content simultaneously in working memory. In the case of consciously mediated deception, this means true information to be suppressed and false information to be promulgated. Deception can be detected by cues associated with cognitive load, such as pausing and simplied sentence structure (Vrij 2004; Vrij et al. 2006). People who are familiar with each others habits can also detect deception via idiosyncratic signs of nervousness, suppression, and cognitive load, because different individuals reveal their mental states in different ways. Despite the cues that are available for the detection of deception, research on lie detection suggests that people perform poorly in detecting the deceptions of others. For example, in their review of research conducted with professional lie detectors (e.g., police ofcers), Vrij and Mann (2005) found an overall detection rate of 55%, barely above chance levels of 50% in these studies. This detection rate is very similar to the 54% that emerged in a huge meta-analysis that included various types of untrained deceivers and detectors (Bond & DePaulo 2006). Many have concluded from these studies that humans are poor lie detectors, which would suggest that the selection pressure on deceptive abilities must be weak. But this conclusion is premature, given that research on detection of deception relies extensively on four conditions that heavily advantage the deceiver. First, most studies involve deception that is of little consequence to the deceiver. As a result, cues associated with nervousness, cognitive load, and suppression are minimized by the fact that unimportant deceptions do not induce these telltale signs (Mann & Vrij 2006; Vrij 2000; Vrij & Mann 2005). Consistent with this possibility, when police ofcers evaluated videotapes of corroborated truths and lies told in actual criminal interrogations, accuracy increased to 72% (Vrij & Mann 2005), and signs of cognitive load and suppression appeared to differentiate liars from truth tellers (Mann & Vrij 2006). Meta-analysis also supports this conclusion, given that several cues to deception (e.g., vocal pitch) are more pronounced when people are more motivated to deceive (DePaulo et al. 2003). Second, most studies do not allow the deceived to question the deceiver. To enhance experimental control and minimize costs, the deceiver is typically presented on videotape, thereby eliminating the possibility of further interaction. This lack of cross-examination minimizes cognitive load, nervousness, and the need for suppression, as the rehearsed lie can be much easier to deliver than a spontaneous answer to an unexpected question.1 Third, a substantial portion of the literature on detection of deception has been conducted with the goal of nding cues that reliably identify most deceivers, perhaps because of the potential applied value of general cues to deception. Thus, idiosyncratic signs of cognitive load, suppression failure, or nervousness tend to be missed in such designs. Fourth, most studies on detection of deception are conducted on people who are unknown to each other. Our ability to detect deception may be poorly suited for this task, because this design feature eliminates the possibility that people can learn to use idiosyncratic cues that might help them determine when a particular individual is lying (DePaulo 1994; Zuckerman et al. 1984). Consistent with this possibility, detection rates increase when people get to know each other (if they feel emotionally close with each other; Anderson et al. 2002) and when lies are told among close friends (DePaulo & Kashy 1998). Such effects are likely to be multiply mediated, but they suggest that cross-examination and idiosyncratic cues could be important in detecting deception. Consistent with the notion that these four factors reduce detection rates in laboratory studies, diary research suggests that people detect deception at rates that are substantially greater than chance. For example, participants in DePaulo et al.s (1996) diary study reported that 15 23% of their lies were detected. This rate is already high enough to pose a signicant threat given the loss in reputation and goal attainment when one is perceived as a liar but DePaulo et al.s (1996) participants also reported that in an additional 16 23% of the cases they were unsure if their lies were detected. We suggest that these reports are biased downward, such that people underreport the degree to which others have or might have detected their own deceptions. This downward bias in reporting is likely to emerge due to an asymmetry in the degree to which people accuse others of deception. On the one hand, targets of intended deceptions do not always indicate that they doubt the deceiver even when they do (Jang et al. 2002). To gain an information advantage and to maintain harmony, people who detect or suspect deception in others sometimes pretend to believe them. On the other hand, it seems unlikely that people indicate that they doubt deceivers when they actually believe them, in part because accusations of deception are a serious charge. This asymmetry suggests that people may often think that others believe their deception when in fact they do not, but will only rarely think that others doubt their veracity when in fact they have accepted their lies.2 Thus, it might be the case that not only do the brunt of these 30 40% of cases reported in DePaulo et al.s (1996) diary study represent detected deceptions, but a portion of the remaining 60 70% might represent detected deceptions as well. For our purposes, it would also be very useful to know what percentage of these detected deceptions emerged from factnding and third-party information, and what percentage were based on information derived directly from the deceiver (e.g., via cross-examination). At this point this remains a topic for future research. In sum, the literature on deception may have grossly underestimated peoples ability to detect deception through
von Hippel & Trivers: The evolution and psychology of self-deception reliance on studies where (a) the deception is of little or no consequence, (b) the deceived has no opportunity to cross-examine the deceiver, (c) deceiver and deceived are strangers to each other, and (d) there are no repeated interactions between deceiver and deceived. If rates of deception detection are, in fact, substantially higher outside the laboratory than in it, we are led back to the notion of a co-evolutionary struggle between deceiver and deceived. Because successful deception can lead to substantial benets for the deceiver while imposing costs on the deceived (DePaulo 2004), and because unsuccessful deception can lead to substantial costs imposed on the deceiver by people they had endeavored to deceive (Boles et al. 2000; Schweitzer et al. 2006), those who would deceive are in a perennial struggle against those who would not be deceived. We propose that self-deception offers an important tool in this co-evolutionary struggle by allowing the deceiver the opportunity to deceive without cognitive load, conscious suppression, increased nervousness, or idiosyncratic indicators that a deception is being perpetrated. To the degree that people can convince themselves that a deception is true or that their motives are beyond reproach, they are no longer in a position in which they must knowingly deceive others. Thus, the central proposal of our evolutionary approach to selfdeception is that by deceiving themselves, people can better deceive others, because they no longer emit the cues of consciously mediated deception that could reveal their deceptive intent. Corollary 1: Cognitive load reveals deception, but it has other costs as well: demands on working memory reduce performance in challenging domains (Schmader & Johns 2003) and disrupt social functioning (von Hippel & Gonsalkorale 2005). When people are forced to maintain both truth and lies in working memory, they are thus likely to show reduced ability to engage in other tasks and access other opportunities. This cognitive load required to maintain conscious deception is difcult to avoid, because many deceptions require the deceiver to keep fact and ction in mind simultaneously to ensure that the former is hidden and the latter is supported. Self-deception provides a way to avoid this cognitive load. To the degree that deceivers can convince themselves that their deception is indeed true, they are no longer required to maintain the real facts of the case in mind while they focus on promulgating the ction. Rather, by believing the ction that they are expressing to others, they can free their mind to concentrate on other matters. Thus, the rst corollary to our central proposal is that by deceiving themselves, people are able to avoid the cognitive costs of consciously mediated deception. Corollary 2: The best-laid plans often go awry, and lies are no exception to this rule; even careful and well-practiced deceptions can be discovered. This ever-present possibility of detection poses a problem for would-be deceivers, as retribution and exclusion are common responses to detected deceptions (Boles et al. 2000; Schweitzer et al. 2006). This retribution appears to have deep evolutionary roots, given that people react with strong feelings of anger and other negative emotions when they realize they are being deceived (Haselton et al. 2005). Such feelings of anger motivate punishment of the offender, even when pun chter ishment exacts a cost for the punisher (Fehr & Ga 4
2002). It is this anger and subsequent punishment that ensures that detection of deception leads to suffering on the part of deceivers, thereby reducing the likelihood of future deception. Because there are many legitimate reasons people may fail to behave as desired or expected, one solution to the threat of retribution when an apparent deception is uncovered is to co-opt such reasons by pleading ignorance or ineptitude rather than deception. Attribution of intent is critical in determining whether the deceived feels anger and seeks retribution or is willing to forgive (Schweitzer et al. 2006; Stouten et al. 2006), and thus those who accompany their deception of others with deception of the self are better placed if discovered to avoid retribution. By arguing that they had not intentionally deceived, selfdeceivers are more likely than conscious deceivers to avoid retribution. Of course, conscious deceivers can also deceive about their original knowledge and intent, but the cues that are present to give away deception are also present to give away deception about deception. Thus, the second corollary to our central proposal is that by deceiving themselves, people can reduce retribution if their deception of others is discovered. Implication: Our evolutionary hypothesis about the role of self-deception in service of interpersonal deception hinges primarily on the argument laid out earlier that despite what the research literature might appear to show, people are actually quite good at detecting deception. This possibility is central to our hypothesis regarding a co-evolutionary struggle and the subsequent origins of self-deception. Just as importantly, this possibility is also central to a proper understanding of deception and its place in social life. Thus, a critical issue for future research in deception and self-deception would be to establish how successful people are at detecting important deceptions occurring in naturalistic settings that allow those who are being deceived to gather further information as they see t. Future research should also consider whether those who deceive in such settings gain an accurate sense of when their deceptions were successful versus unsuccessful, because there are likely to be situations that lead to clear feedback and others that do not. Although studies such as these are likely to be challenging to design and execute, methodologies involving simultaneous experience sampling within groups of closely interacting individuals may be a promising approach. But whatever the hurdles, it should be clear that research on deception must move beyond the closely controlled studies of the sort described earlier if it hopes to answer these fundamental questions about deception and its detection. 3. Self-deception in service of social advancement Self-deception can also facilitate the deception of others in a more general sense, in that it can help us convince others that we are better (e.g., more moral, stronger, smarter) than we really are. Thus, the benets of self-deception go beyond convincing others of specic lies, as self-deception can also help us accrue the more general social advantages of self-ination or self-enhancement. People are impressed by condence in others (Price & Stone 2004; Slovenko 1999). Condence plays a role in determining whom people choose as leaders (Conger &
von Hippel & Trivers: The evolution and psychology of self-deception Kanungo 1987; Shamir et al. 1993), romantic partners (Buss 2009; Miller 2000; Schmitt & Buss 1996), and providers of various social and material services (Brown et al. 1998; de Jong et al. 2006; Westbrook 1980). Condence is also a determinant of social inuence; condent people are believed more, and their advice is more likely to be followed than people who lack in condence (Penrod & Cutler 1995; Zarnoth & Sniezek 1997). To the degree that people can bolster their image of themselves to themselves and enhance their self-condence, they thereby increase the chances that they will be able to inuence others and will be chosen for socially important roles. For this reason, self-enhancement should be ubiquitous and people should believe their own self-enhancing stories. Evidence supports both of these possibilities. With regard to ubiquity, self-enhancing biases are evident in a wide variety of domains and strategies among a wide variety of peoples (for a review, see Alicke & Sedikides 2009). Even East Asians, who value humility and harmony over individualistic self-aggrandizement, show self-enhancement in their claims of the superiority of their collectivist qualities (Sedikides et al. 2003; 2005). Furthermore, like Westerners, East Asians who are lower in depression and stress show this self-enhancement to a greater degree than those who have more psychological problems (Gaertner et al. 2008). People not only self-enhance the world over, but the average person appears to be convinced that he or she is better than average (Alicke & Sedikides 2009). Most of the research on self-enhancement does not allow one to assess whether these aggrandizing tales are self-deceptive or only intended to be other-deceptive, but some of the variables used in this research support the idea that people believe their own self-enhancing stories. For example, in a pair of clever experiments Epley and Whitchurch (2008) photographed participants and then morphed these photographs to varying degrees with highly attractive or highly unattractive photos of samesex individuals. Epley and Whitchurch then presented participants with these morphed or unaltered photos of themselves under different circumstances. In one experiment participants were asked to identify their actual photo in an array of actual and morphed photographs of themselves. Participants were more likely to choose their photo morphed 10% with the more attractive image than either their actual photo or their photo morphed with the unattractive image. This effect emerged to a similar degree with a photo of a close friend, but it did not emerge with a photo of a relative stranger. Because people often perceive their close friends in an overly positive light (Kenny & Kashy 1994), these ndings suggest that people do not have a general bias to perceive people as more attractive than they really are, but rather a specic bias with regard to themselves and close others. In a second study, participants were presented with an array of photos of other individuals, among which was a single photo of themselves (either their actual photo or a photo morphed 20% with the attractive or unattractive image). Epley and Whitchurch (2008) found that people were able to locate photographs of themselves most rapidly if they were morphed with an attractive photo, at an intermediate speed if they were not morphed, and most slowly if they were morphed with an unattractive photo. These ndings suggest that the enhanced photo most closely matches how people see themselves in their minds eye, suggesting that they are deceiving themselves about their own attractiveness. Were they aware of this inaccuracy, they would be unlikely to claim the attractive photo to an experimenter who has the truth at her disposal, and they would be unlikely to locate their more attractive self more rapidly than their actual self. Thus far, the evidence from Epley and Whitchurch (2008) suggests that self-enhancement biases appear to be inaccuracies that are believed by the self. But are they really self-deceptive? That is, do we have any evidence that the participants in their experiments have unconscious knowledge of what they really look like, or that they have prevented themselves from gaining accurate self-knowledge via motivated processes? At this point we do not. But it is the case that the magnitude of the self-enhancement effect documented by Epley and Whitchurch (2008) was correlated with participants implicit self-esteem (as measured by a priming procedure [Spalding & Hardin 1999] and the name letter effect [Nuttin 1985]). This correlation suggests that people systematically distort their self-image not as a function of how much information they have about themselves, but rather as a function of their automatic positivity toward themselves. Nevertheless, the utility of this measure as an indicator of self-deception would be enhanced if it were shown to relate to peoples goals or their biases in encoding of information about their appearance (e.g., differential gazing at attering vs. unattering images of themselves). Along with their self-enhancing stories, people also derogate others. Indeed, self-enhancement and otherderogation are opposite sides of the same coin, as people arrive at their self-image via social comparison (Festinger 1954). For this reason all self-evaluations are relative, and the self can be elevated above others either via selfenhancement or other-derogation. Like self-enhancement, derogation of others can also be an offensive tool used in the service of social advancement, as people often derogate their rivals when they are trying to impress (Buss & Dedden 1990; Schmitt & Buss 2001). As with self-enhancement, derogation of others appears to be both ubiquitous and believed by the self. Some of the best examples of self-deceptive derogation of others can be found in the research of Fein and Spencer. In one of their studies (Fein & Spencer 1997), non-Jewish participants were told either that they did well or poorly on an IQ test. They were then given a chance to watch a videotape of another student being interviewed for a job, and this individual was portrayed as either Jewish or Christian via her ostensible surname and a photo showing her wearing a Jewish star or a cross. When participants watched the individual they believed was Christian, their ratings of this student were not inuenced by whether they had ostensibly failed the IQ test. In contrast, when participants thought they were watching a Jewish student, their ratings were inuenced by how well they had done on the IQ test. Those who thought they had done well showed no sign of prejudice. Those who thought they had done poorly, however, rated the Jewish student negatively on a composite of personality traits. Furthermore, these individuals also showed a rebound in self-esteem compared to people who thought they had failed but watched the Christian woman. These ndings suggest that people
von Hippel & Trivers: The evolution and psychology of self-deception responded to failing an IQ test by denigrating Jews. This denigration appears to have been believed by these individuals, because the more they derogated the Jewish student, the better they felt about themselves. Furthermore, people appeared to have objective information about this person at their disposal, given that they did not rate the Jewish person negatively when portrayed as Christian or when they had not ostensibly failed the test. The research of Fein and Spencer (1997) suggests that people do not derogate others only to make themselves look better in other peoples eyes. Rather, they appear to be self-deceptively making themselves look better in their own eyes as well. This interpretation is corroborated by the ndings of their second experiment, in which participants who had reected on their important values (a process that afrms a sense of self-worth: see sect. 5) evaluated the Jewish student the same as they rated the Christian student even after failure. This sort of threat-induced derogation of others documented by Fein and Spencer (1997) can also take place outside of awareness (Spencer et al. 1998), a nding that further suggests that people truly believe their negative impressions of others when they are led to feel bad about themselves. Because downward social comparison is self-enhancing (Wills 1981), these ndings can be considered the ip-side of the bias documented by Epley and Whitchurch (2008). Thus, our second proposal is that by deceiving themselves about their own positive qualities and the negative qualities of others, people are able to display greater condence than they might otherwise feel, thereby enabling them to advance socially and materially. 4. Self-deception and the non-unitary mind There are a variety of dissociations between seemingly continuous mental processes that ensure that the mental processes that are the target of self-deception do not have access to the same information as the mental processes deceiving the self. For our purposes, these dissociations can be divided into three (overlapping) types: implicit versus explicit memory, implicit versus explicit attitudes, and automatic versus controlled processes. These mental dualisms do not themselves involve selfdeception, but each of them plays an important role in enabling self-deception. By causing neurologically intact individuals to split some aspects of their self off from others, these dissociations ensure that people have limited conscious access to the contents of their own mind and to the motives that drive their behavior (cf. Nisbett & Wilson 1977). In this manner the mind circumvents the seeming paradox of being both deceiver and deceived.
4.1. Explicit versus implicit memory
Substantial research now indicates that people maintain at least two types of information in memory. People retain information that they can consciously recollect (assessed via explicit measures such as recall) and information for which they have no conscious recollection (assessed via implicit measures such as degraded word identication). This dissociation between types of memories has the potential to serve as a basis for an adaptive form of self6
deception, as the conscious memories could be those that are consistent with the ction that the person wishes to promulgate, whereas the unconscious memories could be the facts as originally encountered. By maintaining accurate information in unconscious memory, the individual would retain the ability to behave in accordance with the truth, as implicitly assessed memories have been shown to inuence a variety of behaviors (Coates et al. 2006; Kolers 1976; Lee 2002; Seamon et al. 1995). How might memories selectively retain or be blocked access to consciousness as a function of their utility in self-deception? Although there are probably numerous ways, one possibility is that deception might begin to replace truth in conscious memory simply through the act of reporting the misinformation (via retrieval-induced forgetting; MacLeod & Saunders 2008). Additionally, because the consequences of being caught lying can be severe, and because lies require important deviations from or omissions to actual events, those who practice to deceive often take their practice literally and (at least mentally) rehearse their lies (Vrij 2000). Rehearsal of misinformation can make the source of this information even more difcult to ascertain, with the result that people often come to believe that the inaccurate depiction of events is veridical (Ceci et al. 1994; Zaragoza & Mitchell 1996). According to this possibility, people initiate a deception knowing that they are promulgating a falsehood, but beginning with their initial transmission of the misinformation, they start to unknowingly convince themselves of the truth of their own lie. This process of self-inducing false memories can then be enhanced by other factors that typically accompany intentional deception of others. For example, to the degree that deceivers create an elaborate and concrete image of the lie they are telling, they may unintentionally exacerbate their self-deception, because vivid imagining makes false memories more difcult to distinguish from accurate ones (Gonsalves et al. 2004; Slusher & Anderson 1987). Additionally, social sharing of information can lead to selective forgetting of information that is not discussed (Coman et al. 2009; Cuc et al. 2007) and social conrmation of inaccurate information can exacerbate the false memory effect (Zaragoza et al. 2001). These social effects raise the possibility that when people collaborate in their efforts to deceive others, they might also increase the likelihood that they deceive themselves. Thus, one consequence of retrieving, rehearsing, and telling lies is that people may eventually recollect those lies as if they actually happened, while still maintaining the accurate sequence of events in a less consciously accessible form in memory (Chrobak & Zaragoza 2008; Drivdahl et al. 2009; McCloskey & Zaragoza 1985). Therefore, our third proposal is that the dissociation between conscious and unconscious memories combines with retrieval-induced forgetting and difculties distinguishing false memories to enable self-deception by facilitating the presence of deceptive information in conscious memory while retaining accurate information in unconscious memory.
4.2. Explicit versus implicit attitudes
Just as memories can be relatively inaccessible to consciousness, so too can attitudes. And just as inaccessible memories can inuence behaviors, so too can attitudes
von Hippel & Trivers: The evolution and psychology of self-deception that are relatively inaccessible to consciousness (Greenwald et al. 2009; Nock et al., 2010). Attitudes that are difcult to access consciously can be measured with implicit procedures, for example, via reaction time tests such as the Implicit Association Test (IAT; Greenwald et al. 1998). The study of implicit attitudes is not as well developed as the parallel research program in the domain of memory, but here again the evidence indicates that people maintain two different types of attitudinal information (Fazio & Olson 2003; Wilson et al. 2000). Sometimes implicit and explicit attitudes overlap substantially, and sometimes they differ dramatically (Nosek et al. 2007). Although dissociations between implicit and explicit attitudes tend to be more common in cases where implicit attitudes are socially undesirable, such dissociations emerge across a variety of domains (Hofmann et al. 2005). As is the case with memory, the coexistence of different implicit and explicit attitudes provides fertile ground for self-deception. An example of how dual attitudes play a role in selfdeception can be found in research that shows that socially undesirable implicit attitudes drive behavior when attributions for behavior are ambiguous, whereas socially desirable explicit attitudes drive behavior when attributions are clear. In a demonstration of this effect, Son Hing et al. (2008) found that white Canadians low in both implicit and explicit prejudice toward Asians did not discriminate between white and Asian job applicants who were equally qualied for the job for which they had applied, regardless of the clarity of their qualications. When the applicants qualications were ambiguous, however, people low in explicit but high in implicit prejudice were more likely to hire the white than the Asian job applicant. In such a manner, low explicit/high implicit prejudice individuals are able to hide their prejudiced beliefs and their discriminatory behavior from self and other. Thus, our fourth proposal is that the dissociation between implicit and explicit attitudes lends itself to selfdeception by enabling people to express socially desirable attitudes while nevertheless acting upon relatively inaccessible socially undesirable attitudes when they can maintain plausible deniability.
4.3. Automatic versus controlled processes
to the environment can be executed without awareness (Lakin et al. 2008), people can engage in socially undesirable goal-directed behavior but remain oblivious to that fact. For example, a student whose parents want her to be a physician but who wants to be an artist herself might follow her conscious goal to major in biology and attend medical school, but her unconscious goal might lead her not to study sufciently. As a consequence, she could nd herself unable to attend medical school and left with no choice but to fall back on her artistic talents to nd gainful employment. By deceiving herself (and consequently others) about the unconscious motives underlying her failure, the student can gain her parents sympathy rather than their disapproval. These ndings and possibilities suggest our fth proposal, that the dissociation between automatic and controlled processes facilitates self-deception by enabling the pursuit of deceptive goals via controlled processing while retaining the automatic expression of actual but hidden goals. 5. Varieties of self-deception We begin our review of the evidence for self-deception by describing biases that represent the front end of the information-processing sequence (i.e., information gathering and selective attention). Self-deception at this stage of information processing is akin to failure to tell the self the whole truth. We then discuss varieties of selfdeception that represent the middle of the informationprocessing stream (e.g., memory processes). These are processes that involve obfuscating the truth. We then conclude this discussion with types of self-deception that involve convincing the self that a falsehood is true. An important question that must be addressed with regard to all of these instances of biased processing is whether they reect self-deception or some other source of bias. Two manipulations have proven useful in addressing this issue: self-afrmation (Sherman & Cohen 2006; Steele 1988) and cognitive load (e.g., Valdesolo & DeSteno 2008). When people are self-afrmed, they are typically reminded of their important values (e.g., their artistic, humanist, or scientic orientation) or prior positive behaviors (e.g., their kindness to others). By reecting on their important values or past positive behaviors, people are reminded that they are moral and efcacious individuals, thereby afrming their self-worth. A cornerstone of selfafrmation theory is the idea that specic attacks on ones abilities or morals such as failure on a test do not need to be dealt with directly, but rather can be addressed at a more general level by restoring or reafrming a global sense of self-worth (Steele 1988). Thus, self-afrmation makes people less motivated to defend themselves against a specic attack, as their sense of self-worth is assured despite the threat posed by the attack. Self-afrmation manipulations can be used to assess whether information-processing biases are self-deceptive. If a particular bias represents unmotivated error, then it will be unaffected by self-afrmation. For example, if people rely on a heuristic to solve a problem for which they do not have the knowledge or interest to use the appropriate algorithm, self-afrmation should not reduce their use of this heuristic. In contrast, if the bias represents a self-deceptive process that favors welcome over
Controlled processes involve conscious effort, awareness, and intention and can be stopped at will, whereas automatic processes take place in the absence of effort, awareness, and intention and typically run to completion once initiated (Bargh 1994). It is now apparent that automatic and controlled processes make independent contributions to numerous social tasks, and that these processes can be dissociated (Chaiken & Trope 1999). Research on automatic goal activation has demonstrated that a wide variety of goal-directed behaviors that appear to be consciously directed can take place automatically, often with the goal itself outside of conscious awareness (Chartrand et al. 2008). Because these unconscious goals can sometimes run counter to peoples consciously stated goals, people can consciously hold goals that are socially desirable and supported by their peers or family while simultaneously holding unconscious alternative goals that are socially undesirable or otherwise unacceptable to peers or family (Chartrand et al. 2007; Fitzsimmons & Anderson, in press). Furthermore, because automatically activated responses
von Hippel & Trivers: The evolution and psychology of self-deception unwelcome information, then this bias should be eliminated or attenuated by self-afrmation (see Correll et al. 2004; Sherman et al. 2009). Such an effect for self-afrmation not only provides evidence for the role of motivation in the information-processing bias but also indicates that the person had the potential to process the information in a less biased or unbiased fashion. In this manner, selfafrmation manipulations can test our self-deception criterion that the person is aware of the welcome information but at the same time also had potential awareness of the unwelcome information. It should be noted, however, that our perspective on the interpersonal purpose of self-deception suggests that although self-afrmation should attenuate or eliminate many self-deceptive biases, it should only do so when the self-deceptive bias serves the general goal of enhancing the self. That is, when self-deception is in service of social advancement via self-enhancement, self-afrmation should attenuate or eliminate the self-deception because the afrmation itself satises the enhancement goal. In contrast, when people self-deceive to facilitate their deception of others on a particular issue, self-afrmation should have no effect. Here the goal of the self-deception is not to make the self seem more efcacious or moral, but rather to convince another individual of a specic ction that the self-deceiver wishes to promulgate. Self-afrmation is irrelevant to this goal. Unfortunately, this distinction between general self-enhancement and specic deception is not always easily made. Nevertheless, when deception concerns a specic topic, self-afrmation should not inuence self-deceptive practices unless the afrmation makes people decide that the deception itself is unnecessary or unimportant (e.g., if reminders of their self-worth make people less motivated to deceive others about their errors or poor behavior). A second method for evaluating whether an information-processing bias is self-deceptive involves manipulation of cognitive load. Such manipulations typically require people to keep some information in memory (e.g., an eight-digit number) while they are engaged in the primary task of the experiment (e.g., forming an impression). Although manipulations of cognitive load do not address the motivational issues that underlie selfdeception, they do address the issues of cost and potential awareness. If cognitive load leads to the elimination or attenuation of a particular bias, then the evidence suggests that the biased processing was actually more effortful than unbiased processing. Such a nding suggests that the individual was potentially aware of the unbiased information but was able to avoid it by engaging in the type of mental gymnastics described in the remainder of this section.
5.1. Biased information search 5.1.1. Amount of searching. There are many situations in daily life in which people avoid further information search because they may encounter information that is incompatible with their goals or desires. For example, on the trivial end of the continuum, some people avoid checking alternative products after they have made a purchase that cannot be undone (Olson & Zanna 1979). On the more important end of the continuum, some people avoid AIDS testing out of concern that they might get a
result that they do not want to hear, particularly if they believe the disease is untreatable (Dawson et al. 2006; Lerman et al. 2002). This sort of self-deceptive information avoidance can be seen in the aphorism, What I dont know cant hurt me. Although a moments reection reveals the fallacy of this statement, it is nonetheless psychologically compelling. Similar sorts of biased information search can be seen in laboratory studies. Perhaps the clearest examples can be found in research by Ditto and colleagues (e.g., Ditto & Lopez 1992; Ditto et al. 2003), in which people are confronted with the possibility that they might have a proclivity for a pancreatic disorder. In these studies people expose a test strip to their saliva and are then led to believe that color change is an indicator of either a positive or negative health prognosis. Ditto and Lopez (1992) found that when people are led to believe that color change is a good thing, they wait more than 60% longer for the test strip to change color than when they believe color change is a bad thing. Studies such as these suggest that information search can be biased in the amount of information gathered even when people are unsure what they will encounter next (see also Josephs et al. 1992). Thus, it appears that people sometimes do not tell themselves the whole truth if a partial truth appears likely to be preferable. We are aware of no experiments that have examined the effect of self-afrmation or cognitive load on amount of information gathered, but we would predict that both manipulations would lead to an elimination or attenuation of the effect documented by Ditto and Lopez (1992).
5.1.2. Selective searching. Information search can also
be biased in the type of information gathered. Although one never knows for sure what lies around the next corner, some corners are more likely to yield welcome information than others. Thus, politically liberal people might choose the New York Times as their information source, whereas politically conservative individuals might choose Fox News (Frey 1986). In such a manner, people can be relatively condent that the brunt of the information they gather will be consistent with their worldview, even if they do not know what tomorrows headlines will bring. Laboratory studies have examined this sort of biased information search, in part by assessing the conditions under which people are interested in learning negative information about themselves. One conclusion from this research is that the better people feel about themselves, the more willing they are to face criticism.3 For example, Trope and Neter (1994) told participants that they were going to take a social sensitivity test and asked whether they would like feedback on their assets or liabilities. When participants had just ostensibly failed an unrelated spatial abilities test, or had not taken the test, they showed a slight preference for feedback on their assets. In contrast, when bolstered by the experience of ostensibly having performed very well on the spatial abilities test, participants were more interested in learning about their liabilities, presumably in service of self-improvement. In a related vein, Armitage et al. (2008) demonstrated that smokers were more likely to take an antismoking leaet if they had been self-afrmed by reecting on their prior acts of kindness (see also Harris et al. 2007). These data
von Hippel & Trivers: The evolution and psychology of self-deception implicate potential awareness of unwanted information by showing that people tend to search for welcome information but are capable of searching for unwelcome information when their self-enhancement goals have been met. Thus, it appears that people are often able to avoid telling themselves the whole truth by searching out those bits of truth that they want to hear, but they are also willing to n face uncomfortable truths when feeling secure (Albarrac & Mitchell 2004; Kumashiro & Sedikides 2005).
5.1.3. Selective attention. When information is percep-
tually available and need not be actively discovered, people can still bias their encoding by selectively attending to aspects of the available information that they would prefer to be true. For example, if a person is at a dinner party where one conversation concerns the dangers of smoking and the other concerns the dangers of alcohol, she can choose to attend to one conversation or the other and may do so selectively if she is a smoker or a drinker. In such a case she would likely be aware of the general tone of the information she is choosing not to gather, but by not attending to one of the conversations, she could avoid learning details that she may not want to know. This sort of effect has been documented in a variety of different types of experiments. For example, in a study of proactive coping, Wilson et al. (2004) convinced participants that they might be chosen or were highly unlikely to be chosen for a hypothetical date. When participants believed they might be chosen, they spent slightly more time looking at positive than negative information about their potential partner. In contrast, when they believed that they were highly unlikely to be chosen, they spent more time looking at negative information about their potential partner. Thus, when people faced almost certain disappointment, they directed their attention to information that would make their upcoming rejection more palatable. Although measures such as reading time provide a good indicator of the amount of information processing, attention can be assessed more directly. Eye-tracking studies provide some of the clearest evidence of where people direct their attention, and such studies have also shown that people are often strategic in their attentional decisions (Isaacowitz 2006). For example, older adults look toward positive stimuli and away from negative stimuli when in a bad mood (Isaacowitz et al. 2008). This attentional bias clearly implicates potential awareness, as some encoding of the negative must take place for preferential attention to be directed toward the positive. This effect did not emerge among younger adults, suggesting that older adults are more likely than younger adults to rely on selective attention for mood repair. In a case such as this, it appears that older adults sacrice informational content in service of emotional goals. This strategy might be sensible for older adults who have greater immune challenges than their younger counterparts and thus reap greater benets from maintaining happiness (see sect. 6). As with the strategy of ending information search early, selective attention can allow people to avoid telling themselves the whole truth.
5.2. Biased interpretation
in which such information is nevertheless faithfully encoded. Under such circumstances, unwelcome information can still be dismissed through biased interpretation of attitude-consistent and attitude-inconsistent information. In the classic study of this phenomenon (Lord et al. 1979), people who were preselected for their strong attitudes on both sides of the capital punishment debate were exposed to a mixed bag of information about the efcacy of capital punishment. For example, some of the data with which they were presented suggested that capital punishment was an effective crime deterrent, whereas other data suggested that it was not. Given that the ndings were new to participants, logic would suggest that they would coalesce at least to some degree in their attitudes. In contrast, people ended the experiment more polarized than they began it. Lord et al. (1979) discovered that this attitude polarization was a product of biased interpretation of the data. People who were in favor of capital punishment tended to accept the data as sound that supported capital punishment but reject the data as awed that opposed capital punishment. Those who were against capital punishment showed the opposite pattern (see also Dawson et al. 2002). This selective skepticism appears to be self-deceptive, as it is attenuated or eliminated by self-afrmation (Cohen et al. 2000; Reed & Aspinwall 1998) and cognitive load (Ditto et al. 1998). These ndings suggest that people have potential awareness of an unbiased appraisal, given that they appear to be relying on their motivational and mental resources to be differentially skeptical. Thus, selective skepticism appears to be a form of self-deception rather than simply an objective devaluation of new information to the degree that it is inconsistent with a large body of prior experience (see also, Westen et al. 2006). As a consequence of this selective skepticism, people are able to encounter a mixed bag of evidence but nevertheless walk away with their original beliefs intact and potentially even strengthened. Because they are unaware that a person with a contrary position would show the opposite pattern of acceptance and rejection, they are able to convince themselves that the data support their viewpoint. Thus, it seems that by relying on their considerable powers of skepticism only when information is uncongenial, people are able to prevent themselves from learning the whole truth.
5.3. Misremembering
Despite the strategies just described for avoiding unwelcome information, there remain a variety of circumstances
Even if people attend to unwanted information, and even if they accept it at the time of encoding, this does not guarantee that they will be able to retrieve it later. Rather, information that is inconsistent with their preferences may simply be forgotten or misremembered later as preference-consistent or neutral. Thus, a person might have great memory for the details of his victory in the championship tennis match but very poor memory for the time he lost badly. Indeed, this latter memory might also be distorted to implicate his doubles partner or the unusual talents of his opponent. In section 4, we discussed theoretical mechanisms by which deceptive information might remain in consciousness and truthful information might be relegated to unconsciousness, but what evidence is there that memory is selective in this sort of fashion? Unfortunately, we know of no evidence showing how
von Hippel & Trivers: The evolution and psychology of self-deception intending to deceive others can lead to self-deception as described in section 4, but there is evidence that other motivational sources can lead to selective forgetting processes. First, when people put effort into self-improvement, but the improvement does not materialize, they can manufacture the gains they wish they had made by misremembering how they used to be. For example, Conway and Ross (1984) demonstrated that after taking a study skills class, people misremembered their prior study skills as lower than they rated them originally, thereby supporting their belief that their skills have improved. They then later misremembered their subsequent course performance as better than it was to maintain the ction of improvement. Through processes such as these, people are able to purge their memories of inconvenient truths, thereby preventing themselves from knowing the whole truth, even if they accurately encoded it in the rst instance. Health information can be similarly distorted in memory (Croyle et al. 2006). In Croyle et al.s research, participants were given cholesterol screening, and one, three, or six months later tested for their memory of their results. Respondents showed highly accurate memory of their risk category (89% correctly recalled this information), and this accuracy did not decay over six months. Nevertheless, even in the context of this apparently easy memory task, respondents were more than twice as likely to recall their cholesterol as being lower rather than higher than it really was. This sort of memory bias can also be seen in recollection of daily experiences, whereby people have better recall of their own good than bad behavior but do not show this bias in their recall of the behaviors of others (DArgembeau & Van der Linden 2008). This self-enhancing recall bias is also eliminated by information that bolsters peoples selfimage (in this case, doing well on a test; Green et al. 2008), suggesting that people have potential awareness of both positive and negative information about the self. Thus, peoples memories appear to be self-enhancing, sometimes containing information that is biased to be consistent with preferences and sometimes just failing to contain the whole truth.
5.4. Rationalization
Even if ones prior misdeeds are accurately recalled by self and others, it is still possible to avoid telling oneself the whole truth by reconstructing or rationalizing the motives behind the original behavior to make it more socially acceptable. For example, after eating a second helping of cake that leaves none for those who have not yet had dessert, a person could explain that he had not noticed that there was no other cake, or that he thought more cakes were available elsewhere. Here it is not memory of the misdeed that is critical, but interpretation of the motivation that underlies that deed. Again, laboratory evidence supports this sort of rationalization process. For example, von Hippel et al. (2005) demonstrated that when cheating could be cast as unintentional, people who showed a self-serving bias in another domain were more likely to cheat, but when cheating was clearly intentional, self-serving individuals were no more likely to cheat than others. These data suggest that some types of self-serving biases involve rationalization processes that are also common to some types of cheating. 10
Indeed, people also cheat more when they are told that free will is just an illusion (Vohs & Schooler 2007), suggesting that they rationalize their cheating in these circumstances as caused by life situations rather than their own internal qualities. More direct evidence for this sort of rationalization can be found in the hypocrisy research of Valdesolo and DeSteno (2008). In their study participants were given the opportunity to (a) choose whether to give themselves or another individual an onerous task or (b) randomly assign the onerous task to self versus other. When given this opportunity, nearly all participants chose to give the onerous task to the other participant rather than rely on random assignment. Observers were not asked to make the choice themselves but rather watched a confederate make this same self-serving choice. When asked how fair the choice was, observers rated the act of choosing rather than relying on random assignment as less fair than it was rated by those who had actually made this choice. This hypocrisy in self-ratings shown by those who chose to assign the onerous task to another was eliminated by cognitive load, suggesting that participants have potential awareness of the unfairness underlying their judgments. Research on misattribution reveals evidence of similar rationalization processes. A classic example can be found in Snyder et al.s (1979) research on avoidance of disabled people. In Snyder et al.s experiment, participants chose a seat from two options one next to a person who was disabled and one next to a person who was not disabled. In front of each empty seat there was a television, and the two televisions sometimes showed the same program and sometimes a different program. Snyder et al. (1979) found that when the televisions were showing the same program, the majority of participants sat next to the disabled person, presumably to demonstrate to self and other that they were not prejudiced against disabled people. In contrast, when the televisions were showing different programs, the majority sat away from the disabled person. These data show that people only avoided the disabled person when they could rationalize their behavior as caused by external factors. Similar effects have been documented in differential helping rates for African versus white Americans, with a meta-analysis showing that whites are less likely to help African Americans than fellow whites, but only when there are actual situational impediments to helping, such as distance or risk (Saucier et al. 2005). When people cannot attribute their non-helping to such situational factors, they apparently feel compelled to help African Americans at equal rates to whites. In cases such as these, people are not denying or misremembering their cheating, self-serving choices, avoidance, or lack of helping. Rather, they are denying the socially undesirable motives that appear to underlie their behaviors by rationalizing their actions as the product of external forces.
5.5. Convincing the self that a lie is true
The classic form of self-deception is convincing oneself that a lie is true. This sort of self-deception can be difcult to verify, as it is difcult to know if the person believes the lie that they are telling others, given that situations that motivate lying to the self typically motivate lying to
von Hippel & Trivers: The evolution and psychology of self-deception others. Nevertheless, there are examples of experiments in which this sort of process has been shown. Most of these examples rely on research paradigms in which the experimenter knows the truth, so the participant has little or nothing to gain interpersonally by lying and often has much to lose, given that lying makes the individual look vain, foolish, or deceptive.
5.5.1. Self-deception accompanied by neurological damage. A clear example of this sort of self-deception
can be found in the split-brain research of Gazzaniga and colleagues, which relies on participants who have had their corpus callosum severed and are thereby unable to communicate directly between the two hemispheres. Gazzaniga (1997) described a series of experiments with split-brain patients that suggest that the left hemisphere confabulates when necessary to explain ones own behavior. In one such study a split-brain patient was presented with an apparatus that displayed a chicken foot to the left hemisphere (via the right visual hemield) and a snowy scene to the right hemisphere (via the left visual hemield). The participant was then asked to point with each hand at the picture that most closely matched what was seen. The left hemisphere directed the right hand to point at a chicken head, and the right hemisphere directed the left hand to point at a shovel. When Gazzaniga asked the participant why his left hand was pointing at the shovel, he created a quandary. Because the right hemisphere does not initially have access to speech centers, and thus is functionally mute in early split-brain patients, the participant could not accurately answer this question. The left hemisphere, which has access to speech, did not know why the left hand was pointing at a shovel. Nevertheless, rather than responding that he did not know why he was pointing at the shovel, the participant invented an answer in this case, the plausible story that chickens make a lot of waste and the shovel is necessary to remove this waste. This study reveals an individual who self-deceives only to avoid the uncertainty caused by his lack of awareness of the source of his own behavior (a situation that is likely to be very common; see Nisbett & Wilson 1977). Nevertheless, this motivation appears to be sufcient to cause the person to invent a reason for his behavior and then apparently convince himself of the accuracy of this ction. Other examples from the neuropsychological literature can be found in various types of brain and body damage, in response to which the individual is motivated to maintain certain beliefs that are at odds with reality. For example, anosognosia is a prototypical self-deceptive disorder, in which people who have sustained an injury to some part of their body deny the reality of their injury. Ramachandran (2009) described an anosognosic woman who denied that her left arm was paralyzed. He wrote:
An intelligent and lucid patient I saw recently claimed that her own left arm was not paralyzed and that the lifeless left arm on her lap belonged to her father who was hiding under the table. Yet when I asked her to touch her nose with her left hand she used her intact right hand to grab and raise the paralyzed hand using the latter as a tool to touch her nose! Clearly somebody in there knew that her left arm was paralyzed and that the arm on her lap was her own, but she the person I was talking to didnt know. (Ramachandran 2009)
As can be seen in this description, the patient had awareness that her left arm was paralyzed, as indicated by her use of her right arm to move it, but she appeared to suffer from lack of awareness as well, suggesting selfdeception. Consistent with this interpretation of anosognosia, a recent study (Nardone et al., 2007) presented disabled individuals with neutral and threatening words relevant to their immobility (e.g., walk). Those who were aware of their disability showed rapid disengagement from the threatening words, as indicated by more rapid response to a dot presented elsewhere on the screen in the presence of threatening than neutral words. In contrast, anosognosic individuals were less able to disengage from the threatening words, showing a slower response to the dot probe when paired with threatening versus neutral words. These data highlight implicit awareness of the disability in anosognosia, despite its explicit denial. Cases such as these involve damage to the right hemisphere, which appears to prevent individuals from recognizing the logical inconsistency in their belief systems. Similarly, cases such as the one documented by Gazzaniga also require damage to the brain (the corpus callosum) for the individual to be denied access to information in some parts of the brain that is available to others. In the case of selective access of information to consciousness, however, individuals with no damage to the brain also experience this situation regularly, as reviewed earlier. Thus, there are also studies of individuals deceiving themselves when they are neurologically intact. We turn now to such evidence.
5.5.2. Self-deception unaccompanied by neurological damage. A ubiquitous variety of self-deception can be
found in research on perceptions of control. Perceptions of control appear to be necessary for the maintenance of psychological and physical health (Cohen 1986; Glass & Singer 1972; Klein et al. 1976). When people are deprived of actual control, they often endeavor to regain a sense of control. In a self-deceptive example of this effect, Whitson and Galinsky (2008) found that when people are led to feel low levels of personal control, they perceive illusory patterns in random congurations and are more likely to endorse conspiracy theories to explain co-occurring world events. Importantly, these effects did not emerge when people had self-afrmed, suggesting potential awareness of the absence of patterns and conspiracies. Similar ndings have been documented by Kay et al. (2008), who argued that beliefs in a controlling God and a strong government serve peoples need for control. Consistent with their reasoning, differences in the percentage of people who believe in God between countries can be predicted by the insecurities of existence within countries (e.g., availability of health care, food, and housing), with increased insecurity associated with increased religiosity (Norris & Inglehart 2004). Such a nding suggests the possibility of self-deception on a worldwide scale. Another example of individuals deceiving themselves can be found in the research of Epley and Whitchurch (2008) reviewed earlier, in which people more rapidly located photos of themselves when the photo had been morphed to be more attractive than when the photo was unaltered. This nding suggests that peoples self-image is more attractive than their actual one, as the enhanced self provides a quicker match to their internal template than the actual self. Finally, experiments in cognitive
11
von Hippel & Trivers: The evolution and psychology of self-deception dissonance also suggest that people are facile at lying to others and then coming to believe their own lies. For example, when they believe that they have freely chosen to tell another person that a tedious task is actually interesting, people soon believe that the task really is interesting (Festinger & Carlsmith 1959), and again this effect is eliminated by self-afrmation (Steele & Liu 1983). 6. Who is the audience for self-deception? Thus far we have argued that self-deception evolved to facilitate deception of others, but the examples of selfdeception described in section 5 appear to be directed primarily toward the self. If self-deception evolved to deceive others, why is there so much evidence for self-deception that appears to be intended only for the self? There are three answers to this question. First and foremost, although Trivers (1976/2006) originally suggested that self-deception might have evolved to facilitate the deception of others over 30 years ago, this suggestion has not been taken seriously in the empirical literature. Rather, the tradition in psychology has been to treat self-deception as a defensive response to an uncongenial world (a point to which we return in sect. 7). As a consequence, to the best of our knowledge no one has examined whether self-deception is more likely when people attempt to deceive others. Thus, the theoretical possibility of self-deception in service of other deception remains just that, and the evidence described in section 5 stems primarily from studies in which the motives appear to be more intrapersonal than interpersonal. Second, to the degree that self-deception allows the individual to carry on with greater equanimity and condence, it serves the general interpersonal goal of selfenhancement described in section 3. Most of the cases of apparently self-directed self-deception from section 5 would t under this explanatory rubric. For example, if individuals selectively gather information in a manner that enables them to deny to themselves that they are at risk for a health disorder (as in Ditto & Lopez 1992), then they are better positioned to convince others that they would make reliable and vigorous sexual or coalitional partners. That this pattern of self-deception makes them less capable of dealing with an impending health threat might have been a relatively small price to pay in an ancestral environment where there was little that could be done in any case. A similar interpersonal logic might underlie self-deceptions that help people maintain conviction in their beliefs (e.g., Lord et al. 1979) and give them a sense of control over the world (e.g., Whitson & Galinsky 2008). Other classic examples of intrapersonally oriented selfdeception also have important interpersonal consequences. For example, people tend to be unrealistically optimistic about their future (Armor & Taylor 1998; Taylor & Brown 1988; Weinstein 1980). This optimism can have self-deceptive origins; for example, people high in self-deception measured via Sackeim and Gurs (1979) Self-Deception Scale have been shown to be more likely than people low in self-deception to perceive an upcoming onerous task as a challenge (i.e., within ones capabilities) rather than a threat (i.e., overwhelming ones capabilities; Tomaka et al. 1992). In this sense, self-deceptive optimism 12
appears to create a self-fullling prophecy, as condence in eventual success leads optimists to greater perseverance in the face of difculties (Carver & Scheier 2002; Solberg Nes & Segerstrom 2006). As a consequence of these processes, optimists gain numerous social and nancial benets over their less optimistic counterparts (Assad et al. 2007; Brissette et al. 2002; Carver et al. 1994; Segerstrom 2007). Finally, it is also possible that this system of self-deception that evolved to deceive others becomes applied in intrapersonal domains because of the good feelings that it brings the individual. By analogy, consider masturbation, another intrapersonal activity that has origins in an interpersonal system. Masturbation presumably emerged in primates because we evolved to enjoy copulation (thereby facilitating reproduction), but with the later evolution of hands rather than hooves or claws, we found a way to experience that enjoyment when circumstances conspire against sharing it with others. Self-directed self-deception might be analogous to masturbation in the sense that selfdeception evolved for interpersonal purposes, but people found a way to use it to enhance happiness when circumstances conspire against other methods. So long as the consequences of this self-deception are tness neutral or biologically affordable, self-deception might have leaked into a variety of intrapersonal domains that have important consequences for human happiness. Indeed, the analogy to masturbation remains apt in this regard, as masturbation might be tness neutral or even tness positive among males, because it might enhance sperm quality by shedding older sperm (Baker & Bellis 1993). Thus, primates appear to have found a way to enhance their happiness with little if any cost to inclusive tness. Similarly, although the spread of self-deceptive practices in the pursuit of happiness would clearly bear a cost to the degree that people sacrice information quality for hedonic gains, such practices might also be tness neutral or even tness positive to the degree that people reap the gains brought about by happiness itself (Fredrickson 1998; 2001). As with optimism, happiness has important interpersonal consequences; people experience increased social and nancial success when they are happy (Boehm & Lyubomirsky 2008; Fredrickson et al. 2008; Hertenstein et al. 2009; Lyubomirsky et al. 2005). Others are attracted to happy people and put off by sad people, for a variety of reasons (Bower 1991; Frijda & Mesquita 1994; Harker & Keltner 2001; Keltner & Kring 1998). As the aphorism goes, laugh and the world laughs with you, cry and you cry alone. Because humans are a social species wherein individuals achieve most of their important outcomes through coordinated or cooperative actions with others, attracting others to oneself and ones causes is an important achievement with notable tness consequences (Fredrickson 1998; 2001). Thus, to the degree that people generally maintain a sunny disposition, they are likely to be more effective in their goal pursuits. Happiness is also important for physical well-being, given that there are immune benets to feeling happy and immune costs to feeling sad (Cohen et al. 2006; Marsland et al. 2007; Rosenkranz et al. 2003; Segerstrom & Sephton 2010). Because threats to health loom larger late in life, particularly from cancers and parasites (World Health Organization 2009), happiness may be even more important in late than early adulthood. Consistent with
von Hippel & Trivers: The evolution and psychology of self-deception this possibility, older adults are more likely than younger adults to focus on and remember positive rather than negative information (Mather & Carstensen 2005; Mather et al. 2004). Thus, it appears that late life happiness is maintained in part by the knowledge avoidance form of self-deception, as older but not younger adults look away from negative information and toward positive information when they are in a bad mood (Isaacowitz et al. 2008). Enhanced immune functioning may offset the informational costs of this aging positivity effect.4 7. At what level of consciousness is the self deceived? At the beginning of this article we rejected the classic claim that self-deception must involve two separate representations of reality, with truth preferentially stored in the unconscious mind and falsehood in the conscious mind. Rather, our biased processing perspective suggests that the individual can self-deceive in a variety of ways, some of which prevent even unconscious knowledge of the truth. Nevertheless, such a possibility does not preclude classic forms of self-deception, and here we consider the question of how the truth is represented in different forms of self-deception. We begin with types of self-deception in which conscious and unconscious knowledge are aligned and inaccurate, that is, cases in which individuals believe consciously and unconsciously in the veridicality of the deceptive information. We propose that this sort of self-deception should occur in two types of situations. First, self-deception should exist at both conscious and unconscious levels when individuals prevent themselves from ever encoding the unwelcome truth. For example, Ditto and Lopez (1992) documented how individuals who are pleased with the early returns often stop gathering information before they encounter unwanted information, and thus their self-deception has prevented unwanted information from entering either conscious or unconscious knowledge. Such individuals might be aware that their information gathering strategy could have inuenced the nature of their knowledge, but such awareness is likely to be rare, as (a) there is no set standard for how much information a person should gather in most settings, and (b) there is often no specic reason to believe that the next bit of information would have been contrary to that which had already been gathered. Second, self-deception should also exist at both conscious and unconscious levels in many (although probably not all) cases of self-enhancement. In such cases individuals have a lifetime of gathering and processing information in a manner that favors the self, and it would probably be difcult if not impossible for them to parse out the impact of their long-term processing strategies on their understanding of the world. Support for this possibility can be found in two types of effects. First, as Epley and Whitchurch (2008) demonstrated, people are faster to identify their more attractive self than their actual self in an array of faces. This nding suggests that the enhanced version of the self is likely to be represented in memory below consciousness awareness, as unconscious processes tend to be more rapid than conscious ones (e.g., Neely 1977), and conict between conscious and unconscious processes leads to slower rather than more rapid responses (e.g., Greenwald et al. 1998). Second, research on the convergence of implicit and explicit self-esteem reveals that individuals who show high implicit and high explicit selfesteem appear not to be defensive, whereas individuals who show high explicit but low implicit self-esteem appear to be the most defensive and narcissistic (Jordan et al. 2003). Because defensive and narcissistic individuals tend not to be well liked (Colvin et al. 1995; Paulhus 1998), this research suggests that the interpersonal benets of self-enhancement are most likely to be realized if people believe their self-enhancing stories at both conscious and unconscious levels. In contrast to these situations are classic cases of selfdeception in which the conscious mind is deceived but the unconscious mind is well aware of the truth. This should be common in cases of self-deception intended to facilitate deception of others on specic issues. Under such circumstances individuals often have a limited time frame in which to convince the self of the truth of the deception that they wish to promulgate, and thus the types of memory processes outlined in section 4 are likely to lead to conscious recollection of the deception but implicit memory of the truth as originally encountered. Thus, it seems likely that self-deception can vary in the depth to which the self is deceived, with some processes leading to deception of conscious and unconscious aspects of the self and other processes leading to deception of only conscious aspects of the self. 8. Contrasting our approach with other approaches Our evolutionary approach to self-deception is based on the premise that self-deception is a useful tool in negotiating the social world. According to this viewpoint, self-deception is best considered an offensive strategy evolved for deceiving others. In contrast to this viewpoint, most prior research on self-deception considers it a defensive strategy, adopted by individuals who are having difculty coping with a threatening world. In this research tradition, the hedonic consequences for the self-deceiver are considered to be the primary outcome of self-deception. From an evolutionary perspective, however, hedonic consequences are not an important endpoint themselves, but only a means to an end, such as when they lead to enhanced immune functioning or greater interpersonal success, as described in section 6. The most important consequence of this prior emphasis on hedonic consequences is that the eld has been focused on what we would regard as a secondary aspect of selfdeception. From our perspective, the study of self-deception as if it is a psychological immune system (Gilbert et al. 1998; Wilson & Gilbert 2003) or a suite of positive illusions (Taylor & Brown 1988) intended to enhance or restore happiness is somewhat akin to the study of masturbation as if this is the purpose for which the sexual system was designed. Such an approach to sexuality would lead to worthwhile ndings about some of the affective consequences of sexual behavior, but by ignoring the interpersonal purpose of sexuality, this approach would miss most of the important questions and answers. Thus, in many ways the arguments outlined in this article call for a fundamental change in our approach to
13
von Hippel & Trivers: The evolution and psychology of self-deception the problem of self-deception. The research described in sections 4 and 5 provide ample evidence for the psychological processes involved in self-deception and for the manner in which they protect peoples beliefs and desires from a contrary reality. But because self-deception has been viewed as a defensive response to an uncongenial world, there is virtually no research that considers the interpersonal opportunities that might lead to self-deception. For example, research suggests that heterosexual men enhance their self-presentation when they meet attractive women (Buss 1988), and women rate other women as less attractive when they are themselves ovulating (Fisher 2004).5 The current approach suggests that men should actually believe at least some of their self-enhancement, and women should believe their more negative evaluations of others as well. That is, these strategies should be more effective if people are not just posturing but actually accept their own self-inating and other-deating stories. To test these possibilities, one could assess whether the presence of attractive women causes men to show greater self-enhancement in their information-processing biases and whether ovulation causes women to show greater derogation of rivals in their biases as well. Future research could also address the utility of selfdeception in specic acts of deception of others. For example, people could be put in a situation in which they must lie or otherwise deceive and their use of various self-deceptive strategies reviewed in section 5 could be examined. In such a setting they could be assessed for the degree to which they stop their search early when they encounter initial information that is consistent with their deceptive goals, avoid information that appears potentially inconsistent with an upcoming deception, show biased interpretation of information that is consistent versus inconsistent with an upcoming deception, and show better recall for deception-consistent than deception-inconsistent information. Importantly, the utility of such strategies could be assessed by examining the degree to which these information-processing biases are associated with increased effectiveness in deceiving others. As noted earlier, similar biases should also be invoked by situations that enhance the interpersonal need to appear condent and efcacious. And if these information-processing strategies reect self-deception, then the degree to which people show biases under these circumstances may be moderated by individual differences in the tendency to self-deceive. For example, these effects should be stronger among people who show greater self-enhancement in the Epley and Whitchurch (2008) paradigm than among people who do not show such self-enhancement. The current approach to self-deception is also relevant to a series of debates that have been percolating for some time in psychology, in which self-enhancement (selfdeception) is pitted against self-verication (reality orientation). In the rst of these debates, researchers have asked whether self-enhancement motives are stronger than self-verication motives (see Swann, in press). From an evolutionary perspective this is not a meaningful question to ask, because it is akin to asking whether thirst is stronger than hunger. Just as the desire to eat versus drink is jointly determined by the current needs of the individual and the quality of resources available to satisfy hunger versus thirst, the desire to self-verify versus self14
enhance should be jointly determined by current need states and the quality of available verifying and enhancing opportunities. Indeed, the research on self-afrmation described in section 5 highlights the fact that when peoples self-enhancement needs are met, they become more willing to self-verify in negative domains. Thus, both enhancing and verifying needs are likely to be present nearly all of the time, and their relative strength will vary as a function of the social and material benets they can accrue for the individual at any particular moment. Self-verication allows people to accurately gauge their abilities and behave accordingly, and self-enhancement allows people to gain an edge beyond what their actual abilities provide them. In the second of these debates (e.g., Sedikides et al. 2003 vs. Heine et al. 1999), researchers have argued about whether self-enhancement is pan-cultural or a uniquely Western phenomenon. From our perspective, this debate has gotten off track because of its emphasis on the role of positive self-regard rather than interpersonal goals as the source of self-enhancement. Because self-enhancement induces greater condence, our perspective suggests that it will emerge in every culture in which condence brings social and material gains. If a culture exists in which people benet in various social and material domains by being meek and self-doubting, then individuals in such a culture should not self-enhance in those domains. From an evolutionary perspective, however, people in all cultures should benet from enhancing those qualities that are necessary to win confrontations and secure mates. Because individual differences in abilities and proclivities enable some individuals to win conicts and mates by physical means, others to win them by intellectual means, and still others by displays of kindness, artistic abilities, and so forth, one would expect that self-enhancement should similarly emerge in different forms for different individuals, but should do so in every culture on earth. This debate has also become caught up in issues of measurement focused on how to properly assess selfenhancement in different cultures. Because the rules of social engagement vary widely across different cultures, it should come as no surprise that in some cultures people are unwilling to claim to be better than others. Again, however, the self-deceptive goal of self-enhancement is to believe one is slightly better than one really is, and thus explicit claims are less important than implicit beliefs. Indeed, self-enhancing claims are often likely to be evidence of bad manners or poor socialization, whereas exaggerated beliefs in ones abilities in domains that are necessary to win confrontations and secure mates should be universal. In the third of these debates, people have asked whether enhancing self-views or accurate self-views reect greater mental health (e.g., Colvin & Block 1994 vs. Taylor & Brown 1994). This debate has shades of the second, given that again a balance of the two and knowing when each is appropriate should reect the best strategy for goal achievement, and thus should be reective of mental health. This debate has also gotten bogged down in the misperception (noted by Taylor & Brown 1994) that if a little self-enhancement is a good thing, then more can only be better. Biological systems rely on balance or homeostasis, and too much of even a good thing disrupts this balance. Thus, from an evolutionary
von Hippel & Trivers: The evolution and psychology of self-deception perspective, it is obvious that the benets of self-enhancement depend on proper dosage. Too much self-enhancement might not only be hard to believe oneself but might strike others as preposterous or mentally unbalanced (Colvin et al. 1995; Paulhus 1998) and might also lead to complacency and poor preparation in the face of real problems. This is not evidence, as Colvin et al. (1995) and Paulhus (1998) suggested, that self-enhancement is socially maladaptive. Rather, this nding simply highlights the fact that self-deception remains in permanent tension with veridical perception. Finally, similar to the second debate, our interpersonal perspective on self-deception suggests that the key to understanding whether self-enhancement is a sign of good versus poor mental health is whether people believe their own self-enhancing story. Self-enhancement is useful only to the degree that it is self-deceptive, because only when it is believed by the self will others accept the enhanced self as genuine. If the claims ring hollow or if the claimants appear to be trying unsuccessfully to convince themselves, then this is a sign of failed self-deception and is likely to be an indicator of poor social functioning and poor mental health. This perspective suggests that much of the evidence showing that self-enhancement and other positive illusions are a sign of poor mental health comes from studies in which no differentiation is being made between people who are convinced versus unconvinced by their own self-enhancing claims. As noted earlier, people who are high in explicit but low in implicit self-esteem are the most defensive and narcissistic (Jordan et al. 2003), and thus it seems likely that self-enhancement only brings benets to the degree that it is believed both consciously and unconsciously. For this reason, we would predict that individuals who show convergence in their explicit and implicit measures of self-enhancement will conrm Taylor and Browns (1988; 1994) claims that self-enhancement is a sign of good mental health. Nevertheless, there may still be the occasional individual who shows extreme but convergent self-enhancement, and from an evolutionary perspective, this is not likely to be a successful interpersonal strategy and therefore unlikely to be a sign of good mental health. Such individuals may not come across as defensive, but they are likely to appear deluded. 9. Costs versus benets Finally, it is worth considering the costs versus benets of self-deception. Because self-deception requires a mental architecture that sometimes favors falsehoods, those features of our mental landscape that allow us to selfdeceive are likely to have attendant costs. Although we have focused thus far on the possibility that self-deception can be benecial, it is also worth considering the costs. The most obvious cost of self-deception is loss of information integrity with the resulting potential for inappropriate action and inaction but there are likely to be other costs as well. For example, consider the case of memory discussed in section 4.1, in which we outlined how source confusions can facilitate self-deception in peoples efforts to deceive others. Research suggests that some individuals are more likely than others to have difculty differentiating the source of their memories, with the result that they are more susceptible to the implantation of false memories (Clancy et al. 2000). Such individuals suffer the cost of greater memory failures and therefore poorer foresight and greater manipulation by others. But they should also be more capable of selfdeception in their efforts to deceive others, as retrieval of their lies should make it particularly difcult for them to differentiate the source of the false information. In this manner the costs associated with their poor sourcemonitoring capabilities might be offset by the gains associated with their greater likelihood of deceiving others. At a more general level, the overarching fact that the mind might have evolved to self-deceive and therefore to be tolerant of inconsistencies between information in the conscious and unconscious mind raises the possibility that this weapon of deception might be capable of being turned upon the self. That is, self-deception might be imposed by others. For example, an abusive spouse who rationalizes his abuse as a product of his partners failings might also convince her that she is to blame for his abuse. Consistent with this sort of possibility, system-justication theory as elaborated by Jost and colleagues (Jost et al. 2004; Jost & Hunyady 2005), argues that there are a variety of motivational reasons why people support the status quo, even when they are clear losers in the current system with very little likelihood of improving their situation. Such system-justifying beliefs among those at the bottom of the social ladder serve the purposes of those at the top of the ladder, in part by preventing agitation for social change. This argument suggests that system justication might be considered a variety of selfdeception imposed onto low-status individuals by highstatus people who benet when those beneath them accept their low status as legitimate and do not struggle against it. This argument also suggests that the consequences of self-deception might be wide ranging, as a process that evolved to facilitate the deception of others appears to have effects that manifest themselves from an intrapersonal all the way to a societal level. ACKNOWLEDGMENTS
Preparation of this manuscript was supported by fellowships at the Institute for Advanced Study, Berlin, and by grants from the Australian Research Council and the Biosocial Research Foundation. We thank Michael Anderson, Marzu Banaji, Pablo ol, David Buss, Bella DePaulo, Tony Greenwald, Wilhelm Brin Hofmann, Geoffrey Miller, Srini Narayanan, Steve Pinker, and Constantine Sedikides for helpful comments on an earlier draft of the manuscript. NOTES 1. Cross-examination can also cause deceivers to seem more honest (Levine & McCornack 2001), but further research is needed on the effects of cross-examination in detecting lies that are consequential for deceivers if discovered. 2. It should be noted in this regard that there is evidence for the contrary possibility that people overestimate the degree to which others have detected their lies (Gilovich et al. 1998). However, this research relied on the usual paradigm of people telling trivial lies to strangers. There are few cues in such cases about whether others believe our lies, and thus deceivers might be particularly inclined to assume that whatever cues they are aware they are emitting are equally obvious to observers.
15
Commentary/von Hippel & Trivers: The evolution and psychology of self-deception

3. People with low self-esteem also seek out negative information about themselves even when not feeling good about themselves. This information search strategy appears to involve self-verication strivings, or peoples desire to be known by others as they see themselves (Swann, in press). Such self-verication leads most people to seek out positive information, however, because most people feel positive about themselves. 4. Although selection pressure diminishes in late life, grandparents historically played an important role in the survival of et al. 2004), and thus their contheir grandchildren (Lahdenpera tinued good health and survival is important for their inclusive tness. 5. It should be noted that women are more attractive when ovulating (Thornhill & Gangestad 2009), so this change in ratings of others could reect an unbiased relative judgment. how one can be the deceiver and the one deceived at the same time. VH&T propose another variant of the split-self solution to the paradox. They posit a dissociated mental dualism in which the deceived mind is disconnected from the unconscious mind that knows the truth. Given the richly interconnected neuronal structure of the brain, is this Cartesian physical dualism at odds with what is known physiologically about the brain? The dissociated dualism removes not only the paradox, but also any commerce between the two minds. While the conscious mind is biasing information processing to bolster the self-deception, there is no mention of what the veridical unconscious mind is doing. If it is dissociated, how can it affect the conscious mind? A multitude of neuronal systems is involved in the processing of input information. However, when it comes to action, there is only one body. The diverse systems have to generate a coherent action. There is diversity in information processing but unity of agency in action (Bandura 2008; Korsgaard 1989). Contradictory minds cannot simultaneously be doing their own thing behaviorally. The authors do not explain how the disconnected conicting minds can produce a coherent action. There are other epistemological issues, including the veriability of the central thesis, that need to be addressed. The article presents an excellent review of biased information processing, but it leaves a lot to be desired in conceptual specication. How does one know what the unconscious mind knows? How does one assess the unconscious knowledge? By what criteria does one substantiate truth? Why is the unconscious mind veridical in self-enhancement but self-deceptive in other spheres of functioning? How does one gauge the benets of self-deception, whether in the short term or the long term? Given the evolutionary billing of the article under discussion, what were the ancestral selection pressures that favored self-deception? Claiming that a given behavior has functional value does not necessarily mean it is genetically programmed. People are selective in their information seeking, often misconstrue events, lead themselves astray by their biases and misbeliefs, and act on decient knowledge. However, biased information seeking and processing are not necessarily self-deception. VH&T cite, as an example of self-deception, conservatives favoring Fox News and liberals favoring MSNBC. By their selective exposure, they reinforce their political bias. But that does not mean that they are lying to themselves. The same is true for some of the other forms of biased information processing that are misconstrued as self-deception. In genuine self-deception people, avoid doing things that they have an inkling might reveal what they do not want to know. Suspecting something is not the same as knowing it to be true, however. As long as one does not nd out the truth, what one believes is not known to be false. Keeping oneself uninformed about an unwanted truth is the main vehicle of genuine selfdeception. By not pursuing courses of action that would reveal the truth, individuals render the knowable unknown (Haight 1980). Acting in ways that keep one uninformed about unwanted information is self-deception. Acting in selectively biasing and misinforming ways is a process of self-bias. These are different phenomena. The truth is not harbored in a dissociated unconscious mind. It exists in the information available in the avoided reality. The disconnected unconscious mind cannot know the truth if the evidential basis for it is avoided. VH&T emphasize the benets of self-deception but ignore the social costs to both the self-deceiver and deceived. Being misled is costly to others. Therefore, the deceived do not take kindly to being led astray. Human relationships are longterm affairs. There are limits to how often one can mislead others. After a while, the victims quit caring about whether the deception was intentional or carried out unknowingly. If done repeatedly, the short-term gains of misleading others come with the cost of discrediting ones trustworthiness. Undermining ones ability to exercise social inuence does not have adaptive value.
Open Peer Commentary

Self-deception: A paradox revisited
doi:10.1017/S0140525X10002499 Albert Bandura
Department of Psychology, Stanford University, Stanford, CA 94305. bandura@psych.stanford.edu www.stanford.edu/dept/psychology/abandura
Abstract: A major challenge to von Hippel & Triverss evolutionary analysis of self-deception is the paradox that one cannot deceive oneself into believing something while simultaneously knowing it to be false. The authors use biased information seeking and processing as evidence that individuals knowingly convince themselves of the truth of their falsehood. Acting in ways that keep one uninformed about unwanted information is self-deception. Acting in selectively biasing and misinforming ways is self-bias.
Von Hippel & Trivers (VH&T) present the case that self-deception evolved because it enables individuals to be good deceivers of others. By convincing themselves that their fabrication is true, people can enjoy the benets of misleading others without the intrapsychic and social costs. The authors review a large body of evidence on biased information processing on the assumption that this is the means by which individuals unknowingly convince themselves of the truth of their falsehood. A major challenge to a functional analysis of self-deception is the problematic nature of the phenomenon itself. One cannot deceive oneself into believing something while simultaneously knowing it to be false. Hence, literal self-deception cannot exist (Bok 1980; Champlin 1977; Haight 1980). Attempts to resolve the paradox of how one can be a deceiver fooling oneself have met with little success (Bandura 1986). These efforts usually involve creating split selves and rendering one of them unconscious. Awareness is not an all-or-none phenomenon. There are gradations of partial awareness. Hence, self-splitting can produce a conscious self, various partially unconscious selves, and a deeply unconscious self. In this line of theorizing, the unconscious is not inert. It seeks expression in the intrapsychic conict. The deceiving self has to be aware of what the deceived self believes in order to know how to concoct the self-deception. Different levels of awareness are sometimes proposed as another possible solution to the paradox. It is said that deep down people really know what they believe. Reuniting the conscious and unconscious split selves only reinstates the paradox of
16

Print 1

Încărcat de

Informații document

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Print 1

Încărcat de

Drepturi de autor:

Formate disponibile

How Credit Card Payments Increase Unhealthy Food Purchases: Visceral Regulation of Vices

MANOJ THOMAS KALPESH KAUSHIK DESAI SATHEESHKUMAR SEENIVASAN

VISCERAL REGULATION OF VICES

CONCEPTUAL FRAMEWORK Impulsive Purchase of Vice Products

JOURNAL OF CONSUMER RESEARCH

Pain of Paying in Cash

Individual Differences in Sensitivity to Pain of Payment

Deliberative Purchase of Virtue Products

VISCERAL REGULATION OF VICES

STUDY 1: MODE OF PAYMENT AND HEALTHINESS OF SHOPPING BASKET A FIELD STUDY

JOURNAL OF CONSUMER RESEARCH

VISCERAL REGULATION OF VICES

STUDY 2: MANIPULATING MODE OF PAYMENT

JOURNAL OF CONSUMER RESEARCH

VISCERAL REGULATION OF VICES

STUDY 3: MEDIATION BY PAIN OF PAYMENT

JOURNAL OF CONSUMER RESEARCH

VISCERAL REGULATION OF VICES

JOURNAL OF CONSUMER RESEARCH

STUDY 4: TIGHTWADS AND SPENDTHRIFTS

VISCERAL REGULATION OF VICES

GENERAL DISCUSSION Summary and Implications

JOURNAL OF CONSUMER RESEARCH

Buying Behavior as a Function of Parametric Variation of Number of Choices

Copyright r 2007 Association for Psychological Science

Number of Choices and Buying Behavior

Opportunity Cost Neglect

JOURNAL OF CONSUMER RESEARCH

OPPORTUNITY COST NEGLECT

STUDY 1A: THE EFFECT OF OPPORTUNITY COST SALIENCE ON WILLINGNESS TO PURCHASE

Results and Discussion

OPPORTUNITY COST NEGLECT

Results and Discussion

JOURNAL OF CONSUMER RESEARCH

STUDY 1C: OPPORTUNITY COST SALIENCE IN CONSEQUENTIAL CHOICE

Results and Discussion

Results and Discussion

STUDY 2: EXPLORING A SEGREGATION ACCOUNT OF THIS EFFECT

STUDY 3: A DIRECT MEASURE OF OPPORTUNITY COST SALIENCE

OPPORTUNITY COST NEGLECT

Results and Discussion

STUDY 4: PRIMING OPPORTUNITY COSTS

JOURNAL OF CONSUMER RESEARCH

Results and Discussion

STUDY 5: INDIVIDUAL DIFFERENCES IN SENSITIVITY TO OPPORTUNITY COST CUES

OPPORTUNITY COST NEGLECT

JOURNAL OF CONSUMER RESEARCH

The Psychological Consequences of Money

*To whom correspondence should be addressed. E-mail: kvohs@csom.umn.edu

Downloaded from www.sciencemag.org on March 27, 2013

67.35 84.65 18.00 1.96

Monetary donations (in $)

Downloaded from www.sciencemag.org on March 27, 2013

Supporting Online Material

4.00 1.20 0.83 0.39

2.82 1.00 0.31 0.48

3.10 1.80 0.25 0.45

Downloaded from www.sciencemag.org on March 27, 2013

Contagion and Differentiation in Unethical Behavior

University of North Carolina at Chapel Hill and 2Duke University

behavior of another person increases or decreases individuals dishonesty.

Copyright r 2009 Association for Psychological Science

Francesca Gino, Shahar Ayal, and Dan Ariely

EXPERIMENT 2: EFFECTS OF SALIENCY