Estimating intraclass correlation and its confidence interval in linear mixed models2

N. Demetrashvili, E.R. Heuvel, van den

Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademicpeer-review

Abstract

The methodology proposed in this study is motivated by an example from the medical field. Oncologists delineate organs for radiotherapy and it is essential that the measurements agree in these procedures. To assess the consistency of measurements among oncologists, on a random sample of subjects, the intraclass correlation (ICC) would yield a suitable estimate for studying the agreement. In technical terms, the ICC is a ratio of sum of variances that are related to differences among measured subjects and the total variance. What variance is considered relevant depends on the design of agreement study; respectively, the number of variance components changes in the numerator and the denominator of the ICC. For statistical inference, it is important but challenging to determine the distribution of estimators of such ratios and to construct the confidence intervals. In most literature, the ICC has been studied for one-way and two-way analysis of variance only. Most proposed approximate methods are based on functions of the mean squares which are model-specific (e.g. two factorial) and lack generalization to higher order (e.g. three factorial) models. The objective of this study is to extend the construction of confidence intervals for the linear mixed models, but in particular to our three-way mixed models for delineation of organs. The generalization will coincide with existing methods for two-way and one-way mixed effects models. To obtain an approximate upper and lower confidence limits, we approximate the ICC with a function of F-distributed variable and a Beta distribution. Our proposed methodology is supported by simulation studies.
Original languageEnglish
Title of host publicationInternational Conference on Trends and Perspectives in Linear Statistical Inference (LinStat'12, Bedlewo, Poland, July 16-20, 2012)
Publication statusPublished - 2012
Externally publishedYes
Eventconference; International Conference on Trends and Perspectives in Linear Statistical Inference; 2012-07-16; 2012-07-20 -
Duration: 16 Jul 201220 Jul 2012

Conference

Conferenceconference; International Conference on Trends and Perspectives in Linear Statistical Inference; 2012-07-16; 2012-07-20
Period16/07/1220/07/12
OtherInternational Conference on Trends and Perspectives in Linear Statistical Inference

Fingerprint

Intraclass Correlation
Confidence interval
Factorial
Components of Variance
Mixed Effects Model
Linear Mixed Model
Radiotherapy
Confidence Limits
Beta distribution
Numerator
Methodology
Mixed Model
Analysis of variance
Denominator
Statistical Inference
Mean Square
Simulation Study
Higher Order
Estimator
Term

Cite this

Demetrashvili, N., & Heuvel, van den, E. R. (2012). Estimating intraclass correlation and its confidence interval in linear mixed models2. In International Conference on Trends and Perspectives in Linear Statistical Inference (LinStat'12, Bedlewo, Poland, July 16-20, 2012)
Demetrashvili, N. ; Heuvel, van den, E.R. / Estimating intraclass correlation and its confidence interval in linear mixed models2. International Conference on Trends and Perspectives in Linear Statistical Inference (LinStat'12, Bedlewo, Poland, July 16-20, 2012). 2012.
@inproceedings{d3df5501865f48388d1b8568079c7926,
title = "Estimating intraclass correlation and its confidence interval in linear mixed models2",
abstract = "The methodology proposed in this study is motivated by an example from the medical field. Oncologists delineate organs for radiotherapy and it is essential that the measurements agree in these procedures. To assess the consistency of measurements among oncologists, on a random sample of subjects, the intraclass correlation (ICC) would yield a suitable estimate for studying the agreement. In technical terms, the ICC is a ratio of sum of variances that are related to differences among measured subjects and the total variance. What variance is considered relevant depends on the design of agreement study; respectively, the number of variance components changes in the numerator and the denominator of the ICC. For statistical inference, it is important but challenging to determine the distribution of estimators of such ratios and to construct the confidence intervals. In most literature, the ICC has been studied for one-way and two-way analysis of variance only. Most proposed approximate methods are based on functions of the mean squares which are model-specific (e.g. two factorial) and lack generalization to higher order (e.g. three factorial) models. The objective of this study is to extend the construction of confidence intervals for the linear mixed models, but in particular to our three-way mixed models for delineation of organs. The generalization will coincide with existing methods for two-way and one-way mixed effects models. To obtain an approximate upper and lower confidence limits, we approximate the ICC with a function of F-distributed variable and a Beta distribution. Our proposed methodology is supported by simulation studies.",
author = "N. Demetrashvili and {Heuvel, van den}, E.R.",
year = "2012",
language = "English",
booktitle = "International Conference on Trends and Perspectives in Linear Statistical Inference (LinStat'12, Bedlewo, Poland, July 16-20, 2012)",

}

Demetrashvili, N & Heuvel, van den, ER 2012, Estimating intraclass correlation and its confidence interval in linear mixed models2. in International Conference on Trends and Perspectives in Linear Statistical Inference (LinStat'12, Bedlewo, Poland, July 16-20, 2012). conference; International Conference on Trends and Perspectives in Linear Statistical Inference; 2012-07-16; 2012-07-20, 16/07/12.

Estimating intraclass correlation and its confidence interval in linear mixed models2. / Demetrashvili, N.; Heuvel, van den, E.R.

International Conference on Trends and Perspectives in Linear Statistical Inference (LinStat'12, Bedlewo, Poland, July 16-20, 2012). 2012.

Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademicpeer-review

TY - GEN

T1 - Estimating intraclass correlation and its confidence interval in linear mixed models2

AU - Demetrashvili, N.

AU - Heuvel, van den, E.R.

PY - 2012

Y1 - 2012

N2 - The methodology proposed in this study is motivated by an example from the medical field. Oncologists delineate organs for radiotherapy and it is essential that the measurements agree in these procedures. To assess the consistency of measurements among oncologists, on a random sample of subjects, the intraclass correlation (ICC) would yield a suitable estimate for studying the agreement. In technical terms, the ICC is a ratio of sum of variances that are related to differences among measured subjects and the total variance. What variance is considered relevant depends on the design of agreement study; respectively, the number of variance components changes in the numerator and the denominator of the ICC. For statistical inference, it is important but challenging to determine the distribution of estimators of such ratios and to construct the confidence intervals. In most literature, the ICC has been studied for one-way and two-way analysis of variance only. Most proposed approximate methods are based on functions of the mean squares which are model-specific (e.g. two factorial) and lack generalization to higher order (e.g. three factorial) models. The objective of this study is to extend the construction of confidence intervals for the linear mixed models, but in particular to our three-way mixed models for delineation of organs. The generalization will coincide with existing methods for two-way and one-way mixed effects models. To obtain an approximate upper and lower confidence limits, we approximate the ICC with a function of F-distributed variable and a Beta distribution. Our proposed methodology is supported by simulation studies.

AB - The methodology proposed in this study is motivated by an example from the medical field. Oncologists delineate organs for radiotherapy and it is essential that the measurements agree in these procedures. To assess the consistency of measurements among oncologists, on a random sample of subjects, the intraclass correlation (ICC) would yield a suitable estimate for studying the agreement. In technical terms, the ICC is a ratio of sum of variances that are related to differences among measured subjects and the total variance. What variance is considered relevant depends on the design of agreement study; respectively, the number of variance components changes in the numerator and the denominator of the ICC. For statistical inference, it is important but challenging to determine the distribution of estimators of such ratios and to construct the confidence intervals. In most literature, the ICC has been studied for one-way and two-way analysis of variance only. Most proposed approximate methods are based on functions of the mean squares which are model-specific (e.g. two factorial) and lack generalization to higher order (e.g. three factorial) models. The objective of this study is to extend the construction of confidence intervals for the linear mixed models, but in particular to our three-way mixed models for delineation of organs. The generalization will coincide with existing methods for two-way and one-way mixed effects models. To obtain an approximate upper and lower confidence limits, we approximate the ICC with a function of F-distributed variable and a Beta distribution. Our proposed methodology is supported by simulation studies.

M3 - Conference contribution

BT - International Conference on Trends and Perspectives in Linear Statistical Inference (LinStat'12, Bedlewo, Poland, July 16-20, 2012)

ER -

Demetrashvili N, Heuvel, van den ER. Estimating intraclass correlation and its confidence interval in linear mixed models2. In International Conference on Trends and Perspectives in Linear Statistical Inference (LinStat'12, Bedlewo, Poland, July 16-20, 2012). 2012