An Empirical Study on the Use of Principal Components in overcoming Multicollinearity Problems

Abstract

When the independent variates in a regression model exhibit perfect or high degree of dependence among themselves, the method of least squares fails to provide satisfactory estimates of regression coefficients. The variables are then said to be collinear and the phenomena is termed multicollinearity. In studies pertaining to socio-economic conditions this phenomena is most likely to occur. We are, therefore, not concerned whether or not multicollinearity exists in our data in such situations but upto what degree it exists and affects our estimates, and how the problem can be dealt with, since the presence of severe multicollinearity badly affects the least square estimators and makes them imprecise. Though there are many methods to overcome this problem, we have in our study emphasized the use of Principal Component Analysis as the most suitable method.