Corrected goodness-of-fit index in latent variable modeling using non-parametric bootstrapping

Area of Science:

• Social Sciences
• Educational Measurement
• Psychometrics

Background:

• Latent variable modeling (LVM) is crucial for validating social science tools.
• Traditional model fit evaluation faces challenges with sample size and model complexity.
• Existing methods can be sensitive to statistical artifacts.

Purpose of the Study:

• To develop an R function for assessing LVM fit indices using non-parametric bootstrapping.
• To introduce a new corrected goodness-of-fit index (CGFI) to address limitations in traditional methods.
• To evaluate the performance of the new function with real-world educational data.

Main Methods:

• Implementation of an R function employing non-parametric bootstrapping for fit index assessment.
• Development and application of the corrected goodness-of-fit index (CGFI).
• Analysis of data from the PISA 2022 and PIRLS 2021 studies.

Main Results:

• The CGFIboot function demonstrated differential decision-making compared to sample estimates alone.
• The proposed CGFI offers a potential improvement over existing fit indices.
• Analysis revealed distinct outcomes in evaluating instructional leadership and bullying constructs.

Conclusions:

• The CGFIboot function provides valuable insights for enhancing evaluative criteria in LVM.
• This approach may lead to more robust and reliable model fit assessments.
• The study highlights the importance of advanced statistical techniques in social science research.