Statalist The Stata Listserver

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

st: New version of -metan- package on SSC

From   RJ Harris <>
Subject   st: New version of -metan- package on SSC
Date   Mon, 19 Feb 2007 18:42:48 -0000

Thanks to Kit Baum, an updated version of the -metan- package is available on SSC.

The purpose of this update is twofold: to correct minor bugs and make additions to the -metan- command, and to include the command -metannt-.

Changes to the -metan- command following the December update are minor and mainly concern display capabilities. To avoid tedium, these are catalogued at the end of this email.

-metannt- displays effects in terms of the absolute reduction in risk and number needed to treat. This command was previously part of the -metaaggr- package, and now that the -metan- package includes this the -metaagrr- package is now redundant and therefore no longer available. (Note that the Stata v7 incarnation of the -metan- command is also included under the name -metan7- for those not using Stata v9)

Thank you very much to the Stata users who spotted deficiencies in the -metan- command and took the trouble to make helpful suggestions.

To install -metan-, please type:

-ssc install metan-

and to view the help file

-help metan-

Comments and suggestions are welcome, and should be sent to

Best wishes,

Ross Harris, Roger Harbord and Jonathan Sterne
Department of Social Medicine, University of Bristol UK

-Get original-style solid boxes with the option _classic
-Show diamonds only, no studies with _summaryonly
-Switch off "Weights are from random effects..." with _nowarning

-First variable specified in _lcols() now appears in table
-Option _nobox did not work, now fixed
-Option _nosubgroup did not work, now fixed
-If _xlabel() used without _force and estimates/CI outside specified range, x-axis labels were put up at graph extremes -If any display options changed (e.g., line width for confidence intervals with ciopt()) then any other options were picked at random by Stata (e.g., line colour may have changed)
-In overall test of null hypothesis "Overall I-sq..." was displayed instead of just "Overall"

It has been noted that the text size is rather small with some graphs. In short, there's not much that can be done about it- just use _textsize() to rescale. If you're interested in how the text size is chosen, please read on...

The text size is based on the number of lines and total width of text. Using this, the "natural" aspect ratio of the graph is calculated. However, this is averaged with the usual aspect ratio of a Stata graph, i.e., 0.7. The closer to the natural aspect the graph is displayed as, the larger the text. If your graph is more naturally displayed as a bit wider than normal, you could try forcing the graph to a different shape:

. metan logOR selogOR, aspect(0.5)

I've tried to make it so the aspect ratio is not too extreme if the number of studies is low (or lots of text) or very high (or little text), and the text size is displayed conservatively so the chance of overlap is minimised. Although the results are not always ideal, in my defence this occurred quite frequently in the old metan as well- and not only tiny text, but overlap sometimes.

Ross Harris
Department of Social Medicine
University of Bristol
+44 (0)117 928 7338

* For searches and help try:

© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index