Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Ordered logit estimation and score test

From   "maartenbuis" <[email protected]>
To   [email protected]
Subject   Re: st: Ordered logit estimation and score test
Date   Mon, 17 Jan 2005 10:59:53 -0000

--- "Daniel Schneider" <daniel.schneider@s...> wrote:
> I ran a score test (with the omodel routine) and the result was a 
> bit discouraging as the  Prob > chi2 constantly stayed below .05 
> whatever I did with the modell.
> After I tried several other things like gologit (which is also a bit
> troublesome because of some other assumptions and the fact that in 
> my view it is even harder to make interpretations), creating an 
> index for the same DV with some other variables and then running 
> OLS (which didn't work because the residuals still had some 
> problems, like non-normality and heteroscedasticity).
> So, I am left wondering with what else to do and especially the 
> question how seriously I have to take the score test. I have a 
> sample size with n=1.500, and I was wondering if that might affect 
> the p-level of the test.

-omodel- test the proportional odds assumption that underlies the 
orderd logit model. This assumption is often violated. You could use 
the -brant- command (part of spost) after -ologit- to see which 
variable(s) cause problems and enforce the proportional odds 
assumption for the uproblematic variables in gologit by applying 
constraints to get a more parsimonious (interpretable) model. 

Alternatively, you could estimate a multinomial logit (which relaxes 
the proportional odds assumption for all variables).

Hope this helps,

*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index