SES and Response to Treatment: A Quantile Regression Approach

• In Event: Poster Session 04
  In Poster Session: PS 04 Topic - Education, Schooling

Abstract

In the U.S., a large proportion of children fail to meet established levels of adequate reading performance. Students that do not perform proficiently on measures of literacy important to the development of reading and language skills, are at risk for later reading comprehension difficulties and other academic problems. In response, interventions are often implemented to address deficiencies in these skills. However, as described by child-by-instruction interaction, there are factors that influence how children respond to treatments (Conner et al., 2004). Specifically, socioeconomic status (SES) is a well-known correlate of academic achievement, including reading (Sirin, 2005). To test associations between independent variables of interest and educational outcomes, ordinary least squares (OLS) regression is often used. OLS regression estimates however are limited to focusing on the mean or average effect. Quantile regression is alternative regression framework that allows for the analysis of the distribution of an association conditional on the quantiles (i.e., percentiles) of the outcome (Chernozhukov & Hansen, 2005). In comparison to the traditional OLS regression method, we explore the estimated associations between SES and the distribution of response to treatment using the quantile regression approach. This study used a large, diverse sample of elementary children (N=2,407) who were part of one of eight randomized control trial reading intervention projects over one school year in North Florida schools. The data from the different interventions were combined as part of Project KIDS (Kids and Individual Differences in School) - a federally funded project aimed at exploring how child traits, family environment and familial risk for learning difficulties moderate treatment responsiveness (Daucourt et al., 2018). Response to treatment was measured by residualized gain scores from fall to spring of the school year, on the Woodcock-Johnson Tests of Achievement Letter-Word Identification (LWID) and Picture Vocabulary (PV) subtests. SES was operationalized as free or reduced-price lunch status (FRL), where participants were coded 1 if they did not qualify for free or reduced-price lunch (i.e., had relatively higher socioeconomic status), and 0 if they qualified for or received free or reduced-price lunch (i.e., had relatively lower socioeconomic status). In the quantile regression analyses, children in the lower quantiles were the least responsive to intervention, and those falling in the higher quantiles were the most responsive to intervention. Overall FRL was significantly positively associated with treatment response for both LWID and PV measures. The results indicate that the effect of FRL differs across the distribution of residualized gains in response to treatment. The trends show that the association between FRL and residualized gain scores becomes higher as you move upward in the distribution of residualized gain scores (Figure 1 & 2). Within the 50-70th quantile, results closely resembled estimates of the traditional OLS method. Our findings suggests that having a higher-SES is more highly associated with higher residualized gains scores due to intervention in component skills measured. Incorporating individual differences perspective into our modeling of intervention effects will allow a richer targeted identification of children who will continue to struggle after intervention.

Group Authors

Florida State University

Authors

• Cynthia Ulysse Norris, Florida State Univeristy
  Presenting Author

• Jeffrey A. Shero, Florida State University
  Non-Presenting Author

• Rasheda Haughbrook, Florida State University
  Non-Presenting Author

• Sara Hart, Florida State University
  Non-Presenting Author