Network Perspective on Exploring Factors Affecting Reading Achievement with Data of PIRLS 2016 and 2021

Abstract

<div>Objective: This study applies psychometric network analysis in the Progress in International Reading Literacy Study (PIRLS) 2016 and 2021 data to: 1. explore the strongest explanatory variables for reading achievement among variables from student, family and teacher domains; 2. explore the most influential variables with strongest impact to the network via centrality analysis; 3. explore the network structure and clustering of their interconnections, and potential important mediation pathway for improving reading proficiency; 4. compare the network differences between PIRLS 2016 and 2021.</div><div><br></div><div><p>Method: A total of 310,537 students from 49 countries participated in PIRLS 2016 and 320,546 students from 57 countries in 2021 are included in the analysis. Regularized partial correlation network analysis using the graphical LASSO algorithm is applied. The resulting network reveals how strong different variables are interconnected together while controlling for all variables in the network. The Fruchterman-Reingold and walktrap algorithm is applied to visualize the network clustering pattern such that those variables with stronger relationships are grouped together and those with weaker relationships are pushed apart. Strength, betweenness, and closeness centrality indices are computed to reveal those variables with the highest impact to the network. Network comparison using permutation test will be applied to compare the overall network connectivity and specific network differences between PIRLS 2016 and 2021.</p><p>Results: For student, parent, and teacher domain respectively, the strongest explanatory variables for reading achievement are students’ reading efficacy, children’s book amount at home, and teacher education level; while the variables with highest centrality impact to the whole network are students like reading, children’s book amount at home, and teachers’ collaboration with each other. Both separate and combined networks of respective domains are visualized to reveal the network structure and clustering pattern. Potential important mediation pathways contributing to reading achievement are highlighted in the result.</p><p>Conclusions: Traditional regression analysis mostly focuses on revealing a few most explanatory variables for a single outcome variable, ignoring the complex interplay of different contextual factors. Network analysis offers a more holistic system approach to investigate how all these variables influence each other and ultimately impact reading achievement. Network analysis answers the call for explaining students’ reading competency via the dynamic interaction among different domains of reading development.</p></div>