1
$\begingroup$

I have recently learned about Generalized Ordered Logit regression, which is something in-between ordered logit regression (ordinal regression) and multinomial regression.

It frees us from the proportional odds assumption, but it does account for the dependent variable categories being ordered.

It looks perfect for analysing my data... but.

I only have SPSS, which does not support such analysis.

From what I understood from reading about the actual mathematical workings behind this method, it kind of works as if we did multiple binary logistic regressions for dependent variable collapsed into a series of dichotomous cumulative categories.

But... does it work EXACTLY like this, and could I go around not having proper software by simply re-coding my dependent variable into those dichotomous cumulative categories, and performing the series of binary logistic regressions, and looking at the behaviour of my betas?

Or would I be missing something then?

$\endgroup$
1
$\begingroup$

Close, but not quite. Gologit also constraints the predicted probabilities for all outcomes to add up to one for every observation. Separate logit models cannot do so.

$\endgroup$
  • $\begingroup$ Thank you! This makes sense! I wonder if this means that there is no hope to do this in SPSS... I don't have enough experience with R, and STATA costs money......... $\endgroup$ – Mandarc Mar 22 '18 at 12:00
  • $\begingroup$ I don't know enough about SPSS, it is more than 10 years ago since I last touched it. I vaguely remember that it could do an gologit, but maybe they hid it behind a payed module. $\endgroup$ – Maarten Buis Mar 22 '18 at 13:30
  • $\begingroup$ I don't think you can do gologit in SPSS (as far as I am aware it only supports proportional odds models)... maybe if you use syntax (that I am in no way familiar with). I purchased STATA and hopefully I manage to get a hang of it in a reasonable amount of time... I know there is a program for STATA that can do it and I figured it'd be easier to use it than to learn the SPSS syntax... $\endgroup$ – Mandarc Mar 24 '18 at 12:20
  • $\begingroup$ The program is called 'gologit2', you can install it by typing in Stats: 'ssc install gologit2'. As far as I know there is no menu associated with it, so you can only use it through Stata's syntax ( called do file). $\endgroup$ – Maarten Buis Mar 24 '18 at 18:57
  • $\begingroup$ yes, I already have this program, and I even managed to use it. Now, I have run into another problem - my model does not achieve convergence, and I haven't even introduced interaction terms yet... $\endgroup$ – Mandarc Mar 25 '18 at 13:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.