There seem to be ways to post knitr documents to WordPress blogs, but until that’s set up, I will be publishing them over at RPubs and posting a link here. The post begins like this:

In a post published on his website, Gwern investigates the efficiency of embryo selection. It’s impressive work. In a later revision, he added a simulation that examines the effects of selecting for genetically correlated traits.

He relies on real data about genetic correlations and has selected a set of 35 traits of interest. He finds that when one considers selection on a composite trait of the 35 traits, one can make astonishing gains (mean gain 3.64 Z) which become ever larger when one takes into account the genetic correlations between traits (5.20 Z), which tend to be favorable (i.e. positive correlations between traits we want). Unfortunately, this conclusion is based on a couple of mistakes. One can see this intuitively by considering that if one has 10 embryos to select from, it is not possible to do better than about 1/10 on average, no matter if one selects for a single or composite trait, so there must be a mistake somewhere.