Using generalizability theory to examine the accuracy and validity of large-scale ESL writing assessment

Using generalizability (G-) theory, this study examined the accuracy and validity of the writing scores assigned to secondary school ESL students in the provincial English examinations in Canada. The major research question that guided this study was: Are there any differences between the accuracy and construct validity of the analytic scores assigned to ESL students and to NE students for the provincial English writing examination across three years? A series of G-studies and decision (D-) studies for three years were conducted to examine accuracy and validity issues. Results showed that differences in score accuracy did exist between ESL and NE students when initial (pre-adjudication) scores were used. The observed G-coefficients for ESL students were significantly lower than those for NE students in all three years, indicating that there were less accuracy and increased errors associated with the writing scores assigned to ESL students. Further, there were significantly less convergent validity in one year and less discriminant validity in all three years of the writing scores assigned to ESL students than to NE students. These findings raise a potential question about the presence of bias in the assessment of ESL students’ writing if initial scores were used.

► Differences in writing score accuracy did exist between ESL and native English (NE) students.
► The observed G-coefficients for ESL students were significantly lower.
► ESL writing scores displayed significantly less convergent and discriminant validity.
► These findings raise a potential question about fairness of ESL writing assessments.