This graphs show Probablity of a true 'Yes' on the hoziontal axis, labeled 0, 0.1, 0.2, and so on up to 1.0, and predicted kappa on the vertical axis, labeled 0, 0.2, 0.4, 0.6, 0.8, and 1.0. There are six curves shown, labeled 99% chance correct, 95% chance correct, 90% chance correct, 80% chance correct, 70% chance correct, and 60% chance correct. The curves all meet at probablity of true yes = zero and kappa = zero and at probablity of true yes = one and kappa = zero. They all rise, at at their maxima at probablity of true yes = 0.5, and then descend. They are all symmetrical about a vertical line through 0.5. The curve labeled 99% chance correct rises most sharply and flattens about probablity = 0.1, remaining almost horizontal until probablity = 0.9. The other curves are progressively more rounded, until the 80% chance correct curve which is almost an arc of a circle, as are 70% and 60%. The maxima for kappa are about 0.95 for 99% chance correct, 0.80 for 95% chance correct, 0.64 for 90% chance correct, 0.35 for 80% chance correct, 0.15 for 70% chance correct, and 0.04 for 60% chance correct.
The graph is divided into horizontal zones with kappa = 0 to 0.2 labeled 'Poor', kappa = 0.2 to 0.4 labeled 'Fair', kappa = 0.4 to 0.6 labeled 'Moderate', kappa = 0.6 to 0.8 labeled 'Good', kappa = 0.8 to 1.0 labeled 'Very good'.
Back to Measurement in Health and Disease: Assessing Agreement Using Cohen’s Kappa.
This page maintained by Martin Bland.
Last updated: 1 May, 2008.