|Title:||Scaling Bayesian Latent Variable Models to Big Education Data|
|Principal Investigator:||Merkle, Edgar||Awardee:||University of Missouri, Columbia|
|Program:||Statistical and Research Methodology in Education [Program Details]|
|Award Period:||3 years (08/01/2021 - 07/31/2024)||Award Amount:||$899,456|
|Type:||Methodological Innovation||Award Number:||R305D210044|
Co-Principal Investigator: Bonifay, Wesley
The goal of the proposed project is to develop user-friendly, free, open-source software for estimating the types of Bayesian latent variable models that are often encountered in education: models with multilevel structure, with ordinal variables, and with large sample sizes. This will provide education researchers with tools that allow them to apply state-of-the-art developments quickly and easily in Bayesian statistics to their own datasets. The software will build on the existing R package blavaan for Bayesian structural equation modeling, which relies on the power of Stan for estimation via Hamiltonian Monte Carlo.
The Markov Chain Monte Carlo methods, developed as part of this grant, will rely on theoretical developments related to marginal likelihoods and factor score regression, leading to fast and efficient model estimation. The research team will test these approaches via a series of simulation studies and real-data examples to ensure that they are functioning correctly. The team will then incorporate them into the blavaan software package and test for usability with doctoral students and applied education researchers. In addition to the update to blavaan, the grant team will provide online user support resources, publish in peer-reviewed journals, and give presentations and seminars at major education research conferences.