I am using mplus to conduct a multi-process latent class model (i.e., two LCAs with 5 indicators each which are ordinal in nature -- 2 to 5 response categories for each indicator). The model regresses one LCA on the other and also incorporates an interaction effect (allowing the relation of one LCA to the other to vary across gender groups). Based upon earlier warning messages I have increased the number of random starts substantially.
My question -- the models take extremely long to complete (e.g., over 10-15 days). I recognize that the models are complex (e.g., the contingency table that this model represents would be large and have some sparseness in regions). However, I'm wondering if I'm doing something wrong or if there are ways to speed this process up? I have a dual core processor, but am also using my computer for other tasks. In a recent grant proposal, a reviewer questioned the amount of time I was allotting for various analyses and seemed surprised when I indicated the amount of time necessary to complete some of the models I am proposing (which will also involve LTA rather than LCA approaches).