Post date: Apr 05, 2017 8:10:36 PM
Initial results with entropy ancestry estimates of AIMs had suggested a strong signal whereby the most differentiated SNPs (delta > 0.8) were skewed towards the average ancestry (i.e., more anna like in more anna like populations and vice a versa for melissa). I had been unable to replicate this result with popanc estimates. This was due to an error with the entropy-based analysis whereby I was using the mean ancestry rather than 0.5 as the binomial expectation. Upon fixing this there is still a significant but weaker signal for entropy estimates but not (and even a tendency to reverse) for popanc. This is true despite overall correlations in genome-average and locus-specific ancestry estimates. Rather it reflects shifts in the shape of the locus-specific distributions that affect whether highly differentiated SNPs are on one side or the other of the mean (this makes some sense as popanc is smoothing things more). One result is not clearly better than the other and just presenting one (i.e., the entropy-based results) would give a less than complete picture of things. I may have to drop this or figure out a way to better reconcile the analyses.