การเปรียบเทียบเกณฑ์การคัดเลือกตัวแบบความความถดถอยเชิงเส้นด้วยแนวคิดเชิงแลตทิซ = A comparison on selection criteria for linear regression models based on lattice concept / Surasit Ritsmitchai
The objective of this research is to compare two criterions of linear regression models selection by using the lattice method. The two criterions are Residual Sum of Squares (RSS) and Mean Square Prediction Error (MSPE) of which the number of independent variables in this study equal to 3 and 4. The distribution of error is zero mean normal distribution and standard deviation equal to 1, 2, 3 and 5. The sample sizes are 20, 35 and 50. The data in this research are obtained from computer employing Monte Carlo technique for 500 times. The research shows that when all of correlation coefficients between variables in the model less than 0.55, for using RSS or MSPE criterions, the model which is selected by concurrently comparing all possible models and using the lattice method is the full model. In case of the model which has some two independent variables with correlation coefficient exceeding 0.55, the models selection by using the lattice method is more probable to select reduced model than the models selection by concurrently comparing all possible models. Moreover, MSPE is more probable to select reduced model than RSS. The RSS variation depends on sample sizes and standard deviation, but the MSPE variation just depends only on standard deviation. However, the level of correlation among the independent variables does not affect both criterions.