Re: Degrees of freedom in multinomial regression

Posted by Rich Ulrich on Apr 13, 2014; 8:17pm
URL: http://spssx-discussion.165.s1.nabble.com/Degrees-of-freedom-in-multinomial-regression-tp5725413p5725416.html

Having 4 questions (categories 1-5) give you 16 degrees of freedom for predicting,
when you treat them as categories.

Having 5 categories of Outcome gives you 4 equations - each with 16 d.f.

There's your 64 d.f.  for the likelihood test.

The example that I Googled up doesn't show the Pearson/Deviance tests,
but maybe this is enough clue.

--
Rich Ulrich

> Date: Sun, 13 Apr 2014 09:35:49 -0700

> From: [hidden email]
> Subject: Degrees of freedom in multinomial regression
> To: [hidden email]
>
> Hi,
>
> I am running multinomial regression on:
>
> 4 categorical predictors (responses to 4 questions) - (each of the 4
> questions can be answered by one of 5 categories)
> 1 outcome/dependent variable, which is a rating score (with 5 categories)
>
>
> To see if respondents responses to each of the 4 questions can be used to
> predict the rating score.
>
> There are no covariates.
>
> I am struggling to understand how the df in the likelihood ration tests are
> calculated - this = 64, and then in the Pearson and Deviance tests = 32.
>
> Can anyone assist, or advise where I could find a clear explanation?
>
> Many thanks
>
> VB
>