Study of Multicollinearity on the Effect of Climatic Conditions on Oil Palm Yield at NIFOR Nigeria

ABSTRACT

One of the major problems of multiple linear regression analysis is multicollinearity of the independent variables. The existence of multicollineariity on climate variables such as relative humidity, solar radiation, rainfall, sunshine and temperature on the response of agricultural output may lead to inflation of standard error of the regression coefficients or false non-significant p-value. In this study, monthly data spanning from 1980-2012 obtained from the Nigeria Institute for Oil Palm Research (NIFOR) on relative humidity, solar radiation, rainfall, sunshine, temperature and oil palm yield were used to examine the probable effects of climate conditions/climate change would have on oil palm yield. The estimation of parameters of climatic variables in multiple linear regression appears to have suffered severe distortions due to multicollinearity. This research study resort to principal component regression, ridge regression and stepwise regression to stabilized the parameter estimate. Ridge regression was used to estimate the effect of climate conditions on oil palm yield because it performed better than others due to its lower measure of accuracy. It was observed that average relative humidity and rainfall had positive significant effect while solar radiation, mean sunshine hour and average air temperature had negative significant effect on oil palm yield.