st: Reporting pooled p-values smaller than e-40 with the metan command

[Ömer Gök <oemer-necmi.goek@uniklinik-freiburg.de>]

My name is Oemer Goek and I would like to share with you an informative 
message exchange with Dr. Ross Harris on the topic "reporting pooled 
p-values smaller than e-40 with the metan command".

Dr. Harris is one of the programmers of the "metan" command and kindly 
agreed to post our E-mail exchange regarding this topic on the public 
Statalist.


Sincerely yours,
Oemer Goek, M.D., M.P.H.


---------------------------------------------------------------------------------------------
From: Oemer Goek
To: Ross Harris
Sent: Wed, 4 May, 2011 9:41:02
Subject: Re: Fwd: Re: Re: Metan: pooled p-value reporting

Dear Ross,


Now it works! Fantastic!

The internal z-score allows to compute the more exact p-value, 
corresponding to the p-value exported by metan itself. At the same time, 
it allows to calculate a p-value smaller than e-40.

Thanks a lot!

Sincerely yours,
Oemer


On 04.05.2011 10:20, Ross Harris wrote:
> Dear Oemer
>
> Sorry, I just read the chain further down and realised you've already 
tried using the returned values (I replied in a bit of a hurry 
yesterday). One other option would be to used the z-score from the 
return list, r(z), (which should be fairly accurate as a float, rather 
than the p-value as a float) and obtain the p-value from this in a more 
precise form (i.e., double).
>
> Sorry I can't help with this further...
>
> Best wishes,
> Ross
>
> From: Oemer Goek
> To: Ross Harris
> Sent: Tue, 3 May, 2011 15:30:44
> Subject: Re: Fwd: Re: Re: Metan: pooled p-value reporting
>
> Dear Ross,
>
>
> Thank you very much. We will then use the generate p_z to obtain the 
overall-p-value, which hopefully will report it without rounding to 0, 
even if the overall p-value is smaller than e-40.
>
>
> Kind regards,
> Oemer
>
> On 03.05.2011 11:44, Ross Harris wrote:
>> Dear Oemer Goek
>> Apologies for the delay in replyling; I have just returned from holiday.
>>
>> I'm glad that you're finding the program useful. It would be nice to 
have more control over the reporting of p-values, which is something I 
am considering if I ever get round to an update! However, there is a 
simple enough way round it- metan gives a few "returned" values that are 
available after running the command, including the p-value for the 
overall effect- try typing "return list" after- the overall p-value is 
the one called r(p_z), and may be accessed by doing things like:
>>
>> display r(p_z)
>> local pval = r(p_z)
>>
>> If you're using the "by" option, unfortunately you can't access 
individual subgroup p-vals- to get these, you would have to analyse each 
subgroup in turn and then extract the p-value as in the above.
>>
>> Hope this helps!
>>
>> Best wishes,
>> Ross
>>
>> From: Oemer Goek
>> To:  Ross Harris
>> Sent: Fri, 29 April, 2011 11:55:31
>> Subject: Fwd: Re: Re: Metan: pooled p-value reporting
>>
>> Dear Dr. Harris,
>>
>>
>> I am writing you as a Stata user, who is using your excellent 
"metan" command (version 9) for meta-analysis in Stata.
>>
>> Our problem is that we would like to export the pooled p-value using 
metan into excel, but all exported p-values of e-40 or smaller are 
rounded to zero.
>>
>> If we use the reported z-score from the log file of the metan 
command, the p-values are different from the ones, which are directly 
exported from metan into excel, which is also due to rounding phenomena: 
this time of the reported z-score in the output, which is different from 
the longer z-score used for calculation of the exported pooled p-value, 
as the Technical Support from Stata Corporation kindly explained.
>>
>> They did not have a solution to the rounding problem, which is why 
they recommended me to contact you.
>>
>> All we really need, would be to export the exact pooled p-value from 
metan into excel or any other form, without any rounding of the p-value. 
It would be great, if you could provide some advice, please.
>>
>>
>> Kind regards,
>> Oemer Goek
*
*   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/