Deep neural networks are more accurate than humans at detecting sexual orientation from facial images
Michal Kosinski & Yilun Wang
Journal of Personality and Social Psychology, forthcoming
We show that faces contain much more information about sexual orientation than can be perceived and interpreted by the human brain. We used deep neural networks to extract features from 35,326 facial images. These features were entered into a logistic regression aimed at classifying sexual orientation. Given a single facial image, a classifier could correctly distinguish between gay and heterosexual men in 81% of cases, and in 74% of cases for women. Human judges achieved much lower accuracy: 61% for men and 54% for women. The accuracy of the algorithm increased to 91% and 83%, respectively, given five facial images per person. Facial features employed by the classifier included both fixed (e.g., nose shape) and transient facial features (e.g., grooming style). Consistent with the prenatal hormone theory of sexual orientation, gay men and women tended to have gender-atypical facial morphology, expression, and grooming styles. Prediction models aimed at gender alone allowed for detecting gay males with 57% accuracy and gay females with 58% accuracy. Those findings advance our understanding of the origins of sexual orientation and the limits of human perception. Additionally, given that companies and governments are increasingly using computer vision algorithms to detect people's intimate traits, our findings expose a threat to the privacy and safety of gay men and women.
Physiological Arousal and Self-Reported Valence for Erotica Images Correlate with Sexual Policy Preferences
Amanda Friesen, Kevin Smith & John Hibbing
International Journal of Public Opinion Research, Autumn 2017, Pages 449-470
Individuals do not always accurately report the forces driving their policy preferences. Such inaccuracy may result from the fact that true justifications are socially undesirable or less persuasive than competing justifications or are unavailable in conscious awareness. Because of the delicate nature of these issues, people may be particularly likely to misstate the reasons for preferences on gay marriage, abortion, abstinence-only education, and premarital sex. Advocates on both sides typically justify their preferences in terms of preserving social order, maintaining moral values, or protecting civil liberties, not in terms of their own sexual preferences. Though these are the stated reasons, in empirical tests we find that psychophysiological response to sexual images also may be a significant driver of policy attitudes.
Social Attitudes Regarding Same-Sex Marriage and LGBT Health Disparities: Results from a National Probability Sample
Mark Hatzenbuehler, Andrew Flores & Gary Gates
Journal of Social Issues, September 2017, Pages 508-528
This study examined the health consequences for lesbian, gay, bisexual, and transgender (LGBT) populations of exposure to communities with relatively high versus low levels of support for same-sex marriage. We used data from the Gallup Daily tracking survey, the largest probability-based sample of LGBT-identified adults in the United States (N = 11,949 LGBT respondents; N = 352,343 non-LGBT respondents), which was linked to attitudinal responses on same-sex marriage obtained from the 2012 Cooperative Congressional Election Survey (N = 54,535). Controlling for potential confounders, higher levels of local approval of same-sex marriage lowered the probability that LGBT (and non-LGBT) individuals reported smoking and fair/poor self-rated health; further, LGBT disparities in smoking were lower in communities where residents were most likely to support same-sex marriage. Findings suggest that local attitudes may be related to the health of LGBT individuals and contribute to sexual orientation health disparities, providing further evidence for the role of structural stigma in shaping LGBT health.
Transgender Youth Substance Use Disparities: Results From a Population-Based Sample
Jack Day et al.
Journal of Adolescent Health, forthcoming
Methods: A statewide cross-sectional sample of California middle and high schools collected between 2013 and 2015. This representative sample of students in California included 335 transgender and 31,737 nontransgender youth. Using multivariate linear and logistic regression, we assessed differences between transgender and nontransgender youth in substance use behaviors related to alcohol, cigarette, marijuana, other illicit drugs, polysubstance use, and heavy episodic drinking. Substance use was assessed with lifetime use, age of onset, and past 30-day use for alcohol, cigarettes, and marijuana. Past 30-day use was also assessed for other illicit drugs and polysubstance use. Models were adjusted for demographics and risk factors including victimization, depressive symptoms, and perceived risk of substance use.
Results: The prevalence of substance use was 2.5-4 times higher for transgender youth compared with their nontransgender peers (depending on the substance). Transgender youth were also at greater risk for early age of onset and recent substance use than nontransgender youth. In addition, psychosocial risk factors related to victimization, depressive symptoms, and perceived risk of substance use partially mediated the relationship between gender identity and substance use.
Relationships between the second to fourth digit ratio (2D:4D) and game-related statistics in semi-professional female basketball players
Makailah Dyer et al.
American Journal of Human Biology, forthcoming
Methods: Using a cross-sectional design, 64 female basketball players who competed in the South Australian Premier League were measured in-season for height, mass, and 2D:4D, with game-related statistics collected end-season. Partial correlations (adjusted for age and body mass index) were used to quantify relationships between right and left 2D:4Ds and game-related statistics. Unpaired t-tests were used to quantify differences in mean 2D:4Ds between starting and reserve players.
Results: 2D:4D was a substantial negative correlate of blocks, rebounds, and field-goal percentage; meaning, females with lower 2D:4Ds were generally better defensively as they recorded more blocks and rebounds, and were more efficient scorers, irrespective of their age and body size. Mean 2D:4D differed by position in the starting lineup, as females with lower 2D:4Ds were more likely to be in the starting lineup.
Disparities in Body Mass Index Trajectories From Adolescence to Early Adulthood for Sexual Minority Women
Sarah Wood et al.
Journal of Adolescent Health, forthcoming
Methods: We analyzed data from females in waves I-IV of the National Longitudinal Study of Adolescent to Adult Health. Sexual identity was categorized as heterosexual, mostly heterosexual, bisexual, or lesbian (homosexual/mostly homosexual). We conducted group-based trajectory modeling of BMI with a censored normal distribution and a cubic relationship with age to identify three BMI trajectory groups. Multinomial logit regressions predicted the risk of trajectory membership associated with sexual identity, adjusting for background characteristics.
Results: At wave I, the mean (n = 7,801) age was 15.9 years (95% confidence interval: 15.6-16.1). Subjects were 16.3% African-American; and 80.0% heterosexual, 15.9% mostly heterosexual, 2.5% bisexual, and 1.7% lesbian. Group-based trajectory modeling identified three BMI trajectory groups characterized as (1) minimal obesity (62.2%), (2) developing obesity (29.9%), and (3) progressive obesity (8.0%). In multinomial logit regressions adjusted for age, race, parental obesity and education, sexual abuse, household income, screen time, depressive symptoms, and rural residence, lesbian women had a nearly two-fold higher relative risk of being in the developing obesity trajectory group (relative risk ratio = 1.91, 95% confidence interval: 1.10-3.32) relative to the minimal obesity group, compared with heterosexual women.
Conclusions: Lesbian women were at increased risk of membership in the developing obesity trajectory group compared with heterosexual women. Adjusting for obesity risk factors had minimal impact on the point estimates for this association.