การคัดเลือกตัวแบบไม่ติดกลุ่มอย่างสมบูรณ์ของตัวแบบการถดถอยโลจิสติกแบบ 2 ประเภท โดยใช้ฟังก์ชันโพรบิตเป็นฟังก์ชันเชื่อมโยง / จารุตา ฤทธิ์เดชะ = The strictly non-nested model of binary logistic regression model using probit function as a link function / Jaruta Ridacha
The purpose of this study is to select the strictly non-nested model of binary logistic regression model using probit function as a link function. The factors affecting selected model are ordered pairs of number independent variables in the first model (p1) and the second model (p2); (p1,p2); (5,2),(5,3),(5,4),(5,5),(2,5), (3,5),(4,5),(4,2),(4,3),(4,4),(2,4), (3,4),(3,2),(3,3),(2,3), (2,2), the degree of pair-wise correlation indepedent variables of the level low medium and high and the sample size (n); 50, 100, 150 and 250 the data in all situations are generated using Monte Carlo technique through R-program. The selection criterion is the maximum of the area under ROC curve. The results can be summarized as follow: Factors involved in selecting the strictly non-nested model of binary logistic regression model using probit function as a link function. Area under the ROC curve of the average change for each model. As ordered pairs of number independent variables in the first model (p1) and the second model (p2); (p1,p2) changed but the other factors are kept constant, with the number independent variables in the first model (p1) is greater than the second model (p2) the model chosen is the first model (p1) and the number independent variables in the second model (p2) is greater than first model (p1) the model chosen is the first model (p2). As teh degree of pair wise correlation independent variables of the values changed but the other factors are kept contant, that is, if the degree of pair-wise correlation is at low level, the reliability of the model increases. But if degree of pair-wise correlation is at medium or higher, the reliability of the model decreases. As the sample size changed but the other factors are kept contant, that is, if the size of sample increases, the reliability of the model increases.