Dear all SAS experts, My dataset is a longitudinal one, comprising 7 time points; for each time point, we count the number of online posts under certain topics for 800 users. I intend to discover whether there are different patterns of change in the number of posts across the 7 time points among these users. so, this will be an LCGA (Latent Class Growth Analysis) or GBTM (Group-Based Trajectory Modeling). Exploratory analysis found that our data are overdispersed (variance>>mean) and contained excess zero, and it fit negative binomial (NB) or zero-inflated negative binomial (ZINB) distribution better (preliminary analysis using MPLUS showed that NB and ZINB perform similarly, and both are better than ZIP). For the ease of handling time-varying covariate, I would like to use Proc Traj for GBTM. However, Proc Traj does not support an NB/ZINB model. So, I wonder if there are any ideas on how I can handle my NB data using Proc Traj? I found that some studies used ZIP as an alternative for ZINB (see https://journals.sagepub.com/doi/10.1177/154405910808700112), is this appropriate? Thanks for your answer in advance!
... View more