When a participant has both positive and negatively converged edges between the same two nodes + increasing max_iter and conv_crit to get more models?

[Henry Whitfield]

Hi again Katie,

I'm guessing that when a participant has both positive and negatively converged edges between the same two nodes, that means that both were observed at different times during the time series. Is that correct? Is there anything else this signifies, or that we might discuss about this?

Also for those that don't converge, is it worth trying to tweak the code to make them converge? Increasing the number of iterations.

Chat GPT gave us this: # Load previous output if saved

load("last_known_convergence.RData")

# Modify or pass it into a fresh call
gimme(data = your_data,
     out_name = "retry_model",
     group_paths = last_model$group_paths,   # reuse successful paths
     ind_paths = last_model$ind_paths,
     max_iter = 1000,                        # possibly increase
     conv_crit = 0.0001                      # potentially tighten convergence
)

I'm most interested if this might help us get more models for the CS-GIMME which loses 4/30 participants to having no model at all.

Many thanks,

Henry

[Katie Gates]

The main problem here is that only 4/30 are converging. 

SEM models don't converge for a number of reasons. It could be that the variances differ greatly among the variables, or that there's high collinearity, or that there is no best solution given the model structure (it gets 'stuck'). Sometimes there are just too many paths for the length of data (i.e., too many parameters being requested and not enough information). 

Sometimes increasing iterations can help. We currently don't have this option for the user to manipulate. Given the percent of non-convergence, I don't think this would help as there may be something going on with the data that causes non-convergence. Perhaps explore using lavaan to get a sense of where models break? It seems the data are not suitable for gimme. 

The recursive paths - where two nodes have one directed arrow going one direction and another going the opposite - are interpreted just as any other path. They are across the entire time series. 

Best,

Katie