

Parallel process for nontimestrucut... 

Message/Author 

Anonymous posted on Friday, February 27, 2004  1:59 pm



I am interested in doing a parallel process growth model like the one in example 22.3 on p.215 of the user's guide. However, I have nontimestructured data. I am wondering if TYPE RANDOM like example 5 on p.15 of the addendum can handle parallel process. In particular, does TYPE RANDOM allow regressions among the random effects in the two processes? Thanks. 


A model can have parallel processes with individuallyvarying times of observation and the random effects in the two processes can be correlated. 


How is the intercept term interpreted when a growth model is estimated with individuallyvarying times of observation? And, as a related point, how does one calculate predicted values for the growth trajectories when the times of observation vary across individuals? 

bmuthen posted on Monday, July 05, 2004  9:31 am



The intercept is still interpreted as the systematic part of the growth model at the time when the time variable = 0. A typical example would be a treatment study where time is the number of weeks after treatment. So there is a single time variable measured in weeks, where different individuals have different values on this variable for each of the followup occasions after treatment. For example, the first followup (the second occasion) might be in week 1 for some, week 2 for most, and week 3 for some. Here, 0 would define the intercept at the treatment point. But you could also have the intercept be defined say at 8 weeks, in which case you would subtract 8 from each individual's time variable. Note that some individuals would then not have the time value 0 since not all individuals are observed at exactly week 8, but that's ok, the intercept is still defined at week 8. As for average predicted values of the outcome, you still use the time variable (weeks in the example)  it doesn't matter that specific individuals are observed at different times. To check observed vs fitted curves for individuals, however, it makes sense to use the times the individual was actually observed. 


I am sorry that I cannot find a better suited thread for this question. Can the AT command be used to model the intercept and slope of a within individual covariance relationship that is unrelated to time? So, for example, if we were to ask subjects to rate themselves on a trait, and also ask them to rate how much they value that trait, could we arrive at a depiction of that relation for each individual by using..... i s  trait1 trait2 trait3 AT importance1 importance2 importance3 ? 


That's certainly ok. The model simply specifies a mean and covariance structure. 

Back to top 

