Spatial Autocorrelation Methods in Population Genetics

Spatial autocorrelation methods may be used in population genetics in several ways. They can be tools in exploratory data analysis; they can be used to test for significant correlations in allele frequencies at different locations; and they can be used to understand processes governing allele frequencies. We consider the question of whether spatial autocorrelation methods are useful for determining what processes are governing allele frequencies or for estimating parameter values in models of those processes. We point out that allele frequencies in general and spatial correlograms in particular are affected by several sources of variation: sampling variation, stochastic variation caused by unpredictable events in the history of each allele, parametric variation caused by differences among processes governing allele frequencies at different loci, and variation caused by differences in initial conditions. We review existing population-genetics theory that demonstrates that parametric variation may be very important in populations at equilibrium. We present results from a simulation model that shows that stochastic variation is also important for realistic parameter values. Previous studies of spatial autocorrelations and tests for significant differences in correlograms have not taken all sources of variation into account