Modeling the residual structure in a repeated measure design - better, but not still there

[ACL1972]

Hi,

Thanks for providing this free and intuitively user-friendly package to Jamovi. Much approciated.

I work with mixed models and repeated designs. The latest update was a great improvement in terms of modeling the residual structure by providing an autoregressive option. However, there might be heteroregeneity across the repeated measures that also need to be modeled. The data might also naturally fit some other structures. Thus, the current options are still too limited for research purposes and one might unawarely do both type I and type II errors by sticking to the current options.

I hope that future updates will bring more options for modeling the residual structure!

Best
Christian

[mcfanda]

What other options would you suggest?

[ACL1972]

Thanks for swift reply!

There are a number of important residual structures highlighted in the literature. The choice and relevance depend on the research field. See, e.g., Wolfinger (1996) and Heck and Thomas (2022).

Personally, I misses the ARMA, ARH(1) and CSH options, i.e., autoregressive moving average, heterogeneous autogressive, and heterogeneous compound symmetry.

Best,

Heck, R. H., and S. L. Thomas. (2022). An Introduction To Multilevel Modeling Techniques. MLM and SEM Approaches. Routledge.

Wolfinger, R. D. (1996). Heterogeneous Variance: Covariance Structures for Repeated Measures. Journal of Agricultural, Biological, and Environmental Statistics, 1(2): 205–230.

[mcfanda]

That's doable, the only issue is to let the user to chose from the interface the parameters of the matrix, such as the p and q for ARMA. I'll work on it, thanks

[ACL1972]

Fantastic - and thanks for your dedication to the development of Jamovi. I look forward to test it :-)

NB. In the current version, I am not sure whether jamovi order the data correctly when using the AR(1) option. Clearly, the data need to be ordered in the correct time sequence for a meaningful estimation. This is unimportant if the level 2 cluster is a group such as a school class or a team.

Best,
Christian

[mcfanda]

we use the default ordering, which corresponds to using the order of the observations in the data. So it is up to the user to order the "time" variable is the correct way.