การเปรียบเทียบการแก้ปัญหาข้อมูลผิดปกติ ในการวิเคราะห์ข้อมูลที่ได้จากงานวางแผนการทดลอง โดยใช้อำนาจการทดสอบ / สุรศักดิ์ จินตรัตน์ = A comparison on solving data outlier problem for data analysis obtained from experimental design / Surasak Chintarat
The purpose of this study is to care method of solving outlier data problem obtained from experimental design by rejecting outlier, accepting outlier, correcting outlier by estimation and analyzing data with nonparametric test by using the power of test. The comparisons were made for population with approaching normal distribution and violating from normal distribution. The design of this study were the comparison with the difference between two populations, completely randomized design and randomized complete block design. The data were obtained by simulation technique and computer program which were repeat 1,000 times for each experiment situation. The result of this study can be summarized as follow: 1. For comparison with the difference between two populations and completely randomized design, all methods used for solving outlier problem are capable of controlling the probability of type I error except for randomized complete block design. 2 For comparison with the difference between two populations, parametric test method which estimating outlier by neighbouring value of outlier has the highest power at .01 significance level and accepting outlier is good at only .05 and .10 significance level in case of approaching normal distribution. When the normal distribution is violated, the nonparametric test has the highest power. 3. For completely randomized design parametric test which estimating outlier by neighbouring value of outlier has the highest power at all significance levels when population are nearly normal distribution. When population is not normal distribution, the nonparametric test has the highest power at .05 and .01 significance level. The estimation outlier by neighbouring value of the outlier has the highest power at .01 significance level. 4. For randomized complete block design, the power of test could not be concluded since the probability of type I error for all methods were uncontrollable in all experiment situations.
