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

Re: st: 'Meta' and 'Metareg' Coefficients & CIs Differ

From   "Tom Trikalinos" <>
Subject   Re: st: 'Meta' and 'Metareg' Coefficients & CIs Differ
Date   Mon, 23 Jul 2007 20:17:27 -0400

I assumed that you have roger harbord's _metareg_ (>=v2.2.6 (?) -
10Dec2004). There was a previous version by stephen sharp...  I do not
recall any specifics for the older command though.


On 7/23/07, Tom Trikalinos <> wrote:
In the absence of covariates _metareg_ does a random effects
meta-analysis, indeed.
However, with a different method than _meta_ (using the default
settings for _metareg_):

_metareg_ uses REML to get the tau^2 estimate (between study variance)
_meta_ is using the DerSimonian and Laird method

To get the same results use the following syntax:

metareg beta , wsse(SE) mm z

where mm--> uses the method-of-moments estimate for tau^2, aka the
DerSimonian and Laird method
z --> uses the z and not the t disrtibution to get the SE and the CIs


On 7/23/07, Tiffany Davenport <> wrote:
> I am using the 'meta' and 'metareg' commands for meta-analysis in Stata.  As
> I understand it (and as previously posted) the pooled random effects
> estimate obtained by using the 'meta' command should be the same as the
> constant obtained in meta-regression analysis ('metareg') of a model with no
> covariates.  I am finding that the values of these estimates as well as
> their confidence intervals differ slightly.  I have entered the following
> syntax:
> "meta beta SE" - for the meta command and
> "metareg beta, wsse(SE)" for the meta-regression command
> 1. Shouldn't the pooled random effects estimate from 'meta' be the same as
> the coefficient for the constant from 'metareg,' and shouldn't the
> confidence intervals be the same?  If not, I would appreciate any insight
> into why the two commands would generate different estimates with different
> standard errors.  Are they weighted differently?  Are adjustments to the
> syntax of either command necessary to ensure the intercepts and confidence
> intervals match?
> 2. Is it possible to display more decimal places in the 'meta' output?
> Thanks very much for any help.
> *
> *   For searches and help try:
> *
> *
> *

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index