358 – SPEED: Statistics in Epidemiology
Cumulative vs. Adjacent-Category Logits: Readiness to Quit Smoking Among Cancer Survivors
Zoran Bursac
University of Tennessee Health Science Center
Patricia J. Goedecke
University of Tennessee Health Science Center
Melissa A. Little
University of Virginia
Robert C. Klesges
University of Virginia
When modeling polytomous outcomes with more than two ordered response levels we can apply proportional odds or cumulative logit models, assuming a common set of slopes across the response functions. The resulting odds compare the directionality of higher order response levels to the lower ones. Depending on the data and the purpose of the analysis one might want to compare two neighboring response levels; this is possible with the application of the adjacent-category logit model, which shares some similarities with the cumulative logit. Furthermore, adjacent-category logit allows us to relax the common slopes assumption while maintaining model validity, with predicted probabilities within the [0,1] interval. We illustrate the comparison of the two approaches while modeling a three-level outcome variable; readiness to quit smoking within the next month, one to six months, or longer than six months.