Statalist The Stata Listserver


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

st: RE: Expanded significance values in STATA output?


From   "Nick Cox" <n.j.cox@durham.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: RE: Expanded significance values in STATA output?
Date   Wed, 19 Apr 2006 22:10:52 +0100

I disagree somewhat with your premiss, especially 
in the case of correlation output. 

First, once P is very small you learn more from thinking about 
the magnitude of the correlation or looking at
the associated scatter plot than from trying to 
get a really precise grasp on the P-value. 

Second, are you sure that your variables are pairwise
bivariate normal? and observations mutually independent
(in time and/or space)? The further
out into the tails of the sampling distribution you 
go, the more these assumptions matter. 

Third, P-values for correlations aren't uniquely defined. 
Stata uses a calculation based on t-distributions, but 
many would argue that correlations should be examined 
using Fisher's z transformation. 

Fourth, wanting to use P-values in this way treats them 
as measures of the strength of the relationship. Arguably,
that's not crazy, but it's what the correlation itself
is supposed to do, more directly. 

For these reasons, wanting more decimal places here is 
arguably like trying to look at a misprint with a 
magnifying glass. 

The case of logit is not so extreme, as I understand it. 

That aside, you are correct that Stata does not always 
offer options for giving more decimal places for 
ancillary output. The basic choice is then between

1. retrieving what is left behind in memory and displaying
it with your own format 

and 

2. re-doing the calculation yourself. 

I don't know what you searched, but Statalist over 
the last few days has featured a thread on 
getting the output of -pwcorr- in different ways. In 
essence, you have a choice between extending Bill Gould's
Mata program to show P-values, based on a method of your
chousing, and using a user-written program such 
as Adrian Mander's -matpwcorr- from SSC. 

In the case of -logit-, a subsequent -eret li- shows that the basic 
matrices are saved in e(b) and e(V), which you need to retrieve
so that you can do your own calculations. What to do 
is explained in far more detail in the manuals than 
anywhere on the web. 

Nick 
n.j.cox@durham.ac.uk 

G Elliott Wimmer
 
> is it possible to get more detailed
> significance values (i.e. more digits beyond 0) in STATA output? 
> Currently, the logit and pwcorr (with 'sig') commands that I 
> am using only
> give '0.000' for strongly significant variables.  However, for coming
> publications it would be much more useful to see beyond this limit to
> determine if something is significant at, e.g. p<.0004 or 
> p<.00000001.  A
> reasonably extensive web search from various sites doesn't yield any
> answers.

*
*   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