My colleagues and I are attempting our first foray into growth modeling with longitudinal data on the relationship between intention to quit, quit attempts, and smoking status (N=4600). Here are our dependent variables:
Smoking status: (dichotomous) former smoker or current smoker Quit attempts: (categorical or continuous) Intention to quit: (ordinal -- three categories)
This data was collected over 4 time points.
We are thinking of using an approach similar to Figure 6.13 in the MPlus manual (pg 119 in Version 6), but adding a third parallel process. however, we have the added challenge that smokers who quit at T2 were not asked the intention to quit or quit attempts questions, so we have, in effect, censored data, though the former smokers could return to smoking at a later time point and answer the intention to quit and quit attempts questions. We've thought of splitting up our models so that one is just for the first two time points (everyone is a smoker at T1), and the second model only includes smokers who didn't quit at T2.
Is this the best way to approach this problem? Any advice you have would be greatly appreciated. We've been looking for articles in our field that have taken a similar approach and have yet to find anything.
Sounds complicated to try to juggle all 3 processes simultaneously and the growth function is not clear. Perhaps you are interested in a discrete-time survival analysis of quitting smoking, where you can simplify to first quitting (not accounting for them going back to smoking). Then you can have one of the other processes as a time-varying covariate that influences the quitting probability.