Mixture models for eye-tracking data: A case study

Donna P Ankerst, Michael D. Escobar, John A. Sweeney, Joel Greenhouse

Research output: Contribution to journalArticle

10 Citations (Scopus)

Abstract

Heterogeneity in biomedical data is often a source of great scientific interest and mixture models provide a general framework for modelling the various types that arise in practice. Finite mixture models model discrete subgroups within populations while continuous mixture models inflate the variance to account for over-dispersed data. A potential problem with the application of finite mixture models in practice is that these models may drastically overestimate the number of component densities when there is a lack of model fit. This can have severe consequences, leading the data analyst to attach substantive interpretations to spurious subgroups. For this reason, we propose using the continuous mixture model as an alternative when fitting finite mixture models with an arbitrary number of components. In the context of an example examining a specific oculomotor component of eye-tracking dysfunction in schizophrenia, we demonstrate why the continuous mixture model provides a viable alternative to the finite mixture model for small sample sizes. We present methods for fitting and comparing both models using the parametric bootstrap and EM algorithm, and show that the distinction between the models decreases as the number of component densities in the finite mixture model increases.

Original languageEnglish (US)
Pages (from-to)1365-1376
Number of pages12
JournalStatistics in Medicine
Volume15
Issue number13
DOIs
StatePublished - Jul 15 1996
Externally publishedYes

Fingerprint

Finite Mixture Models
Eye Tracking
Mixture Model
Sample Size
Schizophrenia
Number of Components
Population
Parametric Bootstrap
Discrete Subgroup
Model
Potential Problems
Alternatives
Small Sample Size
EM Algorithm
Subgroup
Decrease
Arbitrary
Modeling
Demonstrate

ASJC Scopus subject areas

  • Epidemiology

Cite this

Mixture models for eye-tracking data : A case study. / Ankerst, Donna P; Escobar, Michael D.; Sweeney, John A.; Greenhouse, Joel.

In: Statistics in Medicine, Vol. 15, No. 13, 15.07.1996, p. 1365-1376.

Research output: Contribution to journalArticle

Ankerst, Donna P ; Escobar, Michael D. ; Sweeney, John A. ; Greenhouse, Joel. / Mixture models for eye-tracking data : A case study. In: Statistics in Medicine. 1996 ; Vol. 15, No. 13. pp. 1365-1376.
@article{7efdf91a806d43fbb2cb9db9e842947a,
title = "Mixture models for eye-tracking data: A case study",
abstract = "Heterogeneity in biomedical data is often a source of great scientific interest and mixture models provide a general framework for modelling the various types that arise in practice. Finite mixture models model discrete subgroups within populations while continuous mixture models inflate the variance to account for over-dispersed data. A potential problem with the application of finite mixture models in practice is that these models may drastically overestimate the number of component densities when there is a lack of model fit. This can have severe consequences, leading the data analyst to attach substantive interpretations to spurious subgroups. For this reason, we propose using the continuous mixture model as an alternative when fitting finite mixture models with an arbitrary number of components. In the context of an example examining a specific oculomotor component of eye-tracking dysfunction in schizophrenia, we demonstrate why the continuous mixture model provides a viable alternative to the finite mixture model for small sample sizes. We present methods for fitting and comparing both models using the parametric bootstrap and EM algorithm, and show that the distinction between the models decreases as the number of component densities in the finite mixture model increases.",
author = "Ankerst, {Donna P} and Escobar, {Michael D.} and Sweeney, {John A.} and Joel Greenhouse",
year = "1996",
month = "7",
day = "15",
doi = "10.1002/(SICI)1097-0258(19960715)15:13<1365::AID-SIM232>3.0.CO;2-J",
language = "English (US)",
volume = "15",
pages = "1365--1376",
journal = "Statistics in Medicine",
issn = "0277-6715",
publisher = "John Wiley and Sons Ltd",
number = "13",

}

TY - JOUR

T1 - Mixture models for eye-tracking data

T2 - A case study

AU - Ankerst, Donna P

AU - Escobar, Michael D.

AU - Sweeney, John A.

AU - Greenhouse, Joel

PY - 1996/7/15

Y1 - 1996/7/15

N2 - Heterogeneity in biomedical data is often a source of great scientific interest and mixture models provide a general framework for modelling the various types that arise in practice. Finite mixture models model discrete subgroups within populations while continuous mixture models inflate the variance to account for over-dispersed data. A potential problem with the application of finite mixture models in practice is that these models may drastically overestimate the number of component densities when there is a lack of model fit. This can have severe consequences, leading the data analyst to attach substantive interpretations to spurious subgroups. For this reason, we propose using the continuous mixture model as an alternative when fitting finite mixture models with an arbitrary number of components. In the context of an example examining a specific oculomotor component of eye-tracking dysfunction in schizophrenia, we demonstrate why the continuous mixture model provides a viable alternative to the finite mixture model for small sample sizes. We present methods for fitting and comparing both models using the parametric bootstrap and EM algorithm, and show that the distinction between the models decreases as the number of component densities in the finite mixture model increases.

AB - Heterogeneity in biomedical data is often a source of great scientific interest and mixture models provide a general framework for modelling the various types that arise in practice. Finite mixture models model discrete subgroups within populations while continuous mixture models inflate the variance to account for over-dispersed data. A potential problem with the application of finite mixture models in practice is that these models may drastically overestimate the number of component densities when there is a lack of model fit. This can have severe consequences, leading the data analyst to attach substantive interpretations to spurious subgroups. For this reason, we propose using the continuous mixture model as an alternative when fitting finite mixture models with an arbitrary number of components. In the context of an example examining a specific oculomotor component of eye-tracking dysfunction in schizophrenia, we demonstrate why the continuous mixture model provides a viable alternative to the finite mixture model for small sample sizes. We present methods for fitting and comparing both models using the parametric bootstrap and EM algorithm, and show that the distinction between the models decreases as the number of component densities in the finite mixture model increases.

UR - http://www.scopus.com/inward/record.url?scp=0029956979&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0029956979&partnerID=8YFLogxK

U2 - 10.1002/(SICI)1097-0258(19960715)15:13<1365::AID-SIM232>3.0.CO;2-J

DO - 10.1002/(SICI)1097-0258(19960715)15:13<1365::AID-SIM232>3.0.CO;2-J

M3 - Article

C2 - 8841647

AN - SCOPUS:0029956979

VL - 15

SP - 1365

EP - 1376

JO - Statistics in Medicine

JF - Statistics in Medicine

SN - 0277-6715

IS - 13

ER -