sample covariance matrix is not positive-definite

456 views

covarianceMatrixlavaan

Skip to first unread message

Jassmyn

unread,

Sep 16, 2019, 10:07:17 AM9/16/19

to lavaan

Hello,

I'm an absolute beginner with R and try to do a SEM.

I had first warnings because of a high variance in my data. I performed a z-standardization and now I have 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
.
But I dont know, how to fix that.

Has somebody a suggestion?

I use the following code:


###------------------------
### Loading packages
###------------------------

library(lavaan)
library(lme4)
library(nlme)
library(reshape)
library(robustbase)
library(robustlmm)
library(WRS2)
library(psych)
library(MASS)
library(reshape2)
library(robmed)
library(semPlot)

###------------------------
### PATH MODEL
###------------------------

sem.MODELL28 <- ' 
ZMEAN_Spende_2 ~ ZMEAN_Schutzbereitschaft + ZMEAN_PMT_Barrieren + ZMEAN_PMT_response_efficacy + ZMEAN_PMT_self_efficacy + ZMEAN_PMT_ ethical_emotions + ZMEAN_PMT_Vulnerability + ZMEAN_PMT_Severity + ZMean_Einstellungen_gesamt + Summe_Quiz_richtig_falsch + PA01 + PA29_01 + PA07
ZMEAN_Schutzbereitschaft ~ ZMEAN_PMT_Barrieren + ZMEAN_PMT_response_efficacy + ZMEAN_PMT_self_efficacy + ZMEAN_PMT_ ethical_emotions + ZMEAN_PMT_Vulnerability + ZMEAN_PMT_Severity + ZMean_Einstellungen_gesamt + ZSumme_Quiz_richtig_falsch + ZPA01 + ZPA29_01 + ZPA07
ZMEAN_PMT_Barrieren ~ ZMean_Einstellungen_gesamt + ZSumme_Quiz_richtig_falsch + ZPA01 + ZPA29_01 + ZPA07
ZMEAN_PMT_response_efficacy ~ ZMean_Einstellungen_gesamt + ZSumme_Quiz_richtig_falsch + ZPA01 + ZPA29_01 + ZPA07
ZMEAN_PMT_self_efficacy ~ ZMean_Einstellungen_gesamt + ZSumme_Quiz_richtig_falsch + ZPA01 + ZPA29_01 + ZPA07
ZMEAN_PMT_ ethical_emotions ~ ZMean_Einstellungen_gesamt + ZSumme_Quiz_richtig_falsch + ZPA01 + ZPA29_01 + ZPA07
ZMEAN_PMT_Vulnerability ~ ZMean_Einstellungen_gesamt + ZSumme_Quiz_richtig_falsch + ZPA01 + ZPA29_01 + ZPA07
ZMEAN_PMT_Severity ~ ZMean_Einstellungen_gesamt + ZSumme_Quiz_richtig_falsch + ZPA01 + ZPA29_01 + ZPA07
ZMean_Einstellungen_gesamt ~ ZSumme_Quiz_richtig_falsch + ZPA01 + ZPA29_01 + ZPA07
'

# fit the model

fitsem <- sem(sem.MODELL28, data=SPSS_DATEI_umbenannte_variablen_gekuerzt_ohne_divers_13_09_gekuerzt_Standardisierte_Variablen, estimator="MLR", test = "bootstrap")

# display summary output

summary(fitsem.MODELL28, fit.measures=TRUE, standardized = TRUE, rsquare = TRUE)

Terrence Jorgensen

unread,

Sep 17, 2019, 5:18:59 AM9/17/19

to lavaan

I had first warnings because of a high variance in my data. I performed a z-standardization

Bad idea, that biases your SEs and inflates Type I error rates.

https://pdfs.semanticscholar.org/3daa/5151f6a0931e9b6587cf06571486013f56ec.pdf

Divide or multiply by a constant (e.g., a factor of 10 or 100) instead of by a sample-based estimate.

now I have 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

If your sample covariance matrix is NPD, there are either redundant variables or some variables with no variance. This should have nothing whatsoever to do with standardizing/transforming your variables. Does this occur when you use test = "standard"? If not, then it probably happens when a bootstrap sample is drawn that coincidentally has all the same values for a particular variable, or in some way produces a NPD matrix. If it does still occur without bootstrapping, the problem is in your original data, so I recommend investigating the correlation matrix among your modeled variables.

Terrence D. Jorgensen

Assistant Professor, Methods and Statistics

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