การเปรยบเทยบวธการคดเลอกตวแปรอสระเขาสตวแบบการถดถอยเชงเสนพหคณโดยใชวธเบสและวธการถดถอยแบบขนตอน

The purpose of this research is to study variable selection for multiple linear regression models with Bayesian Variable Selection by using Gibbs Sampling (GS) and compare the efficiency of variable selection between Gibbs Sampling (GS) and Stepwise Regression (SR). The study employs simulation data which compared both cases with and without multicollinearity. A sample size of 25 and 100 is utilized, and repeated 500 times for each sample size. A significance level of 0.05 is set for selecting independent variables using SR. The comparison criterion includes the percentage of accurate variable selection and the mean squared error (MSE). When simulated dataset is small sample size, the percentage of accuracy with GS is higher than SR in both cases with and without multicollinearity. However, as the sample size increases, percentage of accurate of SR also increases for case without multicollinearity which is equivalent to GS. The MSE for both methods is equal, which is very low. This outcome is due to the similarity between true values and predicted values of the dependent variables obtained from each method. Conversely, when there is multicollinearity, both GS and SR are unable to accurately select variables for the model.