THE NEED FOR INFORMATION on behaviors related to HIV and sexually transmitted disease (STD) transmission and prevention has resulted in questions on sexual behaviors being added to a number of general population surveys. Questions about male same-sex sexual activity (male-to-male sex, or MTMS) were added to the General Social Survey (GSS) beginning in 1988 and now provide measurements at nine time points between 1988 and 2000. The GSS data indicate higher levels of reported MTMS in surveys conducted since 1996. Because MTMS is a very sensitive behavioral topic and likely to be perceived as stigmatized, any survey estimate of this behavior is probably affected by underreporting and needs to be interpreted carefully. 1
Data and Methods
A national household-based probability sample of US adults aged 18 years and older designed to collect information on a variety of topics of social importance, the GSS has been conducted on an annual or semiannual basis since 1972. 2 The overall response rate for the main interview of GSSs conducted between 1988 and 2000 has averaged 77%. Questions on sexual behavior are part of a separate paper-and-pencil self-administered questionnaire (SAQ) that is filled out by respondents following the completion of the main interviewer-administered interview. We present data for the nine most recent surveys (1988, 1989, 1990, 1991, 1993, 1994, 1996, 1998, and 2000) and estimate the percentage of male respondents who reported MTMS in the past year and at any time since age 18 years. From 500 to 1200 adult male respondents provided information on sexual activity in each of these survey years. Past-year measurement is based on a question about the gender of sex partners in the past year (“Have your partners in the past 12 months been exclusively male, both male and female, or exclusively female?”). Adult MTMS is based on a question on the number of male sex partners since age 18 that was first added to the questionnaire in 1989: “Since your 18th birthday (including the past 12 months), how many male partners have you had sex with?”
We have computed percentages, 95% confidence intervals, and significance values using SAS (version 6.09; SAS Institute, Inc., Cary, NC) and SUDAAN software (release 7.0; Research Triangle Institute, Research Triangle Park, NC). In order to examine trends in reported MTMS, we compared the GSSs conducted after 1996 with those conducted earlier. Differences in reported MTMS between the earlier period and the later period were evaluated with use of a test for the difference in percentages between two subgroups that assumes independent samples for the two subgroups (computed by means of the SUDAAN Descript procedure). The subgroups being compared in this case are the earlier and later survey periods. We also conducted logistic regression analysis to determine whether changes in reported MTMS were statistically significant when covariates were controlled. All estimates are based on weighting factors to account for different probabilities of selection, and confidence intervals and significance tests have been adjusted for the complex sample design with an assumed design effect of 1.3.
The questions on gender of adult and past-year sex partners have a considerable level of missing data, so the results must be interpreted with caution. We estimate that 12.6% of respondents between 1988 and 2000 did not fill out the SAQ, and an additional 2.1% did not provide responses to the questions on gender of sex partners in the past year, for an overall level of nonresponse of 14.7% for these items. This measure of combined nonresponse was higher among the oldest respondents, those with the lowest level of education, and black respondents and was also somewhat higher after 1996 (15.8%, compared with 13.8% for the 1988–1994 surveys).
Because the level of item nonresponse is so much higher than the statistic being estimated, there is a potentially large impact of nonresponse on our estimate of MTMS. We present three estimates of the percentage reporting lifetime and past-year MTMS, using different methods of assigning the missing value cases. The first estimate (definition 1) includes all respondents in the denominator, regardless of whether they completed the SAQ. Definition 2 is limited to the 87% of respondents who completed the SAQ section of the interview and includes all of them in the denominator, whether or not they provided usable responses on the gender of sex partners. Finally, definition 3, the most exclusive, is based only on those who completed the SAQ and provided legitimate responses to the questions on gender of sex partners.
The percentage reporting MTMS in the past year was consistently higher in the most recent 3 years in which the GSS was conducted (1996, 1998, and 2000), although confidence intervals were quite wide for individual-year surveys (Figure 1). Comparing the 1996–2000 surveys with the earlier years indicates that reported MTMS in the past year (but not MTMS since age 18) was significantly higher in the 1996–2000 time period, regardless of the definition used (Table 1). Point estimates of reported MTMS in the past year ranged from 3.1% to 3.7% (depending on the definition used) after 1996, compared with 1.7% to 2.0% for the earlier period.
Multivariate logistic regression analysis was conducted to determine if the increased reporting of MTMS was maintained after controlling for background variables that were associated with nonresponse to the sex behavior items (Table 2). A model was constructed of variables predicting reported past-year MTMS. Variables considered as predictors for the model included time period of survey, age, race, education, employment status, region and area of residence, and one-way interaction terms between these variables and the period of survey (before or after 1996). The final model included only statistically significant predictors; no interaction terms were found to be statistically significant. The final model was evaluated for goodness-of-fit with the Hosmer–Lemeshow chi-square statistic, which indicated adequate fit for the model. 3 In the final model, higher reported MTMS was associated with surveys conducted in the later period and with younger age, black race, metropolitan residence, and education beyond high school.
The 1996–2000 GSS results of 3.1% to 3.7% past-year same-sex sexual behavior among men is higher than estimates from previous similar surveys, which have been in the range of 1% to 2%. 4–7 The GSS data provide evidence of a trend toward higher reporting of MTMS in an ongoing survey using the same methodology over a 12-year period. During this time period there have been no changes in the GSS sampling method or in the wording or placement of the SAQ items on gender of sex partners. One possible explanation of the observed results is a trend toward more complete reporting of a stigmatized and previously underreported behavior in the later time period, perhaps related to other changes that were occurring. It is of interest to note, for example, that data from another GSS question indicate a large increase after 1996 in more favorable attitudes among adults toward same-sex sexual behavior (Table 1). The late 1990s also corresponds to a period of change brought about by the introduction of highly effective therapies for HIV infection 8,9 that could possibly have had an effect on attitudes that would make individuals more willing to discuss same-sex sexual behavior. There was no increase in the reporting of MTMS since age 18 years, possibly because this behavior is less sensitive to report (because it may have occurred in the distant past) and therefore is less subject to underreporting than past-year MTMS behavior. If MTMS in the past year is indeed more completely reported than in earlier periods, this could call into question the accuracy of earlier estimates from the GSS and other surveys that have reported 1% to 2% prevalence of this behavior. The differences we observed, if they are in fact due to shifts in the underlying level of underreporting of MTMS, suggest that, in general, survey research estimates of the proportion of the population that engages in MTMS are likely to be too low by some unknown but not inconsiderable factor.
These findings must be treated with a great deal of caution because of the high levels of missing data; the percentage of respondents who did not provide information on gender of sex partners was about four times the number who reported MTMS. We have found that the results are unchanged under different methods of assigning the missing values and when controlling for factors associated with missing data. The use of self-administered methods has been found to result in increased reporting of sensitive items (including MTMS) because of the increased privacy they provide to survey respondents. 1,10 However, the paper-and-pencil approach used by the GSS makes it possible for respondents simply to skip questions, resulting in missing data. Computerized self-administered interviewing methods have a number of advantages, including requiring a response for each question, and have been found to yield higher levels of reporting MTMS than other modes of data collection. 10
The GSS has a number of limitations; it contains few questions on sexual behavior and has a relatively small sample size, for example. Nevertheless, it provides data not available elsewhere because, as limited as the GSS is, there currently is no other national probability-based survey that measures sexual behavior at regular, frequent intervals. This has been recognized by researchers, who have used the GSS as a point of comparison for work with data from other, more detailed surveys. 4,6,7 Because relatively few respondents in general population surveys report MTMS, the ability to study this subpopulation in detail tends to be limited, even in surveys with very large samples. Surveys that use probability sampling methods that specifically target this group are needed for in-depth study of this population. 11
Data collected at 9 points over a 13-year period in the GSS indicate an increase after 1996 in the percentage of adult men who report having MTMS in the past year, compared with earlier estimates from this survey and from other similar surveys. The results suggest that estimates of this type are affected by some unknown level of underreporting and that previous estimates of the size of this population subgroup based on survey research may be too low. Methods are needed that increase survey respondents’ sense of privacy and improve completeness of data collection in order to improve quality of measurement of sensitive sexual behaviors.
1. Catania JA, Turner H, Pierce RC, et al. Response bias in surveys of AIDS-related sexual behavior. In: Ostrow DG, Kessler RC, eds. Methodological Issues in AIDS Behavioral Research. New York: Plenum Press, 1993: 133–162.
2. Davis JA, Smith TW. General Social Surveys, 1972–2000: Cumulative Codebook. Chicago: National Opinion Research Center, 2001.
3. Hosmer D, Taber S, Lemeshow S. The importance of goodness of fit of logistic regression models: a case study. Am J Public Health 1991; 81: 1630–1635.
4. Fay RE, Turner CF, Klassen AD, et al. Prevalence and patterns of same-gender sexual contact among men. Science 1989; 243: 338–348.
5. Billy J, Tanfer K, Grady W, Klepinger D. The sexual behavior of men in the United States. Fam Plann Perspect 1993; 25: 52–60.
6. Turner C, Danella R, Rogers S. Sexual behavior in the United States, 1930–1990. Sex Transm Dis 1995; 22: 173–190.
7. Michaels S. The prevalence of homosexuality in the United States. In: Cabaj R, Stein T, eds. Textbook of Homosexuality and Mental Health. Washington, DC: American Psychiatric Press, 1996: 43–63.
8. Kelly J, Otto-Salaj L, Sikkema K, Pinkerton S, Bloom F. Implications of HIV treatment advances for behavioral research on AIDS: protease inhibitors and new challenges in HIV secondary prevention. Health Psychol 1998; 17: 310–319.
9. Ostrow D, McKirnan D, Vanable P, Hope B. The impact of new combination HIV therapies on knowledge, attitudes and behaviors among Chicago gay men. Presented at the 12th International Conference on AIDS. 1998;898.
10. Turner C, Ku L, Rogers S, Lindberg L, Pleck J, Sonenstein F. Adolescent sexual behavior, drug use, and violence: increased reporting with computer survey technology. Science 1998; 280: 867–873.
© Copyright 2002 American Sexually Transmitted Diseases Association
11. Catania J, Osmond D, Stall R, et al. The continuing HIV epidemic among men who have sex with men. Am J Public Health 2001; 91: 907–914.