Updated GIMME package - new feature

[Katie Gates]

Hey all, 

GIMME 9.2 is now available. 

Main changes: 

- fixed a Confirmatory Subgroup (CS-GIMME) bug that impacted minority of cases

- New argument - user can now define the fit index cutoffs for stopping criteria thanks to the efforts of Björn Siepe of the University of Marburg. 

Best, 

Katie 

[Henry Whitfield]

Hi Katie,
I've noticed that individual models that used to fully converge, now only reach "no additional significant paths".

26/30 used to 'converge normally' now only 3/30 do. Could the update have caused this? Is this a major limitation to the results when most models don't reach  "converged normally"?
Please advise

Many thanks,
Henry

[Katie Gates]

Hi Henry, 

This was a big change from the prior version that will impact some results while not affecting others at all. 

In recent versions, gimme optimized based on fit indices - that is, it would add paths (even if non-significant after Bonferroni correction) until the individual-level model achieved good model fit according to 2 out of 4 fit indices. 

This was an error that occurred due to a mistake in an 'if' loop rule. This does not match the publications and related simulations for most gimme output. In those, paths were only added if they were significant after a Bonferroni correction. 

So, what to do if the results with your data don't match across versions? 

First, choose which optimization you prefer and makes most sense given qualities of your data. In future gimme versions this will be an option. For now, you can either use the most recent version (0.9.3) which prioritizes only adding paths that are significant for that individual or other recent versions if you wish to prioritize model fit. For 0.9.3, if no additional paths are significant, the model search stops for that person even if the fit indices don't meet criteria. 

Earlier (recent) versions keep adding paths even if they are not significant. Here, it prioritizes fit index criteria. This might be a good option if T is small, meaning that you might be under-powered and paths may not reach significance (after the strict Bonferroni correction). This might not be a good option if you have large T. 

As with any software results may change with updates and this causes an issue with reproducibility in the future. As part of the output gimme saves the environment information in "sessionInfo.txt" when output is saved in a directory or $sessionInfo when saved as an R object. This should be provided when providing code for a publication. This shows which version of gimme was used as well as the versions of any other packages loaded into the environment (such as the ones that gimme relies on). 

This is all undesirable, for sure. Just an artifact of free software. Our funding mechanisms don't tend to offer funding to maintain packages. 

Best,
Katie

[Henry Whitfield]

Many thanks Katie, That helps. Is there a way to access the previous version of GIMME. I'm now running it with new exogenous variables so need to perform new analyses with those additional variables and a fairly low T. 

Best wishes,

Henry

[Katie Gates]

Hi Henry, 

We made an update the package that allows for all options. I'll post an email now with details. 

Katie

[Katie Gates]

Henry, I meant to add that to return back to the prior approach (seen in version 0.9.2), you would use the argument "stop_crit" and put 

stop_crit = "model fit"

in the argument list for your gimme runs.