Fairness of Automated Essay Scoring of GMAT® AWA

Overview

This study investigates the fairness of the automated essay scoring from the Analytical Writing Assessment to six subpopulation groups of Graduate Management Admission Test® (GMAT®) test takers: American English vs. non-American English writers, English native speakers vs. English-as-asecond-language speakers, males vs. females, and examinees of three different ethnic groups. Propensity score matching was used to create control groups by matching each member of the studied groups on multiple variables. The study shows that none of the subpopulation groups has an unfair advantage and none has been unfairly punished by the automated essay scoring.