A model for optimal constrained adaptive testing

Wim J. van der Linden, Lynda M. Reese

Research output: Contribution to journalArticleAcademicpeer-review

68 Citations (Scopus)

Abstract

A model for constrained computerized adaptive testing is proposed in which the information in the test at the trait level (0) estimate is maximized subject to a number of possible constraints on the content of the test. At each item-selection step, a full test is assembled to have maximum information at the current 0 estimate, fixing the items already administered. Then the item with maximum in-formation is selected. All test assembly is optimal because a linear programming (LP) model is used that automatically updates to allow for the attributes of the items already administered and the new value of the 0 estimator. The LP model also guarantees that each adaptive test always meets the entire set of constraints. A simulation study using a bank of 753 items from the Law School Admission Test showed that the 0 estimator for adaptive tests of realistic lengths did not suffer any loss of efficiency from the presence of 433 constraints on the item selection process.
Original languageEnglish
Pages (from-to)259-270
Number of pages11
JournalApplied psychological measurement
Volume22
Issue number3
DOIs
Publication statusPublished - 1998

Cite this

@article{bf42b019ec024dce8255614f1df29d65,
title = "A model for optimal constrained adaptive testing",
abstract = "A model for constrained computerized adaptive testing is proposed in which the information in the test at the trait level (0) estimate is maximized subject to a number of possible constraints on the content of the test. At each item-selection step, a full test is assembled to have maximum information at the current 0 estimate, fixing the items already administered. Then the item with maximum in-formation is selected. All test assembly is optimal because a linear programming (LP) model is used that automatically updates to allow for the attributes of the items already administered and the new value of the 0 estimator. The LP model also guarantees that each adaptive test always meets the entire set of constraints. A simulation study using a bank of 753 items from the Law School Admission Test showed that the 0 estimator for adaptive tests of realistic lengths did not suffer any loss of efficiency from the presence of 433 constraints on the item selection process.",
author = "{van der Linden}, {Wim J.} and Reese, {Lynda M.}",
year = "1998",
doi = "10.1177/01466216980223006",
language = "English",
volume = "22",
pages = "259--270",
journal = "Applied psychological measurement",
issn = "0146-6216",
publisher = "SAGE Publications",
number = "3",

}

A model for optimal constrained adaptive testing. / van der Linden, Wim J.; Reese, Lynda M.

In: Applied psychological measurement, Vol. 22, No. 3, 1998, p. 259-270.

Research output: Contribution to journalArticleAcademicpeer-review

TY - JOUR

T1 - A model for optimal constrained adaptive testing

AU - van der Linden, Wim J.

AU - Reese, Lynda M.

PY - 1998

Y1 - 1998

N2 - A model for constrained computerized adaptive testing is proposed in which the information in the test at the trait level (0) estimate is maximized subject to a number of possible constraints on the content of the test. At each item-selection step, a full test is assembled to have maximum information at the current 0 estimate, fixing the items already administered. Then the item with maximum in-formation is selected. All test assembly is optimal because a linear programming (LP) model is used that automatically updates to allow for the attributes of the items already administered and the new value of the 0 estimator. The LP model also guarantees that each adaptive test always meets the entire set of constraints. A simulation study using a bank of 753 items from the Law School Admission Test showed that the 0 estimator for adaptive tests of realistic lengths did not suffer any loss of efficiency from the presence of 433 constraints on the item selection process.

AB - A model for constrained computerized adaptive testing is proposed in which the information in the test at the trait level (0) estimate is maximized subject to a number of possible constraints on the content of the test. At each item-selection step, a full test is assembled to have maximum information at the current 0 estimate, fixing the items already administered. Then the item with maximum in-formation is selected. All test assembly is optimal because a linear programming (LP) model is used that automatically updates to allow for the attributes of the items already administered and the new value of the 0 estimator. The LP model also guarantees that each adaptive test always meets the entire set of constraints. A simulation study using a bank of 753 items from the Law School Admission Test showed that the 0 estimator for adaptive tests of realistic lengths did not suffer any loss of efficiency from the presence of 433 constraints on the item selection process.

U2 - 10.1177/01466216980223006

DO - 10.1177/01466216980223006

M3 - Article

VL - 22

SP - 259

EP - 270

JO - Applied psychological measurement

JF - Applied psychological measurement

SN - 0146-6216

IS - 3

ER -