About link= options in oddsratio calculation

[JokerWang]

Hi all,

To model response variable which has 3 or more categories, the options link = glogit seems able to list all the odds ratio in each response categories, comparing to the reference category; Or I can process my datasets to several binary response dataset, by recoding the other categories to ".", and model them one by one, left the link options by default. 

So which one is the right method to model the response variable with 3 categories (ordinary), and when should I use the link options? Thanks!

[sbxkoenk · Accepted solution]

Hello,

GLOGIT link means response functions called generalized logits.

This is meant for a nominal response (>2 categories).

When your response variable y is ordinally scaled, you need a cumulative logit model.

You can take the detour via multiple binary models, One-vs-All or One-vs-One, and then voting, but I wouldn't do that. The multiple binary models are also better to replace a multi-class Classification with a multinomial response (not so much an ordinal response).

Good luck,

Koen