Participants were 5037 residents (3444 men, 1593 women) in ten sections of a primarily Xhosa-speaking township outside Cape Town, South Africa. All participants were age 18 or older (median = 30 years old). Nearly all (98%) participants identified as Black African, 53% were unemployed, and 50% had not matriculated school.
Research setting and procedures
The study occurred in a township located 20 km outside of Cape Town's central business district. Residents are primarily of Xhosa heritage. Neighborhoods were defined as an area ½ km wide that contained at least one shebeen. Using methods described by Weir et al. [11, 19], we conducted rapid community assessments to identify 10 shebeens separated by ≥ 1 km. All shebeens were visited and we interviewed owners, managers, and patrons to confirm that the shebeens served ≥ 75 patrons per week. Surveys were conducted by 8 indigenous field workers who spoke both Xhosa and English. The field workers approached persons on the street and individuals socializing and drinking in the shebeens, and asked them if they would complete an anonymous, self-administered survey. Individuals who agreed to participate in the study (95%) were given a survey that most completed within 20 minutes. Participants were compensated for their time with a non-monetary item (keychain or shopping bag). Surveys were collected inside (48% of men and 37% of women) and outside (52% of men and 63% of women) of the shebeens. All procedures were approved by the Institutional Review Boards of the University of Connecticut, Syracuse University, and the Human Sciences Research Council or South Africa.
Participants were asked to report demographic characteristics, shebeen attendance, HIV risk history and sexual behaviors including drinking alcohol in conjunction with sex.
Participants reported their age, race, cultural heritage, education, marital and employment status. Participants also indicated whether they had been tested for HIV, their most recent test result, and STI treatment history.
Participants were given a list of all the shebeens in their township section (as well as a response choice for "any other shebeen") and asked how many times they went to these shebeens in the past month; response options were (a) never, (b) 1 to 4 times, (c) 5 to 10 times, (d) 11 to 20 times or (e) 21 or more times.
Sexual risk behaviors
Participants reported the number of male and female sex partners they had in the past month and their number of specific sex acts (vaginal and anal intercourse with and without condoms). We calculated the percent of intercourse occasions protected by condoms for vaginal and anal intercourse separately. We also asked how many times in the past month participants drank alcohol before sex and how many times they had a partner who drank alcohol before sex. We selected a one-month timeframe with open response formats to improve recall accuracy and provide unanchored responses .
Seventy-two men reported sex with male partners in the previous month. Of these men, 41 (56%) reported only male partners and 31 (44%) reported male and female partners; 46 (65%) of these men reported unprotected anal intercourse. Consistent with our primary research question, and to avoid confounding homosexual with heterosexual anal intercourse, we removed these 72 men from further analyses.
We performed descriptive analyses to examine the demographic characteristics, alcohol use, and sexual practices of men and women who engaged in anal intercourse versus those who did not. Preliminary comparisons of persons surveyed inside and outside of shebeens indicated few differences. Data collected inside and outside shebeens were therefore collapsed. We defined risk as engaging in any anal intercourse (because of the tendency toward inconsistent condom use and the potential for condom failures). We first compared men and women on the frequency of engaging in sexual risk behaviors using logistic regressions. We also performed logistic regression analyses to test the hypothesis that multiple independent behaviors would cluster to predict anal intercourse. The model simultaneously tested non-overlapping but significant predictors of anal intercourse from the previous analyses. Results from the logistic regressions are reported as odds ratios (OR) with 95% confidence intervals (CI). All analyses defined statistical significance as p < .05.
We also performed a modeling exercise to estimate the increased risk for HIV acquisition conferred by unprotected anal intercourse to women in our sample. We focused on women because receptive anal intercourse is the highest risk behavior for HIV transmission and because differences in the per-act transmission probabilities for insertive anal intercourse and insertive vaginal intercourse are believed to be minimal.
In these analyses a Bernoullian model of HIV transmission  was used to estimate female study participants' risk of HIV acquisition:
where m denotes the number of sex partners and n
3, and n
4, denote the average number of acts, per partner, of unprotected vaginal intercourse, condom-protected vaginal intercourse, unprotected anal intercourse, and protected anal intercourse, respectively. The following parameter values were used in the base-case analysis: per-act transmission probability for receptive vaginal intercourse, αV = 0.003 [varied from 0.0014 to 0.0063 [3, 22, 23]. in the sensitivity analyses]; prevalence of HIV among sex partners, π = 10% (5% to 20%); and condom effectiveness, ε = 90% (0 to 100%). The per-act transmission probability for receptive anal intercourse was assumed to be 5.7 (2.1 to 14.1) times larger than the transmission probability for receptive vaginal intercourse. Therefore, the per-act transmission probability for receptive anal intercourse ranged from αA = 0.003 to 0.089, with a base-case value of 0.017, consistent with a recent meta-analysis .
Data from the study were used to estimate the average risk of HIV acquisition for study participants who engaged in anal intercourse (riskA) and the average risk of HIV acquisition for participants who did not engage in this activity (riskB). A weighted average was then calculated by multiplying these risk estimates by the proportion of study participants who fell into each of the corresponding categories (pA and pB): average risk1 = pA*riskA + pB*riskB. An additional "hypothetical" risk estimate (riskC) was calculated by replacing all acts of receptive anal intercourse with vaginal intercourse. A second weighted average was then calculated using the hypothetical estimate: average risk2 = pA*riskC + pB*riskB. The ratio of the weighted averages, r = average risk1/average risk2 is an indicator of the amount by which anal intercourse increased participants' risk of HIV acquisition. This ratio also can be translated into an estimate of the risk reduction (1 - 1/r) that would be achieved if all acts of anal intercourse were replaced by vaginal intercourse acts.