“Author's note Table 1 comparing the accuracy of MasterBayes and Pedant appears to be misleading, in that I didn't use the correct summary of the posterior distribution. This led me to the erroneous conclusion that MasterBayes gives more biased estimates than Pedant. In fact, while my estimates of Pedant's bias are accurate, MasterBayes turns out to be unbiased. See this note from Jarrod Hadfield's tutorial on MasterBayes: "Johnson and Haydon [2007] find that MasterBayes overestimates error rates by 31% when error rates are low (compared to a downward bias of 18% for Pedant). However, large scale simulation shows that MasterBayes is unbiased (if such a thing exists in a Bayesian analysis) if the posterior mode is used rather than the posterior mean." p29 of MasterBayes: Maximum Likelihood and Markov chain Monte Carlo methods for pedigree reconstruction, analysis and simulation. J. D. Hadfield, July 21, 2009 http://cran.es.r-project.org/web/packages/MasterBayes/vignettes/Tutorial.pdf (A minor point: I actually used the posterior median, which is likely to be less biased than the mean, but presumably still not as accuarate as using the mode.) I re-ran the analysis for Table 1 on the same simulated data, using the mode instead of the median. The results agree with Jarrod's conclusions: there was no bias in any of the estimates (all 4 mean parameter estimates were within 2 SEs of their target). With hindsight it isn't surprising that MasterBayes is more accurate, given that MasterBayes' error model is more realistic than Pedant's. Pedant makes a few simplifying assumptions (i.e. convenient shortcuts), such as assuming that the only way to mistype an AB genotype as AA is via allelic dropout (AD), whereas in reality it's possible (if unlikely for variable microsatellites) for this error to arise via a false allele (FA) error. This is presumably why Pedant overestimates the number of ADs and underestimates FAs. Paul Johnson 23/10/09”