Central tendency and matched difference approaches for assessing interrater agreement.

In Study 1 of this two-part investigation, we present a “central tendency approach” and procedures for assessing overall interrater agreement across multiple groups. We define parameters for mean group agreement and construct bootstrapped confidence intervals around the mean population parameters for rWG, AD, and ICC(1). In Study 2, we extend assessments of overall interrater agreement by developing a “matched difference approach” and procedures for assessing real versus pseudo agreement in a sample of groups. Here, we use random group resampling and the matched difference between assessments of the respective rWG, AD, and ICC(1) values for actual and pseudo groups, with the establishment of bootstrapped confidence intervals around such differences. In both studies, we employ simulated and real data to demonstrate the accuracy and practical utility of the new procedures for assessing agreement with respect to groups. Notably, to generate simulated data for Studies 1 and 2, we developed a new underlying model for multilevel data and procedure for data generation, and we discuss its potential utility for enhancing research in group-level studies. Moreover, we discuss, relative to current practices, how and why the new inference procedures provide information about mean interrater agreement in the population, which can improve data aggregation decisions and interpretations of findings from group-level studies. (PsycINFO Database Record (c) 2018 APA, all rights reserved)