The purpose of this study was to determine if leniency and halo error in student ratings could be reduced by training the student raters and by using a Behaviorally Anchored Rating Scale (BARS) rather than a Likert scale. Two hypotheses were proposed. First, the ratings collected from the trained raters would contain less halo and leniency error than those collected from the untrained raters. Second, within the group of trained raters the BARS would contain less halo and leniency error than the Likert instrument.

This research clarified existing inconsistencies in self-assessment literature, and added to the body of knowledge for rating effectiveness of organizational assessments by defining relationships among rating effectiveness criteria (ratee, rater, rating scale, rating process) and measures (interrater reliability, halo error, leniency and severity, range restriction) based on extensive literature review. A research framework was developed from this review, and was employed in computing rating effectiveness measures at the individual (i.e., examiner or eight rating scale dimensions) and sector (e.g., Private Manufacturing Sector, Private Service Sector, Public Local Sector, Public State & Federal Sector) levels for a State Quality Award (SQA) using data from the 1998 applications. Interrater reliability (measured by intraclass correlations for each rating scale dimension) was low to moderate, and differed by dimension. Halo error (measured by the determinant of the dimension intercorrelation matrices for each examiner) was present for all examiners. Leniency and severity (measured by presence of statistically significant Rater main effect for each dimension) was present for 11 of 32 cases, and differed by dimension. Range restriction (measured by variance analysis for each dimension) was present for 22 of 32 cases, and differed by dimension. A post-hoc principle component analysis indicated poor internal reliability for the rating scale. To improve, the SQA should replace the existing rating scale and provide in-depth training on all elements of the rating process. The importance of the SQA using boxplots, histograms, and rating effectiveness measures to make fully informed decisions was discussed. / Master of Science

This study focuses on how the presence or absence of an overall rating item on a performance evaluation form affects levels of halo error and satisfaction with the form. Participants included undergraduate college students who were randomly assigned to groups receiving a form with or without an overall rating item at the beginning of the form. A satisfaction item was included on both forms. The analyses included a z-test for correlations from independent samples to determine the difference between the two evaluation forms and a t-test to determine the difference between the satisfaction scores of the two forms. The analyses indicated that the differences between the groups were not significant for levels of halo error or satisfaction.