When analyzing ordinal data, particularly when the proportional odds assumption of the ordered logit model is not met, researchers often face challenges. The proportional odds assumption implies equal odds ratios across all outcome levels, which real-world data often does not satisfy. In such cases, some practitioners persist with the ordered logit model, while others turn to generalized linear models that may not be the most suitable choice. This paper introduces a novel methodology for calibrating category breakpoints, effectively removing the need for the equal distance assumption. By introducing M-1 breakpoints, the model gains additional degrees of freedom, offering a flexible solution for cases where the proportional odds assumption is untenable.
Open-access article — free to read and share.
Continental Scholarly Publications applies rigorous double-blind peer review to every submission. Our expert editorial board ensures your work meets the highest standards of scholarship before reaching an international readership.