Anonymous posted on Monday, September 23, 2002 - 5:34 pm
I'm trying to estimate a 4-class GMM, with 1 class having no growth and y variables equal to zero across time. The trouble is that the estimated covariance matrix for the Y variables in this class cannot be inverted. Is it possible to get a model like this to estimate??? Any help would be much appreciated!
bmuthen posted on Tuesday, September 24, 2002 - 12:12 am
You can represent a class that is at zero throughout your study by having an intercept factor only, with small variance for the intercept factor and a small variance for the residuals at each time point. If you can't estimate the variances (i.e. if they turn zero or negative), you can fix them at small values such as 0.001.
A different situation arises when there are few individuals who are zero at all time points, but there are many individuals at zero at any given time point. If you are interested in this case, I can send you a draft of a paper.