Hi. I have a similar "nonadmissible solutions" situation.
I ran a CFA with 3 factors (20 indicators total) and then the following SEM code for a small sample:
FIT01 <- sem(MYFMODEL, data = MYDATA, mimic = "Mplus", estimator = "ML", missing = "ML", se = "bootstrap", bootstrap = 10000, parallel = "snow", ncpus = cores)
I received the following notes:
2: In lav_model_nvcov_bootstrap(lavmodel = lavmodel, lavsamplestats = lavsamplestats, :
lavaan WARNING: 14 bootstrap runs failed or did not converge.
3: In lav_model_nvcov_bootstrap(lavmodel = lavmodel, lavsamplestats = lavsamplestats, :
lavaan WARNING: 9986 bootstrap runs resulted in nonadmissible solutions.
4: In lav_object_post_check(object) :
lavaan WARNING: covariance matrix of latent variables
is not positive definite;
use lavInspect(fit, "cov.lv") to investigate.
However, I then ran
parameterEstimates(), which returned information for all of the parameters. I did not find any negative error variances. Additionally, when I ran
summary(), it returned the following in the results:
Parameter Estimates:
Standard errors Bootstrap
Number of requested bootstrap draws 10000
Number of successful bootstrap draws 9986Omega results appear strong:
compRelSEM(FIT01, obs.var = FALSE, tau.eq = FALSE, ord.scale = TRUE)
f1 f2 f3 f12
0.879 0.790 0.964 0.910I'm not sure how to interpret all of this. Given the 9986 "successful bootstrap draws" and the returned parameter estimates, could I proceed with the analysis? Or is that all meaningless given the "9986 bootstrap runs resulted in nonadmissible solutions" warning? I would greatly appreciate any insight you may have.