Addressing The Issues of Low Degrees of Freedom and Associated Poor Validation Estimates of Science Education Models: An Advocacy for The Application of Confirmatory Network Analysis

Generally, science education models have a low degree of freedom df, and this psychometric aspect negatively impacts the overall ‎goodness of fit estimates obtained on conducting CB-SEM-based Confirmatory Factor Analysis (CFA) on the hypothesized latent ‎models using the Maximum Likelihood (ML) estimator. To address this issue, the state-of-the-art Confirmatory Network Analysis ‎‎(CNA) approach is advocated, which uses the adjacency matrix in estimation and treats the data type of the data collected using ‎survey questionnaires as ordinal using the appropriate Diagonally Weighted Least Squares (DWLS) estimator to provide the proper ‎scaled goodness of fit estimates of the network. In the present study, the goodness of fit estimates of a science education model ‎involving the perception of 918 secondary school students of India on their science constructivist learning environment, ‎epistemological beliefs in science, science self-efficacy, science academic flow, and science academic achievement were found from ‎the traditional latent variable modeling and the latest network perspectives. Data analysis was conducted using SPSS AMOS Ver ‎‎23.0, several packages of R and through generated using ChatGPT Ver. 5. The scaled goodness of fit estimates of confirmatory ‎network analysis produced a better picture of the reality over confirmatory factor analysis, and provided deeper insights on the ‎interactions among the studied variables as were graphically established by the differences in the Bland-Altman plots of the CFI, ‎TLI, RMSEA and SRMR estimates of the two approaches. Educational and psychometric implications of the study are discussed‎.