3.3. Simulations: variable selection procedure
We compare the proposed procedure GBEAMS to the procedure in [35] (KSIR-GAM in the tables), Group Lasso (available
in R package grplasso) and Group SCAD (available in R package grpreg) in two different setups. In both cases the groups
consist of 5 covariates each. The variables within group i are denoted by Xi1, . . . , Xi5. In the first setup we use the additive
model in Zhu and Li which we call Model 1. In this model, the Xij follow a uniform distribution between 0 and 1, variables
within each of the 10 groups admit a compound symmetric correlation structure with correlation 0.2 while variables in
different groups are independent. The response Y is related to 4 groups through the relation