Simulated replicate trajectories of a Wright–Fisher model for an asexual haploid population of size $N=500$⁠. In the model, a single biallelic locus, with alleles A and B, is under selection. With $X(t)$ denoting the frequency of the A allele in generation t, all trajectories started from an initial frequency of $X(0)=a/N=200/500=0.4$⁠. The selection coefficient of the A allele, relative to that of the B allele is $s=−0.01$⁠, and there are equal forward and backward mutation rates of the A allele of $μ=ν=10−5$⁠. The figure shows only 20 of the stochastic trajectories but 10,000 replicate trajectories were used to calculate statistics at an observation time of $tobs=150$ generations. The 10,000 replicate trajectories represent 10,000 independent loci in the hybridization scenario described in the text. A threshold, acting at a frequency of $z/N=25/500=5%$⁠, renders undetectable any of the focal loci with an A allele frequency of $5%$ or less. For a particular simulation run, a proportion $Pdet≃0.6467$ of all 10,000 trajectories were above threshold at $tobs$⁠, and hence were detectable. For this simulation run the unconditioned mean value of the frequency at the observation time (i.e., calculated from all 10,000 trajectories) was $E[X(tobs)]≃0.1498$⁠, but the corresponding mean value, when conditioned to lie above threshold, was found to be $Ez[X(tobs)]≃0.2249$⁠. Using $Eest[X(tobs)]=Ez[X(tobs)]×Pdet$⁠, as an estimate of the unconditioned mean frequency, leads to $Eest[X(tobs)]≃0.1455$⁠. Thus for this simulation run, the conditioned expected value, $Ez[X(tobs)]$⁠, is approximately $50%$ larger than the unconditioned result $E[X(tobs)]$⁠, while the estimated result, following from $Eest[X(tobs)]$⁠, differs from the unconditioned result by approximately $3%$⁠. More statistics, based on this simulation run, are given in table 1.