I've been asked how to control for regression to the mean effects in a difference of differences analysis on health insurance data.
We're measuring a utilization outcome both pre- and post-intervention for a treatment group and comparing with a control group who didn't receive the intervention program. The concern here is that patients who were high utilization pre-intervention will tend to regress to average utilization following intervention regardless of the effectiveness of the intervention program, and vice versa for the low utilization folks.
My understanding is that techniques such as the James-Stein estimator and its multivariate extension, the ridge regression, control for regression to the mean effects by shrinking regression coefficients to the mean.
My question: Can I use ridge regression (or one of its cousins, lasso or elastic net) as a go-to methodology for minimizing regression to the mean effects?