Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Problem with -metareg-


From   Roger Harbord <rmharbord@googlemail.com>
To   statalist@hsphsun2.harvard.edu, "Tiago V. Pereira" <tiago.pereira@incor.usp.br>
Subject   Re: st: Problem with -metareg-
Date   Tue, 10 Feb 2009 22:03:04 +0000

Dear Tiago,
Hard to tell what the cause of the convergence problem is as clearly i
can't replicate it without the original dataset. The different
behaviour when you re-input the data may be more likely to be due to
the tiny differences between the original stored values and the result
of re-inputting them to 6 or 7 significant figures than the presence
or abscence of additional variables. If you want to email me the
original dataset (to roger.harbord@bristol.ac.uk) i'll investigate
further. Switching to the -mm- option (method-of-moments estimate of
between study variance) instead of the -reml- option (residual maximum
likelihood) should avoid the numerical problems and give the same
answer in this instance where the estimated between-study variance is
zero.

-- 
Roger Harbord
http://www.epi.bris.ac.uk/staff/rharbord.htm

On Tue, Feb 10, 2009 at 6:41 PM, Tiago V. Pereira
<tiago.pereira@incor.usp.br> wrote:
> Dear Statalisters,
>
> Using my whole dataset (with several others variables) I type:
>
> . metareg logor, wsse(selogor) reml
>
> and obtain:
>
> numerical derivatives are approximate
> nearby values are missing
> numerical derivatives are approximate
> nearby values are missing
> numerical derivatives are approximate
> nearby values are missing
> could not calculate numerical derivatives
> missing values encountered
> r(430);
>
> Then, I do the following: clear, inpute the same data but only the two
> variables and obtain sucessfully the result I am looking for:
>
> clear
>
>
> */---------START-------------
> input logor selogor
> .0800427 .8641729
> .2786815 .1604731
> -1.400893 1.19818
> .537752 .5122404
> end
> metareg logor, wsse(selogor) reml
> */--------END--------------------
>
> Meta-regression                                       Number of obs  =
>  4
> REML estimate of between-study variance               tau2           =
>  0
> % residual variation due to heterogeneity             I-squared_res  =
> 0.00%
> With Knapp-Hartung modification
> ------------------------------------------------------------------------------
>       logor |      Coef.   Std. Err.      t    P>|t|     [95% Conf.
> Interval]
> -------------+----------------------------------------------------------------
>       _cons |    .268642   .1496053     1.80   0.170     -.207469
> .7447529
> ------------------------------------------------------------------------------
>
> Does anyone have a guess about what is going on with -metareg- (the most
> recent version)?
>
>
> Cheers!
>
> Tiago
*
*   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