Fig. 3
From: Valection: design optimization for validation and verification studies
All Synthetic Data Simulation Results for Selection Strategy Parameter Combinations. Overall, the best results are obtained using the ‘equal per caller’ method. The ‘random rows’ approach scores comparably except in cases where there is high variability in prediction set sizes across callers. Calls from low-call callers are less likely to be sampled at random and, in cases where none are sampled, it is not possible to get performance estimates for those callers. Failed estimate runs are displayed in grey