I'm running ordered probit regression models with polr in R. This question states that for prediction, polr returns the category with the highest probability for given $x$. I would have expected that the predicted category is equal to the number of estimated thresholds ($\hat{\zeta}$) exceeded by $x' \hat{\beta}$ (what imo would be in line with the model assumptions). Is there a reason, why the polr-approach is "better" then the thresholds-approach?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.