sunny shi posted on Thursday, February 05, 2009 - 5:17 pm
I am trying to fit first order ARMA(1,1)(autogressive and moving average)model to the latent growth modeling with time invariant covariate. There are four time points with outcome variable measured as y1,y2,y3,y4 and time nvariant covariate x. The ARMA (1,1) model looks like the following: y1 = 1*INT + 0*slope+ error1, y2 = 1*INT + 1*SLP + ARlag1*y1 + MAlag1* error1+ error2, y3 = 1*INT + 2*SLP + ARlag1*y2 + MAlag1* error2+error3, y4 = 1*INT + 3*SLP + ARlag1*y3 + MAlag1 error3+ error4, INT = Mean_INT+ beta1*X +disturbance1, SLP = Mean_SLP+ beta2*x+disturbance2;
The sample Mplus program for the model statment is:
time 1: V(error1) time 2: MAlag1^2*V(error1)+V(error2) etc
which means that you have a structure for these residual variance parameters. I think you can handle that using the Constraint= option with MAlag1^2 in the VARIABLE command together with Model Constraint - see the User's Guide.