[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: metan and other meta-analysis commands in Stata
Very helpful, thanks. I look forward to your Stata Journal article.
Not to be a pest, but is there any chance you would add a sentence
discussing clustering to the help file (along with the three Hedges
references I supplied in my prior email)? You could also suggest that
the approximate correction due to Kish (1965) gets you most of the way
to correcting for clustering in almost every case, e.g.
K=sqrt[1+r(b-1)] where b is the number of individuals per cluster (the
assumption of equal cluster size seems not so important as long as
there are no outliers) and r is the coefficient of intraclass
correlation (ICC), and SDcorrected=SD*K. For that to work, of course,
you need to have not only sample size, but number of groups (e.g.
number of students and number of classes, or number of patients and
number of hospitals) for treatment and control in each study.
See page 162 and surrounding of
Kish, Leslie. 1965. Survey Sampling. New York, NY: John Wiley & Sons.
On 12/21/06, Jonathan Sterne <Jonathan.Sterne@bristol.ac.uk> wrote:
Following the recent posting of an updated metan command on the SSC
archive, Austin Nichols <firstname.lastname@example.org> commented:
> I for one would much appreciate a comparison of the various
> user-written packages for meta-analysis included in the help file, as
> a supplement to "Also see" section at the bottom of the help file.
We are currently writing Stata Journal articles on the new metan command
and other new or updated Stata meta-analysis commands. The following is an
attempt to summarise the facilities of some (though not all) of the
user-written commands for meta-analysis in Stata.
1. The meta command
This was the first Stata meta-analysis command. It requires the user to
supply the treatment effect estimate and its standard error for each study.
It uses inverse-variance weighting to derive fixed- and random-effects
summary estimates of the treatment effect estimate.
The meta command has not been updated since 1998, and uses Stata 7
graphics. As explained below, we now regard this command as redundant. We
are considering releasing a new version of meta that acts as a wrapper for
2. The metan command
The original version of the metan comand used as input the cell frequencies
from the 2x2 table for each study (for binary outcomes) or the mean and
standard deviation in each group (for numerical outcomes). It provides a
comprehensive range of methods for meta-analysis, including
inverse-variance weighted meta-analysis, and also creates new variables
containing the the treatment effect estimate and its standard error for
each study. These variables can then be used as input to a number of other
Stata meta-analysis commands.
All the meta-analysis calculations available in metan are based on standard
methods, an overview of which may be found in Chapter 15 ("Statistical
methods for examining heterogeneity and combining results from several
studies in meta-analysis", by Deeks, Altman and Bradburn), in Egger M,
Davey Smith G, Altman DG, eds. Systematic reviews in health care:
meta-analysis in context 2nd ed. London: BMJ Books, 2001."
The metan command has been updated on a number of occasions since it was
originally released. Because it now allows the user to supply the treatment
effect estimate and its standard error for each study, it now posesses
(almost) all the functionality of the meta command. Somewhat confusingly,
the release of metan that added this facility was made available on the SSC
archive in a package called "metaaggr" (meta-analysis of aggregate data).
This may have meant that some users continued with older versions of the
Other important new facilities added since the original metan command was
released include the by() option to conduct meta-analyses in subgroups, and
the recent update to Stata 9 graphics.
3. The metareg command
This command does meta-regression. It was released in 1998, with a major
update made available on the SSC archive in 2004. It requires the user to
input the the treatment effect estimate and its standard error for each
4. The metabias command
This command reports results of the Begg and Mazumdar (1994) and Egger et
al. (1997) tests for funnel plot asymmetry. It also produces funnel plots
and Galbraith plots, but these use Stata 7 graphics. It was released in
1997 and updates have been made available on the SSC archive on a number of
occasions since then. It requires the user to input the the treatment
effect estimate and its standard error for each study.
5. The metafunnel command
This command displays funnel plots. It was released in 2004 and uses Stata
8 graphics. It requires the user to input the the treatment effect estimate
and its standard error for each study.
6. The metatrim command
This command implements the "trim and fill" method to adjust for
publication bias in funnel plots. The most recent release was in 2003. It
requires the user to input the the treatment effect estimate and its
standard error for each study.
7. The metacum command.
This command performs cumulative meta-analyses, and graphs the results. It
does this using repeat calls to the meta command. It was released in 1998,
and has not been updated since then. It uses Stata 7 graphics.
Austin Nichols <email@example.com> also asked:
> Would the authors would be willing to add to the latest release? The
> help file available via view
> is not as helpful as one might expect. The syntax of the required
> varlist is unspecified, but must be inferred from the examples at the
> very end of the help file, e.g. metan n1 m1 sd1 n2 m2 sd2
Austin should ensure that he is looking at the most recent help file for
metan, by typing -view http://fmwww.bc.edu/repec/bocode/m/metan.hlp- in
Stata, or by installing the metan package (type -ssc install metan,
replace- and then -help metan- in Stata). This provides a number of
clickable examples of the use of the command. The required order of the
variables is explained in the description of the command: "When four
variables are specified these correspond to the number of events and
non-events in the experimental group followed by those of the control
We hope that Statalist users will find this information helpful.
With best wishes
Jonathan Sterne, Ross Harris and Roger Harbord
* For searches and help try: