Individual participant data

1 reply [Last post]
jbauer's picture
Joined: 04/14/2014

Hello Mike,

in your 2014 article on MASEM (Behav Res, DOI 10.3758/s13428-013-0361-y) you mention individual participant data (IPD) meta-analysis as a tool for addressing issues in MASEM, such as categorical response variables. I am wondering what would be an advisable analytic strategy for MASEM with IPD (assuming a fixed effects model).

Here are several options I found:
(1) One approach could be to extract a correlation matrix from each study and analyze it with TSSEM in metaSEM.
(2) Cooper and Patall (2009, Psych. Methods) mention that study can be used as a stratification variable.
(3) Curran and Hussong (2009, Psych. Methods) suggest including study membership as a categorical predictor as well as its interactions with other predictors in the model.

From my perspective, approach 1 is straightforward, but I am not sure whether it is applicable to my specific case (models based on categorical items, cluster sampling). Approach 2 is straightforward to implement, e.g. in Mplus, but I am not sure about its drawbacks. It just seems too easy. Approach 3 seems impractical with complex SEMs because including several latent/observed interactions quickly leads to overcomplex models and estimation problems.

What would be the best choice?

Many thanks

Mike Cheung's picture
Joined: 10/08/2009
Hi Johannes, I cannot comment

Hi Johannes,

I cannot comment on (2) and (3) as I haven't gone through the details yet.

When the data are available, it is always better to use the raw data in the analysis. Moreover, the TSSEM approach is based on the assumption that the variables are continuous. Since you have categorical variables, the TSSEM approach may not be appropriate.

When you are analyzing the data, make sure that the scalings are comparable across studies. In the TSSEM approach, we usually work with the correlation matrices. For your data, some sort of standardization in the model may be required.