
Message/Author 

Kira McCabe posted on Monday, February 02, 2015  8:43 pm



Hello, I am running a parallel process model with extraversion and drinking behavior over 4 time points. When I first run the separate, unconditional models, extraversion decreases over time and drinking behavior increases over time. However, once I run the parallel process, the trajectory for drinking appears to change. In the intercepts part of the output, the slope for drinking behavior is negative. When I look at the tech4 output, the mean slope for drinking behavior is still positive. My questions: 1) When interpreting the effects of the intercept on the drinking slope (or any other timeinvariant covariates), should I interpret it with the estimate in the intercepts or with the mean in the tech4 output? In this case, the intercept of extraversion had a negative relationship with the drinking slope, so extraversion predicted slower decreases in drinking over time? 2) A followup question: Can a growth model change from significant to nonsignificant when a parallel process is added? Applying the same example, let's say the unconditioned model showed a significant increase in drinking over time. However, when the growth model of extraversion is added, the intercepts section shows that the drinking slope estimate is no longer significant. When reporting a parallel process, should this drinking slope remain significant in the parallel process model? Thank you for your help! 


1) You should go with the mean in TECH4. See also the plot that you get. 2) The significance should be judged by the TECH4 results. 

Kira McCabe posted on Tuesday, February 03, 2015  5:11 pm



Hello Bengt, Thank you for your reply. I am using an older version of Mplus (Version 7), and I noticed that the tech4 significance results are in the 7.2 update. Is there a way to calculate this information with the variancecovariance matrix? Or is this too difficult all the latent variables in the model? Thank you for your help! 


You can express the tech4 quantities in Model Constraint and thereby get the signif tests, but it may be cumbersome if the model is complex. 

lamjas posted on Thursday, April 09, 2015  8:58 pm



Hello, I am running a parallel process LGM. Both IV and DV have six timepoints. The model with quadratic terms of IV and DV fit well. I am wondering whether I should add the following effects in the model: (1) IV_i > DV_q (2) IV_s > DV_q (3) IV_q > DV_q Are the above effects interpretable in some ways? Or, would you suggest that a model with the following effects is good enough? IV_i > DV_i IV_i > DV_s IV_s > DV_s Thanks for you help! 

Jon Heron posted on Friday, April 10, 2015  6:00 am



Interpreting quadratic terms is usually difficult but the decision to add these terms will depend on your particular application. As an alternative you might consider: (1) A twopart linear spline model (2) Relaxing some of your loadings so the nonlinearity is absorbed by a distortion of the time axis. 

xybi2006 posted on Wednesday, August 05, 2015  10:02 am



Dear Dr. Muthen, I have 5 time points data, with baseline, 1 years, 4 years, 10 years, and 23 years. I ran the LGM with unspecified time points (0, 1 for the first 2 time points, and then estimate the other threes): for one variable, the estimated time points are: 0, 1, 3.646, 4.603, 8.523); and for another variable, the estimated time points are: 0, 1, 1.791, 2.321, and 1.969. Given the estimated time points are so different for the two variables, can I proceed to run the parallel process LGM? Thanks much, 


If these are two separate processes, like height and weight, for example, it seems unlikely that the estimated time scores would be the same. 

Back to top 

