3
$\begingroup$

I am conducting a meta-analysis in psychotherapy research and want to use meta-regression to identify moderators.

The regression outcome is a correlation between a psychotherapy process variable + treatment outcome. I have 3 predictors (potential moderators) with different scales: - Study Quality (continuous) - Type of Treatment (categorial, 3 levels) - Scoring Type in primary study (binary)

I would therefore dummy code Type of Treatment (two dummy variables) and Scoring Type (one dummy variable).

And now I am a bit confused about several questions:

  1. Does it (statistically) make sense to conduct a "hierarchical meta-regression" testing each predictor after one another and only including it in the next step if it was relevant?
  2. Should I center study quality around its mean for interpretation purposes of the categorial predictors?
  3. Theoretically, interaction effects would be possible. They are, however, rarely included in meta-regression. Would you still think it is a good idea do so? (Even a three-way-interaction?). Plus: do I need more power to estimate the additional interaction effects with adequate stability? (I have only 22 "cases", so 3 predictors are already critical)

Thank you very much for your input, it will be greatly appreciated!

$\endgroup$
5
$\begingroup$

I think a hierarchical approach would only make sense if you had a theoretical reason for establishing the order. I would suggest either doing each separately or fitting a full model with all three (which amounts to four as one is categorical with three levels).

Centreing quality will only affect your intercept in a model without interaction and since in a meta-regression the intercept is seldom interesting I would not bother although it would do no harm either.

Unless you have pressing reasons to look for interaction (like a journal referee's request) I would be a bit reluctant with only 22 primary studies. If you are tempted by a three-way interaction I would lie down in a dark room until the temptation has gone away, they are seldom easy to interpret and even less often correspond to anything scientifically interesting.

$\endgroup$
  • 3
    $\begingroup$ I would lie down in a dark room until the temptation has gone away -- I wish I could upvote more than once. $\endgroup$ – Wolfgang Nov 17 '16 at 14:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.