Investigating the Construct Validity of an ESL Test for Young Learners

Abstract

This study investigated the construct validity and reliability of the GEPT-Kids, which is composed of 25 listening items and 30 reading items. Data were collected from 742 test takers from Grades 5 and 6 from eight Chinese-speaking schools in Beijing, Shanghai, Guangzhou, Hong Kong, Macau, Taipei, Taichung, and Kaohsiung. Two CFA models (Model 1: a one-factor model with a general language proficiency construct; Model 2: a two-factor model with listening and reading constructs based on the test design) were proposed and analyzed using AMOS 24.0 to reveal whether the internal structure of the test reflected the test developers’ design for the test. The results showed Model 1 provided an adequate explanation of the data as most of the model fit indices were acceptable, and Model 2 showed that there was a lack of discriminant validity between listening and reading because they were too highly correlated although the fit indices were high. While both models can be considered acceptable based on fit indices, and Model 1 is more parsimonious than Model 2. To sum, the main finding was that GEPT-Kids is valid in terms of score interpretation of young learners’ overall language ability but may lack discrimination in terms of assessing subskills of English language proficiency. Therefore, findings from this study suggest that GEPT-Kids is acceptable as a general language proficiency test but should be used with caveats while using test scores in terms of listening and reading in the classroom teaching-learning context.