A causal model of writing competence

Abstract

Traditionally, assessing competence in English composition involves ignoring most of the steps in the writing process and focusing only on the final submission. The writing process cannot be assessed sufficiently in a traditional setting, and no formal structures exist for improving process-based writing competences. In online distributed learning environments, students’ steps to complete a composition can be recorded at granular levels using content capturing methods. For each step in the writing process, numeric qualities of the text can be determined using natural language processing. These metrics combine into measures of writing competence. The volume of generated data is cumbersome to submit through distributed environments, and time-consuming to process. This research overcomes the difficulty by simulating writing traces from a corpus of completed essays. A writing analytics engine processes the traces, and the resultant metrics are examined using causal inference, producing multiple statistical models of writing competence as systems of interventions.