convergence problem: userModel ran to completion, but no fit indices

[Katerina Zorina-Lichtenwalter]

Hi everyone,

I am running usermodel() on 33 traits. My model specifies 3 factors: a general factor with all 33 traits and 2 specific factors, and I am additionally estimating 29 residual covariances (informed by the observed-implied difference (resid_cov[[2]]) matrix of a simpler model, which I divided by the SEs from LDSCoutput$V and filtered for values > 2). The general factor covariance with specific factors is constrained to 0. The model runs to completion but outputs no fit indices, as shown below. Any advice would be much appreciated!

Thanks,

Katerina

> biquart33_3all_02noCLPOSfit_rep1 <- usermodel(LDSCoutput_regenie, estimation = "DWLS", model = biquart33_3all_02noCLPOSmodel_rep1,  CFIcalc = TRUE, std.lv = T, imp_cov = T)

[1] "Running primary model"

[1] "Calculating model chi-square"

[1] "Calculating CFI"

[1] "Calculating Standardized Results"

[1] "Calculating SRMR"

elapsed

327.839

[1] "The S matrix was smoothed prior to model estimation due to a non-positive definite matrix. The largest absolute difference in a cell between the smoothed and non-smoothed matrix was  0.0163559488681797 As a result of the smoothing, the largest Z-statistic change for the genetic covariances was  Inf . We recommend setting the smooth_check argument to true if you are going  to run a multivariate GWAS."

[1] "The V matrix was smoothed prior to model estimation due to a non-positive definite matrix. The largest absolute difference in a cell between the smoothed and non-smoothed matrix was  7.2402995884434e-11 As a result of the smoothing, the largest Z-statistic change for the genetic covariances was  Inf . We recommend setting the smooth_check argument to true if you are going to run a multivariate GWAS."

Warning message:

In usermodel(LDSCoutput_regenie, estimation = "DWLS", model = biquart33_3all_02noCLPOSmodel_rep1,  :

  A difference greater than .025 was observed pre- and post-smoothing for Z-statistics in the genetic covariance matrix. This reflects a large difference and results should be interpreted with caution!! This can often result from including low powered traits, and you might consider removing those traits from the model. If you are going to run a multivariate GWAS we strongly recommend setting the smooth_check argument to true to check smoothing for each SNP.

> biquart33_3all_02noCLPOSfit_rep1$modelfit

                                             chisq  df  AIC

df The follow-up chi-square model did not converge 444 <NA>

                                                                        CFI

df Either the chi-square or null (i.e. independence) model did not converge

                 SRMR

df 0.0844186911244359