Effects of Modified Deletion Strategies and Scoring Procedures on Cloze Test Performance

Abstract

Most cloze testing research has centered on the traditional model of deleting every-nth word, accepting only exact word replacements in scoring, and providing blanks of standard length. In the present study, alternative cloze forms derived from selected deletion strategies (every-5th versus total random), scoring procedures (verbatim versus synonymic), and blank conditions (standard versus cued) were analyzed for respective effects on the cloze test performance of 80 college-level readers. Increased performance was generally observed across the every-5th, cued, and synonymic levels of the independent variables. Of the eight alternative cloze formats examined, the total random/cued/verbatim condition generated mean scores most similar to the traditional model while averting, in theory, the sampling biases that can accompany approximated deletions. In this format, context distances of less than four were as effective in producing correct responses as context distances of four, the equivalent of an every-5th word deletion pattern. High positive correlations were found to exist between verbatim and synonymic scoring systems in total, and for each of four individual cloze formats.