|Title:||Efficient and Flexible Tools for Complex Multilevel and Latent Variable Modeling in Education Research|
|Principal Investigator:||Gelman, Andrew||Awardee:||Columbia University|
|Program:||Statistical and Research Methodology in Education [Program Details]|
|Award Period:||3 years (7/1/19 – 6/30/22)||Award Amount:||$706,423|
|Type:||Methodological Innovation||Award Number:||R305D190048|
Co-Principal Investigator: Rabe-Hesketh, Sophia
The purpose of this project is to expand the capabilities of Stan, an open-source software environment for Bayesian inference which is widely used in academic and industrial social science applications. Stan was initially developed by this research team through an IES grant, Practical Solutions for Missing Data and Imputation, and was later expanded and rendered more user-friendly through another IES grant, Solving Difficult Bayesian Computation Problems in Education Research Using STAN. In this project, the team will expand the software further by making it computationally more efficient and adding additional statistical algorithms.
The specific plans for methodological expansion of Stan include the addition of parallel computing by distributing computation of the likelihood function across multiple cores, setting up matrix operations on graphical processing units, which could potentially result in a 50-fold improvement in the speed for time-consuming computations, gradient-based marginal optimization for maximum likelihood or Bayesian inference, and expectation propagation for data streaming and splitting for the sorts of models that arise in large multilevel and latent-variable problems in education research settings. To render these advances more approachable to applied education researchers, the researchers will develop full case studies demonstrating the use of these methods. Free and open discussion regarding these tools and case studies will also be available through the online forums for Stan users and developers. The researchers will also publish results and product developments through peer-reviewed journal manuscripts, conference presentations, and workshops at major conferences.