A Comparison of Classification Performances between the Methods of Logistics Regression and CHAID Analysis in accordance with Sample Size

Abstract

The aim of the study is to analyze how classification performances change in accordance with sample size in logistic regression and CHAID analyses. The dataset used in this study was obtained by means of “Attentional Control Scale.” The scale was applied to 1824 students and the analyses were done by randomly choosing the samples from the dataset. Nine classification criteria were determined in order to evaluate classification performances of logistic regression and CHAID analyses, and the results were interpreted in consideration of these criteria. As a result of the analyses, it was found that classification performance in logistic regression showed no change as sample size increased, and performed a better classification in small sample size (N= between 25 and 900) than CHAID analysis. On the other hand, in the method of CHAID analysis it was seen that classification performance improved as sample size increased, and provided stronger findings in large sample size (N= 1000 and above). Moreover, in classification studies logistic regression analysis yielded more reliable results, and CHAID analysis provided stronger classifications. The results of this study are considered to suggest researchers to select the methods in classification studies based on sample size.

Keywords

• logistic regression
• chaid analysis
• classification
• sample size