Statalist


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

Re: st: How does Stata perform with exact tests?


From   "Ben Jann" <ben.jann@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: How does Stata perform with exact tests?
Date   Tue, 9 Oct 2007 10:26:48 +0200

You probably have to give more details. Which statistics are you
talking about? Which algorithms did you implement? Maybe also show us
some critical parts of the code or make your programs available for
download. My own experience with programming exact tests in Stata (see
-ssc describe mgof-) is that results are quite accurate, but this may
also depend on the details of the implementation.
ben

BTW: If, by any chance, you implemented a fast algorithm for
multinomial goodness-of-fit, I'd be very interested...

On 10/8/07, tiago.pereira@incor.usp.br <tiago.pereira@incor.usp.br> wrote:
> Dear all,
>
> I have implemented a number of exact tests to learn specific issues in
> Stata programming that cannot be easily found in the implementation of
> asymptotic tests.
>
> Then, I have compared the results of my impplementations with those
> outputed by Statxact as well as softwares usded in Genetic studies,
> clearly well validated and used during years in the field of Population
> Genetics.
>
> All tests I have implemented do work excellent with few observations, say,
> n < 4 (note that in practice the number of observations will be less than
> 10).
>
> However, the more the number of observations I have, the less precise are
> my exact P values, I mean, larger is the difference between the P value
> generated by my implementation and those outputed by validated programs
> when n increases.
>
> Surprinsingly, when n <=4, my outputs are identical considering a
> precision of 5 decimals. (P = 0.07656 in Statxact and P = 0.07655958 in my
> Stata implementation, for example).
>
> However, when I increase the number of observations (n>4) my P-values tend
> to be smaller than those reported by other softwares (say, for instance,
> 0.05384 in Statxact and 0.05094364 in my Stata implementation).
>
> Thence, I would like to know your experience to have a better picture of
> how well stata performs with exact tests and how robust is the Stata
> programming with exact methods, i.e., summation of several probabilities
> from large factorials, and, etc...
>
> I will be grateful for any comment.
>
> Best regards,
>
> Tiago
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   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