To score an essay response, raters draw on previously trained skills and knowledge about the underlying rubric and score criterion. Cognitive processes such as remembering, forgetting, and skill decay likely influence rater performance. To investigate how forgetting influences scoring, we evaluated raters' scoring accuracy on TOEFL and GRE essays.…

Best linear prediction (BLP) and penalized best linear prediction (PBLP) are techniques for combining sources of information to produce task scores, section scores, and composite test scores. The report examines issues to consider in operational implementation of BLP and PBLP in testing programs administered by ETS [Educational Testing Service].

Many assessments of writing proficiency that aid in making high-stakes decisions consist of several essay tasks evaluated by a combination of human holistic scores and computer-generated scores for essay features such as the rate of grammatical errors per word. Under typical conditions, a summary writing score is provided by a linear combination…

Essay scores generated by machine and by human raters are generally comparable; that is, they can produce scores with similar means and standard deviations, and machine scores generally correlate as highly with human scores as scores from one human correlate with scores from another human. Although human and machine essay scores are highly related…

This paper proposes an alternative content measure for essay scoring, based on the "difference" in the relative frequency of a word in high-scored versus low-scored essays. The "differential word use" (DWU) measure is the average of these differences across all words in the essay. A positive value indicates the essay is using…

A generic approach in automated essay scoring produces scores that have the same meaning across all prompts, existing or new, of a writing assessment. This is accomplished by using a single set of linguistic indicators (or features), a consistent way of combining and weighting these features into essay scores, and a focus on features that are not…

This report evaluates the construct coverage of the e-rater[R[ scoring engine. The matter of construct coverage depends on whether one defines writing skill, in terms of process or product. Originally, the e-rater engine consisted of a large set of components with a proven ability to predict human holistic scores. By organizing these capabilities…

This paper reports on Educational Testing Service research studies investigating the parameters critical to reliability and validity in both the direct and indirect writing ability assessment of higher education applicants. The studies involved: (1) formulating an operational definition of writing competence; (2) designing and pretesting writing…

Four writing samples were obtained from 638 foreign college applicants who represented three major foreign language groups (Arabic, Chinese, and Spanish), and from 60 native English speakers. All four were scored holistically, two were also scored for sentence-level and discourse-level skills, and some were scored by the Writer's Workbench…

A comprehensive review was conducted of writing research literature and writing test program activities in a number of testing programs. The review was limited to writing assessments used for admission in higher education. Programs reviewed included ACT, Inc.'s ACT™ program, the California State Universities and Colleges (CSUC) testing program,…