Abstract: In EFL programs, rating scales used in writing
assessment are often constructed by intuition. Intuition-based scales
tend to provide inaccurate and divisive ratings of learners’ writing
performance. Hence, following an empirical approach, this study
attempted to develop a rating scale for elementary-level writing at an
EFL program in Saudi Arabia. Towards this goal, 98 students’ essays
were scored and then coded using comprehensive taxonomy of
writing constructs and their measures. An automatic linear modeling
was run to find out which measures would best predict essay scores.
A nonparametric ANOVA, the Kruskal-Wallis test, was then used to
determine which measures could best differentiate among scoring
levels. Findings indicated that there were certain measures that could
serve as either good predictors of essay scores or differentiators
among scoring levels, or both. The main conclusion was that a rating
scale can be empirically developed using predictive and
discriminative statistical tests.