There’s debate about the decline of sex, or how the young people these days aren’t what they used to be. There’s plots like this one:
I’m pretty sure there is a newer version of this figure, but I couldn’t find it while writing this post. Here’s one for sex among Californian 18-30s:
This could be attributable to any number of factors, like social media eroding real life connections, dating apps messing with human dating behavior, people being too fat to be attractive etc. Another hypothesis is that it could be attributed to hostility between the sexes, as resulting from increase in left-wing politics:
If mating is strongly assortative for politics — which it is, r = .58 — then increased political gaps between the sexes suggests finding a partner is becoming more difficult, everything else equal. Note the strong divergence in politics between the sexes in South Korea and the country’s recent decline in fertility. There’s one neat way to test this idea, namely to check key predictions regarding sexual minorities:
- If there’s a growing divergence between the ideology of men and women, sexlessness should be increasing among younger heterosexual.
- But there should be no temporal tend for homosexuals.
I was asked to look into the GSS data, which is long-running, but somewhat too small for this. I don’t have any other dataset to try, so let’s see what GSS can tell us.
First, the key variables are:
- sexornt: Which of the following best describes you?: Gay, lesbian, or homosexual; Bisexual; Heterosexual or straight.
- partners: How many sex partners have you had in the last 12 months? The options are various ordinals from 0 to 100+
We begin by plotting the overall distribution by sex and orientation:
Female heteros report somewhat less promiscuity overall, even than the male heteros. This is not entirely impossible because there are more old women than men. This finding is common, even without the old age imbalance issue. Also, overall, non-heterosexuals are more promiscuous, especially the males. However, they are also much younger, so we need to control for age. I fit an ordinal logistic model (ordinals::clm in R), and produced this result:
We still see the hetero sex difference, and we still see the non-heterosexual promiscuity, except for lesbians. And yes, the values for male homo and bi are identical because the model estimates for them are very similar. Male bisexuals appear to be functionally gay insofar as promiscuity is concerned.
With this in mind, we need to look at the changes over time. The data goes from 2008-2022, so before the great awokening to 2 years ago. The prediction being that heterosexuals increased in sexlessness due to sexual conflict in politics, but homosexuals did not. Can we confirm this? Because of the sparse data (e.g. only ~130 non-heterosexuals in the 2022 survey), trying to estimate the full ordinal distribution by year and sex & orientation is perhaps too much. However, can try to estimate just whether someone was having sex or not, simplifying it to a matter of estimating proportions for 6 groups with a change over time. If we further stipulate that the effect of year is a monotonic linear change, we can simplify the model further. It produced this result:
Can you see anything? Neither can the model. There’s no statistically detect effects of year for any group. If we squint and accept the flimsiest evidence of p < .10, the only groups with changes are female bisexuals getting less likely to have 0 partners in 12 months, and male bisexuals getting more likely. The confidence intervals are very wide, so this is not really much evidence against the hypothesis either.
Ideally, we would predict the distribution for each year of data by sex and orientation. Given the sparseness of the data, maybe we need to go full Bayesian. Alas, after fiddling this with this days and delaying this post, I wasn’t able to produce anything sensible. Bayesian L. The conclusion being that the GSS doesn’t have enough sample size to squeeze anything out of it, not even using Bayesian tricks (I also tried simplifying the distribution to those not having enough sex partners, those having the ‘right’ number (1-2) and those having too many (3+). But hopefully some reader can point to another public dataset that has more non-heterosexuals, so that one can test this model.
