Assessing Comprehension beyond Recognition: A Many-Facet Rasch Measurement Approach

Abstract

Purpose: Comprehension is commonly assessed through single-task tests, particularly multiple-choice questions (MCQs). Although MCQs offer many advantages, a growing number of researchers have raised concerns that such measures may overestimate degree of understanding because they are based on recognition. A less-frequently used method are summaries, which are assumed to reflect a higher level of comprehension because they require learners to select and integrate information in order to create a mental representation of the input. This study proposes a combined comprehension measure that integrates summaries (assessing global comprehension) and MCQs (targeting detail-level comprehension) into a single measurement system. The purpose of the study is to collect validity evidence for the use of the combined measure through many-facet Rasch measurement (MFRM).

Methodology: Listening data were longitudinally collected from 290 EFL Japanese high school students over three separate waves, with each involving three measurement points (i.e., Pretest, Posttest 1, and Posttest 2). Comprehension was assessed twice at each measurement point through the combined measure consisting of a summary and a set of five MCQs, which were administered in paper-and-pencil format. The summaries were rated by two expert raters using a five-point rating scale in tandem with a list of main ideas and details that had been previously extracted from the target texts. The MCQs were dichotomously scored. All three waves of data were linked through a Rasch stacking design and were analyzed using MFRM with the analysis involving three facets: persons, items, and raters. Under the theoretical assumption that summaries are more difficult and entail a higher level of comprehension than MCQs, the summaries were given double weight when estimating learner ability.     

Findings: The Wright map confirmed a hierarchy of item difficulty consistent with the theoretical expectation that the summaries were more difficult than the MCQs, providing support for the weighting scheme. The persons showed acceptable fit to the Rasch model with most participants being within parameters. Similarly, all summaries and multiple-choice items fit the Rasch model’s expectations with the exception of only two multiple-choice items, which were slightly above the recommended criteria. The analysis revealed fair person reliability and excellent item reliability, suggesting that the replicability of the person ability and the item difficulty hierarchies were fair and high, respectively. In addition, rater severity did not negatively impact the measurement and the response thresholds suggested that the rating scale functioned as intended. These findings indicate that the combined measure is a valid instrument for comprehension assessment.

Unique Contribution to Theory, Practice and Policy: Practically, this study contributes to the literature by providing a combined measure that mitigates the weaknesses of summaries and MCQs when used separately. In addition, it demonstrates how MFRM can model productive and receptive tasks, which may be differently weighted tasks, within a single measurement system. Regarding policy, this study advocates for tests that move beyond single tasks to provide a more precise picture of learners’ levels of comprehension across different educational settings.