A growing number of researchers have begun to use the Internet to collect behavioral data from men who have sex with men (MSM). 1–5 One of the advantages of using the Internet for behavioral research is that it may provide access to men who would not otherwise be surveyed using traditional methodologic approaches. For example, several studies have reported that men surveyed online are less likely to identify as gay than men surveyed in the community (i.e., offline). 1,4,5 These studies did not, however, examine differences between online and offline samples by HIV status. This is important because online and offline samples may vary in the proportion of men who are HIV-positive, HIV-negative, or who have never been tested for HIV. 5 The behavioral characteristics of MSM are known to vary by HIV status. For example, HIV-positive men are more likely to use recreational drugs and to report high-risk sexual behavior than HIV-negative or never-tested men. 6–8 Consequently, differences in the behavioral profile of men surveyed online and offline could be confounded by HIV status. This merits further investigation.
Research conducted in the United Kingdom, Netherlands, Australia, and the United States among gay men has shown an association, as yet unexplained, between seeking sex on the Internet and high-risk sexual behavior. 8–13 Men who seek sex on the Internet are more likely to report high-risk sexual behavior or a history of sexually transmitted disease than men who do not seek sex in this way. Are men surveyed online, via the World Wide Web, more likely to use the Internet to look for sex than men surveyed offline? If so, differences in the behavioral characteristics of online and offline samples could reflect corresponding differences in their propensity to seek sexual partners online. Clearly this needs to be taken into account when comparing the behavioral profile of online and offline samples.
The objective of this investigation was to compare the social, demographic, and behavioral characteristics of 2 samples of London gay/bisexual men, one surveyed online, the other offline, paying particular attention to the HIV status of men in each sample and whether men used the Internet to look for sexual partners.
In February and March 2002, gay/bisexual men using any one of 7 gyms in central London were asked to complete a confidential, anonymous self-administered questionnaire as part of an annual behavioral surveillance program. All these gyms have a substantial gay male membership: one is exclusively gay while the other 6 estimate that gay men comprise 40–90% of their male membership. All men using the gym during a 1-week period were asked to complete a self-administered pen-and-paper questionnaire that took between 10–15 minutes. Respondents returned completed questionnaires to collection boxes in the gyms or by post to the research team. The methods have been described in detail elsewhere. 7,8,14–19
In May 2002, gay/bisexual men using UK chat rooms or personal profiles on gaydar and gay.com were invited to complete a confidential, anonymous self-administered questionnaire online. Gaydar and gay.com are the 2 most popular Web sites used by gay/bisexual men in the United Kingdom (personal communication, H. Badenhurst, M. Watson). For a 1-month period, a series of pop-ups and banners in the chat rooms and profiles pages advertised the online survey. Clicking on a pop-up or banner took men to the home page of the questionnaire, which took between 15–30 minutes to complete.
The questions for the online and offline surveys were worded identically to ensure direct comparability between the two. Both offline and online, men were asked to provide information on their age, ethnicity, employment, and education (higher education was defined as having had ≥3 years education after the age of 16 years). They were also asked about their sexual orientation, HIV test history (date and result of last test), HIV treatment optimism, 7,20,21 recreational drug use, use of the Internet to look for sex, unprotected anal intercourse (UAI) in the previous 3 months, and the HIV status of their UAI partner(s). UAI was classified as concordant (with a person of the same HIV status), discordant (with a person of opposite HIV status), or status unknown (with a person of unknown HIV status). High-risk sexual behavior was defined as UAI with a discordant partner or with a partner of unknown HIV status. This clearly presents a risk for HIV transmission. Men surveyed online were also asked whether they had completed a questionnaire 3 months earlier in one of the London gyms.
Data from the two surveys were merged into one database and analyzed using SPSS for Windows 10. 22 Men who completed the questionnaire online were compared with men who completed the questionnaire offline. In univariate analysis, χ2 and two sample t tests were used for examining differences in proportions and means. The univariate association between high-risk sexual behavior and mode of questionnaire completion was initially explored in a logistic model. Significant univariate associations between high-risk sexual behavior and being surveyed online or offline were further explored in a multivariate logistic model. In the univariate and multivariate models, high-risk sexual behavior was the dependent variable while the independent variable was being surveyed online vs. offline. To identify potential confounding factors, variables known to be associated with high-risk sexual behavior were compared between the online and offline samples. These variables were age, recreational drug use, being in a relationship, HIV treatment optimism, and seeking sex on the Internet. 6–8,14 Confounding factors were variables that were significantly associated with both the dependent variable (high-risk sexual behavior) as well as the independent variable (being surveyed online or offline). The odds ratios were adjusted by simultaneously entering the independent variable and appropriate confounding factors in the multivariate model.
Number of Respondents
During the survey period, 921 questionnaires were returned by gay/bisexual men in London gyms (estimated response rate, based on earlier surveys, 7,8,14 50–60%). Of these, 879 who provided complete information on their HIV test history, use of the Internet for seeking sex, and sexual risk behavior were included in the analysis: 131 (14.9%) were HIV-positive, 574 (65.3%) HIV-negative, and 174 (19.8%) had never been tested for HIV.
During the survey period, 4974 completed questionnaires were returned electronically; 1250 (25.1%) by men living in London, 3279 (65.9%) by men living elsewhere in the United Kingdom, and 445 (9.0%) by men living outside the United Kingdom. Only the men living in London are included in this analysis to enable a direct comparison with the offline sample of men surveyed in the London gyms. Of the 1250 London men surveyed online, 1218 who provided complete information on their HIV test history, use of the Internet for seeking sex and sexual behavior were included in the analysis; 142 (11.7%) were HIV-positive, 680 (55.8%) HIV-negative, and 396 (32.5%) had never been tested for HIV.
Only 28 London men surveyed online said they had also completed a questionnaire in a gym 3 months earlier; 8 were HIV-positive, 13 HIV negative, 7 never tested. They were initially included in the analysis. The univariate and multivariate analyses were then repeated excluding the 28 men.
Sociodemographic and Behavioral Characteristics of Online and Offline Samples
Men surveyed online were less likely to have been tested for HIV than men surveyed offline (67.5 vs. 80.2%, P < 0.001). Regardless of HIV status, those surveyed online were significantly less likely to have had a higher education and less likely to be in a relationship with a man (for never-tested men, P = 0.07). There were no differences between the 2 samples in ethnicity (Table 1).
For both HIV-negative and never-tested men, the online samples were significantly younger than the offline samples, less likely to define their sexual orientation as gay (HIV-negative men, P = 0.05), less likely to be optimistic in the light of new drug treatments for HIV, and less likely to use recreational drugs (all P < 0.05). Online never-tested men were also less likely to be employed than never-tested men surveyed offline (Table 1). However, for HIV-positive men there were no significant differences between the online and offline samples with respect to age, employment, sexual orientation, recreational drug use, or HIV optimism.
Sexual Behavior of Online and Offline Samples: Univariate Analysis
Men surveyed online were significantly more likely to report high-risk sexual behavior than their offline counterparts; HIV-positive men, 47.2 vs. 42.0%; HIV-negative men, 26.9 vs. 18.6%; never-tested men, 35.6 vs. 19.0% (all P < 0.05) (Table 2). In both samples, HIV-positive men were significantly more likely to report high-risk sexual behavior than other men (P < 0.05).
Online HIV-negative men were less likely to have only had sex with other men (Table 1). This differential was not seen for HIV-positive or never-tested men. Never-tested men surveyed online were least likely to say they only had sex with other men (83.3%).
Men surveyed online were more likely to have used the Internet to seek a sexual partner than men surveyed offline; this was seen for HIV-positive, HIV-negative, and never-tested men alike (P < 0.001, Table 1). In both online and offline samples, HIV-positive men were significantly more likely to have used the Internet to seek a sexual partner than other men (P < 0.01).
HIV-positive and negative men who used the Internet to look for sex were more likely to report high-risk sexual behavior than other men; this was seen for men surveyed online (HIV-positive 47.7 vs. 40.0%; HIV-negative 28.5 vs. 18.7%) and offline (HIV-positive 51.3 vs. 21.8%; HIV-negative 23.7 vs. 14.3%). The differential was significant for men surveyed offline (P < 0.01). For never-tested men there was no association between seeking sex on the Internet and high-risk sexual behavior either for those surveyed online (35.7 vs. 34.9%, P = 0.9) or offline (20.8 vs. 18.3%, P = 0.7).
Sexual Behavior of Online and Offline Samples: Multivariate Analysis
In multivariate analysis, after controlling for recreational drug use and seeking sex on the Internet (confounding factors for HIV-positive men), the association between being surveyed online and high-risk sexual behavior became nonsignificant (P = 0.4) (Table 2).
In multivariate analysis, after controlling for age, HIV optimism, recreational drug use, and seeking sex on the Internet, (confounding factors for HIV-negative men), the association between being surveyed online and high-risk sexual behavior remained statistically significant (adjusted odds ratio [OR] 1.73; 95% CI 1.23–2.42; P < 0.01) (Table 2).
In multivariate analysis, after controlling for HIV optimism and being in a relationship (confounding factors for never-tested men), the association between being surveyed online and high-risk sexual behavior remained significant (adjusted OR 2.45; 95% CI, 1.40,4.29; P < 0.01) (Table 2).
Repeating the univariate and multivariate analyses after excluding the 28 men who completed both the online and offline questionnaires did not alter these findings.
Our study provides further evidence that it is feasible to conduct behavioral research online among MSM. 1,4,5 Over a 4.5-week period, nearly 5000 men completed the online survey, of whom 1250 lived in London.
In London, the online survey attracted a greater proportion of never-tested men than the offline survey. Never-tested men surveyed online were younger and less likely to identify as gay than their offline counterparts but more likely to report high-risk sexual behavior with other men. The online survey appeared, therefore, to reach a group of never-tested men living in London who are at risk for HIV infection yet who may not be reached by traditional, offline, health promotion interventions and survey methods. This demonstrates the value of supplementing behavioral surveillance in the community with data collection via the Internet. It also highlights the Internet’s potential for HIV prevention because of its ability to access hard-to-reach groups at elevated risk of infection.
While clear differences emerged between the characteristics of the online and offline samples, these were most pronounced for HIV-negative and never-tested men and least pronounced for HIV-positive men. The characteristics of HIV-positive men surveyed online were broadly similar to those of the HIV-positive men surveyed offline. On the other hand, HIV-negative and never-tested men surveyed online were less likely to identify as gay and less likely to have only had sex with men. The online survey clearly attracted a greater proportion of behaviorally bisexual men and men who did not identify as gay. While this is in accordance with other research, 1,4,5 ours is the first to highlight the importance of examining these differences by HIV status, a finding that merits further exploration.
Men surveyed online were more likely to report high-risk sexual behavior than the offline sample. They were also more likely to say they had used the Internet to look for sexual partners, which in turn was associated with high-risk behavior for HIV-negative and positive men. For HIV-positive men, the elevated levels of high-risk behavior seen in the online sample were explained entirely by their being more likely to use the Internet to look for sexual partners than the offline sample. On the other hand, the elevated levels of high-risk behavior among HIV-negative and never-tested men surveyed online were statistically independent of their using the Internet to look for sex. After controlling for confounding factors, HIV-negative and never-tested men surveyed online were still more likely to report high-risk sexual behavior than their offline counterparts.
These differentials in high-risk sexual behavior between online and offline samples further highlight the importance of examining by HIV status the behavioral patterns of men surveyed online and offline. They also raise important questions, as yet unanswered, about the association between seeking sex on the Internet and sexual risk behavior. 8,9,11,12,23 Interestingly, the association, previously reported in a number of community and clinic samples, was seen here for the first time among men surveyed online. This association and its underlying processes are not yet fully understood and are currently being investigated as part of a Medical Research Council–funded research project in London.
The elevated levels of high-risk sexual behavior seen in the online sample could reflect an increased willingness to report such behaviors in a Web-based survey because of its perceived anonymity when compared with a pen-and-paper questionnaire. However, in a feasibility study for the second National Survey of Sexual Attitudes and Lifestyles in the United Kingdom, no consistent effects of computer-assisted self-completion interviews were found when compared with pen-and-paper self-completion interviews. 24 Furthermore, the elevated high-risk sexual behavior among HIV-positive men surveyed online, seen in univariate analysis, was explained entirely by their Internet sex-seeking behavior. Since there was no evidence of a mode effect for HIV-positive men, it seems unlikely that this alone could explain the online-offline differentials in risk behavior for the other men in the study. Another explanation may be that online surveys attract some high-risk respondents who only use the Internet to meet other men. If they do not go to bars, clubs, gyms, etc., then, by definition, these men—and their high-risk behavior—cannot feature in offline surveys. A further consideration is that pop-ups and banners advertising the online survey were only placed in gay Internet chat rooms and profiles. Internet users who visit chat rooms and profiles may be different from Internet users who do not. Clearly the potential biases associated with online and offline behavioral surveillance merit further examination.
There are some limitations to this investigation. Because of the high costs of assembling a random or probability sample, behavioral research among gay/bisexual men is often based on convenience samples. 25 In this study, men in the offline sample were all surveyed in London gyms, which clearly do not attract a random sample of gay men. However, the behavioral and demographic characteristics of men surveyed in London gyms are broadly similar to those of men surveyed in bars, clubs, and GUM clinics as part of a parallel behavioral surveillance program. 26,27 The men surveyed in the gyms appear, therefore, to be representative of men using gay venues in London.
Men surveyed online also comprised a convenience sample drawn from a large population of gaydar and gay.com Internet chat room and profile users. Concerns have been expressed about the robustness of convenience samples recruited through the Internet. 28–30 It is impossible to gauge what proportion of chat room and profile users saw the banners and pop-ups promoting the online survey. Nor do we know what percentage of those seeing the pop-ups and banners went on to complete the questionnaire. Based on estimates provided by gaydar and gay.com on the number of people using their Internet chat rooms and profiles during the survey period, it is likely that <1% of all users completed the questionnaire. Inevitably, this raises questions about the characteristics of responders and nonresponders to Web-based surveys. The strength of the online survey is that it reached a greater proportion of behaviorally bisexual men and men who did not identify as gay compared with the offline survey. Some of the online-offline differences reported here for the overall study group have also been reported in another UK study of men surveyed online through the Internet and offline at gay Pride events and other community venues. 5 These similarities provide some face validity for our analysis. The other study did not, however, examine these differences by HIV status.
There were significant differences in the social, demographic, and behavioral characteristics of MSM surveyed online and offline in London. These differences were least pronounced for HIV-positive men and most pronounced for HIV-negative men and those who had never been tested for HIV, highlighting the importance of comparing online and offline samples according to their HIV status.
Clearly the Internet offers important opportunities for conducting behavioral research since it reaches some gay/bisexual men who may not be easily accessed in community or clinic settings yet who are at high risk for HIV and sexually transmitted diseases. There is a strong case for supplementing behavioral surveillance in the community with data collection via the World Wide Web.
This study also throws into sharp focus the Internet’s potential for HIV prevention among gay/bisexual men who may not otherwise be reached by community-based sexual health promotion programs. 31
The authors thank the managers and members of the gyms for their support and participation in the project (offline sample), gaydar and gay.com for promoting the online survey as well as providing technical support, and the many men who took the time to complete the questionnaire online.
1. Ross MW, Tikkanen R, Mansson S-A. Differences between Internet samples and conventional samples of men who have sex with men: implications for research and HIV interventions. Soc Sci Med
2. Bull SS, McFarlane M, Rietmeijer CA. HIV and sexually transmitted infection risk behaviors among men seeking sex online. Am J Public Health
3. Bull SS, McFarlane M, King D. Barriers to STD/HIV prevention on the Internet. Health Educ Res
4. Rhodes S, DiClemente RJ, Cecil H, et al. Risk among men who have sex with men in the United States: a comparison of an Internet sample and a conventional outreach sample. AIDS Educ Prev
5. Reid D, Weatherburn P, Hickson F, et al. Know the score: findings from the National Gay Men’s Sex Survey 2001. London: Sigma Research; 2002. Available at http://www.sigmaresearch.org.uk/reports.html
. Accessed February 5, 2004.
6. Elford J, Bolding G, Maguire M, et al. HIV positive and negative homosexual men have adopted different strategies for reducing the risk of HIV transmission. Sex Transm Infect
7. Elford J, Bolding G, Sherr L. High risk sexual behaviour increases among London gay men between 1998–2001: what is the role of HIV optimism?AIDS
8. Elford J, Bolding G, Sherr L. Seeking sex on the Internet and sexual risk behaviour among gay men using London gyms. AIDS
9. Hospers HJ, Harterink P, van den Hoek K, et al. Chatters on the Internet: a special target group for HIV prevention. AIDS Care
10. Rawstone P, Van de Ven P, Prestage G, et al. Gay Community Periodic Survey Adelaide 2001
. Mongraph 4. Sydney: National Centre in HIV Social Research; 2002.
11. McFarlane M, Bull SS, Rietmeijer CA. The Internet as a newly emerging risk environment for sexually transmitted diseases. JAMA
12. Kim A, Kent C, McFarland W, et al. Cruising on the Internet highway. J Acquir Immune Defic Syndr
13. Benotsch E, Kalichman S, Cage M. Men who have met sex partners via the Internet: prevalence, predictors and implications for HIV prevention. Arch Sex Behav
14. Elford J, Bolding G, Maguire M, et al. Sexual risk behaviour among gay men in a relationship. AIDS
15. Bolding G, Sherr L, Maguire M, et al. HIV risk behaviours among gay men who use anabolic steroids. Addiction
16. Sherr L, Bolding G, Maguire M, et al. Viagra use and sexual risk behaviour among gay men in London. AIDS
17. Elford J, Bolding G, Maguire M, et al. Peer-led HIV prevention among gay men in London. In: Watson J, Platt S, eds. The 4 Gym Project:
I ntervention and Evaluation. Researching Health Promotion
. London: Routledge; 2000.
18. Elford J, Bolding G, Sherr L. Peer education has no significant impact on HIV risk behaviours among gay men in London. AIDS
19. Bolding G, Sherr L, Elford J. Use of anabolic steroids and associated health risks among gay men attending London gyms. Addiction
20. Elford J, Bolding G, Sherr L. HIV optimism: fact or fiction?Focus
21. International Collaboration on HIV Optimism. HIV treatments optimism among gay men: an international perspective. J Acquir Immune Defic Syndr
22. SPSS for Windows version 10.1
. Chicago: SPSS; 1999.
23. Elford J. Surfing for sex. Focus
24. Johnson AM, Copas A, Erens B, et al. Effect of computer-assisted self-interviews on reporting of sexual HIV risk behaviours in a general population sample: a methodological experiment. AIDS
25. McGarrigle D, Fenton K, Gill O, et al. Behavioural surveillance: the value of national coordination. Sex Transm Infect
26. Dodds JP, Nardone A, Mercey DE, et al. Increase in high risk sexual behaviour among homosexual men, London, 1996–8: cross sectional, questionnaire study. BMJ
27. Elford J. Behavioural surveillance in London gyms: trends, comparisons and methodological strengths. London, PHLS CDSC Behavioural Surveillance Collaborators’ Meeting 2002. Available at: http://www.hpa.org.uk/infections/topics_az/hiv_and_sti/behavioural/files/Behav_surv_meeting_2002.pdf
. Accessed February 5, 2004.
28. Mustanski B. Getting wired: exploiting the Internet for the collection of valid sexuality data. J Sex Res
29. Couper MP. The Promises and Perils of Web Surveys
. London: Association for Survey Computing; 2001.
30. Rhodes S, Bowie D, Hergenrather K. Collecting behavioural data using the world wide web: considerations for researchers. J Epidemiol Community Health
31. Ross MW. The Internet as a medium for HIV prevention and counselling. Focus