Investigating Factors Affecting Scientific Literacy with Structural Equation Modeling and Multilevel Structural Equation Modeling: Case of PISA 2015

Akdogdu Yildiz E., Demir M. C., GELBAL S.

CUKUROVA UNIVERSITY FACULTY OF EDUCATION JOURNAL, vol.51, no.2, pp.795-824, 2022 (ESCI) identifier

  • Publication Type: Article / Article
  • Volume: 51 Issue: 2
  • Publication Date: 2022
  • Doi Number: 10.14812/cufej.933101
  • Journal Indexes: Emerging Sources Citation Index (ESCI), TR DİZİN (ULAKBİM)
  • Page Numbers: pp.795-824
  • Keywords: Scientific Literacy, Single-level Analysis, Hierarchical Data, Multi-level Analysis, COVARIANCE STRUCTURE-ANALYSIS, SCIENCE ACHIEVEMENT, SCHOOL RESOURCES, STUDENTS, PERFORMANCE, VARIABLES, MATTER
  • Hacettepe University Affiliated: Yes


There is no empirical evidence in the literature regarding the problems encountered in the use of single-level analyzes on hierarchical data and the implementation of a single-multilevel structural equation model. In this study, the models were created by using Structural Equation Modeling and Multilevel Structural Equation Modeling for the effects of factors such as enjoyment in learning science, instrumental motivation, scientific self-efficacy, hinderances in education, and hinderance to learning which are claimed to predict Turkish students' science performance who participated PISA 2015. The effects of the predictive variables were estimated with two different single-level models constructed by aggregating and disaggregating the data. Then, single-level models are compared with the two-level model in terms of model fit and standardized parameters. As a result, since it was observed that standard error in regression coefficients decreased for the model which disregarded group levels, and variance -within-groups was not included in the model which disregarded individual levels which caused a data loss, the results were biased, and the effectiveness of the statistical test was weakened. In light of the results of this study, some recommendations were suggested for future studies which may consider dealing with analyzing hierarchical data.