It is often to see "THE LATENT VARIABLE COVARIANCE MATRIX IS NOT POSITIVE DEFINITE." when we work with Structural Equation Modeling. What does it mean? Why does it happen?
I have run into this (for factor analysis) when working with missing data and when the correlation matrix was computed using pairwise deletion. My understanding is that a set of correlations can be internally inconsistent/impossible in this situation.
Covariance Matrix is not positive definite means the factor structure of your dataset does not make sense to the model that you specify. The problem might be due to many possibilities such as error in data entry, duplicate input, multiple entry etc.