S-GIMME No Model Convergence / Variance-Covariance Issue

[Megan Hynd]

Hello GIMME team,

I am using S-GIMME in an fMRI analysis and have been getting a strange error that none of the participants' models converged. We have checked each of the suggested troubleshooting steps, here are the answers to each:

1. Ensure that all of your individuals have the same number of variables (columns) in their data sets. We have checked this, and they do have the same number of variables, in the same order.

2. Ensure that all variables have variability (i.e., are not constant). gimme will let you know if this is the case. We have checked this, and none of our variables are constant for any of our participants. Their variation data is also within a similar range to another dataset that we successfully ran gimme on.

3. Ensure your path directories are correct. They are correct, and we have tried using a directory and an RDS list, both of which yield the same result.

4. Ensure that the columns are variables and the rows contain the observations across time. Yes to both.

5. Ensure that variables aren’t perfectly correlated (or nearly perfectly correlated) for all individuals. The variables are not nearly perfectly correlated for the subjects, except for the diagonal self-correlations, which I assume are not the issue.

I have also updated the gimme R package, and that did not help.

Here is the code and error I have been receiving when running S-GIMME:

> # S-GIMME with GFC MVPA-SBC NRAD baseline on WIN BOLD timeseries > gimme_output <- gimme(data        = mydata, +                      header      = TRUE, +                      out         = '/proj/schillerlab/users/mhynd/R Analysis/GNODE/Phase2_WIN/Results/SGIMME_results', +                      sep         = ',', +                      subgroup    = TRUE, +                      sub_feature = "lag & contemp", +                      sub_method  = "Walktrap", +                      groupcutoff = .75, +                      subcutoff   = .51, +                      standardize = TRUE, +                      hybrid      = TRUE) group-level search subgroup-level search

Error in determine.subgroups(base_syntax = c(dat$syntax, grp$group_paths), : gimme ERROR: none of the participants' models converged.

In addition: There were 46 warnings (use warnings() to see them)

> warnings()

Warning messages: 1: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix (Sigma) is smaller than 1e-05; this may cause numerical   instabilities; interpret the results with caution. 2: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix (Sigma) is smaller than 1e-05; this may cause numerical   instabilities; interpret the results with caution. 3: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix (Sigma) is smaller than 1e-05; this may cause numerical   instabilities; interpret the results with caution. 4: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix (Sigma) is smaller than 1e-05; this may cause numerical   instabilities; interpret the results with caution. 5: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix (Sigma) is smaller than 1e-05; this may cause numerical   instabilities; interpret the results with caution. 6: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix (Sigma) is smaller than 1e-05; this may cause numerical   instabilities; interpret the results with caution. 7: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix (Sigma) is smaller than 1e-05; this may cause numerical   instabilities; interpret the results with caution. 8: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix (Sigma) is smaller than 1e-05; this may cause numerical   instabilities; interpret the results with caution. 9: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix (Sigma) is smaller than   1e-05; this may cause numerical instabilities; interpret the results with caution. 10: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix (Sigma) is smaller   than 1e-05; this may cause numerical instabilities; interpret the results with caution. 11: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix (Sigma) is smaller   than 1e-05; this may cause numerical instabilities; interpret the results with caution. 12: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix (Sigma) is smaller   than 1e-05; this may cause numerical instabilities; interpret the results with caution. 13: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix (Sigma) is smaller   than 1e-05; this may cause numerical instabilities; interpret the results with caution. 14: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix (Sigma) is smaller   than 1e-05; this may cause numerical instabilities; interpret the results with caution. 15: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix (Sigma) is smaller   than 1e-05; this may cause numerical instabilities; interpret the results with caution. 16: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix (Sigma) is smaller   than 1e-05; this may cause numerical instabilities; interpret the results with caution. 17: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 18: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 19: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 20: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 21: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 22: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 23: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 24: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 25: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 26: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 27: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 28: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 29: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 30: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 31: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 32: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 33: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 34: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 35: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 36: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 37: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 38: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 39: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 40: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 41: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 42: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 43: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 44: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 45: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution. 46: lavaan->lav_mvnorm_missing_h1_estimate_moments():     The smallest eigenvalue of the EM estimated variance-covariance matrix   (Sigma) is smaller than 1e-05; this may cause numerical instabilities;   interpret the results with caution.

I have also tried turning standardize to FALSE and (on a separate analysis) turning hybrid to FALSE. Both of which yielded the same results. 

I am happy to provide our data, if needed.

Thank you for your help,

Megan Hynd

[Katie Gates]

Hi Megan, 

Sorry to hear of this - so frustrating. 

The last bit of information that might be helpful is: how many variables, and what's the average number of time points (that aren't missing)?

That none of the models converged points to something in the data. I wonder if the Autoregressive effects are very high (like above .90)? This can cause non-convergence. You could use a base R function arima: variable1AR <- arima(my_time_series, order = c(1, 0, 0)).

hope all is going well for you, 

Katie