I want to fit a univariate ALT model on continuous data with 8 timepoints. I have used the syntax below, when I run it I get an error concerning matrix psi. I inspected Tech4 and it turns out that my observed variables are treated as latent variables; there are estimated means for these observed variables and they appear in the latent variables covariance matrix. Could you indicate what I have done wrong in the syntax? Thank you.
title: ALT model; data: file is data.txt; variable: names are y2-y8; analysis: estimator=ml; model: i s | y3@1y4@2y5@3y6@4y7@5y8@6; y3 on y2; y4 on y3; y5 on y4; y6 on y5; y7 on y6; y8 on y7; i with y2; s with y2; savedata: file is FscoresALTmodel.txt; save=fscores; output: modindices; tech4;
There is nothing wrong with the observed y variables ending up as latent variables (perfectly measured). That's just how Mplus represents the model when there is a direct effect pointing to a y variable.
Regarding the TECH4 non-pos-def message, look at the variances to see if any are negative and look at the correlations to see if any are close to 1.
Thanks for your reply. I checked and it turns out the variance of the slope is negative but very close to zero. Can I interpret this finding as meaning that for this particular population, there is no need for allowing the slope to differ across individuals? And if so, am I right to fix the variance of the slope to 0 (i@0) so that each individual has the same slope, namely the mean? Thanks again.