Correcting for sampling bias in quantitative measures of selection when fitness is discrete

Abstract

We show with a simulation that nonrepresentative sampling of two discrete fitness classes leads to biased estimates of selection. Systematic underestimation occurs if the selected class is overrepresented in the sample and overestimation if the unselected class is overrepresented. The bias is greater the stronger the selection intensity, the smaller the true fraction of individuals favored by selected, and the lower the sample size. We present a simple method that allows a posteriori statistical correction in cases of biased sampling given a separate estimate of the actual class representation, describe its practical implementation, and show that it works.