Post date: Nov 20, 2013 9:50:16 PM
I now have 10 chains for each individual experiment and five for each combination. I think this is sufficient. These MCMC results are in projects/lycaeides_hostplant/melGemma/output but I might move them later. I used the perl scripts summarizeGemmaComb.pl and summarizeGemma.pl to obtain point estimates and 90% ETPI for pve, pge, n_gamma (# loci) and the prior inclusion probability for each experiment or combined set with samples from all five or 10 chains. Here are the results for the individual experiments and combined sets. And here are the related plots, (a) individual experiments, (b) combined sets. We see clearly non-zero pve for GLA on Ms (surv), SLA on Ac (surv), GLA on Ac (wgt), Ac (surv), Gla (sruv), overall (surv), and Gla (wgt). I am not quite sure how to interpret these results overall, but they do indicate that there is some segregating genetic variation associated with performance, and that some of this is common across populations or host plants. But I still don't quite trust the SLA on Ac results (we can explain much of the variation, but does this really mean anything). This is where the posterior predictive cross-validation will be useful.