I am fitting a Graded Response Model (ordinal IRT) to evaluate a scale with several questions all using the same 7-point “agreement” scale. The model seems to fit well with no divergences and only a handful of k estimates > 0.7.
The model is:
fmla <- bf(Response ~ 1 + Factor + (1 + Factor |i| Question) + (0 + Factor | ID), disc ~ 1 + (1 + Factor |i| Question))
Factor is the result of a frequentist-based exploratory factor analysis using polychoric correlations between responses to 14 questions.
The pairs plot shows strong correlation between most of the intercepts, other then the last one.
Should I be concerned with the degree of correlation between the intercepts?