In multinomial though you may want a global test whether a covariate has any effect over all of the dependent variable categories. For that you can either do an F test (with and without that variable), or adapt the Wald test to multivariate data.

I will need to do another blog post! See here for a good example in Stata, https://stats.idre.ucla.edu/stata/dae/multinomiallogistic-regression/, I will need to see if you can easily do those tests in SPSS and R though.

]]>Given the latent variable interpretation of ordinal regression I don’t think you *have* to plot the odds ratio (I think odds ratios only make sense for the logit link function). If you plotted the original linear effects they can of course be on a linear scale, and that is applicable to any link function.

]]>