Population Differences in Speed Versus Level of GRE Reading Comprehension: An Exploratory Study
- Wilson, Kenneth M.
- Publication Year:
- Report Number:
- In Gifford, Bernard R. (Ed.), Policy Perspective on Educational Testing. Norwell, MA: Kluwer Academic Publishers, 1992
- Document Type:
- Subject/Key Words:
- English (second language) group differences reading speed test speededness
The concept of differences in "speed" versus "level" of reading comprehension is well established, and tests of reading ability frequently provide separate measures of "speed of reading with understanding" and "level of ability to read with understanding." This study was undertaken to explore possible population differences in speed versus level of GRE reading comprehension, using operational measures computed post hoc from item-level data in GRE files for a pre-October-1977 edition of the verbal test--that is, a version in which 40 GRE reading comprehension (RC) items were included as a separately timed section administered under then current formula-scoring instructions. The "level" (formula) score was defined by performance on the first 20 RC items (RC1), completed by almost all examinees, and the "speed" (formula) score was based on the second 20 items (RC2)--the number of RC2 items "attempted" ranged from 0-20. RC1, RC2, and a formula score based on 28 odd- numbered discrete-verbal items (DVodd) were z-scaled in data for more than 21,000 examinees tested in a scheduled administration of the GRE in October 1976. Patterns of differences between correlated z-scaled means (mean RC2 minus mean RC1) were analyzed for (a) U.S> examinees classified by sex, ethnicity, English-language communication status (English primary language, or EPL, versus English second language, or ESL), and four broad graduate major areas, and (b) non-U.S. examinees classified by sex, language status, and academic area. It was considered plausible that speed/level differences would be present in analyses by graduate major area and by EPL/ESL status, but not in analyses by sex or by ethnicity. Patterns of relative standing on RC2 and RC1 were generally consistent with hypothesis: RC2 > RC1 (higher standing on speed than on level) for humanities and social science majors, and non-U.S. EPL examinees; RC2 < RC1 for physical science majors and bioscience majors, and both U.S. and non-U.S. ESL examinees. For all U.S. ethnic groups studied (African American, Asian American, Hispanic American, and White American), the basic pattern was as expected, namely, RC2 = RC1, and this pattern tended to obtain for males and females. Criterion-related validity for RC1 and RC2 was explored, using self-reported undergraduate GPA (SR-UGPA). RC2 tended to be more highly correlated than RC1 with the SR-UGPA criterion, for subgroups of U.S. examinees, except for Hispanic examinees and ESL examinees. For these two partially overlapping subgroups, and for subgroups o non-U.S. examinees, coefficients for RC1 (level) were consistently larger than those for RC2 (speed). The findings suggest that further exploration of the role of "speed" in measures of GRE reading comprehension, and in other GRE ability measures, is warranted.