Using generalizability theory to investigate the variability and reliability of EFL composition scores by human raters and e-rater

Using the generalizability theory (G-theory) as a theoretical framework, this study aimed at investigating the variability and reliability of holistic scores assigned by human raters and e-rater to the same EFL essays. Eighty argumentative essays written on two different topics by tertiary level Turkish EFL students were scored holistically by e-rater and eight human raters who received a detailed rater training. The results showed that e-rater and human raters assigned significantly different holistic scores to the same EFL essays. G-theory analyses revealed that human raters assigned considerably inconsistent scores to the same EFL essays although they were given a detailed rater training and more reliable ratings were attained when e-rater was integrated in the scoring procedure. Some implications are given for EFL writing assessment practices.