metabias

[Nicole Sofia]

Hello,
I try to assess the bias in meta-analysis. I use the the metabias command, but the "Warning: varlist has 3 variables but option 'ci' not specified; 'ci' assumed." line appears.
What is wrong with my data
And could you please help with assumptions of different tests: Egger, Harbord, Beggs. Which one should be used and how should I interpret the result?
Thanks in advance.

[David Fisher]

Hi Nicole,

I only have a passing acquaintance with metabias, but since no-one else has responded I'll have a go!

I'm not sure which version of metabias you have installed (type "which metabias" to find out), but the most recent one (v2.1, 5th Jan 2009) will not let you supply 3 variables, just 2 (theta + std error) or 4 (2x2 cell counts for binary data). Presumably you are using an older version which will accept 3 variables, but prefers you to confirm this with an additional option "ci". Your options are:

1) See if adding the option "ci" makes the error message go away. Then, check the results by converting your confidence interval into a std error and using the 2-variable syntax (without the "ci" option). The results should be identical; if so, you don't have a problem. (if not, the first step would be to double-check your derivation of the std error!)

2) Update your version of metabias, and use the 2-variable syntax. This may have some advantages, e.g. if bugs have been fixed or additional functionality added, but I wouldn't know about this I'm afraid. However, it is usually good practice to use the most up-to-date version of a program.

Hope that helps,

David.


P.S. Hopefully someone more knowledgeable can help with the test assumptions. If not, there is some information in the Cochrane Handbook at http://handbook.cochrane.org/chapter..._asymmetry.htm, including links to the original papers by Egger, Harbord and Beggs.