Why Retailers Sell More or Less Than Their Fair Share in a Category

Sanjay K. Dhar, Stephen J. Hoch, and Nanda Kumar October 2000

Sanjay K. Dhar is Professor of Marketing and Beatrice Foods Company Scholar at the Graduate School of Business, University of Chicago. Nanda Kumar is Asst. Professor of Marketing at the School of Management, The University of Texas at Dallas. Stephen J. Hoch is John J. Pomerantz Professor of Marketing at the Wharton School, University of Pennsylvania. Partial support for this work was provided by the Beatrice Foods Company Faculty research fund at the Graduate School of Business, University of Chicago.

Abstract Assessing a retailers performance in a category is important to both manufacturers and retailers. The main objective of this paper is to understand the factors that determine across-retailer variation in category performance as measured by the Category Development Index (CDI). Based on data from 19 food categories sold in 85 major supermarket chains operating in the largest 50 retail markets in the U.S, our analysis shows that the best performing retailers: (a) offer broader assortments, (b) have strong private label programs, (c) charge significantly lower everyday prices, and (d) use feature advertising to drive store traffic and display to increase in-store purchases. In addition, the best performing retailers also are more likely to engage in a formal category management process. Finally, we find that the impact of the price, promotion and assortment variables depends importantly on the role (staple, variety enhancers, niche, or fill-in) that the particular category plays in the stores overall portfolio.

1 Why Retailers Sell More or Less Than Their Fair Share in a Category With a substantial amount of research investigating variability in sales at the brand level, there is now growing interest in analyses at the category level. In a cross-category analysis Fader and Lodish (1990) examine various structural characteristics of categories and relate them to the frequency and types of promotions offered by retailers and manufacturers. In earlier studies, Bolton (1989) and Litvack, Calantone and Warshaw (1985) highlight factors that account for differences in brand price elasticities across categories. Other work has shown that despite high brand price elasticities category sales may not change much if promotions and other marketing mix actions primarily lead to brand switching (Gupta, 1988) and/or store switching (Kumar and Leone 1988). This shift from brand level analysis to cross-category analysis (within a retailer) may be attributed to the fact that while an understanding of the variation in brand sales is extremely valuable to the brand managers, it is of considerably less value to the retailer. As Raju (1992) notes , it is possible that a promotional program which may be very desirable from a brand managers perspective, may in fact lead to a reduction in dollar sales of the category if much of the merchandise moves at the promoted price. He then proceeds to analyze the effect of price promotions on the variability in category sales of a large grocery store. To the best of our knowledge the work reported in this paper is the first to analyze variation in category sales across retailers, an analysis valuable to both manufacturers and retailers. First, an analysis of this type can help manufacturers better allocate their scarce marketing resources across retailers and categories. For example, manufacturers can help both better performing and underachieving retailers by reallocating more dollars to the marketing actions that have the biggest impact on a categorys performance (e.g., displays vs feature ads vs depth or breadth of assortment).

2 Second, the findings from such an analysis can help a retailer understand why competitors are doing better or worse than itself (or the market), enabling the retailer to reallocate its resources across categories to improve its overall market position. The main objective of this paper is to relate the variability in category performance across retailers to category characteristics and retailer pricing, promotion, and merchandising strategy, while controlling for the level of manufacturer support (advertising and consumer promotion). We conduct this analysis for 19 different product categories. In answering any performance question, it is mandatory that all parties specify the how well question with an attendant compared to what? Absolute performance figures do not provide an adequate picture since such metrics do not control for anything else. In the work reported here we consider a specific performance measure, one that has a long history in the packaged goods world CDI or Category Development Index. In words, a retailers CDI for a category is calculated by the ratio of its share of the category in the market, compared to its market share across all categories. Consider a retailer with a total market share of 20%. In category 1 they have a 15% share and consequently a CDI of 75, whereas in category 2 they have a 25% share and a CDI of 125. To reiterate, the main purpose of this paper is to understand which actions taken by retailers determine whether they sell more or less than their fair share in a category --- that is, how do retailers that achieve a high CDI set marketing decision variables differently than their underachieving peers. In one sense the answer is fairly obvious. Holding constant the inherent potential of the market that they serve, retailers get more than their fair share by trying harder: charging lower prices, promoting the category more aggressively, and offering more varied assortments. What is less obvious is the exact importance of each of these marketing decision variables and how their impact might vary by category.

3 Specifically because consumer behavior and motivations can differ dramatically given what role the product plays in everyday life, the effectiveness of marketing actions should differ systematically across categories. These issues are key to any formal category management (CM) process where retailers must explicitly define the role that each category plays in the overall store portfolio. The following quote from Johnson (1999) further buttresses these issues The benefits of CM for a supplier organization are maximized when there is a purchase marketing strategy in place for the category. Purchase marketing is essentially using the store as a marketing medium, via tools such as position on shelf, pricing, communications and in-store promotions. Traditionally this has been the prerogative of retailers who own the stores. It is a challenge for the supplier to provide compelling arguments for the retailer to adapt, refine or even change the way they use the tools of the category to benefit the category as a whole and of course the suppliers position within it, thereby satisfying the business interests of both parties. By analyzing the cross-category sales variation across retailers the paper enables manufacturers to make specific recommendations that can help underachieving retailers enhance their performance in categories of mutual interest. In the same article, Johnson (1999) notes that Category Management forces retailers to think about where they want to be a year from now, who their competition is and what they want out of a particular category. The analysis presented in our paper highlights the impact of retailers actions on CDI. This can aid retailers achieve the intended objective once they specify what they want out of a particular category. The rest of the paper is organized as follows. First, we discuss the pros and cons of using CDI as a performance measure from the manufacturers and retailers perspectives in comparison to other candidate category performance measures. Second, we offer a conceptual framework for the role that different marketing decision variables play in influencing CDI and how these roles differ by type of category. Third, we describe the data sources and the variables. Finally, we present the empirical findings of our study based on 19 product categories and then conclude with a brief discussion of future

4 research. CDI as a Measure of Retailer Performance Packaged goods manufacturers and their syndicated data providers use CDI to measure the relative category performance of a retailer. The measure has variously been labeled category development index, CDI, or a fair share index, and is defined as:

CDI =

Retailer i Unit Volume in Category j Market ACV$ X Market Unit Volume in Category j Retailer i ACV$

(1)

ACV stands for All Commodity Volume, or simply total dollar sales for either retaileri (Retaileri ACV$) or the total market (Market ACV$). The market can be specified at any relevant level of geographic specificity (or alternatively demographic segments), from national to regional or an SMSA. CDIs can also be computed using dollar rather than unit sales. CDI is therefore an indicator of a retailers performance in a category relative to their overall performance in the total market. It allows manufacturers to determine whether a retail-account is over-performing or under-performing relative to other accounts in a given market and along with accompanying information on an equivalent measure for their brands performance (called brand development index or BDI ) helps firms to establish strategic objectives for brand growth either through market development and/or brand initiatives. Its usefulness lies in the fact that it provides a standardized basis for comparing category performance for retail accounts either within a given SMSA, region or even on a national basis. On the other hand, other conditional assessments that are either based on a retailers store size (e.g., sales per square foot, sales per store), prior store sales (e.g., category growth rate) are more problematic because it is difficult to control for other confounding factors like differences in population density (e.g., Manhattan vs. Fargo, N.D.), number of stores (e.g., Kroger vs. Ukrops), etc.

5 Although unit and dollar CDI provide useful performance monitoring measures to manufacturers, CDIs are less informative to a retailer. It is true that a cross-category comparison of CDIs may help a retailer to identify categories in which opportunity gaps exist. Moreover, because CDIs benchmark the retailers performance against the total market, they also provide valuable competitive information. At the same time, however, unit and dollar CDIs do not tell the retailer which categories might provide the best return on investment. This requires additional information such as a categorys profit contribution per square or linear foot or something like Direct Product Profit, which also could be expressed as an index similar to CDI. It ends up that a CDI based on profits is difficult to calculate from secondary data. Although retailers have information about their own profit margins and therefore have a sense of what it has cost them to achieve their current level of category performance, they do not have exact knowledge of competitors profit margins. There are other potential problems with relying on CDI as a measure of category level performance. For example, since the denominator of the CDI essentially is the retailers aggregate market share across all categories, the statistic can be misleading when comparing across retail formats because total store sales generated by, for instance, a supermarket is based on a different portfolio of categories than total store sales for a drug chain or mass merchant. Also, CDI can only be used as a relative, not absolute indicator of category performance, where the relative part refers to the retailers performance in all other categories. What this means is that the CDI for any individual category by definition cannot be influenced by any chain level policy that is pursued equally in all product categories and that has the same impact in all categories. The reason for this is that at the chain level every retailers CDI equals 100. And so, for example, CDIs would not be sensitive to the influence of chain wide employee incentive schemes, store equity, positioning strategy, price image, etc., unless these

6 variables influence performance differently in one category versus another, which is of course possible. Furthermore, as noted by a reviewer, it is not in the retailers best interests to take aggressive actions to achieve high CDIs in all categories since an increased CDI in one particular category must come at the expense of some other categories . With scarce marketing resources, the retailer needs to make strategic trade-offs in deciding on which categories to focus to maximize overall profit. Clearly this is a very important issue, but beyond the scope of this paper. Although there are limits to using CDI as a benchmarking statistic, we believe that CDIs represent a reasonable quantification of relative category performance. Understanding the key factors that lead one retailer to sell more or less than their fair share in a category is useful information for both manufacturers and retailers. Conceptual Development Category management has generated incredible interest and activity in the packaged goods industry. Although many of the ideas and processes have been around for years in some form or another, real investments in category management by manufacturers and retailers did not occur until around 1990. As defined in the Category Management Report published by the Joint Industry Project on ECR, category management involves the distributor/supplier process of managing categories as strategic business units, producing enhanced business results by focusing on delivering consumer value. Retailers practicing category management focus on the category as the strategic business unit of analysis. Ensuring category leadership, increasing market share, revenues, and profitability are the key reasons cited by retailers practicing category management (AC Nielsen Seventh Annual Survey of Trade Promotion Practices 1997). Category management involves both front-end activities to enhance category demand and back-door activities to improve supply management and

7 logistics coordination with vendors. A key tenet of category management is that the retailer must decide on the role each category plays in the overall portfolio and then execute towards those goals. By developing clear-cut and distinctive strategies, the retailer avoids dissipation of scarce resources that accompanies trying to be all things to all people with all categories. Retailers influence category volume by taking marketing actions that either: (1) increase store traffic; and/or (2) increase the probability of category purchase by consumers who already are in the store. Although both methods can work, we would argue that the retailers success in generating category demand through either traffic building or in-store tactics depends on the role the category plays in both the consumers and retailers portfolio. One popular classification scheme promoted the Food Marketing Institute (FMI) utilizes consumer-based category roles defined according to the percentage of households that buy the category and the frequency with which it is purchased. Consumer motivation is likely to be quite different depending on whether most households buy the category every week and spend a large part of their shopping budget on it (e.g., bread, carbonated beverages), or if only a few interested consumers make infrequent purchases in the category (e.g., vinegar, yeast). As shown in Figure 1, this approach uses syndicated data on penetration and frequency for each category, something that is easily accessible to retailers (FMI Category Management Guide #1, 1995). Categories are classified into "high" and "low" penetration (percentage of households that purchase the category) and frequency (average number of times per year category is purchased) categories and fall into one of four groups: (1) staples (high penetration/high frequency); (2) niches (low penetration/high frequency); (3) variety enhancers (high penetration/low frequency); and (4) fill-ins (low penetration/low frequency). Since consumer motivations to make purchases in staples will necessarily be different than in fill-ins and similarly among other category groups, we expect the effectiveness of specific marketing actions to

8 differ by category. Fader and Lodish (1990) found that this penetration-frequency distinction produced the most easily interpretable clustering solution in their cross-category analysis of promotional performance. Figure 1: Category Roles
Percent of Households Buying Frequency of Purchase High Frequency High Penetration STAPLE - RTE Cereal - Coffee Low Frequency VARIETY ENHANCERS - Pickles - Rice Low Penetration NICHE - Yogurt - Macaroni & Cheese FILL-IN - Pancake Mix - Syrup

In the next two sections, we discuss how different price, promotion, and assortment policies might affect retailer category performance by either influencing store-traffic or in-store purchase decisions. We utilize the penetration and frequency distinction in Figure 1 to organize our predictions (see Table 1) about how the effectiveness of specific marketing actions differs by category. Promotion Featured items in Best Food Day ads and other store circulars are the most popular method used by retailers to influence out-of-store purchase decisions and to drive store and hence category traffic (Walters and McKenzie 1988; Mulhern and Leone 1990). The extent of feature activity in a category is therefore likely to increase category volume, especially in staples and to some extent in variety enhancers and niches too because these categories are either important to most customers and/or are purchased by them every week. Features are less likely to affect category volume in fill-in

9 categories that have neither the broad appeal nor the purchase frequency of the other category groups (Fader and Lodish 1990). Display activity, on the other hand, enhances relative in-store visibility, drawing consumer attention to the category and therefore positively influencing in-store purchase decisions. This effect is most likely to be observed in fill-in categories with low penetration and frequency and to a lesser extent in variety enhancers and niches, since in these categories retailers tend to allocate less shelf space on a day-to-day basis. Staple categories typically already have a significant shelf-presence, and so display activity is less likely to affect CDI. Price Lower prices, whether regular or promotional, also should have a positive impact on category performance. However, the heavy user effect (e.g., Hoch et al 1995) would predict that price sensitivity and promotion responsiveness are greater in categories where more consumers purchase the category more often, specifically staples and to a lesser extent variety enhancer and niche categories. Fill-in categories should be the least influenced by price and promotion. Since temporary-shelf price reductions (TPRs) are only communicated in the store and therefore do not influence store traffic (Walters and McKenzie 1988), they should be most effective in categories where lots of consumers are likely to see them and possibly make an opportunistic purchase on that shopping trip like in staples and variety enhancers. TPRs should have little impact in low traffic categories that shoppers do not normally visit and in fact with frequent discounting the retailer may simply be giving away margin without getting much in return. Although price sensitivity is likely to be high in staples, it also the case that all retailers know this and therefore are more likely to work extra hard to make sure that they post competitive everyday prices to avoid being saddled with a high price image that might hurt store traffic. Therefore, it is possible that a floor effect reduces variability in staples prices and limits the returns to

10 even lower everyday prices. Assortment The breadth (number of brands) and depth of assortment (number of stock-keeping units) offered in a category helps a retailer to cater to the heterogeneous tastes of their customers. Offering more variety should help a retailer to both attract more consumers to visit the store and the category as well as induce them to make purchases when in the store. Therefore, in general, we expect that retailers who offer a larger number of brands and deeper assortments within each brand should achieve better category performance. Moreover, if the retailers category sales are more evenly distributed across brands (i.e., lower variance/less concentration in brand market shares), the effective number of brands that a consumer perceives will be greater, suggesting to the shopper base that the retailer is more effective at meeting all of their needs (Hoch, Bradlow, and Wansink 1998; van Helpern and Pieters 2000). However, in staple categories, due to the high extent of product proliferation in the last three decades, investment in assortment may provide smaller returns since both breadth and depth of assortment arguably have reached saturation levels (Drze, Hoch, and Purk 1994). This is very much evidenced in the significant efforts made by major consumer packaged goods manufacturers (e.g., P&G in staples like bath tissue) to cut back both on the number of brands as well as the number of skus offered per brand. Recent research on assortment management (Broniarcyzk, Hoyer and McAlister 1998) also suggests that reductions in assortment do not affect consumer intentions to purchase at a store especially when the deleted brands are not favorites Private Label Program Finally, an important part of any retailers assortment is their private label program. Some retailers (e.g., HEB, Wegmans) are successful in using their store brands as a means of differentiation,

11 helping to drive store traffic and increase loyalty due to unique identification with the store. Moreover, store brands allow the retailer to better meet the needs of a growing value-conscious segment of customers. Therefore, we expect a strong private label presence to contribute to better category performance. This effect is likely to be stronger in staple categories that are purchased frequently by a large cross-section of consumers as compared with fill-ins, niches or variety enhancers. Table 1 Predicted Impact of Key Determinants of CDI Influence on CDI + + + + + + Category-Specific Differences fill-ins<others staples, variety enhancers>others fill-ins<others fill-ins>others staples<others staples<others staples<others staples>others

Key Determinants Price and Promotion Regular Price TPR Feature Advertising Display Activity Assortment Depth of Assortment Breadth of Assortment Brand ShareVariance Store Brand Market Share

Summary. Table 1 summarizes our predictions. The second column in Table 1 shows the predicted impact of each variable on CDI and the third column summarizes how the effect might vary across staples, variety enhancers, niches, and fill-ins. For example, in the first row of Table 1, we hypothesize a negative relationship between regular price and CDI and note that this negative impact will be the lowest in fill-ins. Other rows may be similarly interpreted. The Study We utilized a variety of different data sources. A large packaged goods firm provided us with the main database --- syndicated sales data from AC Nielsen for 19 major product categories in their key retail accounts. The category definition is the same as that used by the packaged goods

12 manufacturer for decision-making purposes. Before reporting the results we present a detailed description of the data sources and measures developed. Since the effectiveness of the marketing decision variables might differ based on whether the categories were staples, variety enhancers, niches or fill-ins, we use a random coefficient approach to pool categories into these more homogenous types. Data Sources We use account data for 106 major U.S. grocery retail chains in the 50 Nielsen SCANTRACK markets.1 This includes all retailers with average annual store sales of at least $2 million. These retail chains account for 60% of total supermarket sales in the markets they serve.2 For each retail account, the data set includes monthly brand level sales, price, and consumer and retail promotion information for 19 categories over three calendar years. The categories represent a wide range of both edible grocery and dairy products including major categories such as yogurt, coffee, and cheese, and more minor ones like rice and syrup.3 Nielsen brand totals are used, aggregating all sizes and forms of each brand name. Separate brand totals are reported for each distinct variant; for example, in cereal Kellogg's Corn Flakes and Kellogg's Raisin Bran are reported as separate brands. All store brands are aggregated into one total even if the retailer carries multiple lines.4 The sponsoring firm also provided us with account level information about product facings for each of the categories. This information is collected annually as part of the store audit process. We obtained detailed trading area information from Spectra, Inc. Instead of relying on raw demographic

Each SCANTRACK area covers a designated number of counties, with an average of 30 and a range from 1 to 68. All markets include central city, suburban and rural areas. 2 While the average store volume varies across competing chains, we expect the variation to be considerably less than that if the data were to include smaller retailers (whose annual sales are less than $2 million dollars). As a result the predictor variable coefficients across retailers may be more homogenous in our data set as compared to those from a data set that also includes the smaller retailers. 3 For proprietary reasons, we are unable to divulge all the categories analyzed or report the individual category

13 characteristics, we utilized the data services estimate of category potential for each of the retail accounts which is based on both household level panel data on category consumption matched up to trading area consumer characteristics. Retailer ACV share data for each SCANTRACK market was obtained from Market Scope, a well-known source published annually by Trade Dimensions, a unit of the Progressive Grocer Data Center. Measures Used in the Analysis We developed all measures at the retailer-market level. This means that Kroger in Cincinnati is treated as a separate entity from Kroger in Cleveland. We created yearly aggregates for each measure resulting in three data points per retailer per category. We do so for purposes of stability and the fact that about half of our measures are only available at the annual level. Category Development Index (CDI). We calculate the CDI dependent variable in two ways: (1) on an equivalent unit basis as laid out in Eq. 1; and (2) on a dollar basis, where dollar sales are substituted for unit sales in the same equation. Unit and dollar CDIs were calculated using two different definitions of the total market: (a) national U.S.; and (b) local SMSA market level. Since results were similar irrespective of the definition used, we present results for the U.S. measure only. Pricing (REGPRICE, TPR). Two variables capture different aspects of the retailers pricing policy. REGPRICE is the share weighted average non-promoted price of all items in the category. TPR (temporary price promotion) measures the percent of category volume sold with only a temporary instore price discount (of at least 5%) and no accompanying feature advertising or display. TPR discounts typically are of small magnitude (5-15% deals).

results. 4 The sponsoring firm performed this aggregation to the brand level and we do not have access to upc level data.

14 Retail Sales Promotion (FEATURE, DISPLAY). The extent of retailer initiated sales promotion is described by two variables: the percent of volume sold with feature (FEATURE); and percent of volume sold with any form of display (DISPLAY). Depth, Breadth, and Variance in Assortment (BREADTH, DEPTH, BRNDVAR). BREADTH is the average number of distinct brand names carried by the retailer, a measure of breadth. DEPTH captures the extent of category specific item proliferation and is measured by the average number of stock-keeping units (sku's) per brand carried in that category by a particular retail chain. This measure characterizing depth of assortment was created from another database providing more detailed item level information. A third variable, BRNDVAR, is simply the variance in market shares across brands. Private Label Market Share (PLSHARE). This variable measures the penetration level of the retailers own store brand in each category. Market share is calculated on a unit basis as the ratio of total pound sales for the private label to total pound sales for the whole category. Control Variables (POTENTIAL, FACINGS, RETCOMP, ADVPROMO). We included four control variables in our analysis. POTENTIAL is a proprietary variable developed and estimated by Spectra, Inc. They first use IRI panel data to estimate consumption rates for 54 segments of consumers formed from a multi-way classification based on urban/suburban, geographic region, age, and the presence of children. They then use the composition of the trading area to calculate the percent of population falling into each segment, compute a weighted average of the segment consumption rates, and then standardize by the national averages. Although there may be better ways to control for demand side differences across retailers, here we assume that Spectra has done a much better job than we ever could starting from scratch. And in fact we found that POTENTIAL worked much better than

15 the underlying demographics. FACINGS was calculated as the average total linear feet per store allocated to each category, and was estimated from store audit data by the sponsoring firms sales force. RETCOMP represents the level of local competition that the retailer faces. We utilized the Herfindahl index that is calculated as the sum of the squared market shares. When the retailer faces many similarly sized competitors, the Herfindahl index is smaller and competition is more intense. This measure was constructed using local market data obtained from Progressive Grocers annual Market Scope report. Print and magazine advertising and consumer promotion information is available for all 50 markets (ADVPROMO). Manufacturer investments in their respective brands (and collectively in the category) vary considerably across markets to take advantage of regional differences. Because consumer promotions typically focus on price whereas advertising focuses on attributes, we initially separated these two promotions; but because of multicollinearity (r=.92) and the fact that consumer promotions make up only 14% of spending, we could not estimate separate effects. Consequently, we combined advertising and promotion into a common ADVPROMO measure equal to the sum of number of all newsprint and magazine impressions and consumer promotion averaged across all national brands, and then normalized by the number of households in the market. Model Development and Estimation Since slope coefficients are unlikely to be homogenous across all the 19 categories, in this section we consider methods of pooling information across categories. This will help not only in getting precise estimates of the common pattern but also better summarize central tendencies in the data. Our earlier conceptual development suggested that CDI responsiveness to marketing mix decision variables may vary across categories due to differences in consumer motivations in making purchases in a

16 category i.e., depending on whether they are classified as staples, variety enhancers, niches or fill-ins. We use same-period penetration and frequency estimates for the 19 categories in our dataset from the IRI Factbook, rank-order the categories in terms of both penetration and frequency and perform a median split to classify them into staples, niches, variety enhancers and fill-ins.5 Our classification yielded 7 staples (e.g., coffee), 3 niches (e.g., yogurt), 4 variety enhancers (e.g., rice) and 5 fill-ins (e.g., pancake mix). We specify a regression model for each of the categories in our dataset. This model can be represented as part of a multivariate regression system making the following assumptions and using the following notation: CDIc = X c + ct; c=1,2,19; t=1,2,N (1)

In addition, consistent with a fixed effects specification of a model with slope heterogeneity, we make the following assumptions: c = fill-ins + staples Dstaples + enhancersDenhancers + niches Dniches ; c=1,2,19 (2)

where, Dj = 1 if c belongs to group j and is 0 otherwise, j{staples, enhancers, niches}. The error structure is assumed to be as follows: E[ c] = 0; E[ c2] = 2; E[ ct ds] = 0 if cd or ts,

CDIc is the vector of the CDI values for the cth category expressed in terms of deviations from the mean CDI across retail accounts in a given category. X is the N k matrix of values for the k independent variables, also expressed in terms of deviations from the within-category mean across retail accounts, where N is the number of retail accounts. Notice that retailer specific characteristics might

A cluster analysis based on frequency and penetration also yielded four clusters while a few categories were misclassified the classification based on the cluster analysis and the median-split approach used in the paper were

17 simultaneously determine CDI and some of the explanatory variables. A test for endogeneity (Hausmann, 1978) revealed that facings, assortment related variables (breadth, depth and heterogeneity in shares) and private label share are endogenously related to CDI.6 Since, this can result in erroneous estimates, we endogenize FACINGS, DEPTH, BREADTH, BRNDVAR and PLSHARE and replace them with instruments. We created instruments using the standard two-stage least squares approach, first regressing the current values of the endogenous variables versus the one-year lags and then using the predicted values.7 The instrumented variables do not remain constant from year to year. c is the cth category slope coefficient vector and c is the Nx1 vector of error terms for the cth category. Mean centering by category of both the dependent and independent variables is equivalent to removing the category specific intercepts. Consequently, our analysis seeks to explain the variation in CDI across retail accounts as opposed to identifying factors that might explain differences in mean CDI across categories. In equation (2), fill-ins represents the mean slope coefficients of the fill-in group of categories; j represents the deviation of mean slope coefficients of the group of categories to which category c belongs (j), from fill-ins. Simply put the slope coefficient of any category c, is the mean slope coefficient of the group of categories to which it belongs, which in turn is measured as a deviation from the mean slope coefficient of the fill-in group of categories. We also assume that (a) the error terms are

largely similar. Because of the ease of interpretation of results we adopt the median-split approach to classify the categories into different groups. 6 Technical details on the endogeneity tests may be found in the Appendix. 7 Hausmanns (1978, Johnston 1984) test procedure was used to show that we need to endogenize these variables. Furthermore, in order to show that appropriate instruments were chosen, we selected alternate instruments for the endogenous variables. Since the planning horizon for facings (FACINGS), product stocking (BREADTH, DEPTH, BRNDVAR) and private label policy (PLSHARE) are likely to be well within a year, the further we go back in time to select alternate instruments, the less likely they are to be contemporaneously related with the current error term. Since, two years is the maximum that our dataset allows us to go back to, we use two-stage least squares to create alternate instruments using two year lagged values. Two year lag instruments are not significant in a regression

18 independent across retail accounts within a given category; and (b) that the errors across categories for the same retail account are independent. Our careful selection of the relevant independent variables justifies the first assumption. The second assumption is verified by an analysis of the correlation matrix of the errors across individual category regressions. Formal restriction of the pooling restriction assuming that the slope coefficients are the same across the category regressions is rejected. This is not surprising given large differences in within category variation in CDI. As noted earlier, we use a fixed-effects model to estimate the mean coefficient for the categories in the four sets. The procedure is similar to that used in the pooling literature (Bass and Wittink 1975, 1978; Hoch, Kim, Montgomery and Rossi 1995). We assume that the category coefficient vectors for each of the four sets of categories are drawn from independent super population distributions with different means and variances. This implies that the category coefficient vector within a set, j, has a population mean ( fill-ins + j), j{staples, enhancers, niches}. To characterize the central tendency or commonality within each set, we would like to infer the means for each set. Using the procedure in Hoch et al. (1995), we use the entire data set to obtain consistent estimates of the mean slope coefficient vectors for each of the sets of categories. The estimated asymptotic variance-covariance matrix is used to determine the significance level of the mean slope coefficient vectors. Tables 2 and 3 display the mean slope coefficient vectors for the four sets of categories using both equivalent unit and dollar revenue CDI. Several general observations about the two tables are in order before we go into a discussion of the detailed results. First, the pattern of significant coefficients highlights the fact that the key drivers of category performance differ depending on the role that the category plays in both the consumers and retailers portfolio. Second, with a couple of key exceptions

containing one-year lags (Spencer and Berk 1981), and so we feel comfortable using one-year lagged instruments.

19 (price and display), the coefficients for the equivalent unit CDI analysis are qualitatively similar to that found for the analysis of dollar revenue CDI. This suggests that an analysis of CDIs based on gross profit would yield similar results. The reason for this is that gross profit equals the product of unit sales and gross margin (price variable cost); if we assume that the variable costs of securing product are comparable across retailers, then a CDI based on gross profit will be monotonically related to unit and dollar CDIs. Of course, it is quite possible that incremental operating costs in a category differ significantly (e.g., WalMart vs a traditional supermarket) across retailers, and so CDIs based on operating margins may not be as closely related to unit and dollar CDI. Let us first consider the influence of regular prices and temporary price reductions on unit and dollar CDI. Table 2 shows that lower regular prices lead to higher unit sales performance (a negative coefficient) in all category types, but the effect is much greater in variety enhancers and niches (ps<.05). An examination of the results using the dollar CDI measure, however, shows that the variety enhancer and niche categories may not be price elastic enough to justify charging lower regular prices, as the REGPRICE coefficients flip from negative to positive (with a significant positive effect in the case of variety enhancers). As expected, temporary price reductions increase CDI only in high-traffic staples. This may be due to the fact that price discounts that are communicated in-store cannot drive traffic to a category and only staples experience enough passive traffic to benefit from the opportunistic purchases that small discounts might prompt. In fact, the significant negative TPR coefficient for niches in Table 3 suggests that TPRs are potentially a waste of promotional resources for niches. that are likely to generate store and category traffic have a significantly stronger influence on CDI in variety enhancers and niches than in fill-ins as expected. The differences in the coefficients for the feature and display variables are quite interesting. In

20 both Tables 2 and 3, the pattern of coefficients for FEATURE show that feature ads can influence store or category traffic; feature ads have their biggest impact in staples which have both high penetration and high frequency. Niches and variety enhancers, which have either high penetration or high frequency, also benefit from the store and category traffic building effect of feature ads, though the coefficients are significantly less than that for staples (ps<.05). However, in low penetration-low frequency fill-ins, feature ads do not have the ability to attract much attention or interest and therefore have no significant influence on CDI. Display has its biggest impact in fill-in categories. We attribute this difference to the fact that ancillary, off-shelf display in fill-ins leads to a dramatic increase in space and exposure relative to the regular shelf set which in turn can prompt consumers to make opportunistic purchases that ordinarily would not occur. Staples also benefit from display, but because these categories already start off with larger everyday shelf sets, the retailer is more likely to experience diminishing returns with extra display space. Display does not seem to improve performance very much in variety enhancers and niches and in fact the negative (though not statistically significant) coefficients for dollar CDI may indicate that the display effect on unit sales is not great enough to compensate for the large price discounts that typically accompany in-store display. An examination of the assortment variables indicates that, even after controlling for the number of facings allocated to a category, increasing the breadth and depth of assortment has a positive effect on unit and dollar CDI for variety enhancers, niches and fill-ins. This probably is due to the fact that larger assortments are better able to meet the heterogeneous needs of the retailers customer base. Staples do not benefit from increases in assortment. This may be a ceiling effect because most retailers already have large assortments in these key categories. The significant negative DEPTH coefficient for staples suggests that too large an assortment can actually be detrimental. The positive coefficients for

21 the FACINGS variable also indicates that larger assortments can enhance visibility and draw attention to a category influencing in-store consumer purchases. In addition, a more even distribution of market shares (lower variance between brands) among the brands helps to increase the perceived number of brands or breadth of assortment offered to consumers influencing their purchases in fill-ins, niches, and staples. In contrast, retailers achieve higher CDIs in variety enhancers when brands sales are more concentrated within fewer brands. We have no ready explanation for this findings, though it holds up for both unit and dollar CDI. Finally, a strong private label program (as represented by a higher private label market share) exerts a significant, positive influence on CDI in all four sets.

22 Table 2 Pooled Analysis across Categories (Equivalent Units) Variable Price and Promotion REGPRICE TPR FEATURE DISPLAY Assortment BREADTH DEPTH BRNDVAR PLSHARE Control Variables POTENTIAL FACINGS RETCOMP ADVPROMO a p<0.01, bp<0.05 Staples -0.0200a 0.5838a 1.6629 a 1.2895 a 0.0049 -0.0606 a -1.3029 0.5689 a 0.0061 a 0.0026 a 1.1295 a -0.000016 b Variety Enhancers -0.4752 a -0.0183 0.9616 a 0.6649 a 0.0452 a 0.0459 a 5.4897 a 0.4501 a 0.0117 a 0.0053 a 0.7404 a 0.000132 Niches Fill-Ins

-0.5282 a -0.0387 a -0.2227 -0.0751 1.1070 a 0.1402 0.2487 1.8644 a 0.0760 a 0.0257 a -4.4497 0.2177 b 0.0255 a 0.0239 a -3.0097 a 0.5477 a

0.0085 a 0.0009 0.0037 a 0.0026 a 0.8286 a 1.3454 a -0.000049 0.000002

Table 3 Pooled Analysis across Categories (Dollar Revenue) Variable Price and Promotion REGPRICE TPR FEATURE DISPLAY Assortment BREADTH DEPTH BRNDVAR PLSHARE Control Variables POTENTIAL FACINGS RETCOMP ADVPROMO
a

Staples -0.0059 0.4025 a 1.0562 a 0.9998 a 0.0126 a -0.0275 -1.8285 b 0.2820 a 0.0066 a 0.0021 a 1.1098 a -0.000005

Variety Enhancers 0.1652 a -0.2505 0.3856 -0.1501 0.0401 a 0.0360 a 3.6510 a 0.4467 a 0.0112 a 0.0049 a 0.7188 a 0.000120

Niches

Fill-Ins

0.0600 -0.0072 a -0.4499 a -0.0910 0.7143 a 0.1182 -0.3059 1.2348 a 0.0732 a 0.0151 a 0.0235 a 0.0161 a -7.1188 b -1.3753 b 0.1222 0.1735 b 0.0097 a 0.0032 a 0.9802 a -0.000013 0.0043 a 0.0026 a 0.5490 a 0.000001

p<0.01, bp<0.05

23 Discussion and Concluding Remarks This study helps to understand why retailers sell more or less than their fair share in a category. While this information is important to retailers interested in improving category performance, this is also useful to manufacturers wanting to identify retailers with best practices with whom they want to make specific marketing investments e.g., account-specific promotions. Our analysis helps to determine which specific actions on pricing (regular price), promotion (TPR's, feature, display) and merchandising (national brand assortment - depth or breadth, variance in share, private label strength) help to increase both unit and dollar CDI's. Although lower prices, more promotion, and larger assortments generally improve performance in all categories, the critical drivers of performance systematically vary depending upon the role the category plays in the retailer and consumers portfolios. The key insights are as follows: 1. We find that merchandising variables as a group play a significant role in affecting CDI. Given recent efforts by major national brand manufacturers like P&G in reducing assortment, our results suggest that reductions in the breadth (number of brands) and depth (number of SKU's in terms of size, type of package, flavor) of assortment may meet with resistance in many product categories due to their positive impact on category performance. Our analysis suggests that retailers should be more accepting of reductions in assortment in staples categories where assortment has reached saturation levels and reductions are unlikely to get noticed as much as in niches, variety enhancers and fill-ins. It also appears that retailers should be willing to reduce fringe brands (which lead to higher brand share variance) that are not doing a good job in catering to heterogeneous tastes helps both retailers (based on our results) and manufacturers. 2. This becomes even more important in the context of our finding that a strong store brand program not only leads to higher unit CDI's but also higher dollar CDI's for a retailer. In contrast to conventional wisdom suggesting that retailers rely on national brand assortment to build store traffic (and hence category volume), our results are indicative of the key role a retailer's store brand can play in increasing primary demand for the category, even to the extent of overcoming the lower revenues that may arise from intra-category switching of consumers from higher priced national brands to lower priced store brands. This is further highlighted by retailer's interest in stressing store brands in their product portfolios through multi-tiered offerings e.g., premium versus value private labels (Progressive Grocer Annual Survey 1997). This role of private labels in building store-traffic and hence category volume is much more likely to matter in staple categories which have both a

24 broad-based appeal and constitute an important part of a consumers shopping budget. 3. A lower category price plays an important role in increasing retailer unit category sales and to some extent dollar revenues too in variety enhancer and niche categories. While we expected price to also matter in staple categories, our data suggest that regular price elasticities are low in staples, as competing retailers seem to have similar prices in these important traffic-building categories. In contrast small in-store discounts in the form of TPRs improve performance only in staples and actually have a significant negative impact on dollar revenue (and likely on profits too). 4. Our analysis supports the contention that feature advertising helps build store and hence category traffic while display influences category volume by leading to opportunistic in-store purchasing. Feature based promotions are more effective in increasing CDI in traffic-building staple categories and to some extent in niches and variety enhancers too, while display promotions enhances category volume more so in low-visibility fill-in categories in which displays help to significantly enhance shelfspace relative to the regular shelf-set. Price promotions, without any display or feature support, are mainly effective in high-traffic staples and variety enhancers where they can increase volume by causing opportunistic in-store purchasing. Despite providing a number of interesting insights, our study has several limitations that provide an opportunity for future research. While we focus on the CDI measure due to its popularity in the consumer packaged goods area, for certain decisions like determining a retailers return on investment for a category or in other retailing contexts like apparel retailing in which maximizing space utilization is crucial, other candidate measures of category performance might become more relevant e.g., sales per square or linear foot. It is therefore important in future research to build on our work using other candidate category performance measures to assess the generalizability of our findings. While the standardization of the CDI measure provides a lot of benefits for benchmarking and comparison purposes, it is somewhat limiting in that it does not allow us to determine the impact of retailer decisions whose impact do not vary by category e.g., store price image. Furthermore, while our data is quite comprehensive, it did not provide enough information to incorporate aspects of broader store positioning strategy into our analysis that are not captured by the marketing actions that are included in this study, something that would be very interesting to include in future research.

25 Our analysis suggests that examining the role of store brands to drive store traffic (and hence category volume) may be a worthwhile direction for future research in this area. In the case of retailerowned store brands, the store name is often used as an umbrella over several categories. This combined with the offering of premium quality private labels may help in differentiating a retailer and draw customers to their stores. The President's Choice brand sold by the Loblaw chain in Canada is a case in point on this issue. More research also needs to be conducted in the burgeoning area of category management. An understanding of the decision process that retailers use to make assessments of the trade-offs between demand generation and cost containment in practicing category management is an important area for future research. While our analysis shows that retailers clearly benefit from practicing category management, future research needs to examine the benefits to national brand manufacturers who often collaborate with retailers. Complementing our CDI analysis with an analysis of brand development index (BDI) may also help to provide further managerial insights for national brand manufacturers.

26 References Bass, Frank M. and Dick R. Wittink (1975), Pooling Issues and Methods in Regression Analysis with Examples in Marketing Research, Journal of Marketing Research, 12, 4, 414-425. Bass, Frank M. and Dick R. Wittink (1978), Pooling Issues and Methods in Regression Analysis with Examples in Marketing Research: Some Further Reflections, Journal of Marketing Research, 15, 2, 277-279. Blattberg, Robert C., Edward J. Fox, and Mary E. Purk (1995), Category Management: A Series of Implementation Guides, Food Marketing Institute, Vols. I-IV. Bolton, Ruth (1989), The Relationship between Market Characteristics and Promotional Price Elasticities, Marketing Science, 8 (Spring), 153-169. Broniarczyk, Susan M., Wayne D. Hoyer, and Leigh McAlister (1998), "Consumers' Perceptions of the Assortment Offered in a Grocery Category: The Impact of Item Reduction," Journal of Marketing Research, 35 (May), 166-176. Dhar, Sanjay K. and Stephen J. Hoch (1996), Price Discrimination Using In-Store Merchandising, Journal of Marketing, 60, January, 17-30. Dhar, Sanjay K. and Stephen J. Hoch (1997), Why Store Brand Penetration Varies by Retailer, Marketing Science, 16, 3, 208-227. Drze, Xavier, Stephen J. Hoch, and Mary E. Purk (1994), Shelf Management and Space Elasticity, Journal of Retailing, 70 (Winter), 301-326. Fader, Peter S. and Leonard M. Lodish (1990), A Cross-Category Analysis of Category Structure and Promotional Activity of Grocery Products, Journal of Marketing, 54 (October), 52-65. Gupta, Sunil (1988), Impact of Sales Promotions on When, What, and How Much to Buy, Journal of Marketing Research, 25 (November), 342-355. Hausmann, J. A. (1978), Specification Tests in Econometrics, Econometrica, 46, 6, 1251-1271. Hoch, Stephen J., EricT. Bradlow and Brian Wansink (1999), The Variety of an Assortment, Marketing Science, 18 (4), 527-546. Hoch, Stephen J., Byung-Do Kim, Alan L. Montgomery and Peter E. Rossi (1995), Determinants of Store-Level Price Elasticity, Journal of Marketing Research, 32, 1, 17-29. Hoch, Stephen J. and Leonard M. Lodish (1998), Store Brands and Category Management, working

27 paper #98-012, Wharton School Marketing Department, University of Pennsylvania. Johnson, Maureen (1999), From Understanding Consumer Behavior to Testing Category Strategies, Journal of Marketing Research Society, 41, 3, 259-288. Johnston, Jack (1984), Econometric Methods, New York: McGraw-Hill. Kumar, V. and Robert P. Leone (1988), Measuring the Effect of Retail Store Promotions on Brand and Store Substitution, Journal of Marketing Research, 25 (May), 178-185. Litvack, David S., Roger J. Calantone and Paul R. Warshaw (1985), An Examination of Short-Term Retail Grocery Effects, Journal of Retailing, 3, 61, 9-25. Mulhern, Frank, and Robert P. Leone (1990) "Retail Featured Price Promotions: The Impact of the Number of Deal Items and Size of Deal Discounts on Store Performance," Journal of Business Research, November 1990, Vol. 21 (3), . Nielsen Category Management: Nielsen Marketing Research, Northbrook, IL, 1992. Raju, Jagmohan S. (1992), The Effect of Price Promotions on Variability in Product Category Sales, Marketing Science, 11, 2 (Summer), 207-220. Seventh Annual Survey of Trade Promotion Policies: AC Nielsen, 1997 Spencer, David E. and Kenneth N. Berk (1981), A Limited Information Specification Test, Econometrica, 49, 4, 1079-1086. Supermarket News (1997) Wlaters, Rockney G., and Scott B. MacKenzie (1988). A Structural Equation Analysis of the Impact of Price Promotions on Store Performance, Journal of Marketing Research, 25 (1), 51-63. White, Hal (1980), A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity, Econometrica, 50, 483-499.

28

Appendix
Endogeneity Tests Retailer specific characteristics might simultaneously determine CDI and some of the explanatory variables. To test whether a variable x* is endogenously related to CDI we adopted the following approach. Under the null hypothesis that x* is exogenous we compute the least squares estimator; denote that as b LS. Under the alternative we use an appropriate instrument for x* and re-estimate the model by replacing x* with its corresponding instrument; denote this estimator as b IV. Following Hausman (1978) the test statistic, w = (b LS b IV){Var[b LS] Var[b IV]}-1(b LS b IV)

is the Wald statistic based on the difference of the two estimators and has a degree of freedom of 1. We use the time series nature of the data to test and address possible endogeneity problems. We created instruments using the standard two-stage least squares approach, first regressing the current values of the endogenous variables versus the one-year lags and then using the predicted values as an instrument for the variable in question.8 Tests for endogeneity, conducted in the above manner, revealed that facings, assortment related variables (breadth, depth and heterogeneity in shares) and private label share are endogenously related to CDI. Consequently, we endogenize FACINGS, DEPTH, BREADTH, BRNDVAR and PLSHARE and replace them with their corresponding instruments. Since we lose the first year of data when computing the instruments, we use data for two years for the estimation. Pooling tests show no differences in the values of the estimated coefficients across the two years in any of the category regressions. Therefore, we run regressions by pooling the data across the two years. The results reported in the paper are based on the relationship between CDI and these instruments, other exogenous variables and controls.

In order to show that appropriate instruments were chosen, we selected alternate instruments for the endogenous variables. Since the planning horizon for facings (FACINGS), product stocking (#BRANDS, DEPTH, BRNDVAR) and private label policy (PLSHARE) are likely to be well within a year, the further we go back in time to select alternate instruments, the less likely they are to be contemporaneously related with the current error term. Since, two years is the maximum that our dataset allows us to go back to, we use two-stage least squares to create alternate instruments using two year lagged values. Two year lag instruments are not significant in a regression containing one-year lags (Spencer and Berk 1981), and so we feel comfortable using one-year lagged instruments.