Male bisexuality and condom use at last sexual encounter: results from a national survey.

Relatively little is known about condom use among bisexual men as separate and distinct from exclusively homosexual and heterosexual men. Most previous research on bisexual men has relied on nonprobabilistic, high-risk samples with limited generalizability. We examined the relationship between behavioral bisexuality and condom use in the 2002 cycle of the National Survey of Family Growth (NSFG). Bisexually active men positively differed from heterosexually and homosexually active men on every indicator of confounding risk. Bisexually active men, however, did not report using condoms less often than other men during their last sexual encounters with males and females. Indeed, with female partners, bisexually active men reported higher rates of condom use than did other men. These relationships remained when all sociodemographic and confounding risk factors were held constant. Our results suggest that caution must be used when making assumptions about condom use in the general population of bisexual men from nonprobabilistic samples.

Introduction

Unprotected sexual activity remains the primary mode of HIV transmission in the United States. Despite preventative public health measures, nearly 40,000 individuals are infected with HIV each year (Centers for Disease Control & Prevention, 2004a). Among those most at risk for acquiring and spreading HIV are bisexually active men (Brooks, Rotheram-Borus, Bing, Ayala, & Henry, 2003). These men are of great public health interest, as bisexual activity has been noted as potentially forming a "bridge" from homosexual to heterosexual populations (Chu, Peterman, Doll, Buehler, & Curran, 1992; Doll et al., 1992). As HIV prevention efforts have tended to focus on gay men and/or men who have sex with men (MSM), bisexually active men are a relatively hidden population that is still in great need of research and interventions to address their unique needs (Doll & Beeker, 1996; Stokes, Mckirnan, Doll, & Burzette, 1996; Stokes, Vanable, & McKirnan, 1997). Although research on condom use among bisexually active men is limited, the extant findings suggest that these men may be at increased risk of unprotected sex with both male and female partners.

Background and Significance

Bisexual Behavior and HIV Risk

A number of studies have noted the elevated sexual risk possessed by bisexually active men. These men have been shown to engage in high rates of unprotected anal intercourse (UAI) with men and women, as well as unprotected vaginal intercourse (UVI; Agronick et al., 2004; Doll et al., 1992; Kalichman, Roffman, Picciano, & Bolan, 1998; McKirnan, Stokes, Doll, & Burzette, 1995; Peterson et al., 1992; Stokes et al., 1996). Such behavioral risk also has been noted as occurring concurrent to sex with men and women (McKirnan et al., 1995) and subsequent to HIV infection (Montgomery, Mokotoff, Gentry, & Blair, 2003). Heckman and colleagues (1995) have demonstrated that bisexually active men, in comparison with homosexually active men, express lower intentions to use condoms with male and female partners. The high level of sexual risk purported to exist among bisexually active men warrants investigation of causal mechanisms contributing to the nonuse of condoms.

We sincerely thank Drs. Barbara A. Zsembik and Chuck W. Peek of the university of Florida's Department of Sociology for considerable methodological assistance with this article. The research for this article was conducted while the first author completed a Mcknight Doctoral Fellowship granted by the Florida Education Fund. Correspondence should be addressed to William L. Jeffries, IV, Department of Sociology, University of Florida, P.O. Box 117330, Gainesville, FL, 32611-7330. E-mail: [e-mail]

Confounding Risk Factors for Condom Use Among Bisexually Active Men

Intravenous drug use. Bisexually active men have been found to report high rates of intravenous drug use (IDU)--a substantial risk in its own right (Centers for Disease Control and Prevention, 2004a)--in comparison with exclusively heterosexual or homosexual men (Dodge & Sandfort, 2007; Lewis & Watters, 1994). Intravenous drug use may lead to a decrease in sexual risk cognition and, thus, less condom use. Chu and colleagues (1992) found that a history of IDU among bisexually active men was highly predictive of heterosexual transmission of HIV. As Doll and Beeker (1996) propose, "sexual risk reduction among injecting drug users has long lagged behind changes in their injecting practices" (p. 212).

Sex work. Research suggests that bisexually active men are considerably more likely than heterosexually and homosexually active men to engage in sex for monetary or illicit drug exchange (Boles & Elifson, 1994; Doll & Beeker, 1996; Peterson et al., 1992; Pleak & MeyerBahlburg, 1990; Stokes et al., 1997). Although male sex workers use condoms most frequently with their clients, who are primarily male, they use them less frequently with unpaid male partners and least with their female partners, with whom they typically are more intimate and share greater emotional attachment (Boles & Elifson, 1994; Pleak & Meyer-Bahlburg, 1990; Stokes et al., 1996). The negative relationship between intimacy and condom use has been demonstrated in recent literature (Cusick & Rhodes, 2000; Schiltz & Sandfort, 2000). Men engaging in sex work also report relatively high rates of illicit drug and alcohol use, factors that significantly decrease condom use (Doll & Beeker, 1996; Valdiserri et al., 1988).

The Role of Race/Ethnicity

Samples of bisexually active men tend to have disproportionately high numbers of Black and Latino men (Doll & Beeker, 1996; Stokes, McKirnan, & Burzette, 1993). These men's religious, family, and other social networks are likely to view bisexual behavior with disdain; therefore, these men tend to identify as heterosexual and eschew gay and bisexual identities (Agronick et al., 2004; Doll et al., 1992; Doll & Beeker, 1996; Mays, Cochran, & Zamudio, 2004; Millett, Malebranche, Mason, & Spikes, 2005; Stokes et al., 1996). Consequently, these men may view their same-sex behaviors in a negative light, resulting in elevated levels of internalized homophobia and decreased levels of behavioral disclosure to female partners, which are both determinants of unprotected sex (Kalichman et al., 1998; McKirnan et al., 1995; Stokes et al., 1993, 1996, 1997). When female partners of Black and Latino bisexual men are aware of their partners' behaviors; however, gender differences in relational power (Millett et al., 2005) or the perceived lack of partners within their racial/ethnic group (Malebranche, 2003) may result in these women knowingly engaging in unprotected sex.

Bisexual Identity: An Additional Determinant for HIV Risk?

During the early 1990s, community resources resulted in increased condom use and substantial declines in HIV incidence rates among some gay men. Recent concerns over increases in HIV incidence among bisexually active men and their female partners have been raised due to the degree to which bisexually active men are connected to, or disconnected from, gay-identified communities, which may foster safer sexual practices (Millett et al., 2005; Weatherburn, Hickson, Reid, Davies, & Crosier, 1998). Because they rarely identify as gay, they are less likely to regularly interact within the social networks of gayidentified men and benefit from HIV prevention messages targeted to the gay community (Doll, Myers, Kennedy, & Allman, 1997; Gagnon, 1989; Lever, Kanouse, Rogers, Carson, & Hertz, 1992; McKirnan et al., 1995; Stokes et al., 1997).

The majority of studies of bisexual men and condom use tend to rely solely on behavioral classification schemes in categorizing men as bisexual (Doll et al., 1992, 1996; Lever et al., 1992; Stokes et al., 1993, 1997; McKirnan et al., 1995). Yet, theoretical knowledge about identity in bisexual men begs the question of whether sexual identity has an independent effect upon condom use. Within communities of heterosexual and homosexual men, bisexual men often experience low levels of integration and acceptance (Doll, Myers, Kennedy, & Allman, 1997; Kennedy & Doll, 2001). Such isolation may result in psychosocial stress and inhibit bisexually active men's internalization of safer-sex norms (Dodge & Sandfort, 2007; Hutchins, 1996; Kennedy & Doll, 2001; McKirnan et al., 1995). This is significant given the lack of a bisexual community that could serve as a conduit for public health information (Gagnon, 1989). Moreover, the limited gay-straight dichotomy that is often applied to sexual orientation has resulted in a lack of suitable HIV interventions for these men (Doll et al., 1992; Stokes et al., 1997).

Conceptual and Methodological Limitations of Prior Studies

Studies using non-probabilistic samples are the base of current knowledge of bisexually active men in the United States, and their findings have been used to draw conclusions about the larger population of bisexual men nationwide (Agronick et al., 2004; Kalichman et al., 1998; McKirnan et al., 1995; Stokes et al., 1996; Valdiserri et al., 1988). Yet, all of these studies were based on samples that are urban, which do not properly characterize the sexual risk and geographic distribution of bisexually active men (Denning, Jones, & Ward, 1997). In addition, many of these samples have indiscriminately regarded bisexual men as MSM, without highlighting the sociosexual characteristics, particularly interactions with female partners (Doll et al., 1992), that distinguish them (see Heckman et al., 1995). Moreover, some studies exclusively have relied on samples of HIV-infected men. By doing so, they have groundlessly assumed a level of sexual risk for bisexually active men that has not been empirically tested (Valdiserri et al., 1988).

Hypotheses

This study investigates the relationship between behavioral bisexuality and condom use in a nationally representative sample. Specifically, we explore the roles that sexual behavior and identity play in the likelihood of condom use with men's most recent sexual partners. We proceed with the following hypotheses:

H1: Bisexually active men have lower rates of condom use in comparison to homosexually and heterosexually active men.

H2: Sociodemographic and confounding risk factors mediate the relationship between behavioral bisexuality and condom use.

H3: A nonheterosexual identity is protective against unprotected sex for bisexually active men.

Research Design and Methods

Data and Sample

We use the 2002 cycle of the NSFG. Conducted by the National Center for Health Statistics (NCHS), it is a nationally representative, area probability sample of 4,928 male participants between the ages of 15 and 44 who live in households (Groves et al., 2005). The sampling frame consisted of 121 primary sampling units. Interviewers conducted voluntary and confidential in-person interviews using computer-assisted techniques. Survey instruments were initially pretested between March and August of 2001, and primary data collection occurred from March 2002 to March 2003. The overall response rate for the men was 78%. All participants were compensated $40.

Audio computer-assisted self-interviewing (ACASI) was used on the section of the interview that ascertained sexual identity, sexual behavior, and condom use with male and female partners. Using ACASI allowed participants to enter data on such sensitive issues by using a computer, without the recognition of the interviewer. Interviews averaged one hour in length.

Dependent Variables

Condom use at last sexual encounter. Participants were asked, "Did you use a condom the last time you had vaginal intercourse with a female?" Parallel questions were asked in regard to anal and oral sex with females, as well as anal and oral sex with males. Possible responses were "yes," "no," "refused," and "don't know." Men who answered "don't know" were placed in the "no" category in order that this variable provided a conservative measurement for the certainty of condom use. From these questions, we created dichotomous variables that measured condom use at last sexual encounter with male and female partners.

Independent Variables

Sexual orientation. Behavioral sexual orientation is based upon the gender(s) of sexual partners within the year prior to the participant's interview. Men were asked, "During the last 12 months, how many female sexual partners have you had?" and "During the last 12 months, how many male sexual partners have you had?" We created three dummy variables to categorize sexual orientation. Men with only female partners in the past year were considered heterosexually active; men with only male partners in the past year were considered homosexually active; men with male and female partners in the past year were considered bisexually active. (1)

In order to ascertain sexual identity, participants were asked, "Do you think of yourself as heterosexual, homosexual, bisexual, something else, or don't know?" A dummy variable was created to reflect the presence of a nonheterosexual identity, and it included 24 heterosexually active men who "refused" to answer the question. We adopted this strategy for delineating sexual identity as it would capture the ostensible contrasts between bisexually active men who identify as heterosexual versus those who do not. Some have proposed that the presence of any nonheterosexual identity among bisexually active men has a protective health effect, being conducive to greater rates of condom use (Peterson et al., 1992; Stokes et al., 1993, 1996).

Control Variables

Race/ethnicity. Participants were asked, "Are you of Hispanic origin?" Those who did not indicate being Hispanic were asked about their race. A dummy variable was later created that distinguished non-Latino Whites, non-Latino Blacks, Latinos, and men of all other racial/ethnic categories.

Socioeconomic status (SES). Education was determined by asking participants a series of questions regarding the number of years of completed schooling and the highest degree received. We created a dummy variable to distinguish men with bachelor's degrees or more from those without bachelor's degrees. Rather than the level of high school diploma/GED, a bachelor's degree was deemed the most appropriate cut-off given the variability in educational levels for men over the age of 18. For income, participants were asked, "In the year 2001, did you (or any members of your family living here) receive any wages and salaries, including tips, bonuses, and overtime?" Similar questions were asked in regard to all income from self-employment, social security, disability, retirement, supplemental security income, unemployment, child support, and interest and dividends. All income data were pooled to create an annual household income variable. We created a dummy variable to reflect men with annual household incomes of $50,000 or more in comparison with those with less than $50,000.

Age. In this article, we included age as a control variable because it has been shown to be positively related to condom use (Laumann, Gagnon, Michael, & Michaels, 1994; Valdiserri et al., 1988). Participants were asked, "How old are you?" The men provided their age in years.

Marital status. Participants were asked, "What is your current marital status?" Possible choices were married, cohabitating with an opposite sex partner, widowed, divorced, separated, and never been married. We created a dummy variable to distinguish married men from all others. This was done because marriage has been shown to decrease the likelihood of both condom use and sexual risk (Lever et al., 1992; Stokes et al., 1996; Valdiserri et al., 1988).

IDU. The men were asked, "During the last 12 months, how often have you taken nonprescription drugs using a needle, that is, you took them only for the experience or feeling it caused?" Due to low levels of use (only 32 men in the entire sample reported IDU), we dichotomized responses to "never" and "any use" in the past 12 months. (2)

Sex work. Sex work was determined by asking, "In the past 12 months has a female given you money or drugs to have sex with her?" The men were also asked a parallel question for sex work with men. Participants answered "yes," "no," or "don't know" for both questions. We considered men who answered "yes" or "don't know" to the former question as having engaged in sex work)

Substance use. In this study, being high during sex was ascertained by asking, "In the last 12 months, how often were you 'high' on alcohol or drugs when you had sex with a female?" and "In the last 12 months, how often were you 'high' on alcohol or drugs when you had sex with a male?" We dichotomized the men s responses so that men who were high during sex "about half the time" or more were distinguished from those who were "never" or "sometimes" high. Heterosexually and homosexually active men's data were generated based upon the former and latter questions, respectively. Bisexually active men were placed in the risk category if they were high at least half the time with male or female partners. Men providing responses of "don't know" were placed in the risk category, whereas men who "refused" to answer were excluded from multivariate analyses. (4)

Number of sexual partners. NCHS staff computed the number of female sexual partners by using the NSFG's multiple indicators of heterosexual activity within the past year (see NCHS, 2005). In regard to same-sex encounters, the men were asked, "During the last 12 months, how many male sexual partners have you had?" Based upon the men's responses, we created a variable that distinguished men with three or more sexual partners in the past year from all others. Three or more was chosen as the threshold for the number of sexual partners for three reasons. First, having multiple sexual partners within the past year is predictive of greater rates of condom usage (Feinleib & Michael, 2001; Valdiserri et al., 1988). Second, all of the bisexually active men necessarily had at least two partners in order to be classified as bisexually active. Last, creating a threshold of four or more partners would have significantly diminished variation in this variable.

Interaction Terms

We created two terms to test the interaction of sexual behavior and identity as it relates to condom use at last intercourse. This was done by multiplying the dummy variables for homosexually and bisexually active men by the dummy variable for nonheterosexual identity. These terms describe how condom use is a function of behaviorally defined sexual orientation, but it varies depending upon sexual identity.

Analyses

We began by exploring the univariate and bivariate distributions of sexual identity, sociodemographic correlates, confounding risk factors, and condom use. We compared the subsample proportions of each covariate by iterating three t tests based on behavioral sexuality (heterosexual, homosexual, and bisexual) during the past year. Similar to t tests for differences in means, these tests allow for subsample comparisons of proportion distributions (Agresti & Finlay, 1997; SAS Institute, 1999;

Simple Interactive Statistical Analysis, 2006). To control for type 1 errors, we set our alpha levels to .01, while acknowledging marginal significance at the .05 alpha level. Next, we logistically modeled the relationship between behavioral sexual orientation and condom use. We used two sets of five nested models with odds ratios for behavioral sexual orientation groups, nonheterosexual identity, race/ethnicity, other sociodemographic correlates, and confounding risk factors. In these nested models, bisexually active men are the referent in order to compare them with heterosexually and homosexually active men. Last, we modeled condom use as a function of the interaction of sexual behavior and identity. For these models, however, we made heterosexually active men the referent in order to measure the relative effect of behavioral bisexuality. Data were analyzed using the SAS[R] System for Windows, Version 8.2 (SAS Institute, 1999). (5) All data were weighted in order to adjust for oversampling (NCHS, 2004). (6)

Results

Univariate Analyses

The first column of Table 1 presents the sample mean or proportion for each covariate. About one tenth of the sample identified as nonheterosexual. Latino men, about one-sixth of the overall sample, made up the largest racial/ethnic minority group. One eighth of the sample was Black; about 5% were of another race/ ethnicity minority group. The mean age was 31. Slightly over half of the men were currently married to a woman. One-fifth of the sample had at least a bachelor's degree, and two-fifths had annual household incomes of at least $50,000. The prevalence of IDU and sex work were only 0.5% and 1.7%, respectively. Being high during sex and having three or more partners characterized about one-tenth of the men. Based upon information about bisexually active men's most recent male partner, nearly one-third of the men reported using a condom. Nearly equivalent results were evident when based upon bisexually active men's condom use with their most recent female partner.

Bivariate Analyses

Sexual identity. Bisexually active men were more likely than heterosexually active men (48% versus 7%; [t.sub.(3,640)] = 33.7, p < .001) to have a nonheterosexual identity; yet, they were less likely than homosexually active men (97% [t.sub.(200)] = 9.9, p < .001).

Race/ethnicity. Latino men made up more than one-fourth of the subsample of bisexually active men, marginally greater than the amount of Latinos in the homosexually active ([t.sub.(200)] = 2.3, p = .02) subsample. We observed no other subgroup differences in race/ ethnicity.

Age and marital status. Bisexually active men did not differ in age from heterosexually and homosexually active men. Nearly one-fourth of the bisexually active men were married, however, which is less than heterosexually active men (53%; [t.sub.(3,560)] = 3.6, p < .001) but more than homosexually active men (0.41%; [t.sub.(200)] = 6.3, p < .001).

Socioeconomic status. Among bisexually active men, about 12% had a bachelor's degree or more, which makes them less educated than homosexually active men (27%; [t.sub.(200)] = 2.5, p = .01). About one-sixth of bisexually active men had annual household incomes of $50,000 or more, which is less than incomes for heterosexually active (41%; [t.sub.(3,560)] = 3.2, p = .001) and homosexually active (33%; [t.sub.(200)] = 2.6, p = .01) men.

Confounding risk factors. Bisexually active men positively differed from heterosexually and homosexually active men on every indicator of confounding risk. One-tenth of the bisexually active men participated in IDU in the past year, a proportion greater than heterosexually active (0.4%; [t.sub.(3,553)] = 8.3, p < .001) and homosexually active (0.5%; [t.sub.(200)] = 3.5, p < .001) men. Nearly one-fifth of the bisexually active men reported engaging in sex for monetary or drug exchange, significantly more than heterosexually active (1%; [t.sub.(3,439)] = 9.3, p < .001 and marginally more than homosexually active (9%; [t.sub.(200)] = 2.2, p = .03) men.

Thirty-six percent of bisexually active men were high during sex at least half the time in the past year, which is statistically greater than heterosexually active (10%; t(3,437) = 5.5, p < .001) and homosexually active (15.6%; [t.sub.(200)] = 3.3, p < .001) men. In comparison with heterosexually active men, homosexually active men were marginally more likely to report being high during at least half of their sexual encounters ([t.sub.(3,517)] = 1.9, p = .05).

The vast majority (about 85%) of bisexually active men had three or more sexual partners in the past year. Again, these men statistically differed in comparison with heterosexually active (10%; [t.sub.(3,560)] = 15.6, p < .001) and homosexually active (42%; [t.sub.(200)] = 6.1, p < .001) men. A greater proportion of homosexually active men had three or more partners in the past year in comparison with heterosexually active men ([t.sub.(3,640)] = 9.9, p < .001).

Dependent variables. The first measure for condom use was based upon bisexually active men's last sexual encounter with males. Bisexually active men's proportion (41%) did not differ from that of heterosexually and homosexually active (30% and 44%, respectively). When condom use was based upon bisexually active men's most recent female partner, however, the proportion for bisexually active men (64%) was greater than that of both heterosexually active (30%; [t.sub.(3,461)] = 4.7, p < .001) and homosexually active (44%; [t.sub.(198)] = 2.7, p = .01) men. Homosexually active men were more likely than heterosexually active men to report using a condom during their last sexual encounter ([t.sub.(3,539)] = 2.9, p < .01).

Multivariate Results

Table 2a presents logistic regression models in which condom use, based upon bisexually active men's most recent male partner, is regressed upon behavioral sexual orientation. In model 1, the odds of condom use for heterosexually active men is marginally lower (by 41%) than the odds for bisexually active men, the referent group. Homosexually active men do not differ from bisexually active men. Model 2 contains the addition of sexual identity. Although it is not predictive of condom use, it does diminish the marginal association of behavioral bisexuality. With the addition of race/ ethnicity, other sociodemographics, and confounding risk factors, no sexual orientation differences in condom use are observed. With the inclusion of race/ethnicity dummy variables in model 3, however, it is clear that men of racial/ethnic minority status have higher odds of using condoms at their most recent sexual encounter in comparison with White men. The odds of condom use for Black, Latino, and other men are, respectively, 2.25, 1.43, and 1.34 times the odds of White men.

Model 4 contains the additions of age, marital status, education level, and income. These data indicate that with each one-year increase in age, the adjusted odds of condom use decreases by 6%. Similarly, married men experience a 69% decrease in the odds of condom use in comparison with men who are not currently married. Men who have at least a bachelor's degree are 42% more likely to report using a condom. Income has no effect net of other sociodemographics. The protective effect of race/ethnic minority status essentially remains when other sociodemographics are controlled.

Model 5 is the full model containing all covariates of interest. Intravenous drug use (IDU) and sex work, while controlling for all other variables, are not related to condom use at last sexual encounter. Men who were high during sex at least half the time, however, report 23% lower odds of condom use in comparison with men not reporting this behavior. On the other hand, men who had 3 or more sexual partners in the past year have odds of 42% higher than men who had 1 or 2 partners. The addition of confounding risk factors only slightly diminishes the effects of being Black, Latino, or college educated. Confounding risks diminish neither the significance levels nor the direction of the effect for any sociodemographic correlates. Given the decrease in--2 log likelihood values across models, all sets of covariates, excluding sexual identity, significantly improve model fit.

Table 2b presents logistic regression models in which condom use, based upon bisexually active men's most recent female partner, is a function of behavioral sexual orientation. Model 1 indicates that, in comparison with bisexually active men, heterosexually and homosexually active men have a 77% and 56% lower odds, respectively, of reporting condom use during their last sexual encounters. These odds ratios remain relatively stable in model 2 as sexual identity, which has no net effect, is included. In model 3, race/ethnicity dummy variables are included with behavioral sexual orientation and identity. Being Black, Latino, or of other race/ethnicity, in comparison with being White, is yet predictive of an increase in the odds of condom use during the men's last sexual encounters. Specifically, the odds for Black, Latino, and men of other race/ethnicity are 2.30, 1.43, and 1.32, respectively. Although racial/ethnic ethnic minority status is predictive of condom use, race/ethnicity does little to attenuate the relationship between behavioral sexual orientation and condom use. This is evident given that the adjusted odds ratios are yet significantly lower for heterosexually and homosexually active men (0.25 and 0.47, respectively).

Model 4 contains the additions of age, marital status, and SES. With each one-year increase in age, the adjusted odds of condom use decrease by 6%. Similarly, the odds of condom use for married men are 69% lower than the odds of unmarried men. Men with a bachelor's degree or more have a 37% greater odds of condom use. Income has no effect net of other covariates. The odds ratios for heterosexually and homosexually active men remain stable, with bisexually active men yet reporting greater odds of condom use.

In model 5, we model condom use as a function of sexual orientation, identity, sociodemographic correlates, and confounding risk factors. Intravenous drug use (IDU) and sex work have no apparent effect upon condom use while controlling for other factors. Men who were high during at least half of their sexual encounters however, experience a 19% decrease in the odds of condom use. Yet, men who report three or more sexual partners in the past year had a 42% higher odds of condom use in comparison with those with one or two partners. The appendage of confounding risk factors does little to lessen the effects of behavioral sexual orientation, as heterosexually and homosexually active men continue to have a 65% and 59%, respectively, lower odds of condom use when compared with bisexually active men. Likewise, the presence of confounding risk factors only slightly attenuates the effect of sociodemographic correlates and condom use. In comparison with White men, Black and Latino men are still 83% and 30% more likely, respectively, to report condom use. The odds ratio for men of other race remains the same. Being older and married are yet negatively related to condom use, while having at least a bachelor's degree is positively related to this outcome. Given the decrease in -2 log likelihood values across models, all sets of covariates, excluding sexual identity, significantly improve model fit.

Table 3 presents the nonadditive models for sexual behavior and identity. Given our interests in the relationship among bisexual behavior, sexual identity, and condom use, we chose to make heterosexually active men the referent group. In model 1, as condom use is measured based upon bisexually active men's most recent male sexual partner, the interactions of bisexual and homosexual behavior and nonheterosexual identity are not significant. However, the results of model 2 are more telling. Here, as condom use is measured based upon bisexually active men's most recent female partner, the odds ratio for bisexually active men who identify as heterosexual (1.49) is not significant. The odds ratio for bisexually active men who identify as something other than heterosexual (4.40), however, is significant. Based upon this, the likelihood of condom use for bisexually active men who sexually identify as nonheterosexual (e.g., gay, bisexual, something else, or don't know) is 4.40 times the likelihood of men who are not bisexually active and do identify as heterosexual. In all, the presence of a nonheterosexual identity is protective for bisexually active men. Yet, it has no effect for homosexually active men, whose odds ratio (4.24) is nonsignificant. Model 2 of Table 3 is an extension of model 5 in Table 2b. Given the decrease in -2 log likelihood from model 5 of Table 2b to model 2 of Table 3, the inclusion of the behavior-identity interaction results in significant improvement to the overall model fit.

Discussion

In order to investigate the relationship between behavioral bisexuality and condom use, we used a nationally representative, probabilistic sample to compare bisexually active men with heterosexually and homosexually active men. Also, we investigated the relative effect of a nonheterosexual identity as well as the interaction of behavior and identity. Last, we examined the roles that race/ethnicity, age, marital status, education, and income play as mediators of condom use for bisexually active men while controlling for confounding risk factors.

Contrary to our hypothesis pertaining to behavioral bisexuality and condom use, bisexually active men, based upon their last sexual encounters with males and females, did not use condoms less than heterosexually and homosexually active men. During bisexually active men's most recent same-sex contact, they report rates that are not statistically different from those of heterosexually and homosexually active men. Yet, when condom use was measured in light of bisexually active men's most recent female partners, very different results were observed. Not only were the proportions for bisexually active men statistically higher than those of heterosexually and homosexually active men, the odds for bisexually active men using a condom at their most recent sexual encounter were higher than that of other men. This association remained as all other variables were controlled.

Our finding regarding the mediating effects of socio-demographic and risk covariates was partially supported. In neither main effects regression model, however, did these factors mediate the relationship between sexual orientation and condom use such that bisexually active men had a lower odds of condom use. This finding contradicts prior studies that have suggested that bisexually active men use condoms at lower rates than other men (Doll et al., 1992; Kalichman, et al., 1998; McKirnan et al., 1995; Peterson et al., 1992). Yet, this finding is remarkable given the marked socio-demographic differences between bisexually active men and others. Likewise, confounding risks did not diminish the effect of bisexual behavior upon condom use, despite the fact that bisexually active men were more likely to participate in all risks. It is probable that bisexually active men know the risks associated with their behaviors and, therefore, take necessary precautions to protect themselves and their partners. Of course, this cannot be said of all bisexually active men given that some of them reported unprotected sex with their most recent male and female partners. Yet, to the extent that a greater proportion of bisexually active men, at least in terms of heterosexual contact, used condoms at their most recent sexual encounters, it is possible that bisexual behavior alone functions in a protective way for sexually active men.

As we hypothesized, the interaction between bisexual behavior and a nonheterosexual identity was a positive predictor of condom use with female partners (see Centers for Disease Control and Prevention, 1993). For some bisexually active men who identify as heterosexual, using a condom during sexual encounters with males may force them to psychologically come to terms with their same-sex desires and behaviors (Mays et al., 2004). Therefore, men who already acknowledge, in terms of self-identity, their same-sex desires and behaviors may be more comfortable wearing a condom during sexual encounters. Others have suggested that when bisexually active men possess sexual identities that are congruent with their sexual behaviors, they may internalize the safer-sex messages targeted to gay and bisexual men (Heckman et al., 1995; Stokes et al., 1993).

Although the nonadditive finding for sexual behavior, identity, and condom use is not altogether surprising, it is surprising that the interaction is significant in predicting condom use with females, but not with males. It is possible that these men, being bisexually active and, perhaps, knowing the disproportionate level of risk within this population, see it necessary to take more precautions when engaging in sexual acts with women. More research should investigate how nonheterosexual identities interact with bisexual behavior in order to mediate the use of condoms with bisexually active men's female partners. Qualitative studies exploring bisexually active men's perceptions of gender in regard to condom use also would be helpful in this regard (Dodge, Jeffries, & Sandfort, in press).

Several factors may explain why bisexually active men in our study report greater amounts of condom use with women in comparison with men. Others have found that bisexually active men engage in high rates of insertive and receptive oral sex with men (Heckman et al., 1995; McKirnan et al., 1995; Stokes et al., 1993). Given these findings, it is not wholly unreasonable that these men reported lower rates of condom use with other men. Although the risks for STIs other than HIV are considerable, unprotected oral sex carries substantially less HIV risk than unprotected anal sex. Oral sex also readily occurs in locations not conducive to condom use, such as in cruising venues (Reece & Dodge, 2003).

Alternatively, even if female partners of bisexually active men are unaware of their partners' bisexual behaviors, they may indeed know of these men's confounding risks through IDU, sex work, using drugs during sex, and having multiple partners. As a result, they may perceive their male partners as being of greater risk and mandate that they use condoms during sexual episodes. This alternative also may explain why bisexually active men, even after covariates are controlled, are more likely to use condoms with their female partners than are heterosexually and homosexually active men.

This study does not suggest that bisexually active men in the United States are not at significant risk of contracting HIV and other STIs. To the contrary, the markedly greater prevalence of confounding risks supports existing studies' findings that these men are indeed at greater sexual risk than other men (Peterson et al., 1992; Doll & Beeker, 1996; Doll et al., 1997; Kennedy & Doll, 2001). Thus, there is concern about the role that these behaviors play in the transmission of HIV and other STIs to and from this population. Although these behaviors did not mediate the relationship between bisexual behavior and condom use, all of these men did not report the use of condoms during their most recent sexual encounters. In consideration of the distribution of risk within this population, on occasions that bisexually active men who engage in confounding risks do not use condoms, they may indeed place themselves and their partners at greater risk of HIV/STIs. We advocate for public health interventions to address the high prevalence of confounding risks within this population. Future studies also should address how bisexually active men negotiate condom use in sexual situations that are high risk.

Racial/ethnic minority status was predictive of an increased likelihood of condom use, even while adjusting for factors known to differ between minority and White men, namely, marital status, SES, sexual risk behaviors, and drug use (Laumann et al., 1994; Peterson et al., 1992). This finding is comparable with others' findings that racial/ethnic minority men use condoms at higher rates than White men (Centers for Disease Control & Prevention, 2006; Laumann et al., 1994). It is possible that minority men, like bisexually active men, know the sexual risks purported to exist within their racial/ethnic groups and, thus, use condoms at higher rates. They also may encounter more people infected with HIV and other STIs due to the higher prevalence of these conditions in their communities (Laumann et al., 1994). As a result, they may feel a greater need to protect themselves from such conditions.

As with all research, this study is not without limitations. Ideally, we would prefer measures that took into account the number of occasions the men engaged in UAI and UVI in the past year as well as their types of recent same-sex encounters (Doll et al., 1992; Kalichman et al., 1998; McKirnan et al., 1995; Peterson et al., 1992). This would have afforded us the opportunity to examine a wider range of sexual encounters during the past year and note condom use for specific sexual acts. Such would have increased our understanding of the complex nature of sexual risk for bisexually active men in the United States (Valdiserri et al., 1988). Nevertheless, condom use at last encounter is the only measure provided within the NSFG by which we could compare bisexually active men with heterosexually and homosexually active men. Also, such a measure has been utilized by the Centers for Disease Control and Prevention and can be used for comparisons with their publications (Centers for Disease Control & Prevention, 2004b). Moreover, our results are generalizable only to the population of men between the ages of 15 and 44 within the United States. Last, in this study, we were not able to specify the mechanisms that are conducive to decreased sexual risk for bisexually active men. Future studies should explore social-psychological processes that lead bisexually active men to use condoms at equal or higher rates in comparison with other men. Such will prove vital in the implementation of public health interventions designed to decrease sexual risk within at-risk populations.

Conclusion

Overall, our findings shed new light upon the intricacies of HIV and other STI risks within the overall population of bisexually active men in the United States. Rather than simply relying on nonprobabilistic samples of HIV-infected men or MSM, many of whom are already engaging in risky sexual behaviors, to make implications about all bisexually active men, this study shows the benefits of using a nationally representative sample in order to study the distribution of risk for bisexually active men. In addition, this article highlights the protective factors that are enacted by bisexually active men and yields understanding into how such factors can benefit other populations of at-risk individuals. In all, our results suggest that men recruited as part of probabilistic samples are quite different from those recruited as part of convenience samples. Future nationally representative surveys such as the NSFG should collect in-depth data on same-sex encounters to the extent that such encounters have been known to carry greater HIV/STI risk (Doll et al., 1992; Laumann et al., 1994). By doing so, more accurate sexual risk comparisons can be made among heterosexually, homosexually, and bisexually active men.

References

Agresti, A. & Finlay, B. (1997). Statistical methods for the social sciences (3rd ed.). Upper Saddle River, NJ: Prentice-Hall.

Agronick, G., O'Donnell, L., Stueve, A., San Doval, A., Duran, R., & Vargo, S. (2004). Sexual behaviors and risks among b bisexually- and gay-identified young Latino men. AIDS and Behavior, 8(2), 185-197.

Boles, J. & Elifson, K. W. (1994). Sexual identity and HIV: The male prostitute. The Journal of Sex Research, 31, 39-46.

Brooks, R., Rotheram-Borus, M. J., Bing, E. C., Ayala, G., & Henry, C. L. (2003). HIV and AIDS among men of color who have sex with men and men of color who have sex with men and women: An epidemiological profile. AIDS Education and Prevention, 15(Suppl. A), 1-6.

Centers for Disease Control and Prevention. (1993). Condom use and sexual identity among men who have sex with men--Dallas, 1991. Morbidity and Mortality Weekly Report, 42(1), 13-14.

--. (2004a). HIV/AIDS surveillance report.. Cases of HIV infection and AIDS in the United States, 2004. Retrieved March 8, 2006, from http://www.cdc.gov/hiv/topics/surveillance/resources/reports/ 2004report/pdf/2004SurveillanceReport.pdf

--. (2004b). Morbidity and mortality, weekly report: High-risk sexual behavior by HIV-positive men who have sex with men--16 Sites, United States, 2000 2002. Retrieved December 1, 2005, from http://www.cdc.gov/mmwr/preview/mmwrhtml/mm5338a1.htm--. (2006). Youth risk behavior surveillance--United States, 2005. Morbidity and Mortality Weekly Report, 55(SS-5), 1-112.

Chu, S. Y., Peterman, T. A., Doll, L. S., Buehler, J. W., & Curran, J. W. (1992). AIDS in bisexual men in the United States: Epidemiology and transmission to women. American Journal of Public Health, 82(2), 220-224.

Cusick, L. & Rhodes, T. (2000). Sustaining sexual safety in relationships: HIV positive people and their sexual partners. Culture, Health, and Sexuality, 2(4), 473-487.

Denning, P. H., Jones, J. L., & Ward, J. W. (1997). Recent trends in the HIV epidemic in adolescent and young adult gay and bisexual men. Journal of Acquired Immune Deficiency Syndromes and Human Retrovirology, 16(5), 374-379.

Dodge, B., Jeffries, W., L., & Sandfort, T. G. M. (in press). Beyond the down low: Sexual risk, protection, and disclosure practices among at-risk Black men who have sex with both men and women MSMW (1). Archives of Sexual Behavior.

Dodge, B. & Sandfort, T. G. M. (2007). A review of mental health research on bisexual individuals when compared to homosexual and heterosexual individuals. In B. A. Firestein (Ed.), Becoming visible: Counseling bisexuals across the lifespan (pp. 28 51). New York: Columbia University Press.

Doll, L. S., Petersen, L. R., White, C. R, Johnson, E. S., Ward, J. W., & The Blood Donor Study Group. (1992). Homosexually and nonhomosexually identified men who have sex with men: A behavioral comparison. The Journal of Sex Research, 29(1), 1 14.

Doll, L. S. & Beeker, C. (1996). Male bisexual behavior and HIV risk in the United States: Synthesis of research with implications for behavioral interventions. AIDS Education and Prevention, 8(3), 205 225.

Doll, L. S., Myers, T., Kennedy, M., & Allman, D. (1997). Bisexuality and HIV risk: Experiences in Canada and the United States. Annual Review of Sex Research, 8, 102-147.

Feinleib, J. A. & Michael, R. T. (2001). Reported change in sexual behavior in response to AIDS in the United States. In E. O. Laumann & R. T. Michael (Eds.), Sex, love, and health in America: Private choices and public policies (pp. 302 326). Chicago: The University of Chicago Press.

Gagnon, J. H. (1989, October). The management of erotic relations with both genders. Paper presented at the CDC workshop on Bisexuality and AIDS, Atlanta, GA.

Groves, R. M., Benson, G., Mosher, W. D., Rosenbaum, J., Granda, P., Axinn, W., Lepkowski, J., & Chandra, A. (2005). Plan and operation of cycle 6 of the National Survey of Family Growth. Vital and Health Statistics, 1(42), 1-67. Retrieved February 26, 2006, from http://www.cdc.gov/nchs/data/series/sr_01/sr01_042.pdf

Heckman, T. G., Kelly, J. A., Sikkema, K. J., Roffman, R. R., Solomon, L. J., Winett, R. A., Stevenson, L. Y., Perry, M. J., Norman, A. D., & Desiderato, L. J. (1995). Differences in HIV risk characteristics between bisexual and exclusively gay men. AIDS Education and Prevention, 7(6), 504-512.

Hutchins, L. (1996). Bisexuality: Politics and community. In B. A. Firestein (Ed.), Bisexuality: Psychology and pofitics of an invisible minority (pp. 240-259). Thousand Oaks, CA: Sage Publications.

Kalichman, S. C., Roffman, R. A., Picciano, J. F., & Bolan, M. (1998). Risk for HIV infection among bisexual men seeking HIV-prevention services and risks posed to their female partners. Health Psychology, 17(4), 320-327.

Kennedy, M. & Doll, L. S. (2001). Male bisexuality and HIV risk. Journal of Bisexuality, 1(23, 111-135.

Laumann, E. O., Gagnon, J. H., Michael, R. T., & Michaels, S. (1994). The social organization of sexuality." Sexual practices in the United States. Chicago: The University of Chicago Press.

Laumann, E. O. & Youm, Y. (2001). Racial/ethnic group differences in the prevalence of sexually transmitted diseases in the United States: A network explanation. In E. O. Laumann & R. T. Michael (Eds.), Sex, love, and health in America: Private choices and public policies (pp. 327 351). Chicago: The University of Chicago Press.

Lever, J., Kanouse, D. E., Rogers, W. H., Carson, S., & Hertz, R. (1992). Behavior patterns and sexual identity of bisexual males. The Journal of Sex Research, 29(2), 141-167.

Lewis, D. K. & Watters, J. K. (1994). Sexual behavior and sexual identity in male injection drug users. Journal of Acquired Immune Deficiency Syndromes, 7, 190-198.

Malebranche, D. (2003). Black men who have sex with men and the HIV epidemic: Next steps for public health. American Journal of Public Health, 93(6), 862-865.

Mays, V. M., Cochran, S. D., & Zamudio, A. (2004). HIV prevention research: Are we meeting the needs of African American men who have sex with men? Journal of Black Psychology, 30(1), 78-105.

McKirnan, D. J., Stokes, J. P., Doll, L. S., & Burzette, R. G. (1995). Bisexually active men: Social characteristics and sexual behavior. The Journal of Sex Research, 32(1), 65-76.

Millett, G., Malebranche, D., Mason, B., & Spikes, P. (2005). Focusing down low: Bisexual black men, HIV risk and heterosexual transmission. Journal of the National Medical Association, 97(Suppl. 7), 52S-59S.

Montgomery, J. P., Mokotoff, E. D., Gentry, A. C., & Blair, J. M. (2003). The extent of bisexual behaviour in HIV-infected men and implications for transmission to their female partners. AIDS Care, 15(6), 829-837.

National Center for Health Statistics. (2004). Public use data file documentation, National Survey of Family Growth Cycle 6: 2002. Retrieved February 2, 2006 from www.pop.psu.edu/dataarchive/ codebooks/nsfg/2002/userguide_2002nsfg.pdf

National Center for Health Statistics (NCHS). (2005). ACASI file user's guide and documentation, National Survey of Family Growth Cycle 6: 2002. Hyattsville, MD: United States Department of Health and Human Services.

Peterson, J. L., Coates, T. J., Catania, J. A., Middleton, L., Hilliard, B., & Hearst, N. (1992). High-risk sexual behavior and condom use among gay and bisexual African-American men. American Journal of Public Health, 82(11), 1490-1494.

Pleak, R. R. & Meyer-Bahlburg, H. (1990). Sexual behavior and AIDS knowledge of young male prostitutes in Manhattan. The Journal of Sex Research, 27, 557-587.

Reece, M. & Dodge, D. (2003). Exploring the physical, mental, and social well-being of gay and bisexual men who cruise for sex on a college campus. Journal of Homosexuality, 46(1/2), 111-136.

SAS Institute. (1999). SAS online doc, version 8. Cary, NC: Author.

Schiltz, M. A. & Sandfort, T. G. (2000). HIV-positive people, risk, and sexual behaviour. Social Science and Medicine, 50(11), 1571-1588.

Simple Interactive Statistical Analysis. (2006). T-test, retrieved November l, 2006 from http://home.clara.net/sisa/t-thlp.htm

Stokes, J. P., McKirnan, D. J., & Burzette, R. G. (1993). Sexual behavior, condom use, disclosure of sexuality, and stability of sexual orientation in bisexual men. The Journal of Sex Research, 30(3), 203-213.

Stokes, J. P., McKirnan, D. J., Doll, L. S., & Burzette, R. G. (1996). Female partners of bisexual men: What they don t know might hurt them. Psychology of Women Quarterly, 20(2), 267-284.

Stokes, J. P., Vanable, P., & McKirnan, D. J. (1997). Comparing gay and bisexual men on sexual behavior, condom use, and psychosocial variables related to HIV/AIDS. Archives of Sexual Behavior, 26(4), 383-397.

Valdiserri, R. O., Lyter, D., Leviton, L. C., Callahan, C. M., Kingsley, L., & Rinaldo, C. R. (19883. Variables influencing condom use in a cohort of gay and bisexual men. American Journal of Public Health, 78(7), 801-805.

Weatherburn, P., Hickson, F., Reid, D. S., Davies, P. M., & Crosier, A. (19983. Sexual HIV risk behaviour among men who have sex with men and women. AIDS Care, 10(43, 463-471.

Notes

(1) Of the total 4,928 men in the sample 3,506 were heterosexually active, 141 were homosexually active, 61 were bisexually active, and 1,201 were not sexually active in the past year. Nineteen men had missing data regarding sexual behavior and, therefore, could not be placed into one of these categories. Only 3,501 heterosexually active men, however, had sufficient data to be represented in univariate and bivariate analyses. This number was reduced to 3,376 for multivariate analyses. Of the homosexually active men, 2 had missing data in regard to condom use. Therefore, 139, rather than 141, homosexually active men are included in multivariate analyses. Of the bisexually active men, 2 had missing data in regard to condom use with male partners. Therefore, only 59 bisexually active men appear in Table 2b.

(2) Seven of the heterosexually active men refused to answer the question about IDU. They were, therefore, excluded from multivariate analyses.

(3) Only one heterosexually active participant answered "don't know" to the question about sex work with females. One bisexually active participant answered "don't know" to the question about sex work with males.

(4) Of the heterosexually active men, 3 responded with "don't know" and 2 with "refused" in regard to being high during sex. One homosexually active and one bisexually active man provided responses of "don't know" for this question.

(5) We primarily used dummy variables rather than the ratio or ordinal variables from the raw data due to the small or nonexistent cell frequencies generated by the cross-tabulation of sexual orientation groups (primarily for homosexually and bisexually active men) and other variables. This was done to capture as much variability in condom use at last sexual contact and to facilitate multivariate regression analyses (Agresti & Finlay, 1997).

(6) We created a scaled version of the sample's weight by multiplying the provided weight variable by the ratio of the unweighted sample size to the weighted sample size.

William L. Jeffries IV

University of Florida

Brian Dodge

Indiana University 

Table 1. Univariate and Bivariate Percentage Means for Sexual Identity,
Sociodemographic Correlates, Confounding Risk Factors, and Condom Use
at Last Sexual Encounter

                                                      Heterosexually
                                           Total          Active
                                         (n = 3703)     (n = 3501)
Sexual Orientation
  Nonheterosexual                            9.11           6.5
Sociodemographics
Race/Ethnicity
  Black                                     12.5           12.5
  Latino                                    17.2           17.1
  Other race                                 5.1            5.2
  Age (c)                                   31.3           31.4
  Married                                   51.3           52.8
Socioeconomic Status
  Bachelor's degree or more                 21.2           21.2
  $50,000 or more                           40.7           41.2
Confounding Risks
  Intravenous drug use (any)                 0.5            0.4
  Sex work (for money or drugs)              1.7            1.3
  High during sex                           10.0            9.6
    (at least 50% of the time)
  Three or more partners in past year       11.3            9.7
Dependent Variables
  Condom use at last sex (last male         30.1           29.6
    partner for bisexually active men)
  Condom use at last sex (last female       30.3           29.6
    partner for bisexually active men)

                                         Homosexually    Bisexually
                                            Active         Active
                                          (n = 141)       (n = 61)

Sexual Orientation
  Nonheterosexual                           96.6 (a)      48.7 (a,b)
Sociodemographics
Race/Ethnicity
  Black                                     14.1          16.2
  Latino                                    14.4          27.9 (b')
  Other race                                 1.7           6.9 (b')
  Age (c)                                   31.2          29.9
  Married                                    0.41 (a)     23.7 (a,b)
Socioeconomic Status
  Bachelor's degree or more                 26.9          11.8 (b)
  $50,000 or more                           32.8          15.5 (a,b)
Confounding Risks
  Intravenous drug use (any)                 0.5          10.0 (a,b)
  Sex work (for money or drugs)              8.5          19.3 (a,b)
  High during sex                           15.6 (a')     35.9 (a,b)
    (at least 50% of the time)
  Three or more partners in past year       42.2 (a)      84.9 (a,b)
Dependent Variables
  Condom use at last sex (last male         44.1 (a)      41.4
    partner for bisexually active men)
  Condom use at last sex (last female       44.1 (a)      64.3 (a,b)
    partner for bisexually active men)

Source: National Survey of Family Growth, Cycle 6-2002,
National Center for Health Statistics.

(a) Statistically different from heterosexually active men
([alpha] [less than or equal to] .01).

(a') Marginally different from heterosexually active men
(.01 < [alpha] [less than or equal to] .05).

(b) Statistically different from homosexually active men
([alpha] [less than or equal to] .01).

(b') Marginally different from homosexually active men
(.01 < [alpha] [less than or equal to] .05).

(c) Mean age is presented.

Table 2a. Logistic Regression Models of Condom Use at Last Sexual
Encounter (Male Partner for Bisexually Active Men) Regressed on Sexual
Orientation, Sociodemographic, and Confounding Risk Factors (a,b)

                                    Model 1             Model 2

Sexual Orientation
  Heterosexually active              0.59 ([dagger])      0.62
  Homosexually active                1.12                 1.07
  Bisexually active                  1.00 (ref.)          1.00 (ref.)
  Nonheterosexual identity,                               1.09
Race/Ethnicity
  Black
  Latino
  Other race
  White
Other Sociodemographics
  Age
  Married
  Bachelor's degree or more
  Income of $50,000 or more
Confounding Risk Factors
  Intravenous drug use (any)
  Sex work (for money or drugs)
  High during sex
    (at least 1/2 of time)
  Three or more partners in
    past year
Log Odds of the Intercept           -0.349 (ns)          -0.394 (ns)
Wald [chi square]                   11.52 **             11.89 **
-2 Log Likelihood                 4657.04              4656.68

                                   Model 3              Model 4

Sexual Orientation
  Heterosexually active              0.64                 0.79
  Homosexually active                1.23                 1.04
  Bisexually active                  1.00 (ref.)          1.00 (ref.)
  Nonheterosexual identity,          0.99                 0.88
Race/Ethnicity
  Black                              2.25 ***             1.87 ***
  Latino                             1.43 ***             1.32 **
  Other race                         1.34 ([dagger])      1.46 *
  White                              1.00 (ref.)          1.00 (ref.)
Other Sociodemographics
  Age                                                     0.94 ***
  Married                                                 0.31 ***
  Bachelor's degree or more                               1.42 ***
  Income of $50,000 or more                               0.98
Confounding Risk Factors
  Intravenous drug use (any)
  Sex work (for money or drugs)
  High during sex
    (at least 1/2 of time)
  Three or more partners in
    past year
Log Odds of the Intercept           -0.6081               1.484 ***
Wald [chi square]                   77.52 ***           510.85 ***
-2 Log Likelihood                 4591.58              4066.8

                                   Model 5

Sexual Orientation
  Heterosexually active              0.91
  Homosexually active                1.09
  Bisexually active                  1.00 (ref.)
  Nonheterosexual identity,          0.88
Race/Ethnicity
  Black                              1.80 ***
  Latino                             1.29 *
  Other race                         1.46 *
  White                              1.00 (ref.)
Other Sociodemographics
  Age                                0.94 ***
  Married                            0.32 ***
  Bachelor's degree or more          1.41 ***
  Income of $50,000 or more          0.98
Confounding Risk Factors
  Intravenous drug use (any)         0.47
  Sex work (for money or drugs)      1.36
  High during sex                    0.77 *
    (at least 1/2 of time)
  Three or more partners in          1.42 **
    past year
Log Odds of the Intercept            1.342 ***
Wald [chi square]                  520.88 ***
-2 Log Likelihood                 4051.71

Source: National Survey of Family Growth, Cycle 6-2002,
National Center for Health Statistics.

(a) N = 3,556.

(b) Odds ratios are presented.

([dagger]) .05 < p [less than or equal to] .10,

* .01 < p [less than or equal to] .05,

** .001 < p [less than or equal to] .01,

*** p [less than or equal to] .001.

Table 2b. Logistic Regression Models of Condom Use at Last Sexual
Encounter (Female Partner for Bisexually Active Men) Regressed on
Sexual Orientation, Sociodemographic, and Confounding
Risk Factors (a,b)

                                Model 1               Model 2

Sexual Orientation
  Heterosexually active          0.23 ***              0.24 ***
  Homosexually active            0.44 *                0.41 *
  Bisexually active              1.00 (ref.)           1.00 (ref.)
  Nonheterosexual identity                             1.13
Race/Ethnicity
  Black
  Latino
  Other race
  White
Other Sociodemographics
  Age
  Married
  Bachelor's degree or more
  Income of $50,000 or more
Confounding Risk Factors
  Intravenous drug use
    (any)
  Sex work
    (for money or drugs)
  High during sex
    (at least 1/2 of time)
  Three or more partners
    in past year
Log Odds of the Intercept        0.588 ([dagger])      0.532 ([dagger])
Wald [chi square]               28.82 ***             29.43 **
-2 Log Likelihood             4656.06               4655.43

                                Model 3              Model 4

Sexual Orientation
  Heterosexually active          0.25 ***             0.28 ***
  Homosexually active            0.47 *               0.36 **
  Bisexually active              1.00 (ref.)          1.00 (ref.)
  Nonheterosexual identity       1.02                 0.90
Race/Ethnicity
  Black                          2.30 ***             1.91 ***
  Latino                         1.43 ***             1.32 **
  Other race                     1.32 ([dagger])      1.44 *
  White                          1.00 (ref.)          1.00 (ref.)
Other Sociodemographics
  Age                                                 0.94 ***
  Married                                             0.31 ***
  Bachelor's degree or more                           1.42 ***
  Income of $50,000 or more                           0.98
Confounding Risk Factors
  Intravenous drug use
    (any)
  Sex work
    (for money or drugs)
  High during sex
    (at least 1/2 of time)
  Three or more partners
    in past year
Log Odds of the Intercept        0.339                2.543 ***
Wald [chi square]               96.77 ***           528.37 ***
-2 Log Likelihood             4587.69              4057.35

                                Model 5

Sexual Orientation
  Heterosexually active          0.35 **
  Homosexually active            0.41 *
  Bisexually active              1.00 (ref.)
  Nonheterosexual identity       0.91
Race/Ethnicity
  Black                          1.83 ***
  Latino                         1.30 **
  Other race                     1.44 *
  White                          1.00 (ref.)
Other Sociodemographics
  Age                            0.94 ***
  Married                        0.33 ***
  Bachelor's degree or more      1.40 ***
  Income of $50,000 or more      0.98
Confounding Risk Factors
  Intravenous drug use           0.83
    (any)
  Sex work                       1.4
    (for money or drugs)
  High during sex                0.81 ([dagger])
    (at least 1/2 of time)
  Three or more partners         1.45 **
    in past year
Log Odds of the Intercept        2.295 ***
Wald [chi square]              537.02 ***
-2 Log Likelihood             4044.09

Source: National Survey of Family Growth, Cycle 6-2002,
National Center for Health Statistics.

(a) N = 3,558.

(b) Odds ratios are presented.

([dagger]) .05 < p [less than or equal to] .10,

* .01 < p [less than or equal to] .05,

** .001 < p [less than or equal to] .01,

*** p [less than or equal to] .001.

Table 3. Logistic Regression Models of Condom Use at Last Sexual
Encounter Regressed on Sexual Orientation, Sociodemographic,
Confounding Risk Factors, and Interaction Terms (a,b)

                                    Model 1              Model 2

                               Condom Use at Last   Condom Use at Last
                               Sexual Encounter     Sexual Encounter
                               (male partner for     (female partner
                               bisexually active     for bisexually
                                    men) (c)         active men) (d)

Sexual Orientation
  Bisexually active                   0.74                 1.49
  Homosexually active                 0.31                 0.31
  Heterosexually active               1.00 (ref.)          1.00 (ref.)
  Nonheterosexual identity            0.83                 0.82
Interaction Terms
  Bisexual behavior x                 2.18                 4.40 *
    nonheterosexual identity
  Homosexual behavior x               4.23                 4.24
    nonheterosexual identity
Log Odds of the Intercept             1.264 ***            1.275 ***
Wald [chi square]                   522.43 ***           538.71 ***
-2 Log Likelihood                  4049.55              4039.16

Source: National Survey of Family Growth, Cycle 6-2002,
National Center for Health Statistics.

(a) For parsimonious presentation, we have excluded sociodemographic
and confounding risk factors from this table. These variables are,
nonetheless, in the statistical models and have been displayed in
Tables 2a and 2b. None of the significance levels for sociodemographic
and confounding risk factors change by including interaction terms.

(b) Odds ratios are presented.

(c) N = 3,556.

(d) N = 3,558.

* .01 < p [less than or equal to] .05.