I have a questionnaire administered at two time points. I would like to estimate a latent factor score (theta) at each administration using Item Response Theory. The ideal approach would be to model the change directly (e.g. Longitudinal item response theory models in R), but my items are too highly correlated to estimate a solution.
As a workaround, I could merge time 1 and time 2, estimate the latent trait score for each unique response pattern, then fill the appropriate values for each participant at time 1 and time 2 (based on the response pattern).
I suppose that merging the two can affect the estimation of the the latent scores (which may or may not have important consequences). Any other potential downsides to this approach?