Figure 3.

Model Performance for Choosing K under Two In-model Simulation Scenarios

Boxplots of the difference in the conditional log likelihood from the log likelihood of the true K. The conditional log likelihood was computed on 5,000 held out genotypes (y axis) across 10 for different choices of K (x axis). The highest log likelihood is chosen as the “true” K. For the mixture scenario, the correct K (dashed vertical line) is identified 10 out of 10 times. For the merger scenario, the correct K is identified 9 out of 10 times.