[R-meta] Categorical moderator analysis - subset or mods?

[Viechtbauer, Wolfgang (SP)]

Dear Alina,

The subset approach allows for different tau^2 values per level. You can achieve the same using a single model. See:

http://www.metafor-project.org/doku.php/tips:comp_two_independent_estimates

and esp. the section "Meta-Regression with All Studies but Different Amounts of (Residual) Heterogeneity".

When you use "mods = ~factor -1", then you will get the estimated average effect for each level. The QM-Test then tests whether the aveage true effect is zero for each level.

When you use "mods = ~factor", then one level is the reference level (corresponding to the intercept), and the coefficients for the other levels indicate the *difference* between the reference level and the other two levels. The QM-Test then tests whether the average true effect is the same across all levels. This is typically what we want to test (i.e., whether there are any differences in the average true effect across factor levels).

Best,
Wolfgang

-----Original Message-----
From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces using r-project.org] On Behalf Of Alina Böckmann
Sent: Friday, 10 August, 2018 18:44
To: r-sig-meta-analysis using r-project.org
Subject: [R-meta] Categorical moderator analysis - subset or mods?

Dear all,

I have problems with the evaluation of my moderator hypothesis.
I would like to evaluate a categorical moderator variable with three categories: the development stages "zero", "some" and "considerable experience".

I do not know which of the following functions is the rigth one for my analysis:
- subset  -> one analysis for each of the three moderator categories
- mods= ~factor -1 -> Analysis without intercept
- mods= ~factor -> Analysis with Intercept

I would be very grateful if you would support me on this issue.
Thank you very much.

With kind regards,
Alina Böckmann