1
$\begingroup$

I am conducting a meta-analysis on the efficacy of a specific type of psychotherapy for children and plan to use meta-regression to identify moderators.

The predictor I am interested in is parent psychopathology, but this variable is collected using different measures across studies. For example, Study 1 uses Questionnaire 1 (e.g., Beck Depression Inventory), Study 2 uses Questionnaire 2 (e.g., Depression Anxiety Stress Scales), and both questionnaires use different scales. What information would I need to extract and use for the predictor in the meta-regression from each study? The mean alone doesn't seem appropriate given that different measures are being used from study to study.

Thank you in advance for your input!

$\endgroup$

1 Answer 1

0
$\begingroup$

As a beginner in meta-analysis and here at CV, I write with caution. So far I strongly doubt that mixing scales for a predictor in meta-regression models is valid. I understand that different scales can be mixed if they are pooled as an effect size. E.g.: SMDs in BDI mixed with SMDs in DASS works, they can further be transformed to other ES. However, a predictor variable is not necessarily a standardized effect size and different scales usually don't meet the same metric and item thresholds. I suspect this is the case in your baseline-pathology example.

Thinking of exceptions, I would argue that:

  1. it's okay to mix different scales if they can easily be expressed in a normed metric derived from the same reference population (hypothetically: mixing t-scores of BDI with t-scores of DASS from a representative parent sample).
  2. Similarly, imagine the two scales had the same item anchors (say from 1="not at all" to 6="extremely"), same number of items, and arguably same item difficulties, plus evidence that the raw values behave in similarly.
  3. Finally, effect size measures could predict other effect size measures, e.g.: mixed SMDs from BDI/DASS predicting 1-year relapse proportions.

I'll let you know if someone convinces me of the contrary. Cheers

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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