RE: st: Brant test interpretation with categorical variables

[Massimiliano Volpi <maxvolpi@hotmail.com>]

Thanks for your quick reply, Maarten, 

however, what I meant was something different:

the results I have are something like the ones below (I had  to alter the actual numbers, which are purely an example, because of restrictions imposed by clearance, due to the data I am using).


brant, detail

Estimated coefficients from j-1 binary regressions

                               y>0         y>1
                [...]
                regulation1   .4002      -.3756
                regulation2   .7453       .5645
                regulation3   .5786       .8256
                [...]     

Brant Test of Parallel Regression Assumption

    Variable |      chi2   p>chi2    df
-------------+--------------------------
         All |     
-------------+--------------------------

[.....]
regulation1  |      4.79    0.029     1
regulation2  |      0.30    0.587     1
regulation3  |      0.12    0.733     1
 [....] 
----------------------------------------

A significant test statistic provides evidence that the parallel
regression assumption has been violated.

The regulation variable compares the importance companies assign to a 
specific source of information for innovation when innovation is 
motivated by the need to meet regulatory requirements. Regulatory 
requirements can be classified as  high, medium, low or not applicable.In my model regulation1 compares the low category to "not applicable",
 regulation2 compares medium to not applicable and so forth. 



What I wonder is whether I should rely on the t-test for the low vs not applicable (regulation1) or rather (as I think) on an LR test to test simultaneously all categories (i.e. low vs non-applicable, medium vs non applicable, high vs non applicable, or 1 to 3) given that "regulation" takes on separate values, but this is just the result of it being categorical rather than a continuos variable.

Best Wishes

Max Volpi


----------------------------------------
> Date: Fri, 18 Mar 2011 11:46:47 +0000
> From: maartenbuis@yahoo.co.uk
> Subject: Re: st: Brant test interpretation with categorical variables
> To: statalist@hsphsun2.harvard.edu
>
> --- On Fri, 18/3/11, Massimiliano Volpi wrote:
> > I would like to ask for your help in the interpretation
> > of the results from Brant test (SPost, Scott Long).
> > I am using a number of categorical variables that I expand
> > into dummy variables with the command xi. I have tested my
> > model with Brant and it typically turns out that only one of
> > the dummies within each categorical variable (which in my
> > case represent the importance of motivations: irrelevant, of
> > little importance, important, etc) violates the assumption
> > of parallel lines.
> > So, I would like to ask whether you would suggest to do a
> > LR test to check whether all the dummies within the same
> > categorical variable simultaneously violate the assumption
> > or whether just one "sub-category" (a single dummy value) is
> > enough to consider the hypothesis to be violated (of course
> > the question only makes sense if the LR test does not reject
> > the parallel line assumption for all the dummies in the same
> > category).
>
> The likelihood ratio test and Brant test are asymptotically
> equivalent, so there is little added value of going that
> route. You can use -glogit2- to accomadate your one non-
> paralel variable (type in Stata: -ssc install glogit2-).
>
> Hope this helps,
> Maarten
>
> --------------------------
> Maarten L. Buis
> Institut fuer Soziologie
> Universitaet Tuebingen
> Wilhelmstrasse 36
> 72074 Tuebingen
> Germany
>
> http://www.maartenbuis.nl
> --------------------------
>
>
>
>
> *
> * For searches and help try:
> * http://www.stata.com/help.cgi?search
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
 		 	   		  
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/