lavaan and indicators without error

[scorla]

Hello guys!

I'm trying to do a CFA for simulated data. This data has two latent variables with three indicators each, whereas each indicator has a factor loading of 1.

My model looks like this:

"f1 =~ v1 + v2 + v3
f2 =~ v4 + v5 + v6
f2 ~ f1"

In certain cases, all indicators for f1 and/or f2 are generated without error. Unfortunately, lavaan gives the following error when trying to run a CFA:

sample covariance matrix is not positive-definite

When I fix the residual variance of all indicators of f1 and/or f2, I get another error message:

initial model-implied matrix (Sigma) is not positive definite;
check your model and/or starting parameters.

Is lavaan simply not able to deal with data that comes without measurement error?

[Yves Rosseel]

On 7/13/20 12:45 AM, scorla wrote:
> sample covariance matrix is not positive-definite

Note that you created the sample covariance matrix (in the simulation).

You should check this (eg by checking that all eigenvalues are positive)
before you call lavaan.

Yves.