I'm interested in conducting an individual patient data (IPD) meta-analysis in which the main outcome of interest is a correlation coefficient between and IV and DV.

The IV is measured slightly differently across studies, and conversion to make them all match is not possible. Clearly, I can still do a two-stage IPD meta-analysis based on correlation coefficients calculated for each study individually. I would also like to conduct a one-stage IPD analysis using linear mixed modeling (with random intercepts and/or slopes by study) for a few reasons: to more flexibly model an important covariate, to assess whether the relationship between IV and DV is linear, and to model aggregation bias (per Stewart et al. 2012).

Given the heterogeneity in the IV measures, is this still possible?


Added plot to demonstrate the problem.

Would it make sense, in this case, to standardize the variables within-study before conducting IPD meta-analysis? I'm worried that such an approach would lose across-study effects X.


  • $\begingroup$ It is unclear what "The IV is measured slightly differently across studies" means (can you give an example, how many categories, etc.). Can you dichotomize your IV? What about conducting a 2-stage analysis and then running a meta-regression to test if different measurements of your IV affect the effect sizes? $\endgroup$ – Bernd Weiss Jan 18 '14 at 6:16
  • $\begingroup$ The IV is perceived stress. Most studies used the same validated psychometric scale to measure perceived stress, but they used different methods of creating a single composite score. E.g., some studies use the sum of all the items, some use the mean, and others take a variety of other approaches. I don't have access to the raw responses, so can't create my own consistent composite across studies. Does that help? $\endgroup$ – half-pass Jan 20 '14 at 18:13

Your Answer

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

Browse other questions tagged or ask your own question.