How do different survey methods (behavior vs identity vs attraction) change estimates of men who have sex with men?
Executive summary
Estimates of men who have sex with men (MSM) shift substantially depending on whether surveys measure recent sexual behavior, self-identified sexual orientation, or reported sexual attraction: behavior-based measurements tend to produce different—and often lower or higher depending on sample frame—estimates than identity- or attraction-based questions, and surveillance or venue-based methods usually yield higher risk concentrations than population probability surveys [1] [2]. The choice of method therefore matters for both size estimation and public‑health metrics, because measurement domain, sampling frame and stigma-driven underreporting systematically bias results in predictable directions [1] [3].
1. How “behavior” questions shape the count: transient, contextual and underreported acts
Asking men about recent same‑sex sexual behaviour (for example, sex in the past year) captures an action‑based subset of the male population that may be smaller or larger than identity‑based groups depending on time window and social context; population‑based surveys using behavioral questions produced lower MSM percentages in many studies summarized by the systematic review, and surveillance-derived methods—often tied to diagnosed infections or clinic attendance—produced the highest numerical estimates [1] [2]. Behavioral approaches also miss intermittent or situational MSM (for example, men who have sex with men but only rarely or in certain life periods), and they are sensitive to recall, question framing and the social desirability pressures that lead to underreporting in rural or stigmatizing settings [4] [3].
2. Identity-based measures: stable labels, but blind to practice
Surveys that ask respondents to self‑identify as gay, bisexual or heterosexual map social identity rather than behavior; these identity measures can undercount men who have sex with men but do not adopt gay/bisexual labels and can overcount men whose identity does not reflect recent sexual practice, producing divergent prevalence estimates from behavior‑based questions [1]. Meta‑analytic work pooling multiple population surveys in the U.S. analyzed attraction and orientation domains separately because identity, attraction and behavior are overlapping but non‑congruent dimensions—each yields a distinct numerator for surveillance denominators and thus different disease rates [5] [1].
3. Attraction questions: an intermediate lens that widens the net
Questions about same‑sex attraction (e.g., “mostly/only attracted to men”) often identify a larger group than identity questions but not necessarily the same people who report recent behavior; the JMIR meta‑analysis explicitly coded attraction responses as part of their MSM definition to broaden capture across domains, and found pooled estimates that differed by domain [5]. Attraction questions can surface latent preferences that stigma suppresses in identity reporting, yet they still miss men who act contrary to stated attraction or who are sexually inactive, so estimates remain domain‑dependent [1] [3].
4. Sampling frame and mode: why web, venue, surveillance and census methods diverge
Convenience samples drawn from gay venues or community sites concentrate higher‑risk, more visible MSM and therefore report greater rates of partners, STIs and HIV than probability household surveys, while web‑recruited samples differ again—sometimes reporting fewer multiple partners than venue samples—so mode and recruitment pathway shape both prevalence and size estimates [6] [7] [8]. Census or American Community Survey–based imputation methods use cohabitation markers and urbanicity adjustments to estimate MSM in small areas but must impute hidden populations in rural zones where same‑sex behavior is underreported [4] [3].
5. Implications for policy, surveillance and hidden agendas in reporting
Public‑health programs using behavioral denominators will calculate different HIV or STI rates than programs using identity denominators, which affects resource allocation and perceived burden; researchers emphasizing community convenience samples may highlight high risk and urgent needs, while population probability surveys can be used to argue for smaller overall MSM populations and hence different priorities—both framings reflect methodological choices and potential advocacy agendas [1] [7]. Studies attempting to triangulate surveillance, survey and census data generally find surveillance yields higher estimates and surveys lower ones, underscoring that no single method is definitive and that transparent reporting of domain and sampling frame is essential for policymakers [2] [1].
6. How to interpret conflicting estimates responsibly
Reconciling disparate estimates requires attention to what was measured (behavior, identity, attraction), who was sampled (probability household, venue, web, clinic), and the time window asked; acknowledging stigma and underreporting, using multiple domains in parallel, and applying adjustment or imputation strategies (as in ACS‑based small‑area estimates) are necessary steps to produce defensible MSM population figures for surveillance and intervention planning [4] [5]. Absent a universal standard, the right choice depends on whether the question is epidemiologic risk (favor behavior), social visibility (favor identity) or unmet prevention needs (use multi‑domain triangulation) [1] [9].