Multidimensional Item Response Theory Models with Collateral Information as Poisson Regression Models

Research output: Contribution to journalArticle

Abstract

Multiple choice items on tests and Likert items on surveys are ubiquitous in educational, social and behavioral science research; however, methods for analyzing of such data can be problematic. Multidimensional item response theory models are proposed that yield structured Poisson regression models for the joint distribution of responses to items. The methodology presented here extends the approach described in Anderson, Verkuilen, and Peyton (2010) that used fully conditionally specified multinomial logistic regression models as item response functions. In this paper, covariates are added as predictors of the latent variables along with covariates as predictors of location parameters. Furthermore, the models presented here incorporate ordinal information of the response options thus allowing an empirical examination of assumptions regarding the ordering and the estimation of optimal scoring of the response options. To illustrate the methodology and flexibility of the models, data from a study on aggression in middle school (Espelage, Holt, and Henkel 2004) is analyzed. The models are fit to data using SAS.

Original languageEnglish (US)
Pages (from-to)276-303
Number of pages28
JournalJournal of Classification
Volume30
Issue number2
DOIs
StatePublished - Jul 5 2013

Fingerprint

Poisson Regression
model theory
Poisson Model
Model Theory
Regression Model
Logistic Models
Behavioral Research
Behavioral Sciences
regression
Social Sciences
Aggression
Covariates
Predictors
Methodology
Logistic Regression Model
Location Parameter
Latent Variables
Response Function
Scoring
Joint Distribution

Keywords

  • Constrained optimization
  • Covariates
  • Fully conditionally specified models
  • Log-multiplicative association models
  • Ordinal response scales
  • Polytomous items

ASJC Scopus subject areas

  • Library and Information Sciences
  • Mathematics (miscellaneous)
  • Psychology (miscellaneous)
  • Statistics, Probability and Uncertainty

Cite this

Multidimensional Item Response Theory Models with Collateral Information as Poisson Regression Models. / Anderson, Carolyn Jane.

In: Journal of Classification, Vol. 30, No. 2, 05.07.2013, p. 276-303.

Research output: Contribution to journalArticle

@article{0ce1dba6b9a14ae280d8c50c4a37bad2,
title = "Multidimensional Item Response Theory Models with Collateral Information as Poisson Regression Models",
abstract = "Multiple choice items on tests and Likert items on surveys are ubiquitous in educational, social and behavioral science research; however, methods for analyzing of such data can be problematic. Multidimensional item response theory models are proposed that yield structured Poisson regression models for the joint distribution of responses to items. The methodology presented here extends the approach described in Anderson, Verkuilen, and Peyton (2010) that used fully conditionally specified multinomial logistic regression models as item response functions. In this paper, covariates are added as predictors of the latent variables along with covariates as predictors of location parameters. Furthermore, the models presented here incorporate ordinal information of the response options thus allowing an empirical examination of assumptions regarding the ordering and the estimation of optimal scoring of the response options. To illustrate the methodology and flexibility of the models, data from a study on aggression in middle school (Espelage, Holt, and Henkel 2004) is analyzed. The models are fit to data using SAS.",
keywords = "Constrained optimization, Covariates, Fully conditionally specified models, Log-multiplicative association models, Ordinal response scales, Polytomous items",
author = "Anderson, {Carolyn Jane}",
year = "2013",
month = "7",
day = "5",
doi = "10.1007/s00357-013-9131-x",
language = "English (US)",
volume = "30",
pages = "276--303",
journal = "Journal of Classification",
issn = "0176-4268",
publisher = "Springer New York",
number = "2",

}

TY - JOUR

T1 - Multidimensional Item Response Theory Models with Collateral Information as Poisson Regression Models

AU - Anderson, Carolyn Jane

PY - 2013/7/5

Y1 - 2013/7/5

N2 - Multiple choice items on tests and Likert items on surveys are ubiquitous in educational, social and behavioral science research; however, methods for analyzing of such data can be problematic. Multidimensional item response theory models are proposed that yield structured Poisson regression models for the joint distribution of responses to items. The methodology presented here extends the approach described in Anderson, Verkuilen, and Peyton (2010) that used fully conditionally specified multinomial logistic regression models as item response functions. In this paper, covariates are added as predictors of the latent variables along with covariates as predictors of location parameters. Furthermore, the models presented here incorporate ordinal information of the response options thus allowing an empirical examination of assumptions regarding the ordering and the estimation of optimal scoring of the response options. To illustrate the methodology and flexibility of the models, data from a study on aggression in middle school (Espelage, Holt, and Henkel 2004) is analyzed. The models are fit to data using SAS.

AB - Multiple choice items on tests and Likert items on surveys are ubiquitous in educational, social and behavioral science research; however, methods for analyzing of such data can be problematic. Multidimensional item response theory models are proposed that yield structured Poisson regression models for the joint distribution of responses to items. The methodology presented here extends the approach described in Anderson, Verkuilen, and Peyton (2010) that used fully conditionally specified multinomial logistic regression models as item response functions. In this paper, covariates are added as predictors of the latent variables along with covariates as predictors of location parameters. Furthermore, the models presented here incorporate ordinal information of the response options thus allowing an empirical examination of assumptions regarding the ordering and the estimation of optimal scoring of the response options. To illustrate the methodology and flexibility of the models, data from a study on aggression in middle school (Espelage, Holt, and Henkel 2004) is analyzed. The models are fit to data using SAS.

KW - Constrained optimization

KW - Covariates

KW - Fully conditionally specified models

KW - Log-multiplicative association models

KW - Ordinal response scales

KW - Polytomous items

UR - http://www.scopus.com/inward/record.url?scp=84880339352&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84880339352&partnerID=8YFLogxK

U2 - 10.1007/s00357-013-9131-x

DO - 10.1007/s00357-013-9131-x

M3 - Article

AN - SCOPUS:84880339352

VL - 30

SP - 276

EP - 303

JO - Journal of Classification

JF - Journal of Classification

SN - 0176-4268

IS - 2

ER -