Can simple population genetic models reconcile partial match frequencies observed in large forensic databases?

A recent study of partial matches in the Arizona offender database of DNA profiles has revealed a large number of nine and ten locus matches. I use simple models that incorporate the product rule, population substructure, and relatedness to predict the expected number of matches in large databases. I find that there is a relatively narrow window of parameter values that can plausibly describe the Arizona results. Further research could help determine if the Arizona samples are congruent with some of the models presented here or whether fundamental assumptions for predicting these match frequencies requires adjustments.