When do you use Cohen's kappa (range 0 to 1)?

• A. to measure the level of agreement between 2 or more sets of test results with qualitative outcomes
• B. to measure the linear association between two quantitative tests
• C. to assess test-retest reliability over time
• D. to evaluate sensitivity of a diagnostic test

Answer: (A)

Cohen's kappa is used to quantify how much two raters (or classifiers) agree on categorical, qualitative outcomes beyond what would be expected by chance. The idea is to separate genuine agreement from what would happen just by luck. You look at how often the raters coincide on a category and compare that to the probability they would agree if they were guessing based on their individual tendencies. The resulting score typically falls between 0 and 1, where 0 means agreement no better than chance and 1 indicates perfect agreement.

This is the right situation when you have two people (or systems) classifying items into a limited set of categories and you want a single summary of their concordance that accounts for chance. It’s not about linear relationships between numerical scores—that would be correlation. It’s not about a test’s sensitivity to detect disease, which depends on true disease status. And while kappa can relate to reliability across raters, its purpose is specifically to measure agreement between raters on categorical classifications, not the stability of a measurement over time.