Online Program Home
My Program

Abstract Details

Activity Number: 475 - SPEED: Predictive Analytics with Social/Behavioral Science Applications: Spatial Modeling, Education Assessment, Population Behavior, and the Use of Multiple Data Sources
Type: Contributed
Date/Time: Wednesday, August 1, 2018 : 8:30 AM to 10:20 AM
Sponsor: Social Statistics Section
Abstract #330641 Presentation
Title: Gaussian Variational Estimation for Multidimensional Item Response Theory
Author(s): April Eun Cho* and Gongjun Xu
Companies: University of Michigan and University of Michigan
Keywords: Multidimensional IRT; Variational Inference; Exploratory factor analysis; latent variable model; Metropolis-Hastings Robbins-Monro

Multidimensional Item Response Theory (MIRT) is widely used in assessment and evaluation of education and psychological tests. It models the interaction between individuals' latent traits and their responses. The challenge in parameter estimation is that likelihood involves multidimensional integrals due to MIRT latent variable structure. Various methods have been proposed which involve direct numerical approximations to the integrals or approximations based on Monte Carlo simulation. However, these methods are known to be computationally demanding in high-dimension and dependent on sampling data points from a posterior distribution. We propose a Gaussian Variational EM algorithm which adopts a Variational Inference from machine learning that approximates probability densities through optimization. Its basic strategy is to approximate intractable likelihood by a computationally feasible lower bound. Simulation studies show that the proposed algorithm is computationally more efficient and achieves better performance in parameter estimation than existing approaches. Moreover, the proposed algorithm provides an effective way in estimating dimension of latent traits.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2018 program