Abstract
Theme models were developed for assessing quality of writing within four modes of discourse: argumentative, descriptive, expository, and narrative. Compositions were obtained in each of the four modes at Grades 2, 4, and 6. Using the model themes, three raters rated all themes. In order to locate sources of systematic rater error within the context of these rating procedures, the results were analyzed by analysis of variance (Guilford, 1954) which identifies three sources of rater error: halo, leniency, and rater-trait interaction. At Grade 6, z rater leniency error was found, perhaps reflecting that raters tended to equalize time actually spent in rating themes without proper regard to the time needed to rate themes. It was suggested that attention be focused on rater attitude toward models and the subsequent ratings. Finally, it was noted that the use of different mode models did not affect the variation among raters.