1
$\begingroup$

I am running a three-level meta analysis with about 250 included studies and ~1200 contrasts. I try to get a glimpse on the heterogeneity by running moderation analyses. However, I know that most of my moderators are statistically dependent which is a problem for interpretability (Lipsey, 2003). What would you think would be the best idea for dealing with moderator dependency? Running all moderators of interest within one model and see whether their significance survives, combine levels of moderators to new dummy variables (e.g., moderators: Study Design: A vs. B, Participant type: C vs. D --> Moderators AC, AD, BC, BD), or other solutions?

$\endgroup$

1 Answer 1

1
$\begingroup$

Following the multiple moderator model solution actually provided insightful results. Taking all other pre-specified moderators into account, one moderator did not further explain heterogeneity in effect sizes. I interpret this finding as such that this moderator was only significant due to its interrelation with another meaningful moderator. Maybe not the most elegant solution, yet in my opinion still meaningful.

$\endgroup$

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.