A Comparative Study of Some Estimation Methods for Multicollinear Data

Okeke Evelyn Nkiruka, Okeke Joseph Uchenna

This article compares different estimation methods specially designed to combat the problem induced by multicollinearity using real life data of different specifications and distributions. From the mean squared error of the samples studied we observed that Partial least square came up as the best estimator among the methods we studied. Stepwise regression performs better when the predictor variables are highly correlated. Under the ridge regression study the smallest eigenvalue of the predictor variables of the original data was used in determining the ridge parameter of ridge regression since the variances of some of our samples cannot be estimated by ordinary least squares regression. From our results we found that among all the methods we studied PLSR estimator stands the “best”, followed by the stepwise regression and then the PCR estimator in predicting the response variable. We are not surprise that RR estimator stands the least among the methods since it is known as biasing estimator and more useful in estimating the parameters of the model.  We also wish to state that PLSR is efficient in prediction when the sample size is very small.