0

When running a latent growth curve model in lavaan (see code below) I get the following warning message: lavaan WARNING: 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.lavaan WARNING: the optimizer warns that a solution has NOT been found!

model = '
    # Intercept with fixed coefficients
    i =~ 1*Score.baseline + 1*Score.wave1 + 1*Score.wave2 + 1*Score.wave3
    
    # Slope
    s =~ 0*Score.baseline + 1*Score.wave1 + 2*Score.wave2 + 3*Score.wave3
    
    #Time invariant covaraites
    #regression of time-invariant covariate on intercept and slope factors
    i ~ var.baseline + Age.baseline + Sex.baseline + Education.baseline + Employment.baseline
    s ~ var.baseline + Age.baseline + Sex.baseline + Education.baseline + Employment.baseline
'

lgcm_model = growth(model, data=data_wide, missing = "fiml")
summary(lgcm_model, fit.measures=T)

The error does not occur when the var.baseline is not included. Also, when running the model with the missing argument set to listwise (but var.baseline still included) the error does not occur.

1) What is causing this error? 2) Is there a way the model can be modified to include the var.baseline predictor while using FIML to reduce the chance of this error occurring?

Aepkr
  • 13
  • 3
  • 1
    If "EM" refers to [tag:expectation-maximization] and the code ends up estimating a Gaussian mixture model, then obtaining a zero variance solution is a common problem: https://stats.stackexchange.com/questions/219302/singularity-issues-in-gaussian-mixture-model. AFAIK, there's no widely-accepted solution. – ForceBru Jan 31 '22 at 12:50

0 Answers0