Using Repeated Ratings to Improve Measurement Precision in Incomplete Rating Designs

When selecting a design for rater-mediated assessments, one important consideration is the number of raters who rate each examinee. In balancing costs and rater-coverage, rating designs are often implemented wherein only a portion of the examinees are rated by each judge, resulting in large amounts of missing data. One drawback to these sparse rating designs is the reduced precision of examinee ability estimates they provide. When increasing the number of raters per examinee is not feasible, another option may be to increase the number of ratings provided by each rater per examinee. This study applies a Rasch model to explore the effect of increasing the number of rating occasions used by raters to judge examinee proficiency. We used a simulation study to approximate a sparse but connected rater network with a sequentially increasing number of repeated ratings per examinee. The generated data were used to explore the influence of repeated ratings on the precision of rater, examinee, and task parameter estimates as measured by parameter standard errors, the correlation of sparse parameter estimates to true estimates, and the root mean square error of parameter estimates. Results suggest that increasing the number of rating occasions significantly improves the precision of examinee and rater parameter estimates. Results also suggest that parameter recovery levels of rater and task estimates are quite robust to reductions in the number of repeated ratings, although examinee parameter estimates are more sensitive to them. Implications for research and practice in the context of rater-mediated assessment designs are discussed.