Post date: Mar 04, 2014 11:34:37 PM
The MCMC runs finished and the results are in projects/lycaeides_hostplant/melGemma/mcmc_wgt_predict/. I used the script summarizeWgtPredict.pl to summarize the results. Across all 10 replicates there was a significant positive correlation between true and predicted weight for GLA on Ac (r = 0.312, p < 0.001) and SLA on Ac (r = 0.094, p = 0.026), but not for either population on Ms (GLA r = 0.061, p = 0.432; SLA r = 0.105, p = 0.218). Thus similar to the survival cross-validation results we have some ability to predict phenotype in some experiments, but little to none in others (note however that all r's are positive). Next I need to try the family level cross-validation.