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

st: metareg

From   Roger Harbord <>
Subject   st: metareg
Date   Tue, 03 Jul 2007 13:46:27 +0100

X-Mailer: Mulberry/2.2.1 (Win32)
MIME-Version: 1.0
Content-Type: text/plain; charset=3Diso-8859-1; format=3Dflowed
Content-Disposition: inline
X-Spam-Score: -0.4
X-Spam-Level: /
X-Barracuda-Bayes: INNOCENT GLOBAL 0.0000 1.0000 -2.0210
X-Barracuda-Virus-Scanned: by HSPH Email Quarantine System at hsph.harvard.=
X-Barracuda-Spam-Score: -2.02
X-Barracuda-Spam-Status: No, SCORE=3D-2.02 using global scores of TAG_LEVEL=
=3D1000.0 QUARANTINE_LEVEL=3D3.5 KILL_LEVEL=3D7.0 tests=3D
X-Barracuda-Spam-Report: Code version 3.1, rules version 3.1.21679
  =09Rule breakdown below
  =09 pts rule name              description
  =09---- ---------------------- -------------------------------------------=
Content-Transfer-Encoding: 8bit
X-MIME-Autoconverted: from quoted-printable to 8bit by
  id l63CkVM09250
Precedence: bulk
Apols for repost - seems some non-ascii characters in the subject line mess=
up the message I posted yesterday, so reposting with shorter title.

---------- Forwarded Message ----------
Date: 02 July 2007 14:55 +0100
From: Roger Harbord <>

Dear Tiago,

Stephen Sharp's original version of -metareg- saved the estimate of the
between-studies variance tau-squared in the global macro S_2. The regressio=
coefficients and their variance-covariance matrix are available in the
matrices e(b) and e(V).

I^2 doesn't in fact require tau^2 - it only uses the weighted residual sum =
squares, the analogue of Cochran's Q for meta-regression. I^2 is defined fr=
Cochran's Q in Higgins et al. (2003). So it's the same regardless of the
method used to calculate tau^2. Q is calculated from fixed-effect
meta-regression which can be performed by -vwls-. Q is reported by -vwls- a=
the "Goodness-of-fit chi2" and saved as e(chi2_gf).

I haven't implemented ML or "empirical Bayes" estimation of tau-squared in =
update of -metareg- as I believe that REML is preferable (see Thompson &
Sharp 1999). If Tiago knows of recent evidence to the contrary I'd apprecia=
the references. However, if those methods are requested using the original
syntax then the updated -metareg- simply calls the original program, which =
still present within the same ado-file.


Higgins JPT, Thompson SG, Deeks JJ, Altman DG.  Measuring inconsistency in
meta-analyses.  BMJ 2003;327:557-560.

Thompson SG, Sharp SJ.  Explaining heterogeneity in meta-analysis: A
comparison of methods.  Stat Med 1999;18:2693-2708.


Tiago Pereira wrote :

Dear statalisters,

(Roger will probably reply to this message - I hope).

Well, I was wondering if there is a straighforward way to output (or save)
the results of the old metareg (Sharp=B4s metareg) (sbe 23), beucase I am
unable to get either scalars or macros using the return/ereturn commands.

In this respect, how could one obtain the I^2 using an empirical Bayes
estimate of the between-study variance (tau^2) provided by Sharp=B4s

In relation to the new  -metareg- (Harbor-Steichen=B4s metareg), it would b=
so nice if we (non-statisticians, medical researchers) could obtain the ml
and empirical Bayes estimates of tau^2, since recent evidence have shown
that  these estimates yield more accurate results than either moment or
reml methods, especially for scenarios typical of meta-analyses in
medicine. Hence, their impplementation would be important.

Thx for any help.

Best regards,


---------- End Forwarded Message ----------

*   For searches and help try:

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