Comparison of Some Variable Selection Techniques in Regression Analysis | Biomedgrid

In this research, we analyze critically the performance of four variable selection techniques in the building of a model that best estimate a dependent variable. The variable selection techniques are the direct search on t method, the forward selectio n method, the backward elimination method, and the stepwise regression method. To compare, an economic data of 32 years was collected each on Real Gross Domestic Product, which was the dependent variable used as a measure for economic growth and development, and seven factors which are; Growth Market Capitalization, All-shares index, the Market turn-over, the openness of the Nigerian trade economy, the value of transaction, the total listing of the Nigeria stock exchange, and total new issue. Regressing the Real gross domestic product on the seven factors using the four variable selection techniques, the residual mean square, adjusted R2, and variance inflation factor obtained from the use of each of these techniques which are the criteria for evaluation of best model were compared by ranking them based on the method best satisfying the criteria for evaluation. The result shows that with a mean rank of 1.67 taken across the criteria, the backward elimination method performs best in variable selection based on the sample collected and it supported with the use of the all possible combination method as a control.