|
Activity Number:
|
405
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 6, 2008 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #301855 |
|
Title:
|
Imputation of Missing Items for the Scale Variables Using Item Response Theory Models
|
|
Author(s):
|
Jian Zhu*+ and Trivellore Raghunathan and Raymond Bingham
|
|
Companies:
|
The University of Michigan and The University of Michigan and The University of Michigan
|
|
Address:
|
1420 Washington Heights , Ann Arbor, MI, 48109,
|
|
Keywords:
|
multiple imputation ; scale variables ; item response theory ; Gibbs sampling
|
|
Abstract:
|
Creating scale variables when some items are missing is a common but difficult problem. This paper uses item response theory models as a vehicle for imputing the missing items that incorporates scale-item response structure in the imputation process. We use a Bayesian framework and normal ogive IRT models. Approximately independent draws are obtained from the joint posterior distribution of item parameters, latent scales and missing item responses through Gibbs sampling. We use a longitudinal study with numerous scale-items to illustrate the methodology and contrast it with other standard imputation methods. We also report on a simulation study to assess the repeated sampling properties of various population level estimates derived using this methodology.
|
