東華大學圖書館 |

Language: English

Help

回圖書館首頁

手機版館藏查詢

Back

Switch To: Labeled | MARC Mode | ISBD

An investigation of the robustness o...

Dowling, Norca Maritza.

Linked to FindBook

Google Book

Amazon

博客來

An investigation of the robustness of multilevel item response theory models to violations of distributional assumptions.

Record Type:	Electronic resources : Monograph/item
Title/Author:	An investigation of the robustness of multilevel item response theory models to violations of distributional assumptions./
Author:	Dowling, Norca Maritza.
Description:	632 p.
Notes:	Source: Dissertation Abstracts International, Volume: 67-12, Section: B, page: 7423.
Contained By:	Dissertation Abstracts International67-12B.
Subject:	Psychology, Psychometrics. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3245602

An investigation of the robustness of multilevel item response theory models to violations of distributional assumptions.
Dowling, Norca Maritza.

An investigation of the robustness of multilevel item response theory models to violations of distributional assumptions. - 632 p.

Source: Dissertation Abstracts International, Volume: 67-12, Section: B, page: 7423.

Thesis (Ph.D.)--The University of Wisconsin - Madison, 2006.

The advantages of multilevel techniques for analyzing clustered data have prompted the development of models critical to theory building in education and evaluation. A class of models combining item response theory and multilevel data structure have received considerable attention from psychometricians and applied researchers. These hybrid latent models, known as multilevel item response theory (MLIRT) models, have the advantage of accounting for uncertainty in the measures associated with individual- and cluster-level characteristics, thus increasing the precision of the predictive effects of school, teacher, and student characteristics on student achievement.Subjects--Topical Terms:

1017742
Psychology, Psychometrics.

An investigation of the robustness of multilevel item response theory models to violations of distributional assumptions.
LDR:03357nmm 2200265 4500 001 1834461
005 20071119145656.5
008 130610s2006 eng d
035 $a (UMI)AAI3245602
035 $a AAI3245602
040 $a UMI $c UMI
100 1 $a Dowling, Norca Maritza. $3 1923113
245 1 3 $a An investigation of the robustness of multilevel item response theory models to violations of distributional assumptions.
300 $a 632 p.
500 $a Source: Dissertation Abstracts International, Volume: 67-12, Section: B, page: 7423.
500 $a Adviser: Ronald C. Serlin.
502 $a Thesis (Ph.D.)--The University of Wisconsin - Madison, 2006.
520 $a The advantages of multilevel techniques for analyzing clustered data have prompted the development of models critical to theory building in education and evaluation. A class of models combining item response theory and multilevel data structure have received considerable attention from psychometricians and applied researchers. These hybrid latent models, known as multilevel item response theory (MLIRT) models, have the advantage of accounting for uncertainty in the measures associated with individual- and cluster-level characteristics, thus increasing the precision of the predictive effects of school, teacher, and student characteristics on student achievement.
520 $a Using a Bayesian approach to estimation, a Monte Carlo simulation study investigated the robustness of parameter estimates produced by a random-intercept MLIRT model under different conditions of nonnormality for level-1 and level-2 error components, number of clusters, cluster size, ICC, and test length. Point estimates of the MLIRT model were less reactive to violations of distributional assumptions than classical multilevel models. Fixed and random effects estimates and corresponding posterior standard deviations functioned well under model misspecification. Intercept parameter estimates were positively biased irrespective of error distribution. Violations of distributional assumptions did not noticeably affect the nominal 95% interval coverage of the model's fixed and random components. However, coverages for random components were less optimal and more sensitive to changes in cluster size, number of clusters, and ICC's than those for fixed effects. The power of hypothesis tests on fixed effect parameters did not show a significant reactivity to model misspecifications. The cluster-level parameter produced the lowest power with high ICC's, few clusters, and small cluster sizes regardless of model's error structure. Type I error rates for fixed effect parameters were close to the nominal alpha level of 0.05 under nonnormality of errors. Item parameter estimates showed sensitivity to error distributions especially for items with extreme characteristics. Skewed error distributions were associated with large biases in difficulty and discrimination parameter estimates of extreme items. The modal structure of the error distribution affected the accuracy and efficiency of posterior standard deviations associated with item difficulty parameters.
590 $a School code: 0262.
650 4 $a Psychology, Psychometrics. $3 1017742
690 $a 0632
710 2 0 $a The University of Wisconsin - Madison. $3 626640
773 0 $t Dissertation Abstracts International $g 67-12B.
790 1 0 $a Serlin, Ronald C., $e advisor
790 $a 0262
791 $a Ph.D.
792 $a 2006
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3245602