Ordinal regression models: Problems, solutions, and problems with the solutions

摘要: Ordered logit/probit models are among the most popular ordinal regression techniques. However, these often have serious problems. The proportional odds/parallel lines assumptions made by methods violated. Further, because of way identified, they many same limitations as encountered when analyzing standardized coefficients in OLS regression, e.g., interaction terms and crosspopulation comparisons effects can be highly misleading. This paper shows how generalized ordered (estimated via gologit2) heterogeneous choice/location scale oglm) address concerns ways that more parsimonious easier to interpret than is case with other suggested alternatives. At time, cautions sometimes raise their own researchers need aware know deal with. First, misspecified create worse problems ones were designed solve. Second, estimates implausible, suggesting data being spread too thin and/or yet another method needed. Third, multiple very different interpretations results possible plausible. I will present guidelines for identifying dealing each