Using the Features of Written Compositions to Understand Reading Comprehension

Abstract

AbstractThe current study is an exploration of the association between the characteristics of students' written compositions and their reading comprehension performance. We address the empirical question about the degree to which writing is predictive of reading comprehension by comparing the utility of several popular written composition metrics. These include the 6+1 Trait® Model of Writing, a metric from curriculum‐based measurement in written expression, Coh‐Metrix, and automated counts of word features. The dependent variable for all data analytic models was the reading comprehension score (RC score) from the FAIR‐FS. First, we conducted two random forest regressions, one containing the reading skills, namely, scores on the vocabulary, syntax, word recognition, and reading fluency tasks on the FAIR‐FS and the other containing these reading skills in addition to all of the writing scores calculated from the written composition task. Second, we use the importance indices from the random forest regression to select skills to include in an explanatory regression model. In addition to finding that the writing related skills added to the importance, we further found that the most important skills in the random forest regression were reading‐related followed by the various instantiations of curriculum‐based measures, number of morphemes, and incidence of gerund density. These signify a unique contribution between curriculum‐based measures and reading comprehension when accounting for comprehension‐related reading skills thus furthering our understanding of the interdependency between the components of writing and reading.