Reproducibility of Survey Measurements of Sexual Orientation in the USA

Abstract

AbstractSurvey measurements of sexual orientation have become increasingly common in national population surveys although validation of these measurements is rare and inherently problematic. We instead assess the reproducibility of parallel measurements from two independent samples of the USA population made in the 2008-2018 General Social Surveys and the adult probability subsets of the 2013-2018 National Health Interview Survey (Ns = 12,098 and 190,113).Restricting analysis to the categories gay/lesbian, bisexual, and straight, we obtain similar estimates of the proportion of the U.S. population who consider themselves gay/lesbian (NHIS: 1.59% vs. GSS: 1.93%, p = 0.059) but not bisexual (NHIS: 1.03% vs. GSS: 2.90%, p < 0.001). Fitting multinomial logistic regression models controlling for year, gender, birth cohort, education, and race, we find that compared to the NHIS, the GSS had 1.248 (p=0.022) times higher relative odds of eliciting a response of Gay-Lesbian (vs. Straight) and 2.980 (p<0.001) times higher relative odds of eliciting a response of Bisexual (vs. Straight). Expanding the model by adding 3-way interaction terms for orientation-by-predictor-by-survey, we find that we cannot reject the null hypothesis that trends over time and across subpopulations in reporting of sexual orientation were statistically equivalent for the two survey programs.