Diagnostics in Multinomial Logistic Regression

Dmartin427

Hey EJ - long time JASP super-fan, first time poster. I had a quick question: I can't seem to get any of the diagnostics to populate under Generalized Linear Model if I run it as a Multinomial Logistic Regression. However, the diagnostics do populate if I run the model as another family. Is this a limitation, or am I just missing something?

Thank you!

-Dan

patc3

I thought JASP didn't have multinomial logistic regression? Am I misremembering?

Dmartin427

It does indeed! If you go into Generalized Linear Model under Regression, and then set the family to Other, you can pick between Multinomial, Ordinal and Firth Logistic Regression.

The output for Mutlinomial in JASP is much cleaner/less of a complete mess to read when compared to R, but as it stands right now I have to use both programs to get everything that I want from the analysis. I'm hoping that there's something that I can do to fix the diagnostics problem, unless this is something that universally doesn't for for anyone at the moment.

patc3

Thanks I remember now. I used to know how to get to multinomial regression, but it's so weirdly located that I couldn't find it anymore.

Welp you're definitely right the UI is completely unresponsible to changes in the diagnostics section. It might be that they are not implemented for multinomial logistic regression in JASP (rather than a bug that prevents the diagnostics from showing in the output), but that part of the UI probably shouldn't be shown if that's the case. I'll make a ticket on github.

BTW if you're also using R with the nnet:multinom() function, take a look at broom::tidy(your_model) instead of summary(your_model) to look at the output, and also car::Anova(your_model) to get omnibus tests for categorical predictors.

patc3

See GitHub issue here: https://github.com/jasp-stats/jasp-issues/issues/2926