How to specify prediction limits with the orm() function in R for a bounded outcome measure


In my research field (physical therapy/rehabilitation/sports science), there are many bounded outcome measures ranging from 0 to 1 that are the focus of predictions for making decisions for our patients.

I understand that beta regression can handle such data by restricting the predictions to the bounded range. However, I would like to know if it is possible to implement an ordinary regression model to predict these outcome measures within the orm() function, while ensuring that the predictions, including confidence intervals, remain within the bounded limits.

Thank you in advance,

Rubén Fernández Matías

Ordinal regression using cumulative probability models such as what orm implements do this automatically. And classic prediction intervals, i.e., intervals that expose how difficult it is to predict an individual Y value, can be easily implemented by using the ordinal model to predict e.g. the 0.025 and 0.975 quantiles of Y | X. This is an after-the-fit estimation process as described here.

1 Like

Thank you Frank, I thought it was necessary to specify this aspect in some special way.