Effect of unmodeled demographic changes in the ancestral population on model choice and parameter estimates. In this figure, we report biases in model choice and parameter estimation of simple IM and SC models for IM simulations with $TS=0.05−1.6$ including an ancestral demographic expansion (A) and an ancestral bottleneck followed by expansion (B). On the left of panels (A) and (B) is a graphical representation of the simulation model. In panel (A), the y axis represents the ancestral population size after the expansion as a multiplier of its size preceding the expansion, and the x axis represents the time before T_S at which the ancestral expansion happened (T_AE). In panel (B), the y axis represents the size of the ancestral population after contraction as fraction of the population size preceding contraction and the x axis represents the time before T_S at which the ancestral bottleneck happened (T_AB). Times are given in units of $4NREF$ generations. For (A) and (B) and for each simulated T_S, we present the misestimation of T_S (⁠$TSmis=TSestimated/TSsimulated$⁠), the weight of evidence (W_AIC) for the correct (IM) gene flow scenario as well as the estimated time of strict isolation as a proportion of total divergence time (⁠$TSI/TS$⁠). In each graph, the bottom-left square represents estimates for simulations with no ancestral changes in N_e.