Table 2.

Summary of methods used in simulation study

General approach	Summary Measure of feature clusters	Descriptiona, b
SEPARATE	NA	Regression of the original predictors $\left\{X_1,\dots ,X_p\right\}$ on the response, i.e., no transformation of the predictors is being done here
CLUST	1st PC, average	Create clusters of predictors without using the environment variable $\left\{C_1,\dots ,C_k\right\}$. Use the summary measure of each cluster as inputs of the regression model.
ECLUST	1st PC, average	Create clusters of predictors using the environment variable $\left\{C_{k+1},\dots ,C_{\ell }\right\}$ where $k<\ell <p$, as well as clusters without the environment variable $\left\{C_1,\dots ,C_k\right\}$. Use summary measures of $\left\{C_1,\dots ,C_{\ell }\right\}$ as inputs of the regression model.

^aSimulations 1 and 2 used lasso and elasticnet for the linear models, and Simulation 3 used MARS for estimating nonlinear effects.

^bSimulations 4–6 convert the continuous response generated in simulations 1–3, respectively, into a binary response.

PC: principal component.