Measuring Agreement Between Two Raters for Ordinal Response: a Model-based Approach