Simulation-based distributions of odds ratio estimates from RFPM and logistic regression with 100 features. Simulated distributions of the conditional odds ratios for features with non-null association with the outcome under the simulation model. Under this model, 20 features out of 100 have non-null associations with conditional odds ratios simulated from a N(0.7,0.2) distribution, then multiplied randomly by -1 or 1. The data is generated from a logistic regression model. We find that RFPM estimates are attenuated to the null, whereas logistic regression estimates tend to be biased away from the null. Diamonds denotes the true coefficients under the simulation model.