Am trying to run a two-part (categorical/continuous growth) model over 3 developmentally distinct periods (late adolescence 4 waves, emerging adulthood 5 waves, and early adulthood 3 waves). However, I get a memory error. Am assuming that the model is too large? If this is the case what might be some acceptable alternative ways to run the analyses? Have attmpted to run as a single "piece" with all waves but this model has not fit the data (at least using AMOS) and when it comes to gateway substances. Thoughts?
It sounds like you require numerical integration to estimate your model if you get a memory message. If this is the case, I would fix the variances of all but the intercept growth factor to zero and use INTEGRATION=7; If this is not the case, please send your input, data, output, and license number to firstname.lastname@example.org.
I'm doing a growth mixture model on depression data in the pregnancy/postpartum period, investigating differential onset and time courses for postpartum depression. I have seven data collection points, four during pregnancy and three postpartum. I have done a 2-piece (linear) piecewise model with a transition point right before birth. The problem is that the postpartum slope then includes one pregnancy-point, which is not in accordance with our theoretical model, as I would like to have separate slopes for the pregnancy and postpartum time points. It would then make good sense to have another transition point right after birth as well, making it a three-piece model. This however means that I estimate the linear slopes with four, two and three time points respectively. When I try this I get good fit, the classes and growth factors make good sense etc. But what are the implications of only including two data points for the middle slope, can we then not trust the model as a whole or are the problems restricted to the middle slope? Is there a better way of solving the problem of separating slopes before and after birth in a gmm-model?