Search

DATA FROM THE PARAGRAPH ORGANIZATION PORTION OF THE CEEB ENGLISH COMPOSITION TEST (ECT) WERE CONVERTED TO THE ORIGINAL RANK-ORDER AND WERE THEN RESCORED BY THREE SYSTEMS USING SPEARMAN'S RHO TO DETERMINE WHICH METHOD YIELDED SCORES THAT CORRELATED BEST WITH TOTAL ESSAY SCORES. TWO OF THE METHODS INVESTIGATED, ONE IN WHICH THE NUMBER OF SCORES WAS MULTIPLIED BY THE RANK CORRELATION SQUARED, AND THE OTHER IN WHICH THE NUMBER OF SCORES WAS MULTIPLIED ONLY BY THE RANK CORRELATION, YIELDED CORRELATIONS SLIGHTLY BELOW THE ECT VALUE FOR THE REDUCED SAMPLE. THE THIRD METHOD, USING FISHER'S R-Z TRANSFORMATION AND COMPUTING NZ, YIELDED A CORRELATION WITH THE TOTAL ESSAY SCORE WHICH WAS SLIGHTLY HIGHER. THE IMMEDIATE CONCLUSION DRAWN WAS THAT UNDER THE BEST PREDICTIVE CONDITION, MULTIPLE REGRESSION WEIGHTINGS, THERE IS LITTLE OR NO DIFFERENCE AMONG THE SCORING PROCEDURES. USING TOTAL REARRANGEMENT SCORES TO PREDICT ESSAY SCORES, THE ORIGINAL SCORING IS SUPERIOR. IT WAS DEMONSTRATED, HOWEVER, THAT FACTORS OTHER THAN THE SCORING TECHNIQUES PER SE, INHERENT IN THE DATA AND TESTING CONDITIONS, COULD HAVE SEVERELY RESTRICTED THE PREDICTIVE VALIDITY OF THE TEST. (AUTHOR)