Network Mixing and Network Influences Most Linked to HIV Infection and Risk Behavior in the HIV Epidemic Among Black Men Who Have Sex With Men

Abstract

Objectives. We evaluated network mixing and influences by network members upon Black men who have sex with men.

Methods. We conducted separate social and sexual network mixing analyses to determine the degree of mixing on risk behaviors (e.g., unprotected anal intercourse [UAI]). We used logistic regression to assess the association between a network “enabler” (would not disapprove of the respondent’s behavior) and respondent behavior.

Results. Across the sample (n = 1187) network mixing on risk behaviors was more assortative (like with like) in the sexual network (r_sex, 0.37–0.54) than in the social network (r_social, 0.21–0.24). Minimal assortativity (heterogeneous mixing) among HIV-infected men on UAI was evident. Black men who have sex with men reporting a social network enabler were more likely to practice UAI (adjusted odds ratio = 4.06; 95% confidence interval = 1.64, 10.05) a finding not observed in the sexual network (adjusted odds ratio = 1.31; 95% confidence interval = 0.44, 3.91).

Conclusions. Different mixing on risk behavior was evident with more disassortativity among social than sexual networks. Enabling effects of social network members may affect risky behavior. Attention to of high-risk populations’ social networks is needed for effective and sustained HIV prevention.

The HIV epidemic among men who have sex with men (MSM) has not only grown to alarming levels overall, but it also is one that demonstrates significant and marked racial disparities. In 2008, 28% of MSM with new HIV infection were Black, and among MSM aged 13 to 29 years, the number of new infections in Black MSM was nearly twice that of White MSM.^1,2

Traditional epidemiological approaches have made limited headway in explaining these findings because they tend to focus on the role of individual risk behaviors in shaping rates of HIV infection. The higher rates of HIV among Black MSM may not be explained by individual-level risk behaviors alone, and instead may be attributed in part to social and sexual network factors.^3,4 But efforts to further illuminate these factors have been largely unsuccessful as they have often used sampling methodologies that can distort accurate measurement of existing networks of these MSM (e.g., lack of weighting and focus on most recent sexual partner).^5,6 Furthermore, up until now, network analyses have not examined Black MSM’s nonsexual social networks; such networks may contribute to the disparities observed (e.g., lack of embedded social network members⁷) and might provide opportunities for future interventions.

Some research has explained disparities in HIV rates by examining sexual network mixing patterns within and between racial subgroups.^8,9 Previously, we demonstrated that higher rates of sexually transmitted infections (STIs) within the African American community were related to sexual network mixing patterns.¹⁰ Higher levels of disassortative mixing—core high-risk groups mixing with peripheral low-risk groups—within the African American community, combined with limited interracial mixing, was a major contributor for the disproportionately higher rates of STIs among Blacks than among Whites. Similar sexual network mixing explanations have been demonstrated among Blacks in the Southeastern United States.¹¹ Drug use behavior was found to be highly assortative (like behavior with like), whereas sexual behavior in the form of concurrent (or simultaneous) partnerships was minimally assortative.

In contrast to the attention devoted to sexual^12–17 and drug-use networks,^18–23 comparatively little research has been conducted on how nonrisk social networks comprising MSM’s close friends and family members can affect STI and HIV transmission, with a few notable exceptions.^7,24,25 Social learning and differential association theories^26,27 hold that risky behaviors, including rationalizations for them, diffuse through social networks of close ties. Furthermore, network members influence high-risk behavior by virtue of the behavioral examples they provide, the normative pressures they exert, and MSM’s perceptions of these influences.^28–30 Research has shown in a variety of contexts that risky sexual and substance use behavior is affected by individuals’ perceptions of what their network members do, regardless of whether those perceptions are accurate.^31–33 Studying Black MSM’s normative contexts may help researchers identify not only those social conditions that facilitate risky behavior, but also potential network influences that can be exploited or modified to encourage the spread of HIV prevention behavior through modification of a social network. To date, most work that has examined the indirect role of social networks on the spread of HIV has focused primarily on the role of having social network ties in general, but has not specified the mechanisms through which social network ties affect the risk behavior of MSM.^34,35

Formal social network analysis of high-risk populations has focused on MSM and injecting drug users in general and not specifically on Black MSM.^25,36 One recent pilot study³⁷ demonstrated that sexual partners of Black MSM were mostly introduced through friends. Known risk behaviors associated with HIV infection and that could be “transmitted” through a social network include sex-drug use³⁸ and unprotected anal intercourse (UAI). Moreover, group sexual intercourse has also recently gained increased attention as an important risk practice^39,40 that can complicate network analysis.⁴¹ Important influences and practices such as these, however, have not been previously explored through social network analysis within Black MSM despite this population’s position as a group with the highest risk of HIV infection in the United States. Furthermore, network patterns that potentially confer risk, such as disassortative social mixing, have also not been explored within this population as opposed to the larger Black community.^10,11 We conducted a detailed analysis of close social and sexual networks of Black MSM to determine the salient properties and components of these networks that are most related to HIV risk and preventive behavior among these men.

METHODS

Between January and June of 2010, Black MSM were recruited in Chicago by respondent-driven sampling (RDS)⁴² to participate in the study. All interviews took place at partnering community-based organizations by Black MSM community members trained by the University of Chicago Survey Lab. HIV voluntary counseling and testing were conducted according to procedures and protocols at each organization. Procedures and protocols were approved by institutional review boards at the University of Chicago, Howard Brown Health Center, and National Opinion Research Center. Informed consent was obtained from all respondents and waived for network members listed by respondents.

Study Participants

Eligibility criteria.

Study participants included both study respondents who were interviewed and the network members about whom they reported. Study respondents were eligible for the study if they (1) self-identified as African American or Black, (2) identified as male, (3) were aged 18 years or older, (4) reported anal intercourse with a man within the past 12 months, and (5) were willing and able to provide informed consent at the time of the study visit. Network members were eligible if they were named by respondents during the interview.

Recruitment.

Respondent-driven sampling has been widely applied to study hard-to-reach populations such as injecting drug users, sex workers, and MSM.^43–46 Recent theoretical and empirical work has assessed the strengths and weakness of RDS.^43,47,48 This work has emphasized the importance of careful selection of “seeds” from diverse sources and sufficient iterative rounds of recruitment to penetrate further reaches of the larger social networked population being studied—“recruits.”

We recruited 21 seeds from 4 venues: (1) 4 seeds were recruited from a local federally qualified health center, (2) 8 seeds were referred from existing effective behavioral intervention prevention programs,⁴⁹ (3) 4 seeds were recruited from a substance use treatment program, and (4) 5 seeds were recruited through fliers posted at a lesbian, gay, bisexual, and transgender care center. Seeds were asked to refer up to 4 recruits who were MSM from their social networks, with each subsequent recruit doing the same by using vouchers.

Survey Instruments

Social network assessment.

In designing our Men’s Assessment of Social and Risk Networks (SRN Instrument), we followed an established method of gathering egocentric network data⁵⁰ that is used in several large national surveys, including the General Social Survey,⁵¹ the National Health and Social Life Survey,⁵² and the National Social Life, Health, and Aging Project.⁵³ We asked a “name generator” question during the course of the face-to-face interviews to elicit from each respondent a set of social network members who may indirectly affect the respondent’s risky behaviors. We selected the name generator to identify network “confidants”⁵⁴ who have opportunities, through everyday interactions with the respondent, to exercise normative pressure or informal control, and to exchange information or advice regarding risky behavior: “Let’s make a list of your closest associates with whom you may share information about yourself, your physical and mental health, and your social and sexual lifestyles.” We entered these names into a roster that was recorded for future reference. We then followed up with a series of “name interpreter” questions about each network member’s attributes (e.g., age) and the best descriptor of the nature of his or her relationship with the respondent (e.g., friend) from a list of 18 possible categories. Research has shown that 5 network members is optimal for time and effort to field egocentric network surveys.⁵⁵

We focused on 2 types of confidant subnetworks—(1) social (nonsexual) and (2) sexual—to identify different mechanisms of mixing and influence likely to be driving risk behavior. These networks were determined by a name interpreter who asked respondents to characterize each network member (e.g., friend, acquaintance, sexual partner) in the roster. Because respondents were asked to characterize each network member using 1 descriptor, these networks were nonoverlapping. The sexual network includes the respondent and his nominated network members who were classified into 1 of 4 categories (e.g., nonprimary partner). The social network included the respondent and his nominated network members who were classified into 1 of the other 14 non–sex-partner categories (e.g., friend, relative; Table 1).

TABLE 1—

Respondent (n = 204) and Network Member (n = 983) Attributes and Tie Characteristics: Black Men Who Have Sex With Men and Their Network Members, Chicago, 2010

Attributes	Respondents, No. (%)	Total Network Members, No. (%)	Sexual Network Members, No. (%)	Social Network Members, No. (%)
Participant characteristics
Age, y
< 20	26 (13.3)	112 (11.5)	21 (8.1)	91 (12.8)
20–24	50 (25.6)	219 (22.4)	55 (21.2)	159 (22.3)
25–34	41 (21.0)	256 (26.2)	71 (27.3)	185 (26.0)
35–45	47 (24.1)	217 (22.2)	67 (25.8)	150 (21.1)
≥ 46	31 (15.9)	173 (17.7)	46 (17.7)	127 (17.8)
Race/ethnicity
White	0 (0)	17 (1.7)	7 (2.7)	10 (1.4)
Black	204 (100)	933 (95.0)	246 (93.9)	682 (95.4)
Hispanic/Latino	0 (0)	10 (1.0)	3 (1.2)	7 (1.0)
Other	0 (0)	22 (2.2)	6 (2.3)	16 (2.2)
Educationa
< high school	26 (12.8)	…	…	…
High school	75 (36.8)	…	…	…
Some college	89 (43.6)	…	…	…
College	14 (6.9)	…	…	…
Employed	81 (39.7)	…	…	…
Gender
Male	204 (100)	759 (77.2)	235 (89.7)	519 (72.5)
Female	0 (0)	187 (19.0)	20 (7.6)	167 (23.3)
Trans	0 (0)	37 (3.76)	7 (2.7)	30 (4.2)
HIV status
HIV-positive	88 (44.4)	224 (32.9)b	88 (46.8)	135 (27.7)
HIV-negative	110 (55.6)	457 (67.1)b	100 (53.2)	353 (72.3)
Network tie characteristics
Relationship types
Friend	…	426 (43.6)	…	426 (59.5)
Family	…	176 (18.0)	…	176 (24.6)
Primary partner	…	59 (6.0)	59 (22.5)	…
Ex–primary partner	…	61 (6.2)	61 (23.3)	…
Other sexual partner	…	142 (14.5)	142 (54.2)	…
Otherc	…	114 (11.7)	…	…
Gender of sexual partners
Men	…	703 (72.3)	192 (73.8)	506 (71.6)
Women	…	77 (7.9)	5 (1.9)	72 (10.2)
Both	…	178 (18.3)	63 (24.2)	115 (16.3)
Neither	…	14 (1.4)	0	14 (2.0)
Importance of network members’ advice about sex drugs
Very important	…	418 (43.1)	110 (42.2)	308 (43.6)
Somewhat important	…	354 (36.5)	79 (30.3)	272 (38.5)
Not important	…	198 (20.4)	72 (27.6)	126 (17.8)
How often respondent discusses sexual activity with network member
A lot	…	336 (34.2)	127 (48.5)	207 (29.0)
Sometimes	…	319 (32.5)	82 (31.3)	236 (33.0)
Rarely	…	176 (17.9)	30 (11.4)	146 (20.4)
Never	…	151 (15.4)	23 (8.8)	126 (17.6)
How often respondent discusses drugs with network member
A lot	…	261 (27.9)	76 (30.5)	185 (27.0)
Sometimes	…	218 (23.3)	69 (27.7)	149 (21.8)
Rarely	…	92 (9.9)	20 (8.0)	72 (10.5)
Never	…	363 (38.9)	84 (33.7)	279 (40.7)
How likely is network member to disapprove of unprotected sexual activity
Very likely	…	577 (61.1)	115 (44.6)	457 (67.1)
Somewhat likely	…	185 (19.6)	66 (25.6)	119 (17.5)
Not likely	…	182 (19.3)	77 (29.8)	105 (15.4)

^aEducation status not collected for network members.

^bFrom among those who indicated that respondent knows network members’ HIV status.

^cIncludes drug supplier, drug client, drinking buddy, drug buddy, neighbor, coworker, boss, other.

Sociodemographic, attitude, and behavior measures.

We adapted demographic and HIV status items from the Centers for Disease Control and Prevention’s National HIV Behavioral Surveillance Survey, MSM Cycle⁵⁶ and the visit 51 Core Behavioral Survey of the Multicenter AIDS Cohort Study (available at http://www.statepi.jhsph.edu/macs/macs.html). As in previous work, we measured serodiscordant UAI in response to the item “In the past six months, have you had unprotected anal sex with a male partner of unknown or different HIV status?” and sex-drug use with “Have you ever used any of these substances [from a hand-card of 12 categories] as ‘sex drugs,’ that is to make sex easier, better, last longer, or something similar?”^38,57 We measured group sexual intercourse as “having sex with two or more individuals at the same time.” We assessed behavioral measures in frequency terms over the past year and coded for these analyses as present if they were reported as at least monthly. HIV testing and counseling were offered on site, and HIV-infected clients were referred to appropriate services.

Analysis

Respondent-driven sampling.

We directed our initial analysis toward evaluating the plausibility of assumptions such as random and reciprocal referral, evaluating the degree of homophily with respect to various risk-related outcomes, and developing estimates of the inclusion probabilities for use as weights. Throughout all of our substantive analyses, we utilized the RDS weights and computed design-based standard errors through analytic methods,⁵⁸ and compared these results to those obtained without the weights and assuming independent observations.

Mixing analysis can inform public health researchers as to how behaviors (e.g., sex-drug use) or statuses (e.g., HIV status) are distributed between multiple pairs of individuals within a network. The extent that this network exhibits assortative (i.e., like with like) social or sexual mixing patterns with respect to a given classification can be quantified.^11,59 We calculated an assortativity coefficient to describe the mixing patterns in our sample.⁵⁹ The assortativity coefficient is calculated from the mixing matrix—the proportion of total ties in a cross-tabulation of partnerships between people who do and do not have a risk attribute. This coefficient is defined as:

where r is the assortativity coefficient, Tr e is the trace of the matrix, and e is the matrix whose elements are the cell values, e_ij, of the mixing matrix. This formula gives r = 1 when there is perfect assortative mixing and all partnerships are concordant for the characteristic of interest (e.g., sex-drug use). When the coefficient is 0, this indicates random mixing on the characteristic. This random mixing at r = 0, however, is closer to a perfectly disassortative network and can thus be interpreted in this way.⁵⁹ On the basis of previous work, coefficients of 0.35 or larger were characterized as assortative, 0.26 to 0.34 as moderately assortative, and 0.15 to 0.25 as minimally assortative.¹¹ We computed assortativity coefficients based upon mixing matrices for the presence or absence of a characteristic for the respondent and up to 5 network members for the social and sexual networks, respectively. Characteristics included sex-drug use, UAI, group sexual intercourse, and HIV status. We calculated separate coefficients according to the HIV status for HIV-infected and uninfected respondents.

Risk network effect analysis.

The primary outcomes of this study were risk-related behaviors among respondents: sex-drug use, UAI, and group sexual intercourse. We initially examined these outcomes individually according to the following generalized linear model.⁶⁰

where Y is a measure of risk, X is 1 or more variables characterizing the respondent’s network, and Z are the additional covariates—individual sociodemographics, interview site, and personal network size. Our parameter of interest is b, which describes the association between network characteristics and risk. We created 2 models for each risk behavior outcome. The first examined whether the respondent has at least 1 network member who engages in the risk behavior—a measure of homophily. The second explored the effect of at least 1 network member who would not disapprove of the respondent’s sex-drug use, UAI, or group sexual intercourse—an “enabler”—with adjustment for all variables, including homophily, from the first model.

RESULTS

Twenty-one seeds generated the study respondent sample with each chain averaging 5.8 new individuals (range = 0–42) and up to 9 waves were completed. With subsequent waves of recruitment, the study respondents became younger, were more HIV-negative, reported less sex-drug use and UAI, and were less likely to report network confidants who disapproved of sex-drug use and UAI. We determined the RDS weights and computed design–based standard errors by using analytic methods,⁵⁸ and comparing these results to those obtained without the weights yielded weights of 1 divided by the square root of the respondent’s network degree. We used these weights for all regression analyses.

The sample included a total network (n = 1187) generated from 21 seeds and included respondents (n = 204) and other listed network members (n = 983) as demonstrated in Figure 1. Attributes and tie characteristics of study participants are depicted in Table 1.

FIGURE 1—

Total generated network sample among Black men who have sex with men (MSM) (n = 983): Chicago, 2010.

Note. Stratification of MSM respondent confidant network into social and sexual network was based on respondent characterization of confidants as sexual partners or not.

^aRelationship type missing for 5 network members.

Table 2 depicts the distribution of risk characteristics for respondents and sexual or social network members by self-reported HIV status of respondents.

TABLE 2—

Distribution of Risk Characteristics for Black Men Who Have Sex With Men Respondents and Sexual Partner and Social Network Members by Self-Reported HIV Status of Respondents: Chicago, 2010

	All Respondents		Social Network		Sexual Network
Risk Characteristic	HIV-Positive (n = 89), % (No.)	HIV-Negative (n = 108), % (No.)	HIV-Positive (n = 296), % (No.)	HIV-Negative (n = 397), % (No.)	HIV-Positive (n = 140), % (No.)	HIV-Negative (n = 110), % (No.)
Sex-drug use	47.2	29.6	62.6 (246)	60.6 (307)	68.5 (130)	72.3 (101)
UAI	43.8	28.6 (105)	49.4 (241)	36.3 (300)	71.7 (120)	40.8 (103)
Group sexual intercourse	26.1	18.7 (105)	20.7 (232)	21.0 (300)	35.9 (106)	28.9 (97)
HIV statusa	…	…	46.6 (223)	11.2 (258)	72.7 (110)	7.9 (76)

Note. UAI = unprotected anal intercourse. HIV-positive and HIV-negative in this table refers to respondent.

^aHIV status for network members as reported by respondent (nonmissing cases only).

We generated separate social and sexual network mixing models according to 3 separate risk behaviors and stratified by HIV status (Figure 2). Overall, network mixing on 3 risk behaviors was generally more assortative (like with like) with respect to sexual network members (r_sex, 0.37–0.54; σ_r, 0.28–0.58) than it was with respect to nonsexual social network members (r_social, 0.21–0.24; σ_r, 0.15–0.28). This suggests that Black MSM maintain ties with nonsexual social network members who do not demonstrate the same risk behavior profiles. When we stratified social and sexual networks by HIV status, 1 risk behavior—UAI—demonstrated minimal assortativity among the HIV-infected men to a degree where the coefficients of the distinct sexual and social network overlapped (r_sex = 0.21; σ_r(low-high) = 0.09–0.33 and r_social = 0.17; σ_r(low-high) = 0.08–0.26), respectively.

FIGURE 2—

Mixing within social and sexual networks among Black men who have sex with men, by risk behavior and HIV status: Chicago, 2010.

Note. AC = assortativity coefficient; UAI = unprotected anal intercourse. The AC is calculated from the mixing matrix—the proportion of total ties in a cross-tabulation of ties between people who do and do not have a risk attribute. Social network n = 716. Sexual network n = 262. Nodes and error bars within each behavior category indicate the AC for all dyads (social or sexual) in the sample, further stratified by HIV status of participants. An AC of 1 would indicate perfectly assortative mixing (e.g., respondents who practice a behavior only have sexual relations with or are social with network members who also practice that behavior), whereas respondents who do not practice the behavior only have sexual relations with or are social with network members who also do not practice the behavior. The combined AC indicates a general trend of increasing assortativeness for riskier behaviors (starting with sex-drug use and ending with HIV status [or serosorting]).

We measured the association of network members’ risk behavior with respondent’s behavior and the association between existence of at least 1 enabling network member (not entirely disapproving of risk behavior) and respondent’s behavior. Results are consistent with our expectations in that they suggest that Black MSM whose network members participate in a given behavior are significantly more likely to engage in a particularly risky behavior (Table 3). For example, as shown in the first multivariate model in column 1, MSM respondents who have at least 1 network member in their social network who uses drugs to enhance sexual experience are more than 5.5 times as likely to use sex drugs than men who do not have such a network member (adjusted odds ratio [AOR] = 5.87; P < .01). There is also a potentially stronger association between network members’ attitudes toward respondent’s behavior. For example, Black MSM who report that at least 1 network member is not entirely disapproving of sex-drug use, an enabler, are more than 6 times as likely to use sex drugs than men who do not have such a network member (AOR = 6.35; P < .001) and the previous point estimate of the association between network members’ sex-drug use and respondents’ sex-drug use decreases (AOR = 3.70; P < .05).

TABLE 3—

Two Multivariable Models Each for Social and Sexual Network Relationships With Respondents (n = 204) by Sex-Drug Use and Unprotected Anal Intercourse Among Black Men Who Have Sex With Men: Chicago, 2010

	Social Network		Sexual Network
Characteristic	Model 1, AOR (95% CI)	Model 2, AOR (95% CI)	Model 1, AOR (95% CI)	Model 2, AOR (95% CI)
Sex-drug use
Network member behaviors
Sex-drug use	5.87*** (2.02, 17.05)	3.70* (1.26, 10.90)	8.25** (2.09, 32.61)	5.78* (1.43, 23.41)
UAI	…	…	…	…
Enabling network membera	…	6.35*** (2.20, 18.29)	…	8.59** (1.93, 38.26)
Unprotected anal intercourse
Network member behaviors
Sex-drug use	…	…	…	…
UAI	2.22 (0.95, 5.22)	1.57 (0.64, 3.87)	13.19*** (3.72, 46.75)	11.75*** (3.15, 43.77)
Enabling network membera	…	4.06** (1.64, 10.05)	…	1.31 (0.44, 3.91)

Note. AOR = adjusted odds ratio; CI = confidence interval; UAI = unprotected anal intercourse. Model 1 is an analysis of homophily of risk behaviors between a respondent and his social and sexual networks. Model 2 adds whether at least 1 network member is not entirely disapproving of respondent’s behavior to model 1. Both models control for age, employment status, education, HIV status, size of the social or sexual network, and site of participant recruitment.

^aAt least 1 network member is not entirely disapproving of behavior.

*P < .05; **P < .01; ***P < .001.

DISCUSSION

It is often supposed that interventions to halt the sexual transmission of HIV must engage and work within sexual networks because of the mixing of high- and low-risk individuals.^10,61,62 Ties between high-risk individuals, however, are also critical to disease transmission,⁶³ and interventions at the sexual network level may require behavior change for both members of a high-risk encounter to reduce spread of disease through a network.⁶⁴ Our study participants demonstrated sexual network mixing that was assortative (r ≥ 0.35)¹¹ on specific behaviors and statuses (like with like); contact between high- and low-risk individuals, however, was most apparent within social networks and was found to be minimally assortative (r = 0.15–0.25).¹¹ In further analysis, these minimally assortative social networks demonstrated heterogeneous social influences on behavior; homophily in the network was evident for some behaviors (e.g., sex-drug use) but not others (e.g., UAI). In the case of UAI, however, the presence of an “enabler” within the social network who condoned this behavior increased the likelihood of the respondent practicing UAI, an influence not evident in the sexual networks.

Although these results may not seem surprising—sexual ties are more likely to be based on shared risk behaviors than are social ties—they point to nonsexual social networks as an important medium for interventions targeting high-risk individuals, given the observed behavioral heterogeneity within these networks. These diverse networks exhibited mixing with others of different levels of risk across a behavior (minimally assortative)—but with similar levels of this minimally assortative mixing across behaviors. One notable mixing pattern that differed from other behavioral mixing patterns within social networks was that of HIV status. Assortative mixing on serostatus within social networks was evident and considerably higher than mixing on other risk behaviors within the social network; yet mixing on serostatus within the social network was still less assortative than mixing on serostatus within the sexual network (e.g., serosorting—limiting unprotected sexual activity to partners of same serostatus⁶⁵).

Social influence as measured by the presence of at least 1 enabler within Black MSM’s networks was apparent with respect to social network members. But this influence varied across behaviors and networks. In the example of UAI, presence of an enabler was a significant contributor to an individual’s behavior in the social network, but not in the sexual network, whereas for sex-drug use the enabler was important for both network typologies. This suggests that enablers within networks that are more disassortative—people with different behaviors—may be more consequential to the behavior of individuals who are embedded in such networks. The threshold of 1 enabler out of 5 not entirely disapproving was a low one in our study. Interventions directed at Black MSM networks must grapple with the finding that 1 person, an enabler, may have an impact on an individual’s behavior despite disapproval from other network members. This finding has implications for existing theoretical models of HIV risk behavior such as the AIDS Risk Reduction Model or the Theory of Reasoned Action where an enabler may be additive to existing social norm conceptualizations within such models. If networks were rewired—a controversial network intervention⁶⁶—would the respondent change behavior to conform with network norms (social influence)? Or would he shift to networks not entirely disapproving of the behavior (selection)? Longitudinal social network analysis is needed to help disentangle the contributions to homophily of social influence⁶⁷ and network selection⁶⁸ to develop effective social network interventions.⁶⁹

It is interesting that network members with whom the respondent speaks more frequently about sexual activity were more likely to disapprove of UAI (data not shown). This is somewhat surprising, as one might have expected more discussion about sexual activity with those who are less likely to disapprove, in line with Parson’s theory of deviant behavior⁷⁰ and the stress that this deviance could produce in the respondent. It is also worth pointing out that most respondents reported that the majority of their network members disapproved of these behaviors. Thus, it is not as if respondents’ projection bias—projecting their own attitudes or behaviors onto other network members—was widespread. In fact, respondents appear to have been fairly conservative in their estimates of who was an “enabler.”

Assortative mixing on behaviors and HIV status in the sexual network were evident, with some variability across behaviors. Sex-drug use was the least assortative and mixing on HIV status (serosorting) demonstrated the highest assortativity. This suggests that strong and protective behaviors exist within our sampled sexual network. Compared with Black heterosexuals in the Southeastern United States,¹¹ assortativity coefficients on the drug use variables were similar; however, coefficients of sexual behavior variables in our data were higher. In addition, Doherty et al. found no differences in assortativity by risk behavior when the sample was stratified by HIV status.¹¹ In our sexual network analysis we found similar findings for sex-drug use, but for sexual behavior variables, marked differences were evident with HIV-positive individuals demonstrating highly assortative mixing compared with HIV-negative individuals. Cautious interpretation of these differences between studies is necessary because of the different sexual behaviors measured and the very different populations and settings studied. These findings suggest, however, that in contrast to the larger Black community, Black MSM may be strategically mixing based on serostatus and sexual behavior, and are in line with recent findings that mixing on serostatus is unlikely to be a reason for disparities between White and Black MSM.^71,72

There is, however, 1 alarming finding from the sexual mixing pattern with respect to UAI among HIV-infected respondents in our sample: minimally assortative (disassortative) mixing was evident for this subgroup. In fact, the lack of assortativity on UAI in this group was comparable to the social network assortativity for UAI. Although most current efforts targeting Black MSM suggest increasing awareness of HIV status (e.g., decreasing HIV-positive unaware⁷³), these findings suggest that additional secondary prevention efforts such as Prevention for Positives programs for Black MSM are warranted.⁷⁴ Currently there are no specific secondary prevention interventions targeting this population as existing secondary prevention interventions target MSM in general^75,76 or older Black MSM.^77,78

Limitations

There are several potential limitations to this study. Chicago has the highest proportion of African Americans among the 5 largest cities in the United States⁷⁹ and may not be representative of mixing in other cities that have lower numbers of African Americans in total or as a proportion. For example, the possibility of mixing between races as a reason for higher rates of HIV among Black MSM has been raised in other settings,⁸⁰ but has not been found to be the case in Chicago,¹⁰ and the formal mixing analysis conducted here confirms earlier findings of assortative mixing within African Americans.¹⁰ Without longitudinal data, it is impossible to disentangle direct social influence by network members, or their self-selection into enabling social environments.⁶⁹

In addition, all data were self-reported by respondents (typical of egocentric analysis) and, thus, are susceptible to projection bias.⁸¹ If study respondents accurately report on their perceptions of network members’ behavior, however, this could be even more important than the network members’ actual behavior.⁸² Individuals may to some extent, however, justify their actions by believing that others support and encourage those actions.⁸³ We focused on descriptors of network members that best describe them, recognizing that overlapping categories might exist. We also focused on strong ties by utilizing confidants who may not be representative of all sexual partners. Strong ties, however, have the greatest effect on influence⁸⁴ rather than information flow, which is more apparent through weak ties.⁸⁵ Finally, the seeds and recruits may not reflect the larger population of Black MSM and the small sample size decreases the precision of our estimates.

Conclusions

This study is the first published quantitative network analysis of social and sexual mixing patterns among Black MSM. Our findings highlight important enabling effects of social network members as a mechanism through which social contacts may affect risky behavior, above and beyond participants’ perceptions of network members’ own risk-related behaviors. In addition, different levels of mixing assortativeness according to network typology was evident with social networks on the whole being more disassortative than sexual networks across a diverse set of risk behaviors and HIV statuses. Notably, however, there was much more disassortative sexual mixing for UAI as practiced by HIV-infected men. Future HIV prevention interventions may be made more effective by incorporating and potentially altering social networks. This might include additional focus on the norms of the network as well as fostering relationships with specific individuals within the network. In addition, incorporating secondary prevention strategies, such as Prevention for Positives, for spread of HIV from positive to uninfected Black MSM, may complement the numerous existing HIV-prevention programs for uninfected Black MSM.