Are differences in the posterior linear predictor meaningful/useful?

I have fitted a cumulative ordinal model (logit) on responses from a 5-point Likert scale.

As means, and thus mean differences, aren’t particularly meaningful or appropriate for ordinal responses. I’m wondering if differences in the latent scale are meaningful or appropriate.

Am I correct that the values returned from posterior_linpred are the untransformed values of the combined predictors? Are between-group comparisons in these values at all similar to mean comparisons? For example, on a Disagree to Agree scale, would positive differences between groups indicate “more agreement”?

1 Like

Yes, you can interprete the latent scale like this and posterior_linpred indeed gives you the latent scale. For more details, you can also take a look at

Thanks Paul!