Sunteți pe pagina 1din 12

Journal of Counseling Psychology 2009, Vol. 56, No.

1, 44 55

2009 American Psychological Association 0022-0167/09/$12.00 DOI: 10.1037/a0013498

Within-Group Differences in Sexual Orientation and Identity


Roger L. Worthington
University of Missouri

Amy L. Reynolds
University at Buffalo

The purpose of this investigation was to examine within-group differences among self-identified sexual orientation and identity groups. To understand these within-group differences, 2 types of analysis were conducted. First, a sample of 2,732 participants completed the Sexual Orientation and Identity Scale. Cluster analyses were used to identify 3 types of bisexual women, 3 types of bisexual men, 2 types of heterosexual women, and 2 types of gay men. Lesbians and heterosexual men were clustered into singular groupings. In addition, a subsample of 838 participants also completed measures of sexual identity development and lesbian, gay, and bisexual knowledge and attitudes. Within-group differences were found for bisexual men and women, gay men, and heterosexual women along these dimensions. Findings supported the existence of distinct subgroups of bisexual men, bisexual women, gay men, and heterosexual women. Keywords: sexual orientation, sexual identity, sexual identity development, homonegativity, cluster analysis

Since Kinsey (Kinsey, Pomeroy, & Martin, 1948; Kinsey, Pomeroy, Martin, & Gebhard, 1953) first published his influential works on the sexual behavior of men and women, there has been a great deal of scholarly debate about the nature of sexual orientations and identities. Sexual orientation refers to ones actual sexuality related predispositions (e.g., arousal, attraction, fantasy), whereas sexual orientation identity refers to ones conscious recognition, identification, and self-labeling (e.g., gay, lesbian, bisexual, queer, heterosexual) with respect to ones sexual predispositions (e.g., Broido, 2000; Chung & Katayama, 1996; Diamond, 2000; Diamond & Savin-Williams, 2000; Weinberg, Williams, & Pryor, 1994; Worthington & Moreno, 2005). Self-identification has become increasingly complicated as outdated terms are discarded and new terms are added to the lexicon of an expanding list of possible identities (e.g., heteroflexibility, metrosexual, bicurious, polyamorous, queer). Nevertheless, content analyses of the research on lesbian, gay, and bisexual (LGB) issues in psychology have indicated that the most common method of assessing the sexual orientation of participants is to request self-identification as a gay man, a lesbian, a bisexual, or a heterosexual (or some variation), often by forced-choice formats (Morin, 1977; Phillips, Ingram, Smith, & Mindes, 2003). This method provides categorical self-identification, which is only a global proxy variable for the cognitive, behavioral, emotional, and physiological bases underlying sexual orientation. The Kinsey Scale, the original assessment tool for sexual orientation, also has been criticized for di-

chotomizing heterosexuality and homosexuality along a single dimension and promoting the notion of same-sex and other-sex attractions as bipolar opposites rather than as multifaceted and multidimensional (Sell, 1997). As a result, potential within-group variations in sexual identity and behavior have been neglected in human sexuality research due to limitations arising from problematic assessment of target variables (McConaghy, 1999; Sell, 1997). The purpose of this investigation was to examine within-group differences among self-identified sexual orientation groups. This investigation has implications for both theory and measurement of sexual orientation and identity.

The Nature and Variety of Sexual Orientations


A host of different authors have explicated the complexities associated with demarcating the boundaries between sexual orientation groups (e.g., Broido, 2000; Kinsey et al., 1948; Kinsey et al., 1953; Klein, 1993; Phillips et al., 2003; Sell, 1997; Weinrich & Klein, 2003; Worthington & Moreno, 2005). Worthington and Moreno (2005), in an effort to further our understanding of sexual orientation and identity, demonstrated that orientation to men and orientation to women are related but independent constructs that should be measured separately from heterosexual identity and LGB identities. As inquiry regarding LGB issues increasingly relies on constructivist notions about the nature of sexual orientation and sexual identity (e.g., Broido, 2000), ideas about how to define sexual orientations and identities are becoming more complex. Thus, measurement of the various constructs that converge on sexual orientation identity is a formidable task (Sell, 1997). An example of the inability to fully understand the complexity of sexual orientation is evident in the research on bisexuality. Despite the fact that Kinsey and colleagues (1948; 1953) long ago advanced the notion that bisexuality, in terms of behavior and attraction, was much more common than previously expected, the social and scientific treatment of bisexuality has often been disparaging. Commonly held assumptions regarding the gaystraight
44

Roger L. Worthington, Department of Educational, School, and Counseling Psychology, University of Missouri; Amy L. Reynolds, Department of Counseling, School, and Educational Psychology, University at Buffalo. We thank Matthew V. Moreno for his contribution to the data collection for this research. Correspondence concerning this article should be sent to Roger L. Worthington, Office of the Deputy Chancellor, 217 Jesse Hall, University of Missouri, Columbia, MO 65211. E-mail: worthingtonr@ missouri.edu

SPECIAL ISSUE: CLUSTER ANALYZING SEXUAL ORIENTATION AND IDENTITY

45

binary have been particularly pernicious. In fact, current definitions of biphobia incorporate the outdated notion that all people must or do view their sexual orientations in bipolar terms (Eliason, 1997; Mohr & Rochlen, 1999). For example, two commonly held beliefs regarding bisexuality are that it is either (a) a transitional phase of sexual identity development within which many persons become stuck and unable to fully achieve a lesbian or gay identity or (b) a safe middle ground for same-sex attracted individuals who can pass for heterosexual while attempting to avoid the stigma and discrimination associated with minority status and protect privileges associated with majority status (Eliason, 1997; Klein, 1993; Mohr & Rochlen, 1999; Weinberg et al., 1994). As is true of all stereotypes, these assumptions tend to minimize within-group differences while manufacturing or exaggerating between-groups differences. Furthermore, it has been estimated that many bisexual research participants might be misidentified as gay, lesbian, or heterosexual in human sexuality research because a substantial proportion of behaviorally bisexual individuals might label themselves as heterosexual, gay, or lesbian (e.g., Kinsey et al., 1948; Masters & Johnson, 1979) or are presumed by researchers to belong to one of the other groups (Phillips et al., 2003). In a recent study, Rieger, Chivers, and Bailey (2005) suggested that male bisexuality might not exist. They drew this conclusion after finding that bisexual men in their study were not equally sexually aroused by both men and women. Although the researchers advertised for participants who were either gay, bisexual, or heterosexual (e.g., self-labeled), they were identified in the study as belonging to one of these groups on the basis of Kinsey Scale scores (sexual attraction and behavior), not on the basis of selfidentification. As a result, 8% of their participants were reclassified into a different sexual orientation grouping (i.e., 2% gay men, 14% bisexuals, and 7% self-identified heterosexuals; G. Rieger, personal communication, July 25, 2005). After eliminating between 30% and 34% of their participants in each group for failure to become measurably aroused in a laboratory setting, the researchers reported that those identified as bisexual (on the basis of Kinsey Scale scores) showed patterns of arousal only after exposure to sex between two men (e.g., similar to gay men) or only after exposure to sex between two women (e.g., similar to heterosexual men), but not in response to both of these conditions. Their findings might be an artifact of several factors: the extent to which self-identification was discrepant from categorizations resulting from Kinsey Scale ratings, the rate of attrition associated with men in each group failing to achieve arousal to either form of stimulus, or the use of arousal stimuli that included only homoerotic content. Furthermore, the 68 participants in their study could be classified into seven different groups: (a) self-identified gay men classified as gay men, (b) self-identified gay men classified as bisexual men, (c) self-identified bisexual men classified as bisexual men, (d) self-identified bisexual men classified as gay men, (e) selfidentified bisexual men classified as heterosexual men, (f) selfidentified heterosexual men classified as heterosexual men, and (g) self-identified heterosexual men classified as bisexual men. Thus, due to problems of conceptualization and measurement, the Reiger et al. study seems to obscure rather than clarify bisexuality in men. In their influential research on bisexuality, Weinberg et al. (1994) point out that becoming bisexual involves the rejection of not one but two recognized categories of sexual identity (p. 26). They describe a stagewise model of bisexual identity development

that traverses experiencing initial confusion, finding and applying the label, settling into the identity, and experiencing continued uncertainty, which indicates that a substantial amount of bisexual identity development involves confusion, exploration, and uncertainty. Nevertheless, although larger proportions of their bisexual research participants expressed ongoing and past uncertainty about self-identification than did heterosexuals, lesbians, and gay men, the vast majority of bisexuals expressed comfort and certainty with their bisexual identities. Furthermore, this research demonstrates several important aspects of bisexuality that counteract stereotypes: (a) bisexuality is a unique and legitimate identity, (b) substantial external pressures to conform to the gaystraight binary may result in substantial confusion, exploration, and uncertainty, and (c) there are important within-group differences among bisexual individuals that have critical influences on sexual identity development. In another study of sexual orientation, Weinrich and Klein (2003) used cluster analysis with the Klein sexual orientation grid (KSOG) to identify 10 distinct groups in a large sample of gay, lesbian, bisexual, and heterosexual participants. When participant orientation, sexual attraction, behavior, and fantasies are classified, emotional preferences, socializing, lifestyle, and self-identification are equally considered on the KSOG. In addition, on the KSOG, participants are asked to consider past, present, and ideal when ranking all of these other variables. Clustering was conducted separately by participant gender but resulted in identical fivecluster outcomes: (a) heterosexual participants, (b) bistraight participants, (c) bi bi participants, (d) bi gay/bilesbian participants, and (e) lesbian and gay male participants. Similar to Weinrich and Klein (2003), Weinberg et al. (1994) identified several different types of bisexuality that included the pure, mid, heterosexual-leaning, homosexual-leaning, and varied types. Weinrich and Klein (2003) provided an analysis of the means and standard errors for each cluster (by gender) to demonstrate the probable validity of these groupings. These researchers also promised a follow-up study that would provide data correlating these clusters with variables not used in the clustering process, which was never conducted (F. Klein, personal communication, July 7, 2005). Thus, unfortunately, the findings of Weinrich and Klein (2003) cannot be assumed to be definitive until (a) future research replicates their findings and (b) there is some evidence that the subgroups they found were not artifacts of measurement based solely on the KSOG (e.g., that the subgroups differ in some other meaningful ways on important variables). Nevertheless, their initial findings indicate that on the basis of their multidimensional method of measuring sexual orientation, there may be subgroups of bisexual men and women who closely resemble heterosexuals, gay men, or lesbians yet who self-identify as bisexual. Furthermore, a substantial number of scholars have suggested that there are a variety of ways in which self-identified heterosexual, gay, and lesbian individuals might exhibit bisexual behavior, cognitions, or emotions without identifying as bisexual (e.g., Diamond, 2000, 2003; Diamond & Savin-Williams, 2000; Gonsiorek & Weinrich, 1995; Klein et al., 1985; McConaghy, Buhrich, & Silove, 1994; Storms, 1980). Thus, whereas Rieger et al. (2005) suggested that male bisexuality may not exist (at least in terms of sexual arousal in a tightly controlled laboratory setting), the broader trend in the literature on the nature of sexual orientations

46

WORTHINGTON AND REYNOLDS

and identities suggests that there may be more sexual orientation subgroups than is commonly assumed. Models of sexual identity development provide an additional perspective about the nature and variety of sexual orientation identities (e.g., Fassinger & Miller, 1996; McCarn & Fassinger, 1996; Worthington, Savoy, Dillon, & Vernaglia, 2002). Arguably, the most advanced developmental models of lesbian and gay identity are those of McCarn and Fassinger (1996) and Fassinger and Miller (1996). Their models describe both individual and social processes of identity development within four statuses: awareness, exploration, deepening and commitment, and internalization and synthesis. Similarly, Worthington et al. (2002) advanced a model of heterosexual identity development that was designed to incorporate many of the same processes and statuses of the sexual minority identity models of Fassinger and colleagues (Fassinger & Miller, 1996; McCarn & Fassinger, 1996). Using the Marcia (1987) model of identity development, Worthington et al. (2002) hypothesized that sexual identity could be conceptualized along the dimensions of exploration and commitment. They further asserted that individual identity development occurred in relation to six dimensions of human sexuality: sexual needs, sexual values, modes of sexual expression, preferred characteristics of sexual partners, preferred sexual activities and behaviors, and sexual orientation identity. Thus, one unique aspect of the Worthington et al. (2002) model is the recognition that sexual orientation identity is only one of several dimensions of sexual identity. In addition, Worthington et al. (2002) identified two components of social identity: attitudes toward sexual minorities and group membership identity. Bieschke (2002) suggested that the Worthington et al. (2002) model could be advanced and extended to apply to individuals of any sexual orientation identity, which was subsequently accomplished via scale development research (Worthington et al., 2008). Although these models have not been conceptually applied to bisexual individuals, research on the development of the Worthington et al. (2008) measure provided evidence for validity with people of any sexual orientation identity. Homonegativity may help to explain a portion of the withingroup differences in the variations between orientation and identity among bisexual, gay, lesbian, and heterosexual individuals (Kashubeck-West, Szymanski, & Meyer, 2008). For many years, heterosexual attitudes toward lesbians and gay men have been conceptualized along a single, bipolar, cognitive continuum ranging from condemnation to tolerance (Herek, 1984, 1998). However, Hudson and Ricketts (1980) and Logan (1996) argued that homonegativity is multidimensional, with three components: cognitive (attitudinal), affective, and behavioral. Furthermore, Worthington, Dillon, and Becker-Schutte (2005) found five different dimensions of heterosexual knowledge and attitudes toward LGB individuals: Hate, Religious Conflict, Civil Rights Attitudes, Internalized Affirmativeness, and LGB Knowledge. They found that the most commonly used unidimensional measure of homophobia (i.e., Herek, 1984) corresponds most closely to attitudes regarding LGB civil rights and only moderately correlates with the other dimensions. Worthington et al. (2005) used the scale to demonstrate that although there were differences between LGB and heterosexual individuals on all five dimensions measured by the scale, effect sizes were very large for Religious Conflict, Internalized Affirmativeness, and LBG Knowledge, but effect sizes were small for Civil Rights Attitudes and Hate. These find-

ings support the use of the scale for persons of any sexual orientation identity group and indicate that the scale could help distinguish between members of different sexual orientation identity groupings, especially on the basis of Religious Conflict, Internalized Affirmativeness, and LGB Knowledge. The purpose of this investigation was to examine within-group differences among self-identified sexual orientation identity groups. This research was intended to build on the work of three important studies (i.e., Diamond, 2008; Weinberg et al., 1994; Weinrich & Klein, 2003), which, through divergent methodologies, identified different types of sexual orientation identities. Because these prior investigations used differing instrumentation, research designs, and samples, in the current study we did not make any a priori hypotheses about the nature or numbers of clusters that would result from our research. Using cluster analysis, we sought to identify subgroups within groups of self-identified lesbians, gay men, bisexual men and women, and heterosexual women and men. Once subgroups were produced via statistical procedures, we sought to examine the viability of those subgroups on the basis of between-groups differences on a new set of variables. Specifically, to what degree would clusters differ on the extent to which participants within each cluster responded to measures assessing sexual identity development processes, knowledge regarding LGB individuals, and attitudes regarding LGB individuals? In addition, how would the cluster groupings from the current data compare with those of Weinrich and Klein (2003)?

Method Participants
There were 2,372 total participants, including 1,189 women (50.1%) and 1,183 men (49.9%). Ages ranged from18 years to 89 years (M 32.79, SD 12.94). Participants self-identified as 349 bisexual women (14.7%), 349 bisexual men (14.7%), 319 lesbians (13.4%), 440 gay men (18.5%), 521 heterosexual women (22.0%), and 394 heterosexual men (16.6%). There were 89 African American (3.8%), 60 Asian/Pacific Islander (2.5%), 75 biracial/ multiethnic (3.2%), 1,793 European American (75.6%), 106 Latino/a (4.5%), 95 international/non-U.S. citizen (4.0%), 21 Native American Indian (0.9%), and 133 other (5.6%) individuals. The largest percentage of participants categorized their educational level as some college (n 968, 40.8%), a 4-year college degree (n 566, 23.9%), or masters degree (n 432, 18.2%), whereas smaller numbers reported the following categories: less than high school diploma (n 25, 1.1%), high school diploma (n 121, 5.1%), doctoral degree (n 171, 7.2%), medical or law degrees (MD, JD), or other professional degree (n 61, 2.6%), and other (n 28, 1.2%). The largest percentage of participants indicated that their annual income was less than $20,000 (n 674, 28.4%), with the remainder distributed across the remaining categories as follows: between $20,000 and $39,999 (n 405, 17.1%), between $40,000 and $59,999 (n 290, 12.2%), between $60,000 and $79,999 (n 371, 15.6%), between $80,000 and $99,999 (n 226, 9.5%), and $100,000 and above (n 280, 11.8%), with some not reporting income (n 126, 5.3%).

SPECIAL ISSUE: CLUSTER ANALYZING SEXUAL ORIENTATION AND IDENTITY

47

Instruments
The Sexual Orientation Identity Scale (SOIS; Worthington & Moreno, 2005). The SOIS was developed with a combination of item types and response scales from four earlier models (e.g., Kinsey et al., 1948, 1953; Klein et al., 1985; Shively & DeCecco, and Storms,1980). The instrument is a 14-item, self-report measure of sexual orientation identity that contains the following four subscales: Orientation to Females, Orientation to Males, Heterosexual Identity, and LGB Identity. Items have a Likert-type scale ranging from 1 (never) to 6 (always). Higher scores on each subscale are indicative of higher levels of the construct measured by that subscale. Orientation items are divided in two sets so that participants can rate themselves independently for men and women on the basis of the following dimensions: sexual arousal, sexual attraction, sexual fantasies, sexual behavior, and romantic relationships. Identity items were also divided into two sets to assess internal and external dimensions of identity as heterosexual and LGB. Examples include the following items: To what extent do you experience sexual attraction to males? To what extent do you experience sexual attraction to females? To what extent do you inwardly identify as gay, lesbian, or bisexual? and To what extent do you outwardly identify as heterosexual? Results of the scale development study with exploratory and confirmatory factor analysis with men and women of a variety of different sexual orientation identities indicated that the scale was measuring two dimensions of sexual orientation identity: orientation and identity. These two factors were further divided conceptually to establish the four different subscales. The SOIS subscales were highly correlated in the total sample; however, when subsamples (e.g., lesbians, gays, bisexuals, heterosexuals) were examined, the correlations were moderate. Orientation to Females correlated highly with Orientation to Males (r .83) as did heterosexual identity and LGB identity (r .93). However, when subscale correlations are examined by subgroup, these strong correlations are attenuated (e.g., the correlations between Orientation to Males subscale and Orientation to Females subscale drops to r .20 and r .37 for bisexual men and women, respectively, and the correlations between heterosexual and LGB identity subscales drops to r .41 and r .56 for heterosexual women and men, respectively. This is keeping with the conceptualization and design of the SOIS, which was intended to distinguish or predict group differences. Worthington & Moreno (2005) reported the following findings as evidence for the validity of the scale: (a) significant correlations between all four subscales and the Kinsey Scale, (b) nonsignificant correlations with a measure of social desirability, and (c) very high correct classification rates with the SOIS subscales in discriminant function analysis to predict self-identified sexual orientation identity in two separate samples. The internal consistency estimates for the current study on the Orientation to Females, Orientation to Males, Heterosexual Identity, and LGB Identity subscales were as follows: .96, .96, .91, and .90. LGB Knowledge and Attitudes Scale for Heterosexuals (LGBKASH; Worthington et al., 2005). The LGB-KASH is a 28-item, self-report measure of heterosexual attitudes toward LGB individuals that contains the following five subscales: LGB Civil Rights (5 items), Internalized Affirmativeness (5 items), LGB Knowledge (5 items), Hate (6 items), and Religious Conflict (7 items). Items are rated on a Likert-type scale ranging from 1 (very uncharac-

teristic of me or my views) to 6 (very characteristic of me or my views). Higher scores on each subscale are indicative of higher levels of the construct measured by that subscale. Examples of items include the following: I think marriage should be legal for same-sex couples (LGB Civil Rights); Feeling attracted to another person of the same sex would not make me uncomfortable (Internalized Affirmativeness); I am knowledgeable about the history and mission of the PFLAG organization (LGB Knowledge); I sometimes think about being violent toward LGB people (Hate); and I keep my religious views to myself in order to accept LGB people (Religious Conflict). Worthington et al. (2005) reported evidence for the validity of the scale via findings of hypothesized relationships with social dominance orientation, sexual identity development, homophobia and biphobia, age, gender, and sexual orientation identity. They report internal consistencies from .73 to .94 for the various subscales in three separate studies. Although the scale was originally developed to measure heterosexual attitudes, Worthington et al. (2005) successfully administered the scale to a sizable sample of LGB participants and compared the scores obtained from that sample with those of a heterosexual sample, finding large effect sizes for differences between heterosexual participants and LGB participants on Internalized Affirmativeness, Religious Conflict, and LGB Knowledge. Significant differences were also found between groups on the Hate and LGB Civil Rights subscales, but with smaller effects sizes. The internal consistency estimates for the current sample were as follows: Hate ( .59), LGB Knowledge ( .92), LGB Civil Rights ( .87), Religious Conflict ( .72), and Internalized Affirmativeness ( .83) subscales. Measure of Sexual Identity Exploration and Commitment (MoSIEC; Worthington, Navarro, Savoy, & Hampton, 2008). The MoSIEC was developed to measure sexual identity development within the framework described by Worthington et al. (2002). The instrument contains 22 items designed to measure four dimensions of sexual identity development: Exploration (8 items), Commitment (6 items), Sexual Orientation Identity Uncertainty (3 items), and Synthesis (5 items). Items are rated on a Likert-type scale ranging from 1 (very uncharacteristic of me) to 6 (very characteristic of me). Higher scores on each subscale are indicative of higher levels of the construct measured by that subscale. Examples of items include the following: I am actively trying to learn more about my own sexual needs (Exploration), I have a clear sense of the types of sexual activities I prefer (Commitment), My sexual orientation is not clear to me (Sexual Orientation Identity Uncertainty), and The ways I express my self sexually are consistent with all of the other aspects of my sexuality (Synthesis). Worthington et al. (2005) reported evidence for the validity of the scale via findings of hypothesized relationships with sexual conservatism, sexual self-monitoring, sexual assertiveness, awareness of sexual appeal, age, gender, and sexual orientation identity. They report internal consistencies from .74 to .89 for the various subscales. Cronbachs alphas for the current investigation were .89, .84, .73, and .83 for the Exploration, Commitment, Sexual Orientation Identity Uncertainty, and Synthesis subscales, respectively.

Procedures
Data for the current study were obtained from two sources: 828 cases of data were collected specifically for the purpose of this

48

WORTHINGTON AND REYNOLDS

investigation, and 1,544 cases of data from an existing investigation were used (Worthington & Moreno, 2005). These two groups of cases were combined and used to provide a large sample for cluster analysis. Participants were recruited via two methods: (a) online participation via e-mail announcements posted to lesbian, gay, bisexual, and transsexual (LGBT) resource-center community listservs at a large number of major universities across the United States and Canada and on Internet bulletin boards serving the bisexual community and (b) classroom administration of the survey to students at a large Midwestern university. For the classroom administration, students were given paper copies of the instruments. Caveats of Internet-based data collection are the possibilities that participants can submit their completed surveys more than once and that web-based data collection is susceptible to malicious or random responding. As recommended by Schmidt (1997), Smith and Leigh (1997), and Mohr and Rochlen (1999), duplicate surveys were identified with the date, time, and origin of submission or Internet protocol address. In the event that duplicate surveys appeared to be submitted accidentally from the same Internet protocol address (e.g., two identical cases submitted within a minute or 2), one survey from the pair of duplicate surveys was eliminated from the data set. Schmidt noted that World Wide Web based survey methodologies are particularly susceptible to respondents who intentionally supply incorrect survey data to undermine the research. This danger may be especially great when conducting LGB research because of the pervasive societal intolerance about LGB issues. Furthermore, because the survey-taking environments of Internet users are highly variable, respondents may supply incorrect data because of inattentiveness and distractions. In addition, we used several survey items to check for random responding (e.g., Please do not respond to this item, Please click the button at the far right of the scale). We examined the demographics of all cases in the sample to identify identical demographic data (e.g., age, gender, race or ethnicity, sexual orientation identity, religious affiliation). Identical procedures for data collection via the Internet were used for both samples, except that there were differing combina-

tions of instruments offered. Specifically, the earlier investigation offered the SOIS in combination with either one or both of the Balanced Inventory of Desirable Responding (Paulhus, 1991) and the Kinsey Scale (along with demographic items), whereas the current data collection effort included the measures listed in the Instruments section. The survey Web page also was posted on two public-access Internet sites containing links to psychological research studies. The survey Web page provided an informed consent page, followed by the survey questionnaire. Prospective participants were told that the goal of this project was to investigate the quality of a new instrument to measure sexual orientation identity. Debriefing information was contained at the end of the survey questionnaire. There were no incentives offered to participants for participation.

Results Preliminary Analysis


Preliminary analyses were conducted to examine whether there were any significant differences on important demographic variables (e.g., race, gender, age, sexual orientation, education, and income) as well as outcomes on the SOIS subscales for the two samples combined for this study (N 838; N 1,894). Analysis revealed no significant relationships among samples and race or gender and no significant difference between the samples on participant age, Orientation to Females, and Orientation to Males. There were significant differences between the two samples on income, educational level, Heterosexual Identity, and LGB Identity, but the effect sizes for these differences were small, F(1, 2245) 98.66, p .001, 2 p .042; F(1, 2 2367) 3.96, p .05, 2 p .002; F(1, 2371) 29.21, p .001, p 2 .012; and F(1, 2372) 26.96, p .001, p .011, respectively. Given that the overall sample size was so large for this study (N 2,732) and the effect sizes were small, these significant differences were not enough to rule out combining the two subsamples into one larger sample. A bivariate correlation matrix for the central study variables is provided in Table 1. Results of the bivariate correlations for the

Table 1 Correlation Matrix for Subscales of the Sexual Orientation and Identity Scale With Subscales of the Measure of Sexual Identity Exploration and Commitment and the LGB-Knowledge and Attitudes Scale
Variables Exploration Commitment Synthesis Uncertainty Hate Knowledge Civil Rights Conflict Affirmativeness Orientation to Males .132* .039 .014 .115* .105* .088 .129* .020 .092* Orientation to Females .115* .006 .021 .036 .151* .124* .206* .056 .219* Heterosexual Identity .189* .141* .236* .147* .225* .661* .453* .343* .720* LGB Identity .257* .145* .231* .189* .244* .657* .472* .339* .754*

Note. LGB lesbian, gay, and bisexual; Exploration Exploration subscale of Measure of Sexual Identity Exploration and Commitment (MoSEIC); Commitment Commitment subscale of MoSEIC; Synthesis Synthesis subscale of MoSEIC; Uncertainty Sexual Orientation Identity Uncertainty subscale of MoSEIC; Hate Hate subscale of LGB-Knowledge and Attitudes Scale (LGB-KASH); Knowledge LGB Knowledge subscale of LGB-KASH; Civil Rights LGB Civil Rights subscale of LGB-KASH; Conflict Religious Conflict subscale of the LGB-KASH; Affirmativeness Internalized Affirmativeness subscale of LGB-KASH. * p .01.

SPECIAL ISSUE: CLUSTER ANALYZING SEXUAL ORIENTATION AND IDENTITY

49

total sample indicated that there were a number of moderate to high correlations between the subscales of the SOIS and those of the MoSIEC and LGB-KASH, especially with respect to the Heterosexual Identity and LGB Identity subscales. These correlations indicate that sexual identity development and LGB knowledge and attitudes were likely to be good predictors of variations in sexual orientation and identity.

Cluster Analysis
This research focused on identifying clusters of sexual orientation identities based on Orientation to Women, Orientation to Men, LGB Identity, and Heterosexual Identity. Cluster analysis of SOIS scores was used to identify sexual orientation identity subgroups. Cluster analysis consists of a family of statistical techniques that group objects with respect to a set of defined attributes such that subgroups are distinguished by heterogeneity (maximal variance between subgroups) and homogeneity (minimal variance within subgroups). Wards (1963) method of cluster analysis was chosen because of its wide use in social science research (Aldenderfer & Blashfield, 1984; Borgen & Barnett, 1987) and because of its superior accuracy compared with other clustering methods (Blashfield, 1976; Milligan, 1981). The Ward method, which is a hierarchical type of cluster analysis, was designed to optimize the minimum variance within clusters by joining clusters that result in the minimum increase in within-cluster variance. According to this method, a cluster is defined as a group of cases in which variance among the members is relatively small. Cluster formation was based on squared Euclidean distance. Establishing a meaningful protocol for conducting the cluster analysis in advance contributes to the success of the analysis (Everitt, 1980). Specifically, there is a variety of different approaches to how cluster analysis could be applied to the entire sample or subsamples on the basis of gender or sexual orientation identity groups. Conducting a single cluster analysis with the entire sample would likely result in clusters containing both men and women of varying self-identified sexual orientation identities. Alternatively, because the SOIS has been demonstrated to more effectively classify respondents into preexisting self-identified sexual orientation identity groups when gender is controlled (Worthington & Moreno, 2005), one approach might be to conduct separate cluster analyses for men and women (cf., Weinrich & Klein, 2003). However, two factors indicated that a different approach was warranted: (a) Conducting analyses by gender without regard for sexual orientation identity would be likely to result in clusters that were homogenous for gender but heterogeneous with respect to self-identified sexual orientation identity (e.g.,

intermingling self-identified gay and bisexual men, as in Rieger et al., 2005), and (b) the potential for lower classification rates for the SOIS found in earlier research could be controlled if self-identified sexual orientation identity was held constant. Thus, separate cluster analyses for each of the major sexual orientation identity groups (bisexual, gay and lesbian, and heterosexual) were conducted with the expectation that the resulting clusters could be divided by gender post hoc. Thus, three identical cluster analyses were conducted with each of the 14 item ratings as entry variables (cf., Weinrich & Klein, 2003), specifying a range of possible solutions resulting in three to seven clusters for each self-identified sexual orientation identity grouping. An a priori decision was made to include only clusters that included at least 15% of the sample (c.f., Rochlen, Milburn, & Hill, 2004). As a result of applying this criterion, cluster solutions with more than three groups were eliminated. None of the solutions was viable beyond 3 clusters for any of the self-identified sexual orientation identity subsamples (i.e., solutions of 4 clusters or more produced very small clusters that were unlikely to be reliably reproduced in future research; n 50). The results of each 3 cluster solution (by gender) are presented in Table 2. Together, these solutions produced 12 different clusters of sexual orientation identities by gender: 3 clusters of bisexual women, 3 clusters of bisexual men, 1 cluster of lesbians, 2 clusters of gay men, 2 clusters of heterosexual women, and 1 cluster of heterosexual men. Means and standard deviations on the SOIS subscales for each group are presented in Table 3. A total of 18 cases (0.7%) of gay men, lesbians, and heterosexual women and men were misclassified as a result of the cluster analyses and dropped from further analyses. Bisexual women. The first cluster of bisexual women (Group 1, n 160) was characterized by moderate and relatively equal Orientation to Males and Females, low Heterosexual Identity, and very high LGB Identity. The second cluster of bisexual women (Group 2, n 112) was characterized by moderately high Orientation to Males, moderate Orientation to Females, and moderate and relatively equal Heterosexual Identity and LGB Identity. The third cluster of bisexual women (Group 3, n 66) was characterized by moderately high Orientation to Females, moderate Orientation to Males, moderate Heterosexual Identity, and moderately high LGB Identity. Bisexual men. The first cluster of bisexual men (Group 4, n 56) was characterized by moderately high Orientation to Males, moderate Orientation to Females, low Heterosexual Identity, and very high LBG Identity. The second cluster of bisexual men (Group 5, n 89) was characterized by moderate and relatively

Table 2 Results of Cluster Analyses for Self-Identified Sexual Orientation Identity Groups by Gender
Sexual orientation and gender Bisexual Cluster 1 2 3 Women 160 112 66 Men 56 89 192 Lesbian and gay Women 307 0 2 Men 3 242 173 Heterosexual Women 346 152 8 Men 3 2 368

50

WORTHINGTON AND REYNOLDS

Table 3 Means and Standard Deviations for SOIS Subscale Scores by Cluster Group
Orientation to Females Cluster group Bisexual women Group 1 Group 2 Group 3 Bisexual men Group 4 Group 5 Group 6 Lesbians Group 7 Gay men Group 8 Group 9 Heterosexual women Group 10 Group 11 Heterosexual men Group 12 Note. n 160 112 66 56 89 192 307 242 173 346 152 368 M 3.97 3.53 4.74 3.75 3.48 5.00 5.40 1.35 1.38 1.35 2.34 5.37 SD 0.83 0.70 0.60 1.15 0.80 0.52 0.52 0.37 0.42 0.34 0.78 0.64 Orientation to Males M 4.40 4.57 3.00 4.63 4.16 3.36 1.67 5.64 5.52 5.32 5.26 1.31 SD 0.74 0.72 1.00 0.76 0.68 0.87 0.55 0.39 0.53 0.71 0.52 0.61 Heterosexual Identity M 1.83 3.72 2.31 1.88 3.99 3.86 1.38 1.15 2.07 5.88 5.23 5.66 SD 0.97 0.99 1.14 1.06 0.87 1.43 0.55 0.34 0.77 0.28 0.91 .67 LGB Identity M 5.43 3.74 4.82 5.42 3.39 3.59 5.54 5.85 4.90 1.18 1.88 1.32 SD 0.60 0.93 1.06 0.61 0.87 1.25 0.54 0.29 0.70 0.34 0.84 0.58

N 2,263. SOIS Sexual Orientation Identity Scale; LGB lesbian, gay, and bisexual.

equal Orientation to Males and Females, and moderate and relatively equal Heterosexual Identity and LGB Identity. The third cluster of bisexual men (Group 6, n 192) was characterized by moderately high Orientation to Females, moderate Orientation to Males, and moderate and relatively equal Heterosexual Identity and LGB Identity. Lesbians. The single cluster of lesbians (Group 7, n 307) was characterized by very high Orientation to Females, very low Orientation to Males, very low Heterosexual Identity, and very high LGB Identity. Three cases of self-identified gay men also were classified in this group by cluster analysis and were subsequently dropped from further analyses. Gay men. Both clusters of gay men (Group 8, n 242; Group 9, n 173) were characterized by very low Orientation to Females, very high Orientation to Males, very low Heterosexual Identity, and very high LGB Identity. However, Group 9 had slightly higher Heterosexual Identity and slightly lower LGB Identity than did Group 8. Two cases of self-identified lesbians also were classified into Group 9 by cluster analysis and were subsequently dropped from further analyses. Heterosexual women. Both clusters of heterosexual women (Group 10, n 346; Group 11, n 152) were characterized by low Orientation to Females, high Orientation to Males, high Heterosexual Identity, and low LGB Identity. However, Group 11 had slightly higher Orientation to Females, slightly lower Heterosexual Identity, and slightly higher LGB Identity. Three cases of heterosexual men also were classified into Group 10, and two cases of heterosexual men were classified into Group 11 by cluster analysis and were subsequently dropped from further analyses. Heterosexual men. The single cluster of heterosexual men (Group 12, n 368) was characterized by very high Orientation to Females, very low Orientation to Males, very high Heterosexual Identity, and very low LGB Identity. Eight heterosexual women

also were classified into this group by cluster analysis and were subsequently dropped from further analyses.

Comparisons on Sexual Identity Development and Homonegativity/Affirmativeness


There were 828 participants included in the cluster analyses who completed instruments measuring sexual identity development and LGB affirmativeness/homonegativity. A series of four one-way multivariate analyses of variance (MANOVAs) were conducted to evaluate the extent to which members of the groups formed by cluster analyses differed on these variables. In order to control for the familywise error rate, a high threshold for interpreting significant differences was established by requiring medium effect sizes (2 .08). Means and standard deviations for MoSIEC and LGB-KASH subscale scores by cluster group are presented in Table 4. Self-identified bisexual women. The one-way MANOVA for self-identified bisexual women, with the three cluster groups as the grouping variable and the four subscales of the MoSIEC (Exploration, Commitment, Synthesis and Sexual Orientation Identity Uncertainty) and the five subscales of the LBG-KASH (Hate, Religious Conflict, LGB Civil Rights, LGB Knowledge, and Internalized Affirmativeness) as the dependent variables, resulted in a significant omnibus test, Wilks (18, 202) .696, p .005, 2 .166. Univariate F tests for Sexual Orientation Identity Uncertainty (2 .105), LGB Civil Rights (2 .139), LGB Knowledge (2 .099), and Internalized Affirmativeness (2 .140) were also significant ( ps .005). GamesHowell post hoc comparisons showed that Group 2 (moderate and equal Orientation to Males and Females; moderate and equal Heterosexual Identity and LGB Identity) was higher on Sexual Orientation Identity Uncertainty and lower on LGB Civil Rights and LGB Knowledge

SPECIAL ISSUE: CLUSTER ANALYZING SEXUAL ORIENTATION AND IDENTITY

51

Table 4 Means and Standard Deviations for MoSIEC and LGB-KASH Subscale Scores by Cluster Group
Explore Cluster group Bisexual women Group 1 Group 2 Group 3 Bisexual men Group 4 Group 5 Group 6 Lesbians Group 7 Gay men Group 8 Group 9 Heterosexual women Group 10 Group 11 Heterosexual men Group 12 n 58 32 22 27 33 85 121 103 65 108 30 98 M 5.00 4.89 4.74 4.58 5.06 5.10 4.01 4.15 4.15 3.52 4.02 3.36 SD .85 1.12 .58 .94 .76 .68 1.09 1.09 1.07 1.38 1.13 1.29 Commit M 4.97 4.57 4.84 5.29 4.62 4.90 5.21 5.23 4.84 4.92 4.57 4.96 SD .93 1.07 .88 .80 .93 .88 .83 .71 1.00 .97 .86 .85 Synthesis M 4.98 4.66 5.03 5.28 4.46 4.88 5.31 5.35 5.02 4.83 4.82 4.85 SD 0.84 1.15 0.92 0.95 1.16 1.01 0.78 0.70 0.88 0.89 0.91 1.04 Uncertainty M 2.92 3.71 2.95 2.38 3.52 2.98 2.37 2.29 2.49 1.88 2.52 1.81 SD 1.04 1.12 0.90 0.68 0.98 0.86 0.83 0.69 0.85 0.78 0.97 0.89 M 1.16 1.14 1.15 1.11 1.36 1.30 1.12 1.10 1.42 1.20 1.18 1.61 Hate SD 0.42 0.35 0.31 0.28 0.48 0.51 0.35 0.24 0.57 0.45 0.38 0.73 Knowledge M 4.11 3.16 4.37 4.59 2.76 2.98 4.78 5.06 4.02 2.35 2.93 1.92 SD 1.30 1.69 1.39 1.39 .96 1.64 1.07 .93 1.37 1.40 1.24 1.11 Rights M 5.94 5.61 5.97 5.76 5.30 5.20 5.91 5.90 5.71 5.20 5.49 4.18 SD 0.18 0.68 0.09 0.77 0.99 1.24 0.49 0.27 0.49 1.28 0.73 1.55 Conflict M 1.55 1.81 1.61 1.67 2.42 2.18 1.53 1.67 2.27 2.07 1.72 2.61 SD 0.69 0.84 0.68 0.70 1.11 0.93 0.61 0.71 0.95 0.97 0.96 1.06 Affirm M 5.64 5.00 5.51 5.24 4.09 4.20 5.69 5.67 4.96 3.51 4.54 2.33 SD 0.46 0.93 0.80 0.93 0.97 1.20 0.47 0.48 0.96 1.26 1.13 1.05

Note. N 838. MoSIEC Measure of Sexual Identity Exploration and Commitment; LGB lesbian, gay, and bisexual. LGB-KASH LGB Knowledge and Attitudes Scale; LGB lesbian, gay, and bisexual; Explore Explore subscale of MoSEIC; Commit Commit subscale of MoSEIC; Synthesis Synthesis subscale of MoSEIC; Uncertainty Sexual Orientation Identity Uncertainty subscale of MoSEIC; Hate Hate subscale of LGB-KASH; Knowledge LGB Knowledge subscale of LGB-KASH; Rights LGB Civil Rights subscale of LGB-KASH; Conflict Religious Conflict subscale of LGB-KASH; Affirm Internalized Affirmativeness subscale of LGB-KASH.

than were the other two groups, and Group 2 was lower on Internalized Affirmativeness than was Group 3 (moderately high Orientation to Females, moderate Orientation to Males, and relatively equal Heterosexual Identity and LGB Identity). Self-identified bisexual men. The one-way MANOVA for selfidentified bisexual men, with the three cluster groups as the grouping variable and the four subscales of the MoSIEC and the five subscales of the LBG-KASH as the dependent variables, resulted in a significant omnibus test, Wilks (18, 268) .660, p .001, 2 .188. Univariate F tests for Sexual Orientation Identity Uncertainty (2 .155), LGB Knowledge (2 .169), and Internalized Affirmativeness (2 .129) were also significant ( ps .001). GamesHowell post hoc comparisons showed that Group 4 (moderately high Orientation to Males, moderate Orientation to Females, low Heterosexual Identity, and very high LGB Identity) was lower on Sexual Orientation Identity Uncertainty and higher on LGB Knowledge and Internalized Affirmativeness than were the other two groups. Group 5 (moderate and equal Orientation to Males and Females, moderate and equal Heterosexual Identity and LGB Identity) was higher on Sexual Orientation Identity Uncertainty than was Group 6 (moderately high Orientation to Females, very low Orientation to Males, relatively equal Heterosexual Identity and LGB Identity). Self-identified gay men. The one-way MANOVA for selfidentified gay men, with the two cluster groups as the grouping variable and the four subscales of the MoSIEC and the five subscales of the LBG-KASH as the dependent variables, resulted in a significant omnibus test, Wilks (9, 158) .725, p .001, 2 .275. Univariate F tests for Hate (2 .133), LGB Knowledge (2 .170), Religious Conflict (2 .114) and Internalized Affirmativeness (2 .197) were also significant ( ps .001).

Group 8 (low Orientation to Females, very high Orientation to Males, very low Heterosexual Identity, and very high LGB Identity) was lower on Hate and Religious Conflict and higher on LGB Knowledge and Internalized Affirmativeness than was Group 9 (low orientation to Females, very high Orientation to Males, very low Heterosexual Identity, and very high LGB Identity with slightly higher Heterosexual Identity and slightly lower LGB Identity than Group 8). Because of a low reliability (.59) for the Hate subscale, these significant findings should be interpreted with caution. Self-identified heterosexual women. The one-way MANOVA for self-identified heterosexual women, with the two cluster groups as the grouping variable and the four subscales of the MoSIEC and the five subscales of the LBG-KASH as the dependent variables, resulted in a significant omnibus test, Wilks (9, 128) .814, p .001, 2 .186. Univariate F tests for Sexual Orientation Identity Uncertainty (2 .094) and Internalized Affirmativeness (2 .108) were also significant ( ps .001). Group 10 (low Orientation to Females, high Orientation to Males, high Heterosexual Identity, and low LGB Identity) was lower on Sexual Orientation Identity Uncertainty and Internalized Affirmativeness than was Group 11 (low Orientation to Females, high Orientation to Males, high Heterosexual Identity, low LGB Identity with slightly higher Orientation to Females, slightly lower Heterosexual Identity, and higher LGB Identity than Group 10).

Discussion
The findings of this investigation partially replicated and substantially extended the research of Weinrich and Klein (2003), which used cluster analysis with the KSOG to produce 10 distinct

52

WORTHINGTON AND REYNOLDS

sexual orientation identity groups in a large sample of gay, lesbian, bisexual, and heterosexual participants. In addition to some similar outcomes, the two studies also produced outcomes that were not alike. For example, Group 3 bisexual women were somewhat more similar to lesbians on orientation to men and women than were the other two groups, potentially similar to the Weinrich and Klein grouping identified as bilesbian. However, Group 1 bisexual women were more similar to lesbians on LGB and heterosexual identity than were the other two groups, which could also be indicative of a bilesbian grouping. Group 2, on the other hand, was much more clearly consistent with the bi bi grouping identified by Weinrich and Klein, but this study did not produce results that reflected a hetero bi group of bisexual women. A somewhat different pattern emerged for bisexual men, in which Group 5 could be identified as a bi bi group, whereas Group 4 looked very similar to the clusters of gay men on LGB and heterosexual identities. However, Group 6 tended to be more similar to heterosexual men on orientation to men and women, despite being very similar to Group 5 with respect to LGB and heterosexual identities. Furthermore, the distinct groupings of heterosexual women and gay men appeared to be more similar to each other along SOIS dimensions than to any of the other cluster groupings. Nevertheless, both studies produced six different clusters of bisexual men and women, and the current study also produced separate clusters of gay men and heterosexual women. Lesbians and heterosexual men did not produce separate clusters in either study. There were some distinct and important differences in the procedures of the two studies that may account for differences in outcomes. Although both instruments contain similar dimensions hypothesized to underlie sexual orientation and identity, the SOIS has the advantage of items designed to tap four distinct factors (i.e., Orientation to Males, Orientation to Females, Heterosexual Identity, and LGB Identity), resulting in very high correct classification rates for self-reported sexual orientation identities (Worthington & Moreno, 2005). As a result, whereas Weinrich and Klein (2003) conducted two separate cluster analyses by gender, the high predictive validity of the SOIS allowed us to conduct three separate cluster analyses by sexual orientation identity grouping, in which gender groupings were conducted post hoc. This procedure was chosen in order to maximize the identification of within-group clusters and avoid cross-classifications among self-identified sexual orientation identity groups. An important advancement of the current study over that of Weinrich and Klein (2003) was that the current investigation assessed whether the cluster groupings varied along additional dimensions of sexual identity development and homonegativity/ affirmativeness. These findings provided strong support for the outcomes of the cluster analyses. For example, in addition to being higher on Heterosexual Identity and lower on LGB Identity than the other two groups of bisexual women, Group 2 was higher on Sexual Orientation Identity Uncertainty, lower on LGB Knowledge and Civil Rights attitudes, and lower than Group 3 on Internalized Affirmativeness. Thus, Group 2 appeared to be less certain of their commitment to their self-identified sexual orientation identity and more open to sexual identity exploration than were the other groups. Furthermore, in addition to reporting high LGB Identity and low Heterosexual Identity, Group 4 bisexual men also were lower on Sexual Orientation Identity Uncertainty and higher on LGB Knowledge and Internalized Affirmativeness

than were the other two clusters of bisexual men. However, despite very similar, moderate scores on LGB and Heterosexual Identities and stronger Orientation to Females than Orientation to Males (like heterosexual men), Group 6 bisexual men scored lower on Sexual Orientation Identity Uncertainty than did Group 5. Thus, among both bisexual men and women, the two groups most similar to Weinrich and Kleins bi bi groups were higher on Sexual Orientation Identity Uncertainty, reflecting a tendency to avoid commitment and engage in exploration of sexual orientation identity. For gay men, there were no differences between groups on the sexual identity development variables, but an interesting pattern of findings emerged with respect to homonegativity/affirmativeness. Specifically, gay men in Group 9 (who were slightly higher on Heterosexual Identity and lower on LGB Identity than Group 10) were lower on Internalized Affirmativeness and LGB Knowledge and were higher on Religious Conflict and Hate attitudes. The central differences between the two groups of gay men on the basis of SOIS scores were reflected in internalized homonegativity/ affirmativeness rather than on sexual identity development. Findings for heterosexual women indicated that in addition to differences in Orientation to Females and LGB Identity and Heterosexual Identity, Group 11 was higher than Group 10 on Sexual Orientation Identity Uncertainty as well as LGB Internalized Affirmativeness. That is, this group of heterosexual women had more favorable attitudes toward their own same-sex thoughts and feelings and was also more likely to express a tendency to avoid commitment and engage in exploration of sexual orientation identity. Descriptive statistics also produced interesting patterns of findings for sexual identity development and homonegativity/ affirmativeness among lesbians and heterosexual men. Group 7 (lesbians) and Group 12 (heterosexual men) tended to be both lower on sexual identity exploration and uncertainty and higher on commitment and synthesis than were the majority of other sexual orientation identity groups. Whereas lesbians were very high on LGB Knowledge and Internalized Affirmativeness, heterosexual men were very low on those variables and slightly higher than all of the other sexual orientation identity groups on Hate. These findings again support the outcomes of the cluster analysis and replicate earlier research that produced similar between-groups outcomes (e.g., Worthington et al., 2005, 2008).

Limitations
Several limitations of the present study should be noted. Some of the limitations center around measurement concerns. First, the associations observed in the present study are vulnerable to inflation because of the common method variance resulting from concurrent self-report data across the measures. As was discussed previously, there is much discussion in the literature about the most effective and meaningful ways to measure sexual orientation. The sexual orientation instrument, SOIS, used in this study assessed diverse factors that typically make up sexual orientation such as sexual behavior, fantasies, arousal, relationships, among others; however, participants were not asked to consider actual sexual relations within a given time period. It is possible that attention to explicit behavior within a specific time frame might change participant responses. An additional measurement concern

SPECIAL ISSUE: CLUSTER ANALYZING SEXUAL ORIENTATION AND IDENTITY

53

is the result of the low reliability (.59) of the Hate subscale of the LGB-KASH. Although it is likely that some constriction of range limited the reliability of the Hate subscale and previous research has resulted in higher reliability estimates for the Hate subscale (Worthington, Dillon, & Becker-Schutte, 2005), it is important that the findings be viewed with some caution. Despite having a large sample for this study, it was not feasible to split the sample and run a confirmatory cluster analysis. Providing this type of analysis could add meaning and value to the results. The existence of the six subgroups (gay men, lesbians, bisexual women, bisexual men, heterosexual women, and heterosexual men) would have resulted in small ns for each subgroup, thus making it difficult to split those subgroups into distinct clusters that were replicable. Aldenderfer and Blashfield (1984) suggested using replication with alternative samples to ascertain the generality and stability of a particular cluster structure; future research would benefit from such efforts. In addition to measurement, as is common with the vast majority of research involving sexuality research, self-selection bias in the sample could have played a role in the findings. Furthermore, the data collection method used via the Internet is susceptible to certain types of malicious responding that are not found in other forms of data collection (e.g., submission of multiple surveys, intentional fabrication of responses). Although several standard steps were taken to control for and eliminate these types of responses, it is impossible to know whether these efforts were completely effective, but the combination of a large sample size along with the data cleaning method would indicate that malicious responding would have little impact on the overall findings. Finally, it is critical to note that the SOIS is difficult to apply to transgender and intersex individuals because of wording that refers exclusively to men or women in the items. During data collection, a number of transgender and intersex individuals contacted Roger L. Worthington with concerns about the wording and their inability or refusal to participate, and the design of the research analyses required that transgender and intersex participants be excluded for statistical or conceptual reasons.

Implications for Research, Measurement, and Theory


Although this investigation partially replicated the earlier findings of Weinrich and Klein (2003), discrepant findings between the two studies need to be reconciled via additional research possibly with the use of both scales to classify subgroups of participants. Furthermore, replication in a large-scale populationbased sample will increase confidence in these outcomes. Nevertheless, findings from the current study, in combination with those of Weinrich and Klein suggest important implications for the measurement and interpretation of findings with respect to sexual orientation identity in future research. A corollary outcome of this investigation was that the data indicated that the SOIS could be used with substantial confidence in its psychometric properties; it will measure sexual orientation and identity with greater precision and reliability than earlier instruments have offered. The findings of this investigation also provide additional support for the independent measurement of identities apart from orientations. This is a critical advance in the measurement of sexual orientation identity in that, although the literature has reflected this important distinction for decades, ex-

tant instrumentation has not. Although it seems appropriate to continue simpler assessment methods (e.g., self-identification or the Kinsey Scale) when researchers are interested in simply reporting sexual orientation identity as a demographic variable, future research should take these important measurement issues into account when investigating sexual orientation identity as an independent or dependent variable. Although identity and orientation go hand in hand for most people, there is a substantial proportion of the population for which this may not be true (e.g., people with same-sex orientations prior to the coming out process). Furthermore, there are individuals for whom internal (experienced) and external (expressed) identities may be out of sync (e.g., a person who internally recognizes and acknowledges same-sex orientation yet prefers to express an outwardly heterosexual identity). Confusion of these two discrete dimensions of sexual orientation and identity have the potential to result in erroneous assumptions and claims in scholarship and professional practice (e.g., Nicolosi, 1991; Rieger et al., 2005; Spitzer, 2003). For example, it is likely that there is an incongruence between identity and orientation and/or internal versus external identity among individuals who seek sexual reorientation therapies or ex-gay ministries (e.g., Beckstead, 2003; Carlson, 2003; Cohen & Savin-Williams, 2003; Diamond, 2003; Hartmann, 2003; Herek, 2003; Hill & DiClemente, 2003; Strassberg, 2003; Vasey & Rendall, 2003; Worthington, 2003, 2004). In addition, there has been a longstanding debate about the prevalence rates of bisexuality in the population (e.g., that many people who identify as heterosexual, lesbian, or gay may in fact have a bisexual orientation or, alternatively, that people who identify as bisexual actually have exclusively other-sex or same-sex orientations). Contrary to recent assertions questioning the existence of bisexuality among men (Rieger et al., 2005), findings of the current study indicate that there may be as many as six distinct types of bisexuality, as well as two distinct types each for heterosexual women and gay men, which were determined on the basis of a multidimensional operationalization of sexual orientation and identity. These divergent results for the various sexual orientation groups further demonstrate the complexity of sexual orientation and identity. For example, different numbers of clusters were identified for major self-identification groups (e.g., heterosexual men and lesbians were only classified into one cluster each, whereas heterosexual women and gay men each had two clusters). In addition, there were distinct differences in how these groups responded to measures of identity development and attitudes toward LGB (e.g., some cluster groups of bisexual men and women and heterosexual women endorsed sexual orientation identity uncertainty in the MoSEIC, whereas others did not; one cluster of gay men endorsed lower levels of religious conflict and higher levels of internalized affirmativeness than did the other cluster). Although it is not possible to fully understand why these groups were different without further study, these results provide support for a constructivist conceptualization of sexual orientation and identity. Our views of sexual orientation have evolved across time and place (Broido, 2000). For example, although there have always been men who have been sexual partners with other men, gay identity did not fully exist pre-Stonewall. After many years of gay identity politics and AIDS, researchers and practitioners have had to acknowledge that there are men who sleep with men and who do not and may never claim a gay identity. As Fassinger and Arsenau

54

WORTHINGTON AND REYNOLDS prejudice against lesbians, gay men, and bisexuals. Thousand Oaks, CA: Sage. Hill, C. A., & DiClemente. (2003). Methodological limitations do not justify the claim that same-sex attraction changed through reparative therapy. Archives of Sexual Behavior, 32, 440 442. Hudson, W. W., & Ricketts, W. A. (1980). A strategy for the measurement of homophobia. Journal of Homosexuality, 5, 357372. Kashubeck-West, S., Szymanski, D., & Meyer, J. (2008). Internalized heterosexism: Clinical implications and training considerations. The Counseling Psychologist, 36, 615 630. Kinsey, A. C., Pomeroy, W. B., & Martin, C. E. (1948). Sexual behavior in the human male. Philadelphia: W. B. Saunders. Kinsey, A. C., Pomeroy, W. B., Martin, C. E., & Gebhard, P. (1953). Sexual behavior in the human female. Philadelphia: W. B. Saunders. Klein, F. (1993). The bisexual option (2nd ed.). Philadelphia: Haworth Press. Klein, F., Sepekoff, B., & Wolf, T. J. (1985). Sexual orientation: A multi-variable dynamic process. Journal of Homosexuality, 11, 35 49. Logan, C. R. (1996). Homophobia? No, homoprejudice. Journal of Homosexuality, 31, 3153. Marcia, J. E. (1966). Development and validation of ego identity status. Journal of Personality and Social Psychology, 5, 551558. Masters, W. H., & Johnson, V. E. (1979). Homosexuality in perspective. Boston: Little, Brown. McCarn, S. R., & Fassinger, R. E. (1996). Revisioning sexual minority identity formation: A new model of lesbian identity and its implications for counseling and research. The Counseling Psychologist, 24, 508 534. McConaghy, N. (1999). Time to abandon the gay/heterosexual dichotomy? Archives of Sexual Behavior, 34, 12. McConaghy, N., Buhrich, N., & Silove, D. (1994). Opposite sex linked behaviors and homosexual feelings in the predominantly heterosexual male majority. Archives of Sexual Behavior, 23, 565577. Milligan, G. W. (1981). A review of Monte Carlo tests of cluster analysis. Multivariate Behavioral Research, 16, 379 407. Mohr, J. J., & Rochlen, A. B. (1999). Measuring attitudes regarding bisexuality in lesbian, gay male, and heterosexual populations. Journal of Counseling Psychology, 45, 353369. Morin, S. F. (1977). Heterosexual bias in psychological research on lesbianism and male homosexuality. American Psychologist, 32, 629 637. Nicolosi, J. (1991). Reparative therapy of male homosexuality. Northvale, NJ: Jason Aronson. Paulhus, D. L. (1991). Measurement and control of response bias. In J. P. Robinson, P. R. Shaver, & L. S. Wrightsman (Eds.), Measures of personality and social psychological attitudes (pp. 1759). New York: Academic Press. Phillips, J. C., Ingram, K. M., Smith, N. G., & Mindes, E. J. (2003). Methodological and content review of lesbian-, gay-, and bisexualrelated articles in counseling journals: 1990 1999. The Counseling Psychologist, 31, 25 62. Rieger, G., Chivers, M. L., & Bailey, J. M. (2005). Sexual arousal patterns of bisexual men. Psychological Science, 16, 579 584. Rochlen, A. B., Milburn, L., & Hill, C. E. (2004). Examining the process and outcome of career counseling for different types of career counseling clients. Journal of Career Development, 30, 263275. Schmidt, W. C. (1977). World-wide web survey research: Benefits, potential problems, and solutions. Behavior Research Methods, Instruments and Computers, 29, 274 279. Sell, R. L. (1997). Defining and measuring sexual orientation: A review. Archives of Sexual Behavior, 26, 643 658. Shively, M. G., DeCecco, J. P. (1977). Components of sexual identity. Journal of Homosexuality, 3, 41 48. Smith, M. A., & Leigh, B. (1977). Virtual subjects: Using the Internet as an alternative source of subjects and research environment. Behavior Research Methods, Instruments, and Computers, 29, 496 505.

(2006), Diamond and Savin-Williams (2000), and others have suggested, our ability to understand and affirm the various expressions and self-conceptualizations of sexual orientation and identity is essential to being effective practitioners and researchers.

References
Aldenderfer, M. S., & Blashfield, R. K. (1984). Cluster analysis. Thousand Oaks, CA: Sage. Beckstead, A. L. (2003). Understanding the self-reports of reparative therapy successes. Archives of Sexual Behavior, 32, 421 423. Bieschke, K. (2002). Charting the waters. The Counseling Psychologist, 30, 575581. Blashfield, R. K. (1976). Mixture model tests of cluster analysis: Accuracy of four agglomerative hierarchical methods. Psychological Bulletin, 83, 377388. Borgen, F. H., & Barnett, D. C. (1987). Applying cluster analysis in counseling psychology research. Journal of Counseling Psychology, 34, 456 468. Broido, E. M. (2000). Constructing identity: The nature and meaning of lesbian, gay, and bisexual identities. In R. M. Perez, K. A. DeBord, & K. J. Bieschke (Eds.), Handbook of counseling and psychotherapy with lesbian, gay, and bisexual clients (pp. 1333). Washington DC: American Psychological Association. Carlson, H. M. (2003). A methodological critique of Spitzers research on reparative therapy. Archives of Sexual Behavior, 32, 425 427. Chung, Y. B., & Katayama, M. (1996). Assessment of sexual orientation in lesbian/gay/bisexual studies. Journal of Homosexuality, 30, 49 62. Cohen, K. M., & Savin-Williams, R. C. (2003). Are converts to be believed? Assessing sexual orientation conversions. Archives of Sexual Behavior, 32, 427 429. Diamond, L. M. (2000). Sexual identity, attractions, and behavior among young sexual-minority women over a 2-year period. Developmental Psychology, 36, 241250. Diamond, L. M. (2003). Reconsidering sexual desire in the context of reparative therapy. Archives of Sexual Behavior, 32, 429 431. Diamond, L. M. (2008). Female bisexuality from adolescence to adulthood: Results from a ten year longitudinal study. Developmental Psychology, 44, 514. Diamond, L. M., & Savin-Williams, R. C. (2000). Explaining diversity in the development of same-sex sexuality among young women. Journal of Social Issues, 56, 297313. Eliason, M. (1997). The prevalence and nature of biphobia in heterosexual undergraduate students. Archives of Sexual Behavior, 26, 317326. Everitt, B. (1980). Cluster analysis. London: Oxford University Press. Fassinger, R. E., & Arsenau, J. R. (2006). Id rather get wet than be under that umbrella: Differentiating the experiences and identities of lesbian, gay, bisexual, and transgender people. In R. M. Perez, K. A. DeBord, & K. J. Bieschke (Eds.), Handbook of counseling and psychotherapy with lesbian, gay, and bisexual clients (2nd ed., pp. 1333). Washington DC: American Psychological Association. Fassinger, R. E., & Miller, B. A. (1996). Validation of an inclusive model of sexual minority identity formation on a sample of gay men. Journal of Homosexuality, 32, 5378. Gonsiorek, J. C., & Weinrich, J. D. (1995). Definition and measurement of sexual orientation. Suicide and Life-Threatening Behavior, 25, 40 51. Hartmann, L. (2003). Too flawed: Dont publish. Archives of Sexual Behavior, 32, 436 438. Herek, G. M. (1984). Attitudes toward lesbians and gay men: A factoranalytic study. Journal of Homosexuality, 10, 39 51. Herek, G. M. (2003). Evaluating interventions to alter sexual orientation: Methodological and ethical considerations. Archives of Sexual Behavior, 32, 438 439. Herek, G. M. (Ed.). (1998). Stigma and sexual orientation: Understanding

SPECIAL ISSUE: CLUSTER ANALYZING SEXUAL ORIENTATION AND IDENTITY Spitzer, R. L. (2003). Can some gay men and lesbians change their sexual orientation? 200 participants reporting a change from homosexual to heterosexual orientation. Archives of Sexual Behavior, 32, 403 417. Storms, M. D. (1980). Theories of sexual orientation. Journal of Personality and Social Psychology, 38, 783792. Strassberg, D. S. (2003). A candle in the wind: Spitzers study of reparative therapy. Archives of Sexual Behavior, 32, 451 452. Vasey, P. L., & Rendall, D. (2003). Sexual diversity and change along a continuum of bisexual desire. Archives of Sexual Behavior, 32, 453 455. Ward, J. H. (1963). Hierarchical grouping to optimize an objective function. Journal of the American Statistical Association, 58, 236 244. Weinberg, M. S., Williams, C. J., & Pryor, D. W. (1994). Dual attraction: Understanding bisexuality. New York: Oxford University Press. Weinrich, J. D., & Klein, F. (2003). Bi-gay, bi-straight, and bi-bi: Three bisexual subgroups identified using cluster analysis of the Klein Sexual Orientation Grid. Journal of Bisexuality, 2, 111139. Worthington, R. L. (2003). Heterosexual identities, sexual reorientation therapies and science. Archives of Sexual Behavior, 32, 460 461. Worthington, R. L. (2004). Sexual identity, sexual orientation, religious identity, and change: Is it possible to depolarize the debate? The Counseling Psychologist, 32, 741749.

55

Worthington, R. L., Dillon, F. R., & Becker-Schutte, A. M. (2005). Development, reliability and validity of the LGB Knowledge and Attitudes Scale for Heterosexuals (LGB-KASH). Journal of Counseling Psychology, 52, 104 118. Worthington, R. L., & Moreno, M. V. (2005, August). Beyond Kinsey and Klein: Measuring Sexual Orientation and Identity. Paper presented at the Division 44 symposium of the 113th Annual Convention of the American Psychological Association, Washington, DC. Worthington, R. L., Navarro, R. L., Savoy, H. B., & Hampton, D. (2008). Development, reliability and validity of the Measure of Sexual Identity Exploration and Commitment (MoSIEC). Developmental Psychology, 44, 2233. Worthington, R. L., Savoy, H., Dillon, F. R., & Vernaglia, E. R. (2002). Heterosexual identity development: A multidimensional model of individual and group identity [Monograph]. The Counseling Psychologist, 30, 496 531.

Received October 7, 2007 Revision received July 6, 2008 Accepted July 7, 2008

S-ar putea să vă placă și