The document compares several regression methods, including ordinary least squares (OLS), principal component regression (PCR), ridge regression (RR), and partial least squares (PLS) regression. It finds that PLS regression generally yields better predictive models than the other methods when applied to real data with multicollinearity among predictor variables. The document provides theoretical background on each method and evaluates their performance on a real dataset using measures like root-mean-square error of cross-validation (RMSECV).