Re: st: Ologit question

[Maarten Buis <maartenlbuis@gmail.com>]

On Wed, Oct 31, 2012 at 4:00 PM, Anat (Manes) Tchetchik  wrote:
> so If the null is rejected for all tests (which is just what I received now:
> . test _b[cut1:_cons]=_b[cut3:_cons], accum
>
>  ( 1)  [cut1]_cons - [cut2]_cons = 0
>  ( 2)  [cut1]_cons - [cut3]_cons = 0
>
>            chi2(  2) =   48.22
>          Prob > chi2 =    0.0000

On Wed, Oct 31, 2012 at 2:23 PM, Anat (Manes) Tchetchik wrote:
> I ran an Ologit and received the following cutoffs:
>
>                 Coef.         Std. Err.                   95% Conf. Interval]
>  /cut1 |  -1.798009   2.900799                 -7.483471    3.887453
>  /cut2 |   .7973023   2.793147                 -4.677166    6.271771
>  /cut3 |   2.902347   2.794626                  -2.57502    8.379713

So how is it possible that the confidence intervals for the cut values
are so wide and clearly overlap, but a test rejects the null
hypothesis that the cut values are equal?

The answer is that the confidence intervals don't use the covariances
in the sampling distribution of these parameters. How could they?
Confidence intervals refer only to their parameter not to any other
parameters. However a test of equality of one or more of these
parameters should take that covariance into account. If you type
-estat vce- after you estimated your -ologit- model you'll see that
the covariances between the cut values are substantial. This is the
reason why -test- and the confidence intervals give so different
conclusions and why you should use -test- to perform such tests and
not look at overlapping confidence intervals.

-- Maarten

---------------------------------
Maarten L. Buis
WZB
Reichpietschufer 50
10785 Berlin
Germany

http://www.maartenbuis.nl
---------------------------------
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/