Problems with computational methods in population ge - netics

Advances in computational power, statistical methodology, and population genetics theory have recently had an impact on the way in which information is extracted from molecular genetic data. Several computationally intensive schemes have already been developed for analysis of such data under simple models (see for example Gri ths and Tavar e, 1994; Kuhner et al., 1995; Wilson and Balding, 1998), and their popularity will no-doubt increase as they become able to incorporate more biologically-realistic assumptions. It is important to recognize that these methods are being used to tackle di cult problems which are extremely challenging, even using the sophisticated theory and technology now available. Since these methods are likely to be widely applied by people who are not experts in computational statistics, it is particularly important to develop methods which are easily used to produce reliable results. This piece aims to describe one sense in which some current approaches are more likely to provide reliable results than others.