My sample size is 48. At 4 time points I collected data on 3 variables. For each participant I have 12 scores. For my model, I am testing if the growth across var1 and var2 impact the growth in var3. When I run the model, I get a
"WARNING: THE LATENT VARIABLE COVARIANCE MATRIX (PSI) IS NOT POSITIVE DEFINITE"
It also tells me to check the var2_slope variable. The output indicates that the variance in var2_slope "may be" 0 (-0.012 with p value 0.548).
I'm not sure if I should run the model again and constrain the variance to 0 or remove var2_slope from the model and the implications. If I run it with the constraint, I get an error message involving var2_slope
"THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD NOT BE COMPUTED."
Should I run the constrained the model? If I do, how do I proceed with the error message? If I remove it from the model completely, is this appropriate?
Thanks for the quick reply. When I run the processes separately. Var1 works fine no error messages. For var2 and var3 (var3 is the dependent var) I get the same error messages. "THE LATENT VARIABLE COVARIANCE MATRIX (PSI) IS NOT POSITIVE DEFINITE."
For var2, it tells me there is a problem involving var2_slope. for var3, it tells me there is a problem involving an observed variable, var3_t4 (var3 at timepoint4).
I looked at the residual variance estimates for both of these. Both have values greater than .05. var2_slope: est=-0.012; p=0.548. var3_t4: est=-0.009; p=0.898
The issue with var3_t4 was my second question. I was hoping that the process for addressing var2_slope would help me address the problem with var3_t4.
Should I remove var2_slope from the full model or constrain it to 0?