lavaan ERROR: sample covariance matrix is not positive-definite

[Jade Ding]

I am using lavaan for a model I created:

mod.3 <-'

s_cred ~ Med_Alg

s_cred ~ Med_Mdia

s_cred ~ Med_Fact

s_cred ~ Med_Uni

s_exp ~ Med_Alg

s_exp ~ Med_Mdia

s_exp ~ Med_Fact

s_exp ~ Med_Uni

m_credibility ~ s_cred + s_exp

post_att ~ m_credibility

s_cred ~~ s_exp

'

fit.mod.3 <-sem(mod.3, data = fact_pass1, std.lv = TRUE, missing = "fiml")

summary(fit.mod.3, standardized = TRUE, fit.measures = TRUE, rsquare = TRUE)

I got the error: 

Error in lav_samplestats_icov(COV = cov[[g]], ridge = ridge, x.idx = x.idx[[g]],  : 

  lavaan ERROR: sample covariance matrix is not positive-definite

I checked the sample covariance matrix:

             [,1]        [,2]        [,3]         [,4]         [,5]         [,6]         [,7]        [,8]

[1,]  1.692791957 -0.10917026 -0.21861477 -0.234459695 -0.009085666  0.028148062  0.009349616 -0.02841201

[2,] -0.109170265  2.22850978  1.29626675  0.710606040 -0.016972996 -0.129350567 -0.041114109  0.18743767

[3,] -0.218614775  1.29626675  2.80818255  1.150875217 -0.018927323 -0.026051670 -0.025861506  0.07084050

[4,] -0.234459695  0.71060604  1.15087522  2.662285491 -0.004239921 -0.008223286 -0.015055770  0.02751898

[5,] -0.009085666 -0.01697300 -0.01892732 -0.004239921  0.094536038 -0.030570077 -0.030826969 -0.03313899

[6,]  0.028148062 -0.12935057 -0.02605167 -0.008223286 -0.030570077  0.205557417 -0.084331248 -0.09065609

[7,]  0.009349616 -0.04111411 -0.02586151 -0.015055770 -0.030826969 -0.084331248  0.206576125 -0.09141791

[8,] -0.028412012  0.18743767  0.07084050  0.027518977 -0.033138992 -0.090656092 -0.091417908  0.21521299

I don't know how to solve this problem, does anyone have any ideas?

[Terrence Jorgensen]

fit.mod.3 <-sem(mod.3, data = fact_pass1, std.lv = TRUE, missing = "fiml")

Error in lav_samplestats_icov(COV = cov[[g]], ridge = ridge, x.idx = x.idx[[g]],  : 

  lavaan ERROR: sample covariance matrix is not positive-definite

I can only guess that the estimated covariance matrix is NPD because there is a lot of missing data.  I'm not sure there is a solution if FIML cannot find a proper solution when fitting the saturated (h1) model.  You could try using multiple imputation instead, but I'm not sure the same problem wouldn't manifest as the pooled (i.e., average) sample covariance matrix being NPD.

Terrence D. Jorgensen

Assistant Professor, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam

http://www.uva.nl/profile/t.d.jorgensen