TUBITAK 2236 | Co-funded Brain Circulation Postdoctoral Fellow
Ezgi Aytürk, PhD | Primary Investigator
Koç University, Department of Psychology
Improving mediation modeling by handling nonignorable item nonresponse: A large-scale simulation study and an illustration using the cross-national PISA data
The project aimed to advance mediation modeling by incorporating a promising new method to identify and remedy the most problematic and largely overlooked missing data type, missing-not-at-random (MNAR). The project involves a large-scale Monte Carlo simulation study and the application of methods to the analysis of the Program for International Student Assessment (PISA) dataset. The simulation study, for the first time, demonstrated the effect of MNAR items in single- and multi-group mediation models with latent variables. The simulation study was relevant to both observational and experimental studies with applications in treatment/prevention research.
Additionally, using the PISA dataset, a theory-based mediation model tested across countries to illustrate the sensitivity of group (i.e. country) comparisons on mediation estimates to different assumptions about missing data mechanisms. Analysis of the PISA data had both methodological and substantive relevance to education research.
The project will contribute to the European Commission`s vision for “evidence-based policy- making in education” by advancing the analytical methods to assess educational processes accurately.