2D:4D ratio is the length of someone’s index finger divided by the length of their ring finger on the same hand. It seems to correlate with some kind of prenatal hormone exposure, which makes it a unique way to explore the effects of purely-biological factors on various traits without having to do hormone assays or go through an ethics board.
Past research suggests that men generally have lower 2D:4D ratios than women, and that within each sex people with various stereotypically-male characteristics have lower 2D:4D ratios than people with various stereotypically female characteristics. For example, aggressiveness, penis length in men, and lesbianism in women have all been correlated with more masculine 2D:4D ratio. But there have also been other studies that challenge some of these results, and the whole field is maddeningly inconsistent.
Maybe the weirdest paper along these lines is Madison et al’s 2014 paper showing that feminist activist women have more masculinized 2D:4D ratios than other women, which suggests something about how “neurological gender” influences the way people respond to social gender roles. Or something.
I wanted to look at this more, so I included questions about digit ratio in the 2014 Less Wrong Survey and in the 2014 Slate Star Codex Survey. Both surveys also included demographics questions, questions about sex and gender identification, and questions about political beliefs.
Methods and Results
407 people on the Less Wrong Survey and 122 people on the Slate Star Codex Survey gave plausible digit ratio measurements (I rejected measurements outside the range of 0.8 – 1.3 as implausible), for a total of 529 people. Of these, 454 were biologically male and 75 biologically female. Although participants gave digit ratios from both their right and left hands, the differences were not significant and I averaged them together to create a less noisy measure.
The average male digit ratio was 0.972; the average female digit ratio was 0.975. The difference was not significant, didn’t trend toward significance and actually was the opposite direction on right vs. left hands.
Average Bem Femininity was 42.2 for men and 45.8 for women. Average Bem Masculinity was 42.3 for men and 40.5 for women. The sex difference in femininity was significant (p = 0.01) but masculinity wasn’t (p = 0.19) though it trended in the expected direction.
On a scale from one (strongly disagree) to five (strongly agree), the average male opinion of feminism was 3.4; the average female opinion was 4.1; the difference was significant at a p < 0.001 level. Other political opinions also had gender differences, always with women more liberal, but the gender difference in feminism was the strongest.
In a subsample limited to men, digit ratio did not correlate with Bem sex roles at all, but did correlate with positive opinion of feminism (p = 0.001). Bem sex role did corelate with feminism; more masculine men had lower opinions of feminism (r = -0.277, p < 0.001) and more feminine men had higher opinions of feminism (r = +0.277, p < 0.001).
In the opposite subsample limited to women, digit ratio correlated strongly with Bem masculinity (r = -0.381, p = 0.005) but not with Bem femininity (r = +0.123, p = 0.38). Opinion of feminism did not correlate with either sex role or digit ratio and there wasn’t even a real trend.
I looked at eight other political issues not clearly related to gender. None of these had a clear connection with digit ratio the same way feminism did. One issue, immigration, had a single correlation significant at the p = 0.05 level, but was likely a false positive.
There was a limited sample of transgender people in the study; because most participants were biologically male I compared cismen to transwomen. Difference in digit ratio was not significant and showed no clear trend. Bem sex role inventory found transwomen to be much less masculine and much more feminine than cismen (and in fact less masculine and more feminine than ciswomen), but because of the low sample size the trend didn’t quite reach significance. Transwomen were more positive towards feminism than cismen (p = .047) but not quite as positive as ciswomen. I tried something similar with straight men versus non-straight men and the trends were arguably in the expected direction but didn’t come close to significance. There weren’t enough biologically female people to be worth running the analyses with that sample.
A binary autism variable (does vs. does not identify as autistic) had little correlation with sex roles or opinion of feminism, but approached significance for digit ratio in men (p = 0.051) and achieved significance once men and women were combined (p = 0.047). Autistic people had slightly lower (more masculine) digit ratios. I am wary of this result since it was so weak and not replicated on the left hand.
This was weird.
The whole point of digit ratio work is that men are supposed to have lower digit ratio than women, and then you use that to determine whether other characteristics are associated with more male-typical or female-typical hormone balances. But we couldn’t even get the basic finding of men being more male-typical than women.
In fact, our entire sample was heavily feminized compared to those of every other study; our men had more feminine digit ratios than other studies’ women. This is really weird because it’s a sample selected for very high mathematical ability – average IQ of 139, many math and physics PhDs, computer-related jobs as the most common occupations – but mathematical ability is usually linked to more masculinized digit ratio. Likewise, presumably this sample’s much more autistic than normal, but that’s also supposed to be more masculinized. I think maybe the measurement technique on the survey predisposed towards overreporting the ratio. Or maybe I screwed up somewhere and divided when I was supposed to multiply. Really that’s all I can think of.
Despite our failure to pick up what should have been the most drop-dead obvious finding, we still got strong signals on psychological traits that should have been much more subtle. The two clear results were a correlation between digit ratio and opinion-of-feminism in men, and between digit ratio and masculinity in women. P-values for both were very low and unlikely to be coincidences despite the multiple tests performed. But the pattern is hard to explain.
I don’t understand the biology of digit ratio very well, but it’s certainly only one part of sexual differentiation; after all, nearly all studies show wide overlaps between men and women in digit ratio, but except for a few intersex people the sexes still clearly differentiate anyway. So maybe women’s level of masculinity is determined mostly by the thing behind digit ratio, and men’s level of masculinity is determined by that plus various other male-specific masculinizing processes?
More annoying is the apparent partial-disconnect between digit ratios, sex roles, and feminism. In men, digit ratio affected feminism but not sex roles. In women, digit ratio affected sex roles but not feminism. The results weren’t ambiguous either. But why would digit ratio make men less feminist if not by making them less gender-conforming themselves? And is it really plausible that digit ratio makes women much more masculine without shifting their opinions toward feminism at all? And remember, this whole area was started by someone who found a connection between digit ratio and feminist activism in women. While it’s possible that the original study was really finding that masculine-digit-ratio women were more activist rather than more feminist per se, that seems like kind of a stretch. I’m going to blame this one on low sample size and high measurement error until somebody forces me otherwise.
In conclusion, this study was a mess, but somehow managed to find clear signals in weird places anyway. I don’t know.
If you want to see for yourself, you can find the public Less Wrong survey data here and the public SSC survey data here. You could also get my haphazard combination of the two here but I would warn against that if you’re really trying to double-check the results as the combination process was one of the most potentially error-prone steps and I’d rather see it independently replicated.