Unfolding the phenomenon of interrater agreement: a multicomponent approach for in-depth examination was proposed.

OBJECTIVE The overall objective was to unfold the phenomenon of interrater agreement: to identify potential sources of variation in agreement data and to explore how they can be statistically accounted for. The ultimate aim was to propose recommendations for in-depth examination of agreement to improve the reliability of assessment instruments. STUDY DESIGN AND SETTING Using a sample where 10 rater pairs had assessed the presence/absence of 188 environmental barriers by a systematic rating form, a raters × items data set was generated (N=1,880). In addition to common agreement indices, relative shares of agreement variation were calculated. Multilevel regression analysis was carried out, using rater and item characteristics as predictors of agreement variation. RESULTS Following a conceptual decomposition, the agreement variation was statistically disentangled into relative shares. The raters accounted for 6-11%, the items for 32-33%, and the residual for 57-60% of the variation. Multilevel regression analysis showed barrier prevalence and raters' familiarity with using standardized instruments to have the strongest impact on agreement. CONCLUSION Supported by a conceptual analysis, we propose an approach of in-depth examination of agreement variation, as a strategy for increasing the level of interrater agreement. By identifying and limiting the most important sources of disagreement, instrument reliability can be improved ultimately.