TABLE 5.

Summary and recommendations for stability measures

Stability measure	What it should estimate (interpretation of the estimand)	How it should be estimated (resampling approach)	Rationale and limitation
Variable inclusion frequency	Probability of a variable to be selected in the final model	Subsampling with m = 0.5N (S0.5)	Consistent and efficient estimation for nonpredictor
Model selection frequency	Probability of a specific combination of variables to constitute the final model	Subsampling with m = 0.5N (S0.5)	Consistent and efficient estimation
Relative conditional bias	Bias of a selected regression coefficient relative to the true regression coefficient	Bootstrap	Consistent and efficient estimation
Root mean squared difference ratio	Inflation of the SE of a regression coefficient induced by the variable selection procedure	Bootstrap	Consistent for uncorrelated predictors Overestimated for correlated predictors and for nonpredictors