The Potential for Automated Text Evaluation to Improve the Technical Adequacy of Written Expression Curriculum-Based Measurement

Abstract

Written-expression curriculum-based measurement (WE-CBM) is used for screening and progress monitoring students with or at risk of learning disabilities (LD) for academic supports; however, WE-CBM has limitations in technical adequacy, construct representation, and scoring feasibility as grade-level increases. The purpose of this study was to examine the structural and external validity of automated text evaluation with Coh-Metrix versus traditional WE-CBM scoring for narrative writing samples (7-min duration) collected in fall and winter from 144 second- through fifth-grade students. Seven algorithms were applied to train models of Coh-Metrix and traditional WE-CBM scores to predict holistic quality of the writing samples as evidence of structural validity; then, external validity was evaluated via correlations with rated quality on other writing samples. Key findings were that (a) structural validity coefficients were higher for Coh-Metrix compared with traditional WE-CBM but similar in the external validity analyses, (b) external validity coefficients were higher than reported in prior WE-CBM studies with holistic or analytic ratings as a criterion measure, and (c) there were few differences in performance across the predictive algorithms. Overall, the results highlight the potential use of automated text evaluation for WE-CBM scoring. Implications for screening and progress monitoring are discussed.

Validity of Automated Text Evaluation Tools for Written-Expression Curriculum-Based Measurement: A Comparison Study

“…First, the findings of greater validity and diagnostic accuracy for all scoring approaches when based on three versus one screening sample per student build on prior generalizability theory studies demonstrating that multiple writing samples are needed for adequate reliability (Keller-Margulis et al, 2016;Kim et al, 2017). Second, the findings of comparable validity and diagnostic accuracy across complex (Wilson et al, 2016) and (b) comparable relations to holistic quality for 7 min samples scored with free automated text evaluation and complex WE-CBM metrics (Mercer et al, 2019). These findings illustrate that it may be difficult to improve validity of automated text evaluation beyond complex WE-CBM scoring; instead, investigating the optimal number of writing samples and writing duration may yield more benefits.…”

“…Consistent with recent calls for adopting open science practices in special education (Cook et al, 2018), developing and evaluating open-source tools can facilitate critical review and refinement of scoring models, replication efforts by other research teams, and greater usage of tools by removing cost barriers. For example, Mercer et al (2019) examined the validity of scoring models based on Coh-Metrix (Graesser et al, 2014), a free, but proprietary, tool for originally designed to predict reading comprehension difficulty of texts, relative to WE-CBM scores for 7-minute narrative writing samples from students in second through fifth grade. Results indicated that composite scores based on Coh-Metrix could predict raters' holistic quality ratings on the screening samples, both for the samples used to generate the Coh-Metrix scores and on similar samples collected 3 months apart, and that correlations with holistic quality were similar for composites based on Coh-Metrix (r = .73 -.81) and WE-CBM scores (r = .74 -.77).…”

“…Coh-Metrix calculates 108 indices in 11 categories: descriptives, text easability principal component scores, referential cohesion, latent semantic analysis, lexical diversity, connectives, situation model, syntactic complexity, syntactic pattern density, word information, and readability. Prior work has found Coh-Metrix indices useful in predicting writing performance in elementary (Mercer et al, 2019) and middle school (Wilson et al, 2017) students.…”

“…Efforts to improve writing skills hinge on the ability to identify those students who are at-risk for poor performance and to measure their response to changes in instruction (McMaster et al, 2020); however, existing approaches to scoring student writing samples are either too elaborate for use in screening or do not offer the technical adequacy required for decision-making about student risk status (e.g., Gansle et al, 2002;McMaster & Espin, 2007). Advances in automated text evaluation may offer a viable alternative to address these measurement limitations (Mercer et al, 2019). The purpose of this study was to examine and compare the validity and diagnostic accuracy of several automated approaches to scoring written expression screening samples to determine whether automated scoring yields scores with improved technical adequacy.…”

Design, Implementation and Evaluation of a Support System for Educators and Therapists to Rate the Acquisition of Pre-Writing Skills

“…As the theory developed further, computational tools such as Coh-Metrix were presented to produces indices of the linguistic and discourse representations of pieces of text. Mercer et al [5] studied the validity of automated text evaluation with this tool -using a number of algorithms to train different models-and found it generally valid for written-expression curriculum-based measurement.…”