การศึกษาเปรียบเทียบการวิเคราะห์ความถดถอยเชิงซ้อน และการวิเคราะห์จำแนกกลุ่ม เพื่อใช้ในการแยกกลุ่ม 2 กลุ่ม / วิไลลักษณ์ ชุลีวัฒนกุล = A comparative study between multiple regression and discriminant analysis in classifying 2 groups / Wilailuck Chulewatanakul
In comparative study between multiple regression analysis and discriminant analysis for the purpose of classifying 2 groups, two types of comparison are between both analyses. Firstly, independent variables are inserted into the equation by direct method to both analyses. Secondly, they are inserted into the equation by stepwise regression method to the former analysis and stepwise discriminant of Mahalanobis method to the latter. These methods are calculated by statistical package for the social sciences (SPSS…..) for the data comprising various types of proportion of value 1 and 0 of dichotomous dependent variables. The types of proportion are .50: .50, .55: .45, .60: .40, .65.35, .70:.30, .75:.25, .80: .20, 85: .15, .90:10 and .95:.05 each with the sample sizes of 10, 20, 30, 40 and 60. The results of the study show that, with all types of proportion of dependent variables and all sample sizes, the suitable criteria in separating the predicted values obtained from multiple regression equation into 2 groups is (multiple regression equation)where y [superscript ^ ] [subscript 1i ]and y [superscript ^ ] [subscript 2i ]are the predicted values and n[subscript 1 ] and n[subscript 2 ] are the number of sample sizes when the dependent variables have values of 1 and 0 respectively. By using this criteria, the results of both analyses are similar. Moreover, from the equation obtained from multiple regression analysis and discriminant analysis with different proportions of dependent variables and all sample sizes, it was found that some relationship exist, namely: the order of significance of independent variables are associated, the proportion of coefficients in multiple regression is similar to that in discriminant analysis, and the signs of coefficients obtained from these two methods, whether the signs are different of similar, leads to a similar and corresponding result for both methods. Apart from this, in analyzing by stepwise regression method and stepwise discriminant of Mahalanobis method at the same level of significance, the independent variables selected are the same. Furthermore, in analyzing a large number of dummy variables, both analyses can be used to separate the groups similar results.