# Re: st: strange test statistic with ologit

 From jpitblado@stata.com (Jeff Pitblado, StataCorp LP) To statalist@hsphsun2.harvard.edu Subject Re: st: strange test statistic with ologit Date Tue, 11 Jul 2006 11:30:56 -0500

```Stephan Rudolfer <stephan.rudolfer@manchester.ac.uk> reported a problem with
the estimation table header from -ologit-.

> Consider the following output:
>
> . * fit PO model +/- ulnar sensories
> . ologit drdiag *N*
>
> ...
>
> Ordered logistic regression                       Number of obs   =        958
>                                                   LR chi2(28)     =    1275.68
>                                                   Prob > chi2     =     0.0000
> Log likelihood = -481.66917                       Pseudo R2       =     0.5698
>
> ------------------------------------------------------------------------------
>       drdiag |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
> -------------+----------------------------------------------------------------
> ...
> -------------+----------------------------------------------------------------
>        /cut1 |  -17.09747   4.765666                       -26.438   -7.756936
>        /cut2 |  -12.30247   4.741235                     -21.59512   -3.009824
>        /cut3 |  -7.826134   4.744828                     -17.12583    1.473559
> ------------------------------------------------------------------------------
> Note: 8 observations completely determined.  Standard errors questionable.
>
> . predict p1-p4,p
>
> . drop *N_us* p1-p4
>
> . ologit drdiag *N*
>
> ...
>
> Ordered logistic regression                       Number of obs   =        958
>                                                   F(  22,      .) =          .
>                                                   Prob > F        =          .
>
> ------------------------------------------------------------------------------
>       drdiag |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
> -------------+----------------------------------------------------------------
> ...
> -------------+----------------------------------------------------------------
>        /cut1 |  -16.51915   4.513598                     -25.36564   -7.672665
>        /cut2 |  -12.14342   4.486871                     -20.93752   -3.349314
>        /cut3 |  -7.916497   4.498279                     -16.73296    .8999674
> ------------------------------------------------------------------------------
> Note: 8 observations completely determined.  Standard errors questionable.
>
> I am astonished that Stata has given the F-statistic instead of the
> chi-squared one in the second model fitted.  This seems to have occurred
> just because I deleted the last six variables of the first model.  Having
> asked two Stata experts, who told me that they had never seen anything like
> this, I am hoping that someone on statalist will be able to clarify the
> situation for me.  Many thanks in advance.

If Stephan will send me a private email with a dataset that will reproduce
the above, I will investigate.  I would also appreciate a log containing the
output from the following commands:

***** BEGIN:
update query
which ologit
which _coef_table
***** END:

--Jeff