Allele frequency models

There are two basic models for the allele frequencies. One model assumes that the allele frequencies in each population are independent draws from a distribution that is specified by a parameter called $\lambda$. That is the original model that we used in Pritchard et al. (2000a). Usually we set $\lambda=1$; this is the default setting. More recently, we have implemented a model with correlated allele frequencies. This says that frequencies in the different populations are likely to be similar (probably due to migration or shared ancestry). Further details are given below.⁷ The independent model works well for many data sets. Roughly speaking, this prior says that we expect allele frequencies in different populations to be reasonably different from each other. The correlated frequencies model says that they may actually be quite similar. This often improves clustering for closely related populations, but may increase the risk of over-estimating $K$ (see below). If one population is quite divergent from the others, the correlated model can sometimes achieve better inference if that population is removed.