Message/Author 


Hi, I'm trying to fit a parallel process latent growth curve model with a latent moderator as a predictor of one of the slopes. The latent moderator is the intercept multiplied by a latent variable (measured at time one). Basically, this is what I've asked of the model: slopeY on interceptX; slopeY on latentvariable; (main effect) interaction  interceptX XWITH latentvariable; slopeY on interaction; (interaction effect) I get this error message: THE ESTIMATED COVARIANCE MATRIX COULD NOT BE INVERTED.COMPUTATION COULD NOT BE COMPLETED IN ITERATION 16.CHANGE YOUR MODEL AND/OR STARTING VALUES. THE MODEL ESTIMATION DID NOT TERMINATE NORMALLY DUE TO AN ERROR IN THE COMPUTATION. CHANGE YOUR MODEL AND/OR STARTING VALUES. I'm wondering if it's possible to create this kind of interaction (latent variable X latent intercept)? Any help, advice, suggestions appreciated. Thank you! 


This should be possible. Try correlating residuals across processes at each time point. 


Thank you Dr. Muthen. I appreciate your advice. I was correlating the residuals across each process at each time point, but it seems there was another model specification error that I was missing until now. The model is running normally now. Thank you. 


Dear Dr. Múthen, I am trying to estimate a latent growth model with a two way interaction over 4 weeks. I am using 3 continuous variable A C and V (C*V on A over 4 weeks). This is the syntax I am building. Can you please help me by telling me if am I going the right way? VARIABLE: (...) CENTERING = GRANDMEAN (C1C4 Ve1Ve4); DEFINE: C1xVe1 = C1 XWITH Ve1; C2xVe2 = C2 XWITH Ve2; C3xVe3 = C3 XWITH Ve3; C4xVe4 = C4 XWITH Ve4; ANALYSIS: TYPE = RANDOM; ALGORITHM = Integration; MODEL: i s  A1@0 A2@1 A3@2 A4@3; inter  i XWITH C1xVe1@0 C2xVe2@1 C3xVe3@2 C4xVe4@3; s ON i C1xVe1@0 C2xVe2@1 C3xVe3@2 C4xVe4@3 inter; i ON C1xVe1@0; OUTPUT: SAMPSTAT STANDARDIZED TECH1 TECH8 modindices(ALL); 


XWITH is not used in the DEFINE command. What are you trying to do? Which variables are you trying to create interactions for? 


Hello Dr Muthen, My model has 3 continuous variables, A, C and V. What I want is to test whether the intercepts and slopes for the interaction (if any) between C and V over 4 weeks (4 time points)explains variable A intercept and slope over this same 4 weeks. In addition, I want to include 3 control variables (2 continuous and 1 dichotomous. Best regards and thanks in advance Pedro 


C and V are observed variables. You create the interactions as: DEFINE: C1xVe1 = C1 * Ve1; C2xVe2 = C2 * Ve2; C3xVe3 = C3 * Ve3; C4xVe4 = C4 * Ve4; What do you intent with the statement below. inter  i XWITH C1xVe1@0 C2xVe2@1 C3xVe3@2 C4xVe4@3; 


Hello again! I intent to a) test the slope and intercept for the interaction between C and V over 4 weeks (4 time points) and I than b) I want to see if the slope and intercept for this interaction have any effect on the slope and intercept for A over 4 weeks. I think this would be a parallel LGM in which one trajectory regards variable A and the other C*V Does this makes sense? 


So in a), are you saying that you are considering a growth model for the interaction C*V? 


Hi, Yes, that is it! 


Yes, you can do a parallel growth model. Although I don't quite understand how one can think of growth in an interaction variable. 

Daniel Lee posted on Monday, September 01, 2014  7:57 pm



Hi Dr. Muthen, I am fitting a parallel process latent growth model (3 waves) with several covariates (e.g., gender, ses). I would like to extend this model, however, and investigate if the latent intercept and the latent slope of the first latent variable (religiosity) is conditioned on gender as it effects the intercept/slope of my second latent variable (depression). Is this possible? If so, would I just specify in the model that gender will interact with the intercept and slope of religiosity? Thank you! Dan 


You can test the interaction is a multiple group model with gender as the grouping variable or using XWITH. 

Ann Farrell posted on Tuesday, November 19, 2019  7:42 am



Hello, I am trying to see if the interaction between two latent growth curves predict an outcome: iB sB  bul5@1 bul6@0 bul7@1 bul8@2 ; [iB@0 sB]; iM  mach6@0 mach7@1 mach8@2 ; [iM@0]; ibxiM  iB XWITH iM; indirect9 on IB (b1); indirect9 on IM (b2); indirect9 on SB ; indirect9 on ibxim (b3) ; model constraint: NEW (lowbi modbi highbi); lowbi = b1+b3*(1) ; modbi = b1+b3*(0); highbi =b1+b3*1; 1) Is the syntax for the interaction correct? 2) Is the syntax for the model constraint correct to determine significance of simple slopes? 3) Is it correct to set the intercept of growth curves at zero to interpret standardized simple effects? Would the interaction be interpreted at "when one standard deviation above the mean of the intercept of the growth curve (zero)" 4) I saw the sample syntax on the short course topic 3 hand out slide 165. Why are the intercepts of the measured variables at each time point constrained to be equal? Do I need to do this for the interaction? Thank you! 


1) Don't fix the intercept means at zero. 2) Because you have a latent moderator, label its mean and variance in the Model command: [ib] (IBmean); IB (IBvar); and then use that label in the Model Constraint command: lowbi = b1 + b3*(IBmeansqrt(IBvar); modbi = b1 + b3*IBmean; highbi = b1 + b3*(IBmean+ sqrt(IBvar); 3) No. The interpretation is + and 1SD away from the mean of the intercept growth factor. 4) Q1: That was done as one way of treating an interaction (not needed; and note that the intercepts were freed). Q2: No 

Ann Farrell posted on Thursday, November 21, 2019  6:52 am



Thank you very much for your response! That was very helpful. I have switched the moderator variable. I wanted to confirm that the following final syntax is correct? iB sB  bul5@1 bul6@0 bul7@1 bul8@2 ; iM  mach6@0 mach7@1 mach8@2 ; [iM] (IMmean); IM (IMvar); ibxiM  IB XWITH IM; indirect9 on IB (b1); indirect9 on IM (b2); indirect9 on SB; indirect9 on ibxim (b3) ; model constraint: NEW (lowIM modIM highIM); lowIM = b1 + b3*(IMmeansqrt(IMvar); modIM = b1 + b3*IMmean; highIM = b1 + b3*(IMmean+ sqrt(IMvar); 


Looks correct. 

Back to top 