Using Meta-Analysis and Propensity Score Methods to Assess Treatment Effects Toward Evidence-Based Practice in Extensive Reading

• Main Author: Akira Hamada
• Format: Article
• Language: English
• Published: Frontiers Media S.A. 2020-04-01
• Series: Frontiers in Psychology
• Subjects: evidence-based practice; quantitative methods; treatment effect assessment; meta-analysis; propensity score analysis; extensive reading
• Online Access: https://www.frontiersin.org/article/10.3389/fpsyg.2020.00617/full

This study aimed to depict the assessment process of treatment effects of extensive reading in a second language (L2) toward the establishment of an evidence-based practice. Although standardized mean differences between treatment and control groups have been applied to interpret the magnitude of treatment effects in observational studies on L2 teaching, individual effect sizes vary according to differences in learners, measures, teaching approaches, and research quality. Prior research on extensive reading has suffered from methodological restrictions, especially due to a lack of appropriate comparison between treatment and control groups. For these reasons, a retrospective meta-analysis including only studies that ensured between-group equivalence was conducted in Study 1 to estimate the effect sizes of extensive reading expected in specific teaching environments. When the focused skill of the one-semester program was reading comprehension, its effect size was predicted as d = 0.55. However, the moderator analysis showed that this treatment effect was overestimated due to selection bias in the analyzed studies and adjusted the effect size from 0.55 to 0.37. In Study 2, propensity score analysis was applied to minimize selection bias attributed to observed confounding variables in the comparison between non-randomized treatment and control groups. Data were collected from 109 Japanese university students of English who received in-class extensive reading for one semester and 115 students who attended another English class as the control group. Various types of matching were attempted, and in consideration of balancing the five covariates that might affect treatment effect estimation, the best solutions were nearest neighborhood matching without replacement, nearest neighborhood matching with replacement, and full matching. The results showed that the average treatment effects of extensive reading on all the participants (d = 0.24–0.44) and on the treated individuals (d = 0.32–0.40) were both consistent with the benchmark established in Study 1. Pedagogical implications and methodological limitations are discussed for decision-making regarding the implementation of L2 teaching practices based on research evidence.