Reading Comprehension and Vocabulary Relationships Across Comprehension Skill Levels in Grade 4: A Quantile Regression Approach

Reading Comprehension and Vocabulary Relationships Across Comprehension Skill Levels in Grade 4: A Quantile Regression Approach

First Author: Dawna Duff -- University of Pittsburgh, Communication Sciences and Disorders
Keywords: Vocabulary, Reading comprehension, Quantile regression, Response to Intervention, Individual Differences
Abstract / Summary: 

Purpose: The effect of vocabulary (Vocab) intervention on reading comprehension (RC) would be expected to vary with the extent to which Vocab limits RC for a given reader. Previous research indicates substantially larger effect sizes of Vocab intervention on RC for children with/at risk of reading disabilities (Elleman, Lindo, Morphy & Compton, 2009, Duff, 2015). This leads to the hypothesis that Vocab may constrain RC more for readers with poor comprehension, relative to readers with higher comprehension ability. This study investigates whether the relation between RC and Vocab is the same across the RC distribution.

Methods: Participants in 4th grade (n=570) from an epidemiological sample (Tomblin & Nippold, 2014) completed standardized tests of vocabulary knowledge (PPVT-R, CREVT-E) and reading comprehension (WRMT-R passage comp, GORTIII, DAB-2). Composite scores for Vocab and RC were derived. Quantile regression (QR) analysis was conducted to test the relation between RC and Vocab across RC levels. Comparison was made with ordinary least squares regression (OLS) analysis of the same data.

Results: As expected, OLS regression analysis indicated that Vocab significantly predicted RC (r= .706, t(568)= 23.77, p<0.000 ). Quantile regression analysis indicated that Vocab had the strongest relationship with RC at the low end of the RC distribution, and the weakest relation with RC at the high end of the RC distribution. The estimated slope coefficient for Vocab was 0.85 at the 0.05 quantile, and 0.61 at the 0.95 quantile. Hence, a quantile regression approach reveals a more complex RC Vocab relationship than an OLS regression approach alone.

Conclusions: At lower RC skill levels, Vocab is more strongly related to RC. This provides one account for larger effect sizes for Vocab intervention on RC for poorer readers. Implications for a theoretical account of individual differences in reading comprehension are discussed.