Detecting latent subpopulations in international large-scale assessments by fitting MixIRT models using NUTS

Abstract The focus of this study is to use the mixture item response theory (MixIRT) model while implementing the no-U-turn sampler as a technique for investigating the presence of latent classes (i.e., subpopulations) among eighth-grade students who were administered TIMSS 2019 mathematics subtest...

Full description

Saved in:
Bibliographic Details
Main Authors: Rehab AlHakmani, Yanyan Sheng
Format: Article
Language:English
Published: SpringerOpen 2024-11-01
Series:Large-scale Assessments in Education
Subjects:
Online Access:https://doi.org/10.1186/s40536-024-00226-7
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850179743707561984
author Rehab AlHakmani
Yanyan Sheng
author_facet Rehab AlHakmani
Yanyan Sheng
author_sort Rehab AlHakmani
collection DOAJ
description Abstract The focus of this study is to use the mixture item response theory (MixIRT) model while implementing the no-U-turn sampler as a technique for investigating the presence of latent classes (i.e., subpopulations) among eighth-grade students who were administered TIMSS 2019 mathematics subtest in paper format from the gulf cooperation council (GCC) countries. One-, two-, and constrained three-parameter logistic MixIRT models with one to four classes were used to fit to the data, where the model data fit was assessed using Bayesian fit indices. The results indicate that multiple latent classes or subpopulations can better reflect the mathematical proficiency of eighth graders from the four GCC countries, and specifically the two-class constrained three-parameter MixIRT model provides a relatively better fit to the data. The results also indicate that when a mixture of several latent classes present, the conventional unidimensional IRT model is limited in providing information for multiple latent classes and shall be avoided. In addition to adding to the existing literature on MixIRT models for international large-scale assessments such as TIMSS on its heterogenous subpopulations from a fully Bayesian approach, this study sheds light on the limitation of conventional unidimensional IRT models and subsequently directs attention to the use of the more complex MixIRT model for such assessments.
format Article
id doaj-art-bed81f10e4bc402ca36ddd6f3f7658e2
institution OA Journals
issn 2196-0739
language English
publishDate 2024-11-01
publisher SpringerOpen
record_format Article
series Large-scale Assessments in Education
spelling doaj-art-bed81f10e4bc402ca36ddd6f3f7658e22025-08-20T02:18:25ZengSpringerOpenLarge-scale Assessments in Education2196-07392024-11-0112112710.1186/s40536-024-00226-7Detecting latent subpopulations in international large-scale assessments by fitting MixIRT models using NUTSRehab AlHakmani0Yanyan Sheng1Emirates College for Advanced EducationThe University of ChicagoAbstract The focus of this study is to use the mixture item response theory (MixIRT) model while implementing the no-U-turn sampler as a technique for investigating the presence of latent classes (i.e., subpopulations) among eighth-grade students who were administered TIMSS 2019 mathematics subtest in paper format from the gulf cooperation council (GCC) countries. One-, two-, and constrained three-parameter logistic MixIRT models with one to four classes were used to fit to the data, where the model data fit was assessed using Bayesian fit indices. The results indicate that multiple latent classes or subpopulations can better reflect the mathematical proficiency of eighth graders from the four GCC countries, and specifically the two-class constrained three-parameter MixIRT model provides a relatively better fit to the data. The results also indicate that when a mixture of several latent classes present, the conventional unidimensional IRT model is limited in providing information for multiple latent classes and shall be avoided. In addition to adding to the existing literature on MixIRT models for international large-scale assessments such as TIMSS on its heterogenous subpopulations from a fully Bayesian approach, this study sheds light on the limitation of conventional unidimensional IRT models and subsequently directs attention to the use of the more complex MixIRT model for such assessments.https://doi.org/10.1186/s40536-024-00226-7Mixture item response theory modelsLatent classesNo-U-turn samplerBayesian fit indicesTIMSS
spellingShingle Rehab AlHakmani
Yanyan Sheng
Detecting latent subpopulations in international large-scale assessments by fitting MixIRT models using NUTS
Large-scale Assessments in Education
Mixture item response theory models
Latent classes
No-U-turn sampler
Bayesian fit indices
TIMSS
title Detecting latent subpopulations in international large-scale assessments by fitting MixIRT models using NUTS
title_full Detecting latent subpopulations in international large-scale assessments by fitting MixIRT models using NUTS
title_fullStr Detecting latent subpopulations in international large-scale assessments by fitting MixIRT models using NUTS
title_full_unstemmed Detecting latent subpopulations in international large-scale assessments by fitting MixIRT models using NUTS
title_short Detecting latent subpopulations in international large-scale assessments by fitting MixIRT models using NUTS
title_sort detecting latent subpopulations in international large scale assessments by fitting mixirt models using nuts
topic Mixture item response theory models
Latent classes
No-U-turn sampler
Bayesian fit indices
TIMSS
url https://doi.org/10.1186/s40536-024-00226-7
work_keys_str_mv AT rehabalhakmani detectinglatentsubpopulationsininternationallargescaleassessmentsbyfittingmixirtmodelsusingnuts
AT yanyansheng detectinglatentsubpopulationsininternationallargescaleassessmentsbyfittingmixirtmodelsusingnuts