Fig. 2.

Scatterplots showing the relationship between the NRI statistic (× 100) and β̂₁–α̂₁ in 1000 simulated datasets generated according to the scenario shown in the second to last row of Table 1. The coefficients are calculated by fitting the models logit P(D = 1|X) = α₀ + α₁X and logit P(D = 1|X, Y) = β₀ + β₁X + β₂Y to the training data. The NRI is calculated using the test dataset.