Figure 4

The proportions of a patient variable assigned to each of the four alternative models using the AIC (left hand panels) or BIC (right hand panels). Data shown is for a sample size of 800. The red point/lines/shading indicate the regions corresponding to the true model and the gray lines/shading indicate the model is incorrect. The “true” models include the covariate in neither (model 0), the E (model E), U (model U) or both (model EU) sub-models as indicated.