Evaluating Dispersal Estimates Using mtDNA Data: Comparing Analytical and Simulation Approaches

Abstract

Abstract: Using genetic data to estimate dispersal is attractive because direct estimation is often infeasible. Unfortunately, analytical tools remain rudimentary. Various standard population genetic formulae estimate dispersal rates between populations but require unrealistic assumptions, such as equal population size. Recent increases in computer speed enable simulations of genetic processes in populations. We designed simulation models to evaluate the adequacy of using analytical formulae to estimate dispersal rates needed for conservation decisions. Our models incorporated genetic stochasticity. They adhered closely to the assumptions of the standard formulae except that we relaxed the assumption that all populations were equal in size. We found that measures of population differentiation, such as the statistic ΦS T , varied greatly over time, reflecting the stochastic nature of genetic systems. In some cases, however, even large amounts of variation over time made no difference in defining population structure for management purposes, whereas in other cases the opposite was true. Statistics of genetic differentiation were so variable that management decisions based on analytical formulae would often be incorrect. Adjoining populations with little dispersal between them that should be managed separately were by chance alone so genetically similar that they appear panmictic, incorrectly implying that they should be managed as a single unit. The form of distributions of these statistics was complex and did not conform to standard statistical distributions. Furthermore, differences between these distributions for equal and unequal abundance cases indicated that pursuits of simple adjustments to analytical formulae are unlikely to be fruitful. In contrast, our simulation technique can be used in a decision analysis framework because it incorporates uncertainties into a probability distribution. We conclude with a case in which a manager wants to know whether current human‐caused mortality levels can be sustained while maintaining the local population at healthy levels. The example demonstrates how decision analysis for population structure can use the probability distributions generated from simulations to incorporate uncertainty and estimate over‐ and underprotection errors.