Abstract:
The objective of this research is to compare the Bayesian selection methods for best regression equation with conjugate gamma prior distribution. The three methods in this comparison are Bayesian Model Averaging method using Markov Chain Monte Carlo model composition , Optimal Predictive Model Selection (OPM) and Stepwise Regression method (SR). The criterion of comparison is average of mean square error (AMSE) to compare the efficiency of these three methods. In this study, the numbers of independent variables in regression model are 3 5 10 and 15. The size of the samples are 15 30 50 and 100. The standard deviation of random errors are 0.25 0.50 and 2.50. The constant of Bayesian approach for and OPM are (1,5) (1,10) (10,100) and (10,500). All data for this research are generated through the Monte Carlo simulation technique and repeating 500 times for each case. The result of this study can be summarized as follow: 1. The AMSE from the three methods rank from minimum to maximum are, OPM and SR for all cases. The has AMSE lower than OPM but SR has AMSE higher than and OPM for all cases. 2. The factors that affect AMSE of all methods are the numbers of independent variables, sample size and the standard deviation of random errors. The AMSE of all methods directly vary to the numbers of independent variables and the standard deviation of random errors but the AMSE of all methods inversely vary to sample size. 3. The constant of Bayesian approach affected AMSE of both and OPM. The AMSE of the two methods directly vary to the constant of Bayesian approach.