Article

Bayesian Lasso Regression for Identifying Home and Parental Predictors of Reading Achievement: Evidence from PIRLS 2021 Türkiye

---

Abstract

We analyze PIRLS 2021 Türkiye, 4,883 Grade-4 students in 166 schools, using a weighted multilevel Bayesian LASSO that respects plausible values and school clustering. After shrinkage, four predictors remain: student gender, parental education, children’s book counts at home, and study supports. Boys score about 0.15 SD lower than girls. Homes with 0–10 children’s books are roughly 0.32 SD below homes with 200+ books. Children of parents with only primary schooling score about 0.46 SD below children of university-educated parents. High study support predicts about +0.18 SD relative to low support. We fit each of the five plausible values, pool posteriors, apply probability weights, and include a school random intercept. Convergence diagnostics meet standard thresholds. Effects are reported in PIRLS SD units, with probability-of-direction and ROPE summaries to convey direction and practical magnitude. Broader socio-economic indicators and many attitudinal variables show no credibly non-zero effects. Results highlight book access and study environments as practical levers.

Keywords: PIRLS, reading achievement, Bayesian LASSO, priors

How to Cite:

Ozkan, F., (2026) “Bayesian Lasso Regression for Identifying Home and Parental Predictors of Reading Achievement: Evidence from PIRLS 2021 Türkiye”, Practical Assessment, Research, and Evaluation 31(1): 12. doi: https://doi.org/10.7275/pare.3544