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

Abstract Reading proficiency is foundational to future educational success and socio‑economic mobility. Despite widespread recognition that children’s reading outcomes are shaped by both the home environment and broader socio‑economic contexts, the relative contributions and complex interactions of these factors remain incompletely understood. This study analyses data from the fourth‑grade sample of the 2021 Progress in International Reading Literacy Study (PIRLS) for Türkiye. We examine associations between reading achievement and a comprehensive set of home, parental and socio‑economic variables. The data are cross‑sectional, exposures and outcomes are measured at the same time, so our findings describe associations rather than causal effects. We employ Bayesian LASSO regression to handle multicollinearity and identify a parsimonious set of variables. Consistent with PIRLS methodological guidelines, we estimate separate models for each of the five plausible values of reading achievement and pool the posterior distributions using Rubin’s rules. Analyses incorporate PIRLS student sampling weights and account for clustering at the school level through random intercepts. Posterior summaries include medians, 95% credible intervals, the probability of direction (PD)—the proportion of the posterior distribution on the median’s side of zero—and the percentage of posterior mass lying within a region of practical equivalence (ROPE) set to ± 0.1 standardised units. Findings indicate that girls outperform boys by roughly 12 points after accounting for school‑level clustering; the availability of children’s books, parental education and study supports show the clearest positive associations with reading scores. Many other variables (e.g., socio‑economic status categories, parental attitudes and school‑facing parental involvement) display wide credible intervals, PD values near 0.55–0.70 and substantial ROPE mass, signalling indeterminate associations. We discuss these results in light of theoretical frameworks and the international literature, highlight policy implications for Türkiye and other contexts, and outline directions for future research.