Skip to main content
Article
Bayesian Hierarchical Modeling with 3PNO Item Response Models
American Journal of Mathematics and Statistics (2013)
  • yanyan Sheng, Southern Illinois University Carbondale
  • Todd Christopher Headrick, Southern Illinois University Carbondale
Abstract
Fully Bayesian estimation has been developed for unidimensional IRT models. In this context, prior distributions can be specified in a hierarchical manner so that item hyperparameters are unknown and yet still have their own priors. This type of hierarchical modeling is useful in terms of the three-parameter IRT model as it reduces the difficulty of specifying model hyperparameters that lead to adequate prior distributions. Further, hierarchical modeling ameliorates the noncovergence problem associated with nonhierarchical models when appropriate prior information is not available. As such, a Fortran subroutine is provided to implement a hierarchical modeling procedure associated with the three-parameter normal ogive model for binary item response data using Gibbs sampling. Model parameters can be estimated with the choice of noninformative and conjugate prior distributions for the hyperparameters.
Keywords
  • IRT,
  • Three-parameter Normal Ogive Model,
  • MCMC,
  • Gibbs Sampling,
  • Hyperparameter,
  • Fortran
Publication Date
August, 2013
Citation Information
yanyan Sheng and Todd Christopher Headrick. "Bayesian Hierarchical Modeling with 3PNO Item Response Models" American Journal of Mathematics and Statistics Vol. 3 Iss. 5 (2013)
Available at: http://works.bepress.com/todd_headrick/26/