Talk:Ordered logit

merge with ordered probit?
I propose a merge with ordered probit. Neither article has any information, but there is certainly no reason to consider them separately. Pdbailey 15:43, 13 June 2007 (UTC)

Does anyone have an idea for where they could merge to?

A bit of a late response, but how about ordinal regression? Or alternatively one could combine with multinomial logit (ordered logit is a special case of multinomial logit) to create multinomial regression? -3mta3 (talk) 00:32, 8 April 2008 (UTC)

Clarity of Text
I wish someone would more thoroughly explain the proportional odds assumption here, or at its own page. Briancady413 (talk) 23:44, 14 October 2009 (UTC)


 * I can see why, the existing text was so awful as to be counterproductive. PDBailey (talk) 02:47, 15 October 2009 (UTC)

Wondering about the $$\mu$$ variables, are those to be inferred? — Preceding unsigned comment added by 94.191.159.1 (talk) 21:38, 3 March 2012 (UTC)


 * I have now replaced the phenomenally vague account of the proportional odds assumption that was given here. Michael Hardy (talk) 17:55, 11 April 2014 (UTC)

Once again, we have an article in which the author presents an equation (and inequalities) without defining all the terms. This bothers me no end; equations are meaningless if their terms are not defined. Can whoever wrote this please define the $$ \mu_i, i \in {1, 2,...,N} $$ terms? Further, the $$ \epsilon $$ term in the model assumption is undefined (statisticians will assume it's the error term, but you must define it), and its distribution unstated. This makes the proposed model useless to the reader. If those two items are fixed it will help a lot. Thank you Chafe66 (talk) 17:15, 6 May 2016 (UTC)


 * I clarified the distribution of the error term. Anton Nils (talk) 14:05, 16 October 2022 (UTC)

"accuracy"
I've put a "factual accuracy" tag in place because I don't have time to do any editing at the moment. I think the technical details are in some places confused. I'll be back. Michael Hardy (talk) 21:34, 15 April 2014 (UTC)


 * it's been more than a year and you still haven't described any specific problem with the article. Q VVERTYVS (hm?) 06:16, 12 August 2015 (UTC)


 * I was wondering if if I'm misunderstanding something regarding your explanation of the proportional odds assumption and wanted to ask before editing. If the logarithms form an arithmetic sequence, this would mean that the difference between the intercepts in the model is equal. However, the proportional odds assumption means that the coefficients are the same across pairs of categories. In other words, the change in the log-odds that corresponds to a change in the linear combination of the parameters is the same for every pair that is listed. I also could not find any reference to the idea of an arithmetic sequence of the log-odds in the source you cited. There are sources that took it from here, though. If the probabilities ( $$ p_1, p_2, ... p_5 $$ ) here refer to those predicted by the model, than the log-odds definitely do not form an arithmetic sequence. Ofyalcin (talk) 20:16, 24 April 2022 (UTC)
 * I agree that the explanation of the proportional odds assumption was faulty and I have revised it. The logarithms do not need to form an arithmetic sequence; instead, the crucial part is that the differences are the same regardless of x. Anton Nils (talk) 14:10, 16 October 2022 (UTC)
 * Great, thanks! Ofyalcin (talk) 20:20, 22 October 2022 (UTC)

Proportional odds model
The proportional odds model is only one form of ordinal logistic regression; there is also, e.g., the partial proportional odds model, see eg. Peterson & Harrell. I will try to come back and do more with this. PeterLFlomPhD (talk) 23:22, 23 July 2015 (UTC)