การเปรียบเทียบอำนาจการทดสอบของการทดสอบเทียบความกลมกลืนสำหรับตัวแบบการถดถอย / ทิพย์วัลย์ กันทอง = A comparison on the power of goodness of fit test for regression models / Thipwan Kunthong
The purpose of this research is to compare test statistic on goodness of fit test for regression models. The test statistics are F statistic (F). Kolmogorov-Smirnov statistic (KS) and Cramer-von Mises statistic (CvM). The analysis was performed in case of independent variables having replications and having no replications. The distributions of random error are normal and lognormal distributions. The significant levels are 0.01, 0.05, and 0.10; sample sizes are 10, 15, 20, 25, 30, 50 and 70; regression coefficients are 1, 3, and 5. The linear model with one and two independent variables, polynomial model of degrees 2 and regression model with 2 independent variables having interaction are the four regression models considered in this study. The data of this research was obtained by Monte Carlo Simulation Technique and the program was designed to calculate the probability of type I error power of tests through 1,000 times of simulation for each specified situation. The results of this research are as follows: In case of independent variables having replications. The F test statistic has the highest power for almost specified situations when the distributions of random error are normal and lognormal distributions. In case of independent variables having no replications. The KS test statistic has the highest power in the case that the distribution of error is normal and for all sample sizes. If the distribution of error is lognormal, the CvM test statistic has the highest power when sample size is less than 20 and the KS test statistic has the highest power when sample size is greater than 20.
