In this imulation study , we simulate the data and use it to compare three Penalized Least Squares Estimators:
- L1-norm penalty LASSO.
- Smoothly Clipped Absolute Deviations penalty (SCAD).
- Minimax Concave Penalty (MCP).
We perform 50 independent replications and choose tunning parrameters by cross-validation.
We answer to the question in which case SCAD and MCP mitigate bias better than LASSO, by examining the error variance.