I have a question for the mathematicians and statisticians. If I want to measure the reliability of agreement between multiple observers, with multiple nominal categories, how should I proceed? Should I use weighted kappa coefficient or perhaps Fleiss's kappa? Any ideas?