Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: probit in Stata and in Some Alternative Software


From   Maarten buis <maartenbuis@yahoo.co.uk>
To   stata list <statalist@hsphsun2.harvard.edu>
Subject   st: probit in Stata and in Some Alternative Software
Date   Tue, 16 Sep 2008 22:37:39 +0100 (BST)

--- Andy Call wrote to me privately:
> I found your info at the following link: 
> 
> http://www.stata.com/statalist/archive/2006-05/msg00925.html

> If you have a moment, I have a question for you.

It is better to ask such question on the list and not privately for
reasons discussed in the Statalist FAQ, point 5:
http://www.stata.com/support/faqs/res/statalist.html

> I see from the above that I am not the only one scratching my head
> as to why I can run the same probit model in Stata and in SAS and
> get different results.
> 
> You answered above that one should draw the same conclusions from 
> either the Stata or SAS output.  Yet I do not see how this could be,
> given that I have a variable that is highly significant when I run
> the model with one package, and only marginally significant when I
> run the model with another package.
>
> In general, is Stata or SAS preferred (more accurate) when running
> probit models? 

This particular answer involved the ordered probit model and not the
probit model. The ordered probit model has a number of equivalent
parameterizations, which could lead to parameter estimates that look
very different, including the level of significance, but if interpreted
properly mean exactly the same thing. I am not aware of a similar
difficulty with the probit model.

Comparing 
http://www.ats.ucla.edu/stat/sas/output/SAS_probit.htm
with 
http://www.ats.ucla.edu/stat/stata/output/Stata_probit.htm

seems to result in very similar results. The one thing that is
different is the test statistic (Called Wald in SAS and z in Stata):
SAS reports (b/se)^2 and compares with a Chi square distribution with
one degree of freedom, while Stata reports b/se and compares it with
the standard normal distribution. This looks very different, but a
standard normal squared is a Chi-squared with one degree of freedom, so
these two are equivalent (as can be seen by the very similar p-values)

I know Stata, and I completely trust the numerical properties of
Stata's -probit-. I have no experience with SAS, but what I have heard
is that SAS is, like Stata, a very high quality packages. Moreover,
probit regression is a well established statistical procedures. So, I
have no reason to believe that the reason for your problem is due to
differences in the quality of the software. If you find differences
than that in all likelihood means that you are either not doing exactly
the same thing in both packages or you are misinterpreting the output
in one of the two packages. Following the links I gave above might help
you track down the problem.

Hope this helps,
Maarten

-----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands

visiting address:
Buitenveldertselaan 3 (Metropolitan), room N515

+31 20 5986715

http://home.fsw.vu.nl/m.buis/
-----------------------------------------


      
*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index