Modelling errors in scoring the darkness of staining

Abstract

Objective: To construct a statistical model for the agreements and disagreements between two observers on darkness of staining. Study Design: Data from an earlier observer-agreement study by van Diest et al were reanalyzed. Results: A model in which the random variation in error is permitted to depend upon the true darkness of staining was found to fit the data much better than does one in which the random variation is constant. Conclusion: For the dataset analyzed, error tends to be greater (that is, correlation between observers tends to be less) when staining is darker.