The Double-Lasso Method for Principled Variable Selection.

The decision of whether to control for covariates, and how to select which covariates toinclude, is ubiquitous in psychological research. Failing to control for valid covariates can yieldbiased parameter estimates in correlational analyses or in imperfectly randomized experimentsand contributes to underpowered analyses even in effectively randomized experiments. Weintroduce double-lasso regression as a principle method for variable selection. The double lassomethod is calibrated to not over-select potentially spurious covariates, and simulationsdemonstrate that using this method reduces error and increases statistical power. This methodcan be used to identify which covariates have sufficient empirical support for inclusion inanalyses of correlations, moderation, mediation and experimental interventions, as well as to testfor the effectiveness of randomization. We illustrate both the method’s usefulness and how toimplement it in practice by applying it to four analyses from the prior literature, using bothcorrelational and experimental data.