Magnitude and moderators of bias in observer ratings: A meta-analysis

Bias in observer ratings compromises generalizability of measurement, typically resulting in attenuation of observed associations between variables. This quantitative review of 79 generalizability studies including raters as a facet examines bias in observer ratings in published psychological research and identifies properties of rating systems likely to place them at risk for problems with rater bias. For the rating systems studied, an average of 37% of score variance was attributable to 2 types of rater bias: (a) raters' differential interpretations of the rating scale and (b) their differential evaluations of the same targets. Ratings of explicit attributes (e.g., frequency counts) contained negligible bias variance, whereas ratings of attributes requiring rater inference contained substantial bias variance. Rater training ameliorated but did not solve the problem of bias in inferential rating scales.