Deficiency of Ridge Regression in Double Sampling for Regression using Two Auxiliary Variables

Abstract

Double sampling for regression is an advanced sampling and estimation method in Statistics. Sahoo’s Chain regression in 1993 was a case study of double sampling for regression where two auxiliary variables ( and ) are used. Authors have established that ridge regression performs better than Ordinary Least Squares (OLS) estimation in the presence of collinearity problem in linear regression with more than one independent variable. However, since double sampling for regression with more than one auxiliary variable could as well be faced with collinearity, hence, there is need to verify the validity of ridge regression in estimating the regression coefficient for use in double sampling for regression. This research applied ridge regression estimation method to Sahoo double sampling chain regression estimator when there exists high collinearity between the two auxiliary variables and . The empirical results show that two auxiliary variables maintains higher efficiency over one auxiliary variable in double sampling for regression; it was further established that ridge regression not only performed poorly than OLS estimation but may also cause Heywood case (negative variance which is an abnormality) in double sampling for regression with two auxiliary variables in the atmosphere of multicollinearity. Further investigations show that removal of outliers in the data may solve collinearity problem, hence, OLS estimation can further be used instead of Ridge Regression in double sampling for regression with two auxiliary variables.