Fast stable relative risk regression using an overparameterised EM algorithm

Mark W. Donoghoe; Ian C. Marschner

Fast stable relative risk regression using an overparameterised EM algorithm

Research output: Chapter in Book/Report/Conference proceeding › Conference proceeding contribution › peer-review

Abstract

Relative risk regression models can be fitted using a log-link binomial GLM, however standard algorithms can suffer convergence problems. Combinatorial EM (CEM) algorithms that provide stable convergence can be computationally intensive, particularly for large models. We present a new approach using an EM algorithm with an overparameterised model that retains the stability of the CEM algorithm but greatly reduces computing time. This is demonstrated with a small example in which modified Fisher scoring fails to converge to the MLE, and a bootstrap analysis of data from a clinical trial in heart attack patients.

Original language	English
Title of host publication	Proceedings of the 31st International Workshop on Statistical Modelling
Editors	Jean-François Dupuy, Julie Josse
Place of Publication	Amsterdam, the Netherlands
Publisher	Statistical Modelling Society
Pages	93-98
Number of pages	6
Volume	1
Publication status	Published - 2016
Event	International Workshop on Statistical Modelling (31st : 2016) - Rennes, France Duration: 4 Jul 2016 → 8 Jul 2016 Conference number: 31st

Conference

Conference	International Workshop on Statistical Modelling (31st : 2016)
Abbreviated title	IWSM2016
Country/Territory	France
City	Rennes
Period	4/07/16 → 8/07/16

Keywords

Binomial regression
EM algorithm
Relative risk

Cite this

@inproceedings{232606dc6eae4e4cbc1566ae9149f633,

title = "Fast stable relative risk regression using an overparameterised EM algorithm",

abstract = "Relative risk regression models can be fitted using a log-link binomial GLM, however standard algorithms can suffer convergence problems. Combinatorial EM (CEM) algorithms that provide stable convergence can be computationally intensive, particularly for large models. We present a new approach using an EM algorithm with an overparameterised model that retains the stability of the CEM algorithm but greatly reduces computing time. This is demonstrated with a small example in which modified Fisher scoring fails to converge to the MLE, and a bootstrap analysis of data from a clinical trial in heart attack patients.",

keywords = "Binomial regression, EM algorithm, Relative risk",

author = "Donoghoe, {Mark W.} and Marschner, {Ian C.}",

year = "2016",

language = "English",

volume = "1",

pages = "93--98",

editor = "Jean-Fran{\c c}ois Dupuy and Julie Josse",

booktitle = "Proceedings of the 31st International Workshop on Statistical Modelling",

publisher = "Statistical Modelling Society",

note = "International Workshop on Statistical Modelling (31st : 2016), IWSM2016 ; Conference date: 04-07-2016 Through 08-07-2016",

}

Donoghoe, MW & Marschner, IC 2016, Fast stable relative risk regression using an overparameterised EM algorithm. in J-F Dupuy & J Josse (eds), Proceedings of the 31st International Workshop on Statistical Modelling. vol. 1, Statistical Modelling Society, Amsterdam, the Netherlands, pp. 93-98, International Workshop on Statistical Modelling (31st : 2016), Rennes, France, 4/07/16.

Fast stable relative risk regression using an overparameterised EM algorithm. / Donoghoe, Mark W.; Marschner, Ian C.

Proceedings of the 31st International Workshop on Statistical Modelling. ed. / Jean-François Dupuy; Julie Josse. Vol. 1 Amsterdam, the Netherlands : Statistical Modelling Society, 2016. p. 93-98.

Research output: Chapter in Book/Report/Conference proceeding › Conference proceeding contribution › peer-review

TY - GEN

T1 - Fast stable relative risk regression using an overparameterised EM algorithm

AU - Donoghoe, Mark W.

AU - Marschner, Ian C.

N1 - Conference code: 31st

PY - 2016

Y1 - 2016

N2 - Relative risk regression models can be fitted using a log-link binomial GLM, however standard algorithms can suffer convergence problems. Combinatorial EM (CEM) algorithms that provide stable convergence can be computationally intensive, particularly for large models. We present a new approach using an EM algorithm with an overparameterised model that retains the stability of the CEM algorithm but greatly reduces computing time. This is demonstrated with a small example in which modified Fisher scoring fails to converge to the MLE, and a bootstrap analysis of data from a clinical trial in heart attack patients.

AB - Relative risk regression models can be fitted using a log-link binomial GLM, however standard algorithms can suffer convergence problems. Combinatorial EM (CEM) algorithms that provide stable convergence can be computationally intensive, particularly for large models. We present a new approach using an EM algorithm with an overparameterised model that retains the stability of the CEM algorithm but greatly reduces computing time. This is demonstrated with a small example in which modified Fisher scoring fails to converge to the MLE, and a bootstrap analysis of data from a clinical trial in heart attack patients.

KW - Binomial regression

KW - EM algorithm

KW - Relative risk

M3 - Conference proceeding contribution

VL - 1

SP - 93

EP - 98

BT - Proceedings of the 31st International Workshop on Statistical Modelling

A2 - Dupuy, Jean-François

A2 - Josse, Julie

PB - Statistical Modelling Society

CY - Amsterdam, the Netherlands

T2 - International Workshop on Statistical Modelling (31st : 2016)

Y2 - 4 July 2016 through 8 July 2016

ER -

Fast stable relative risk regression using an overparameterised EM algorithm

Abstract

Conference

Keywords

Fingerprint

Cite this