| Multicollinearity can be very difficult to deal with in regression analysis. Horel and Kennard (1970) proposed the use of ridge regression (RR) and generalized ridge regression (GRR) to deal with problems with highly correlated predictors. In recent years, the theory and application have developed rapidly. After introducing the definition and some useful properties of RR and GRR, a faster algorithm of GRR is discussed. Theoretically, it reduces complexity in some situations (especially when n<p). In the end, we also use data simulations to make comparisons between ordinary least squares and ridge regression in different situations. It shows that RR works better when strong multicollinearity is present among regressor variables. |
