Optimal Scaling: Regularization Including Ridge, Lasso, and Elastic Net Regression

Abstract

In the previous chapter we showed that linear regression, although commonly used in clinical research for analyzing the effects of predictors (x-variables) on outcome variables (y-axis variables), has a problem because consecutive levels of the predictor variables are assumed to be equal while in practice this is virtually never true. We showed that optimal scaling can largely improve the sensitivity of testing, but in order to fully benefit from this methodology the risk of overfitting, otherwise called overdispersion, of the data should be accounted.