Using change scores to examine dynamics in reading research

Using change scores to examine dynamics in reading research

First Author: Lee Branum-Martin -- Georgia State University
Keywords: Growth Modeling, Structural equation modelling, Language Development, Spanish, African Americans
Abstract / Summary: 

Dual change score models or latent difference scores can accommodate nonlinear individual processes, but have not seen wide application in reading research. Bivariate and trivariate extensions can capture differential change across constructs as leading or lagging influences. These models open interesting questions not only of co-development, but of treatment, summer learning, bilingualism, and disability. The current symposium highlights innovative applications in language and reading. The first presentation gives an overview of change score models, their relation to standard growth models, and an application of a bivariate model for reading real words versus nonwords. The second presentation examines summer setback for students learning Spanish and English. The third presentation examines intervention effects for language and self-regulation. The fourth examines school differences in the effect of African American dialect on the co-development of reading and writing. The fifth presentation examines performance discrepancies in three phonological awareness skills for students with disabilities.

Symposium Papers: 

Latent change score modeling: Applications for reading research

First Author/Chair:Jamie Quinn -- Florida State University
Additional authors/chairs: 
Laura Steacy; Donald Compton

Purpose: The purpose of this presentation is to provide a brief overview of latent change score (LCS) modeling and its applications for reading research. LCS modeling is used to investigate the dynamic relations of co-developmental processes. In this presentation, I will describe LCS modeling to prepare researchers to apply this method to their own longitudinal research data and to provide an introduction for the presentations to follow. 

Method: I will cover the statistical application of univariate and bivariate LCS modeling to repeated measures data. The history and relevant recent applications of LCS modeling will be discussed. I then will briefly examine data structure assumptions, describe typical coefficients in a LCS model, and will introduce vector field plots. Using data from a four-year cohort-sequential longitudinal study, I will fit and describe a bivariate LCS to repeated measures of real-word reading and nonword reading in a sample of n = 1,387 students to investigate leading and lagging effects of these two important reading skills.

Results: The bivariate LCS model fit to the longitudinal data on real and nonword word reading will be discussed. Model parameters, univariate growth figures, and bivariate growth figures (i.e., vector plots) will be presented and explained.

Conclusions: Bivariate LCS modeling is a useful tool for studying the co-development of related constructs. Trivariate models will be briefly introduced in preparation for the remaining presenters. Attendees will leave with an understanding of LCS modeling and how this method can be used to study the development of reading-related skills and abilities.

Developmental changes in language proficiency of bilingual children during the academic year and summer

First Author/Chair:Yusra Ahmed -- University of Houston
Additional authors/chairs: 
David Francis; Shiva Khalaf; Jeremy Miciak

Purpose: The effect of summer gap has been studied extensively with respect to changes in students’ oral language proficiency and early literacy skills. Although many studies include English language learners (ELs), they fail to examine this population in relation to the language of instruction and language use at home. This study investigated the growth in vocabulary, listening comprehension and verbal analogies for students in English vs. Spanish programs.

Method: We modeled growth in oral language during the academic year and drop during the summer. The sample included 3,946 ELs attending a bilingual (n = 2,649) or English (n = 1,297) instructional programs in grades K-2. Listening comprehension, picture vocabulary and verbal analogies were assessed twice a year in English and Spanish using the Woodcock-Johnson Test of Oral Language. We investigated multivariate and univariate models (composite scores and latent variables), and competing approaches to parameterize academic year and summer growth.

Results: The results indicated that the latent change score (LCS) models fit better than growth curve models for each language outcome. However, a univariate LCS model with a composite oral language proficiency score fit as well as a multivariate model with three parallel processes.

Conclusion: Students experienced more setback during the initial summer month entering grade 1 compared to the summer entering grade 2. When examining summer setbacks, a more complex model may not necessarily be more informative.

School effects in the dynamics of reading, writing, and African-American dialect

First Author/Chair:Lee Branum-Martin -- Georgia State University
Additional authors/chairs: 
Cynthia Puranik; Julie Washington

Purpose: We explored the longitudinal relations across reading, writing, and a measure of dialect density for African American students. While African American dialect is generally associated with negative language and reading outcomes, longitudinal details of how this might change over time are not well understood.

Method: Using an accelerated cohort design, 869 African American students in grades 1-5 from 7 schools were measured in two subsequent grades each. The three measures were the Woodcock-Johnson passage comprehension, the Test of Written Language, and a measure of dialect density from the Diagnostic Evaluation of Language Variation. Following univariate models, a trivariate latent difference score model was fit for the three constructs across the five grades.

Results: The model fit reasonably well, with dialect decreasing over time and reading and writing increasing, but both slowing down in later grades. Dialect density related negatively to reading and writing. Dialect had a negative effect on change in reading and writing. Schools had a range of differences, both in level as well as in rate of expected change.

Conclusions: While reading and writing slowed down, they appeared to have a mutually reinforcing relation. The effects of dialect, while negative, were not uniform across different levels of the other two constructs. The school effects suggest that there may be instructional and environmental differences in schools which may change how language and literacy relate over time. These results suggest that language, reading, and writing may co-develop differently in different educational environments.

A bivariate latent difference score analysis of intervention effects on language and self-regulation coupling

First Author/Chair:Janelle Montroy -- Children's Learning Institute
Additional authors/chairs: 
Bethanie Van Horne; Kaitlyn May

Purpose: Early language and self-regulation are important for later reading success. We explored intervention effects across language and self-regulation in children attending Early Head Start. Language development contributes to self-regulation by giving children mental representations that help them manage behavior. Self-regulation development may enhance language development, as children need to regulate their behavior to engage in learning. How these skills and relations change via intervention is underexplored. We utilize a multi-group bivariate latent difference score (LDS) model to evaluate intervention effects on self-regulation and language coupling pre-to-post.

Method: Across two cohorts (3rd cohort to come), participants included Early Head Start teachers (n = 61), parent and child dyads (n = 124). Intervention teachers and parents received online responsive caregiving programs. Language and self-regulation were assessed pre/post via PLS-5 auditory comprehension, an observational language measure (Landry et al., 2006), MacArthur-Bates, and Snack Delay.

Results: We created language factors that were scalar invariant. We tested group differences in the LDS model via metric invariance. Preliminary results indicated language and self-regulation increased with potentially greater increases with intervention (language factor d = 0.46, self-regulation d = 0.37, p = 0.19). Intervention children with lower pretest self-regulation had significantly greater change in self-regulation and language. There was a trend relation between higher pretest language and self-regulation change. Control children demonstrated a significantly different pattern. 

Conclusions: Caregiver interventions may impact child language and self-regulation by affecting how these variables couple. Results also suggest the promise of multi-group LDS models for evaluating multivariate intervention effects.     

Performance discrepancy in phonological awareness tasks

First Author/Chair:Ryan Bowles -- Michigan State University
Additional authors/chairs: 
Lori Skibbe; Sarah Goodwin; Gary Troia

Purpose: Phonological awareness (PA) is a critical early literacy skill, strongly predicting later reading skills. PA is a unidimensional skill, with numerous methods of ascertaining a child’s level of PA, including assessments of rhyming, segmenting, and blending. However, children may find certain PA tasks relatively easier or more difficult. Using a latent difference score model to examine discrepancies in performance (de Haan et al., 2014), we considered whether children with speech and/or language impairments exhibited greater discrepancies across tasks compared to children with typical language.

Method: The ATLAS test of PA (Skibbe et al., 2019), which was designed to be a valid measure of PA both for children with disabilities and for children with typical development, was administered to 938 children with typical language development and 227 children with IEPs associated with speech and/or language impairments. The three subtests of the ATLAS-PA, Rhyming, Segmenting, and Blending, were put on a common scale using Rasch measurement, and then children’s levels of PA were estimated separately for each subtest.

Results: Speech and/or language impairments were associated with greater discrepancies in PA levels, such that children with speech and/or language impairments performed substantially better on rhyming relative to both segmenting and blending, compared to children with typical language.

Conclusions: Rhyming is a relative strength for children with speech and/or language impairments compared to segmenting and blending. This study also demonstrates the potential for latent difference score models in understanding discrepancies in performance across tasks measuring the same skill.