Comparing Binary and Ordered Logistic Regression in Studies of Education and Learner Behavior

Abstract:

This study compares the effectiveness of binary logistic regression and ordered logistic regression in analyzing office workers’ readiness to pursue postgraduate education. A simulated survey dataset generated by ChatGPT includes 400 observations encompassing motivational, barrier, and support factors. Two models were applied: binary logistic regression (combining levels 4–5 as "ready" and levels 1–3 as "not ready") and ordered logistic regression (retaining all five ordinal levels). The ordered model identified a significant negative effect of family- related barriers on readiness (β = -0.468, OR = 0.63, p = 0.021), which was not detected in the binary model. Both models showed low overall fit (Pseudo R2 < 3%). The findings suggest that ordered logistic regression better utilizes ordinal information, while binary regression offers a simpler, more intuitive approach. The study recommends field surveys to validate these results.