Relations Between Cohesion in Constructed Responses and Individual Differences in Reading Literacy Skills (Poster 6)

• In Event: Leveraging Constructed Responses to Understand Text Comprehension

Abstract

1. Objectives. Computational approaches to analyze CRs made during reading have promise to shed insights into individual differences that are related to comprehension outcomes (Authors, 2003). Computational approaches may be sensitive to features of CRs that are reflective of coherence building, which may be difficult to assess based on expert coding which tends to focus on the presence and quality of comprehension strategies. Specifically, computational approaches are adept at assessing the cohesion of a set of protocols produced over the course of reading a text (Authors, 2015). Cohesion refers to the ways in which ideas are linked (Graesser et al., 2011). Cohesion analyses involve assessing the degree to which CRs overlap with each other. In other words, this approach examines the degree to which readers establish connections across the content within their CRs. In this poster, the research team will present data from two recently published studies that explore the extent that different approaches for assessing the cohesion of CRs for texts are related to individual differences in foundational skills of reading, reading skill, and working memory (Authors, 2021; Authors, in-press).

2. Perspectives. There are two theoretical perspectives that guide this research. This study is grounded in theories of comprehension that assume that comprehension is based on the coherence of a mental model for a text (Authors, 2009b). Additionally, reading systems perspectives of reading assume that the efficiency of foundational skills of reading and reading skill affect coherence building processes (Perfetti & Stafura, 2014).

3. Method. CRs were analyzed to assess in a variety of ways to determine their cohesion. One approach involved assessing how readers established connections with themselves (i.e., to other CRs they produced). The other approach involved assessing connections between the CRs and the texts that were read. Additionally computational measures of lexical (use of word matching algorithms) and semantic (use of high dimensional semantic spaces) overlap were employed. Measures of individual difference in the vocabulary knowledge, reading skill, and working memory were administered. The poster will highlight both linear modeling and machine learning approaches to using computational measures to predict individual differences

4. Data Sources. Archival data for CRs collected from college readers across two and four year institutions was used in these studies (n = 119 for Authors, 2021; n = 560 for Authors, in-press).

5. Results. Across the two studies, the results showed with both analytic approaches that computationally measure of cohesion are predictive of vocabulary knowledge, reading skill, and working memory capacity. Authors-a showed stability in these relationships across multiple texts and with participants from two- and four-year institutions, whereas Authors-b found differences as a function of text type. The stability of the finding will be discussed.

6. Significance. These findings are consistent with theoretical frameworks that assume that the proficiencies in foundational skills of reading affect coherence building. Moreover, they indicate that theories of comprehension that have typically ignored the role of individual differences in comprehension should be revised to account for them.

Authors

• Lauren Flynn, University of Minnesota
  Presenting Author

• Joseph Magliano, Georgia State University
  Non-Presenting Author

• Laura K. Allen, University of Minnesota
  Presenting Author