We are so excited to have the option for including both bidirectional and directed contemporaneous relations. The option, called hybrid-GIMME, allows for covariances among residuals (somewhat similar to graphical VAR and traditional VAR) while also allowing for directed relations among the observed variables in the search space. Check it out!

GIMME now provides another option for arriving at model structures. This option does not use Granger causality to test the directionality of paths. Rather, it allows for multiple final models with both directions followed in model building. The researcher can then choose from the final models.

Details can be found in:

Beltz, A. M., & Molenaar, P. C. (2016). Dealing with multiple solutions in structural vector autoregressive models. Multivariate behavioral research, 51(2-3), 357-373.

with a tutorial found here: https://gimme.web.unc.edu/63-2/230-2/

Recent work has shown that GIMME can reliably find the relations among latent variables – be they factor score estimates or PCA component scores.

The option to conduct factor analysis (using MIIV-2SLS via the MIIVsem package; Fisher, Bollen, Gates, & Rönkkö, 2017 ) is now available within GIMME.

Using lavaan (Rosseel, 2018) style syntax, users can indicate the measurement model structure for indicating how observed variables relate to latent factors. Factor scores are then generated using this model and gimme searches for the directed lagged and contemporaneous paths among these factors. Final estimates are obtained via MIIVsem.

Researchers can now specify which variables can only be predictors (and not be predicted). This is necessary when a researcher wants to see how a variable that occurs outside the system influences variables inside the system. For instance, weather may predict self-reports of mood that occur across days, but mood will never predict the weather.

This option is essential for fMRI studies. Here, a task vector of onsets can be included. GIMME will convolve this vector using a smoothed FIR and include it in analysis. Researchers can indicate any interaction effects (e.g., the task times cerebellum activity across time) to include in the search procedure to see if relations among brain regions change during task.

We’ve recently made our GitHub repository public for programmers and users that would like to contribute to the code: https://github.com/GatesLab/gimme