Length of Gay Fingers

Preliminaries

Perinatal androgens are crucial in molding sex-related distinctions in the brain and behavior across mammalian species, by modulating gene expression patterns within the developing brain (e.g., Xu et al., 2012). Examinations of individuals with and without varied endocrine disorders suggest androgens contribute similarly to the evolution of human sex disparities in brain function and behavior (Shirazi et al., 2022; Swift-Gallant et al., 2022, 2023). However, disentangling the straightforward consequences of androgens on brain development from other biological or environmental elements, such as varying treatment from guardians or medical practitioners, remains a hurdle. This difficulty, along with the ethical improbability of experimental investigations on humans, has ignited considerable interest in scrutinizing retrospective indicators of early androgen influence (Swift-Gallant et al., 2020, 2023).

One possible biomarker is 2D:4D, the proportion between the lengths of the second (2D) and fourth (4D) manual digits. 2D:4D is consistently smaller in men (a shorter index finger relative to the ring finger; for meta-analysis, see H&x000F6;nekopp and Watson, 2010), and convergent evidence ties prenatal androgens to its development (reviewed in Puts et al., 2008; Swift-Gallant et al., 2020, 2023). For instance, digit ratios are reduced (more male-like) among women with congenital adrenal hyperplasia, wherein prenatal androgens are elevated (Richards et al., 2020a), whereas digit ratios are higher (more female-like) among chromosomal males with insensitivity to androgens (e.g., androgen insensitivity syndrome; Berenbaum et al., 2009; Van Hemmen et al., 2017).

Certain evidence supports a relationship between 2D:4D and sexual orientation, one of the most markedly sexually differentiated human psychological traits (Balthazart, 2011; Hines, 2011; Kostic and Scofield, 2022). However, reported links between 2D:4D and sexual orientation have been inconsistent, with studies citing lower ratios, elevated ratios, or no considerable correlation (reviewed in Swift-Gallant et al., 2020). These varied outcomes stimulated a prior meta-analysis (Grimbos et al., 2010), which detected reduced (i.e., more male-typical) 2D:4D among women with same-sex proclivities than their heterosexual counterparts, but no association of 2D:4D with male sexual orientation.

Notwithstanding its contribution, the Grimbos et al. (2010) meta-analysis didn't incorporate unpublished datasets, leaving it susceptible to the "file drawer problem," wherein adverse results may be less likely to be published, a worry frequently mentioned in criticisms of digit ratio research (e.g., McCormick and Carr&x000E9;, 2020). Grimbos et al. also treated sexual orientation dichotomously, uniting bisexual individuals with those strictly oriented toward same-sex partners, and therefore couldn't test whether bisexual individuals are intermediate between heterosexual and homosexual orientations or more comparable to either. Furthermore, since the last meta-analysis by Grimbos et al. (2010), there's been a substantial increase in 2D:4D research, with the quantity of publications more than doubling between 2010 and 2020 (Figure 1). We've identified forty-four datasets (ten unpublished) for male sexual orientation and thirty-four datasets (five unpublished) for female sexual orientation, a surge of twenty-six male and twenty-one female datasets from those included in Grimbos et al. (2010). We consequently performed a fresh meta-analysis on this larger sample that encompasses unpublished data to mitigate the ramifications of publication bias, consider intermediary sexual orientations, augment the accuracy of effect size estimations, and better assess the robustness of any associations.

Methods

Adhering to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Page et al., 2021), we carried out a systematic literature search and extracted relevant data from sixty published studies utilizing two prominent electronic databases, PubMed and Google Scholar. In addition to consulting published works, we reached out to researchers who've published on 2D:4D, irrespective of whether they evaluated this marker concerning sexual orientation. Using keywords such as &x201C;2D:4D&x201D; and &x201C;digit ratio,&x201D; we pinpointed two hundred and ninety-six unique corresponding authors with publications employing this marker. From these authors, we requested data on sexual orientation for their published datasets where this information was initially left out. Concurrently, we inquired about any unpublished datasets including the required information.

Articles and unpublished data were eligible for inclusion if they reported data on 2D:4D by sex and sexual orientation. Studies were excluded if they did not report an effect size or mean 2D:4D and standard deviation (SD) or standard error (SE), broken down by sex and sexual orientation. Studies were also excluded if they concentrated exclusively on heterosexual or homosexual individuals, or employed or reanalyzed previously published data that had already been selected for inclusion. Out of sixty published papers and thirteen unpublished datasets assessed for eligibility, forty published studies and eleven unpublished datasets (comprising ten male and five female unpublished data sets) were determined to be suitable for inclusion in the meta-analysis (i.e., twenty-two were excluded; see Supplementary Table S1), encompassing information from a total of two hundred twenty-seven thousand six hundred forty-eight participants (Figure 2). In addition to recording effect sizes for right and left hand 2D:4D from these studies, we documented methodologically pertinent variables and study characteristics for planned moderation analysis, including publication status (published or unpublished) geographic location (North American, UK/Europe, Asia, or other) and digit measurement method (direct, self-report, photocopy/scan, mixed, or unknown). The collated data, encompassing effect sizes and moderator variables, is provided as a supplementary file.

Search Strategy and Study Selection

We conducted an article search from two thousand to two thousand twenty-four using PubMed and Google Scholar. Two search strategies were utilized on PubMed employing the terms: (1) (((((2D:4D and Sexual Orientation)); and (2) ((((((Digit ratio) OR (2D:4D)) OR (2D4D)) OR (finger length)) OR (digit length)) OR (finger) OR (digit)) AND ((((sexual orientation) OR (lesbian)) OR (bisexual)) OR (heterosexual)). Likewise, on Google Scholar, the search included the terms: (1) &x201C;2D:4D and Sexual Orientation&x201D; and (2) &x201C;Digit ratio&x201D;|&x201C;2D:4D&x201D;|&x201C;2D4D&x201D;|&x201C;finger length&x201D;|&x201C;digit length&x201D;|&x201C;finger&x201D;|&x201C;digit&x201D; and &x201C;lesbian&x201D;|&x201C;bisexual&x201D;|&x201C;sexual orientation&x201D;|&x201C;heterosexual.&x201D; PubMed retrieved a total of four hundred sixty-one reports, while Google Scholar yielded thirty-three thousand eight hundred reports; all Pubmed studies and the top four hundred articles returned by Google Scholar were reviewed for inclusion. Furthermore, we contacted the corresponding authors (n&x202F;=&x202F;296) of 2D:4D studies to request information on the sexual orientation of participants, as well as inquire about unpublished data on both 2D:4D and sexual orientation of participants.

Statistical Analyses

Meta-analyses were performed using a random effects model implemented via the "metafor" package (version 4.2-0; Viechtbauer, 2010) in R (version 4.3.0). Standardized effect sizes were calculated using Hedges's g via the "escalc" function, and the random effects models were tested with the "rma.mv" function. Additionally, leave-one-out analyses were executed utilizing the "leave1out" function to probe the robustness of results and their dependency on any individual study. To address publication bias, Duval and Tweedie's trim-and-fill tests were applied using the "trimfill" function. The data and analysis scripts for all tested models are accessible in the supplementary file.

Following Grimbos et al. (2010), we omitted Manning et al. (2007) from primary analyses owing to its potential to exert undue influence on meta-analytic results because of its size (&x0003E;200,000 participants). Results of analyses incorporating Manning et al. (2007) are presented in section Supplementary Results.

Results

Sex differences in digit ratios

Digit ratios manifested expected sex differences: Heterosexual men possessed lower 2D:4D than heterosexual women for both the right hand (g&x202F;=&x202F;&x02212;0.49, p&x202F;&x0003C;&x202F;0.001; Figure 3) and left hand (g&x202F;=&x202F;&x02212;0.43, p&x202F;&x0003C;&x202F;0.001).

Male sexual orientation and digit ratios

Exclusive heterosexual vs. exclusive homosexual men

Right 2D:4D

Exclusively heterosexual and homosexual men didn't significantly differ in right hand 2D:4D (g&x202F;=&x202F;&x02212;0.15, p&x202F;=&x202F;0.051; Figure 4A). Leave-one-out analysis yielded Hedge's g values ranging from &x02212;0.12 to &x02212;0.20 (Supplementary Figures S1&x2013;S4). No consequential moderators were identified for the right hand heterosexual and homosexual comparisons (Table 1). However, trim-and-fill analysis imputed seven studies and omitted one (Rahman, 2005), leading to a modified point estimate where 2D:4D is lower in heterosexual men than in homosexual men (g&x202F;=&x202F;&x02212;0.17, p&x202F;&x0003C;&x202F;0.001; Supplementary Figures S5&x2013;S10).

Left 2D:4D

Exclusively heterosexual men had a lower left hand 2D:4D than exclusively homosexual men (g&x202F;=&x202F;&x02212;0.18, p&x202F;&x0003C;&x202F;0.001; Figure 4B). Leave-one-out analyses produced Hedge's g values ranging from &x02212;0.16 to &x02212;0.20, suggesting that the difference between exclusive heterosexual and homosexual men does not rely on the inclusion of any specific study (Supplementary Figure S2). This relationship was moderated by publication status and only present in published studies, intimating a propensity for statistically significant effects to be disseminated (Table 1). Two missing studies were imputed during trim-and-fill analysis (Supplementary Figure S8), leading to a point estimate of g&x202F;=&x202F;&x02212;0.20. Geographical location, measurement type, and publication status were significant moderators of left hand digit ratios, but no pairwise comparisons were significant (Table 1).

Heterosexual vs. non-heterosexual men

Right 2D:4D

Exclusively heterosexual men had a reduced right 2D:4D than non-heterosexual (bisexual plus homosexual) men (g&x202F;=&x202F;&x02212;0.10, p&x202F;=&x202F;0.018; Figure 5A). This relationship was present in published studies, whereas the point estimate in unpublished studies was close to zero, suggesting a tendency for statistically significant effects to be published (Table 2). However, following trim-and-fill analyses, which imputed seven studies and excluded one (Rahman, 2005), the relationship remained significant, and the effect size increased (adjusted g&x202F;=&x202F;&x02212;0.17, p&x202F;&x0003C;&x202F;0.001; Supplementary Figure S9). Leave-one-out analyses generated Hedge's g values ranging from &x02212;0.08 to &x02212;0.11, suggesting that difference between heterosexual and non-heterosexual men is robust to the exclusion of individual studies (Supplementary Figure S3). Measurement type was a significant moderator, with a significant difference between photocopy/scan and mixed or unknown measures, suggesting that mixed methods or studies that did not report how they measured ratios were more likely to find higher right 2D:4D among heterosexual men than non-heterosexual men (Table 2).

Left 2D:4D

Exclusively heterosexual men also displayed a lower left hand 2D:4D than non-heterosexual (bisexual plus homosexual) men (g&x202F;=&x202F;&x02212;0.13, p&x202F;=&x202F;0.006; Figure 5B). Publication status was a significant moderator; however, the point estimate was almost identical for published (g&x202F;=&x202F;&x02212;0.13, p&x202F;=&x202F;0.013) and unpublished (g&x202F;=&x202F;&x02212;0.12, p&x202F;=&x202F;0.267) studies, and no studies were imputed in trim-and-fill analysis. Leave-one-out analyses for the left hand produced Hedge's g values ranging from &x02212;0.11 to &x02212;0.15, indicating these results are robust to the exclusion of any particular study (Supplementary Figure S4). Measurement type moderated left hand comparisons, with substantial differences observed between direct and mixed or unknown measures, between self-reported measures and mixed or unknown measures, and between photocopy/scan and mixed or unknown measures (Table 2). The effect size was in the converse direction for studies reporting mixed or unknown measurement methods compared to direct, self-report and photocopy/scan methods. Geographical location moderated relationships, but no pairwise comparisons attained significance.

Comparisons of heterosexual, bisexual, and homosexual men

We further tested whether relationships between sexual orientation and 2D:4D varied across comparisons between heterosexual and bisexual men, bisexual and homosexual men, and heterosexual and homosexual men throughout the eight samples for which these comparisons were feasible (Supplementary Table S4). In left 2D:4D, homosexual men manifested a higher (more female-typical) 2D:4D than heterosexual men, whereas bisexual men differed from neither heterosexual nor homosexual men. A comparable non-significant trend was evident for right 2D:4D.

Female sexual orientation and digit ratios

Exclusive heterosexual vs. exclusive homosexual women

Right 2D:4D

Exclusively heterosexual women possessed a higher right 2D:4D than exclusively homosexual women (g&x202F;= 0.26, p&x202F;=&x202F;0.016; Figure 6A). No moderators, including publication status, were significant (Table 3), and no studies were imputed or removed in the trim-and fill analysis (Supplementary Figure S8). Leave-one-out analysis produced Hedge's g values ranging from 0.14 to 0.30 (Supplementary Figures S13).

Left 2D:4D

Exclusively heterosexual women furthermore had a higher left 2D:4D than exclusively homosexual women (g&x202F;= 0.17, p&x202F;=&x202F;0.006; Figure 6B). Publication status moderated this effect: the effect was larger in unpublished (g&x202F;= 0.31, p&x202F;=&x202F;0.040) than published (g&x202F;= 0.14, p&x202F;=&x202F;0.033) studies (Table 3), although both effects were significant in the same direction. Following trim-and-fill analyses, no study was imputed and one study was removed (Kraemer et al., 2006), resulting in a significant adjusted main effect (g&x202F;= 0.16, p&x202F;=&x202F;0.010; Supplementary Figure S8). Leave-one-out analyses produced Hedge's g values ranging from 0.12 to 0.20, also suggesting that the results are robust to the exclusion of any particular study (Supplementary Figure S14). Geographical location moderated left-hand digit ratios. Pairwise comparisons revealed that UK/Europe differed significantly from North America, with UK/Europe having a larger positive effect size (lower left 2D:4D in exclusive homosexual women compared to heterosexual women) than North America (Table 2). Measurement type was a significant moderator, but no pairwise comparisons were significant.

Heterosexual vs. non-heterosexual women

Heterosexual women had a higher digit ratio than non-heterosexual women in both right (g&x202F;= 0.17, p&x202F;=&x202F;0.012; Figure 7A) and left (g&x202F;= 0.27, p&x202F;=&x202F;0.005; Figure 7B) hands. While publication status was a significant moderator for both hands (Table 4), effect sizes were comparable for published and unpublished studies (right hand: published g&x202F;= 0.17, p&x202F;=&x202F;0.023; unpublished g&x202F;= 0.18, p&x202F;=&x202F;0.324; left hand: published g&x202F;= 0.28, p&x202F;=&x202F;0.010; unpublished g&x202F;= 0.26, p&x202F;=&x202F;0.333). Following trim-and-fill analyses, no studies were imputed, and one study was removed for the right hand, resulting in an adjusted estimate of g&x202F;= 0.15, p&x202F;=&x202F;0.019 (Supplementary Figure S9). For the left hand, six studies were imputed and one study was removed, resulting in a non-significant adjusted estimate g&x202F;= 0.06, p&x202F;=&x202F;0.472 (Supplementary Figure S9). Leave-one-out analyses for right (Supplementary Figure S15) and left (Supplementary Figure S16) 2D:4D produced Hedge's g values of 0.09 to 0.18 and 0.18 to 0.29, respectively, indicating that differences between heterosexual and non-heterosexual women are robust to the exclusion of any particular study. Geographical location and measurement type were significant moderators for left, but not right, hand 2D:4D; however, no pairwise comparisons reached significance (Table 4).

Comparisons of heterosexual, bisexual, and homosexual women

We additionally tested whether relationships between sexual orientation and 2D:4D differed across comparisons between heterosexual and bisexual women, bisexual and homosexual women, and heterosexual and homosexual women across the six samples for which these comparisons were feasible (Supplementary Table S5). In both hands, heterosexual and bisexual women exhibited higher (more female-typical) 2D:4D than homosexual women but did not differ from one another.

Discussion

Though relatively few studies were accessible for comparing heterosexual, bisexual, and homosexual individuals separately, a tendency materialized in both sexes: 2D:4D ratios tended to be more alike between bisexual and heterosexual individuals than between either grouping and homosexual individuals. There was also a propensity for exclusively heterosexual and homosexual individuals to exhibit the utmost differences. These results indicate that the approach deployed in prior studies comparing heterosexual to non-heterosexual individuals may be less instructive than analyses comparing exclusively heterosexual and homosexual individuals. We hence center our discussion on comparisons between exclusively heterosexual and homosexual individuals and contemplate heterosexual/non-heterosexual comparisons in this light.

Our results replicate the main finding from a preceding meta-analysis (Grimbos et al., 2010) exhibiting an association between 2D:4D and women's sexual orientation: Homosexual women tend to possess lower (more male-typical) digit ratios in both hands than heterosexual women. However, the incorporation of unpublished data and supplementary published studies in the present meta-analysis, as well as comparisons of more homogenous groups (exclusive heterosexual vs. homosexual), appears to contribute in two crucial ways.

Firstly, the present data seem to be less influenced by publication bias than those in Grimbos et al. In the previous meta-analysis, adjusted effect sizes following trim-and-fill (right: 0.13, left: 0.07) were less than half of unadjusted values (right: 0.29, left: 0.23). In the present meta-analysis, the effect size for right 2D:4D (0.26) was unaltered following trim-and-fill, and the variance between adjusted (0.16) and unadjusted (0.17) effects for the left hand was negligible. Secondly, adjusted effect sizes were around twice as large in the present meta-analysis as in Grimbos et al. These results augment certainty that associations between sexual orientation and 2D:4D are real and meaningful (see below). Results from comparisons of heterosexual to non-heterosexual women were also positive but exhibited greater evidence of publication bias.

Contrary to Grimbos et al. (2010), our findings exhibited that exclusively homosexual men tend to have higher (more female-typical) 2D:4D ratios than exclusively heterosexual men. This association was statistically significant in the left hand previous to correction for publication bias, and in both hands following trim-and-fill analysis, which also slightly enhanced effect size estimations from &x02212;0.15 to &x02212;0.17 (right hand) and from &x02212;0.18 to &x02212;0.20 (left hand). Somewhat smaller, but statistically significant, relationships were observed in comparisons between heterosexual and non-heterosexual men.

Comparable to Grimbos et al. (2010), we excluded Manning et al. (2007) from primary analyses due to its potential to exert undue influence on meta-analytic results because of its size (&x0003E;200,000 participants). Nonetheless, this study is included in the Supplementary Results. Notably, the effect sizes remain nearly identical whether or not this study is included. The only exception is the unadjusted right hand comparison of heterosexual and homosexual men: with Manning et al. included, the effect is significant (p&x202F;=&x202F;0.045), whereas it is not when excluded (p&x202F;=&x202F;0.051); in both cases, the effect size is g&x202F;=&x202F;&x02212;0.15. All other effect sizes differ by g&x202F;=&x202F;0.02 or less, with no changes in significance.

It is notable that Manning et al. (2024) conducted a follow-up to their earlier work, Manning et al. (2007). In the initial analysis, Manning et al. (2007) compared distinct sexual orientation categories (i.e., homosexual, bisexual, and heterosexual), whilst the 2024 study assessed sexual attraction scores on a seven-point Likert scale. The 2007 findings revealed significant differences in men, with homosexual and bisexual men exhibiting higher 2D:4D ratios compared to heterosexual men, aligning with the results of the present meta-analyses. However, no substantial relationships