Date: Mon, 8 Aug 2005 15:21:04 -0400
Reply-To: Marc Scoppettone <mscoppettone@IAGR.NET>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Marc Scoppettone <mscoppettone@IAGR.NET>
Subject: Re: multinomial vs. ordinal
Ordinal regression is a special case of multinomial logistic regression.
Both require a dependent variable that is categorical with more than two
categories. However, the ordinal model also requires that the categories
be ordinal. The ordinal regression is more commonly referred to as the
proportional odds model.
For example, suppose you have five categories. The multinomial logisitc
model will have a base outcome (say, outcome 5), and then estimate the
odds ratio of any other outcome relative to outcome 5. It makes not
assumptions about the relation between, say outcome 1 and outcome 2 except
the restriction that all of the probabilities add up to 1.
However, the proportional odds model will estimate the odds of being in,
say, outcome x or better versus the outcome of being in outcome y or
worse. This goes for all x > y (in terms of ordinality).
The proportional odds model assumes the same affect of an explanatory
factor for all levels. However, the general multinomial logistic model
will give one parameter estimate for each explanatory factor at each of (n-
1) levels of the dependent variable.
Alan Agresti's "Introduction to Categorical Data Analysis" has a nice
simple discussion (much much clearer than mine!!!)
On Mon, 8 Aug 2005 12:12:18 -0700, Winny Chi <email@example.com> wrote:
>Would anyone here kindly advise me the difference between multinomial and
>ordinal regression analysis in terms of what type of DV and IV fit? It
>to me that they both deal with categorical DV that has more than two
>possible values. Any comments?