The rmsb library allows to fit ordinal models for ordered response variables. I would like to know what is the best way to accommodate ordinal predictors in these models. For example, the response variable can be quality of life (ordered from 0 to 100), while the predictors (or independent variables) can be items using a 5-level Likert scale. In this sense I want to assume that the metric scale is not correct for these predictors, and that the distance between the ordered predictor levels is unknown and variable.
The only package of any kind that I know has implemented a really good way of handling ordinal predictors is the brms
package.
1 Like
Thank you Frank for your valuable feedback.
I have read the interesting article on monotonic effects and I think it is a cool method:
As far as I have understood (taking into account the limitations of an oncologist), if I use this approach in the framework of an ordinal regression model (i.e., assuming POs), the exponentiated slope would be equivalent to an odds ratio, while the simplex parameters would only make sense in the log scale (therefore, should no be exponentiated). If so, I have a very nice idea for applying this method.