A basic hypothesis is that cultural evolutionary processes sustain differences between groups, these differences have evolutionary relevance and they would not otherwise occur in a system without cultural transmission. The empirical challenge is that groups vary for many reasons, and isolating the causal effects of culture often requires appropriate data and a quasi-experimental approach to analysis. We address this challenge with historical data from the final Soviet census of 1989, and our analysis is an example of the epidemiological approach to identifying cultural variation. We find that the fertility decisions of Armenian, Georgian and Azeri parents living in Soviet-era Russia were significantly more son-biased than those of other ethnic groups in Russia. This bias for sons took the form of differential stopping rules; families with sons stopped having children sooner than families without sons. This finding suggests that the increase in sex ratios at birth in the Caucasus, which began in the 1990s, reflects a cultural preference for sons that predates the end of the Soviet Union. This result also supports one of the key hypotheses of gene–culture coevolution, namely that cultural evolutionary processes can support group-level differences in selection pressures that would not otherwise occur in a system without culture.