Nonconvergence/Unstable solutions

[Dina Dajani]

Hi,

After I run gimme with subgroup estimation on 26 people, I see that 3 individuals have 'nonconvergence' or 'unstable solution' status in the summaryFit.csv. Is it safe for me to use the group, subgroup and individual level results (except for the nonconvergence/unstable folks) from this run or should I re-do the analysis without those 3 subjects?

thanks

Dina

[Katie Gates]

HI Dina, 

I'd keep these analyses. This is just the status of their final model. They could have contributed to the identification of group (and subgroup) level paths. Keeping them allows you to use whatever usable information they have for those aspects of the models. 

Katie

[Henry Whitfield]

Thanks for this Katie, 

And what about the new "no additional significant paths". This seems to have replaced the previous 'last known convergence'. I'm trying to determine how significant or problematic it is if the majority of individuals are  "no additional significant paths". Since GIMME updated to a new version,  most of my previously "Converged normally"  are now "no additional significant paths"
Apologies for posting something similar in a different thread yesterday but I realise this is a more appropriate thread, and my question is more nuanced here.
Many thanks,

Henry

[Katie Gates]

Please see response to other message, which I saw first. 

"Last known convergence" is also an potential outcome. That occurs when adding paths lead to nonconvergence, so the algorithm rolls back and estimates the last-known model that converged for that data. This might mean fit indices that are not excellent or good.