Fig. 2
From: Causal models and prediction in cell line perturbation experiments
Comparison of performance of LR and CSR/Cellbox on simulated data. The x-axis is the true response and the y-axis is the predicted response. LR and CSR perform similarly for RF validation. For RF with B misspecified, CSR, which uses B, performs poorly. LR is unaffected by misspecified B because B is not used in the LR estimator. For LODO validation, LR performs poorly because it cannot model the effect of the left out drug on the responses