Marginal structural models

Much ado about (almost) nothing

Eyal Shahar, Doron J. Shahar

Research output: Contribution to journalArticle

2 Citations (Scopus)

Abstract

Marginal structural models were developed to account for a so-called time-dependent confounder and to estimate the presumed effect of 'treatment regime' (treatment over time). We present a set of causal axioms, according to which the problem of time-dependent confounding does not exist, and 'treatment regime' affects nothing. Per our axiomatization, marginal structural models do not introduce a new idea of deconfounding, but simply estimate a weighted average of effects. Whenever a weighted average and the weighting scheme can both be rationalized, the models are acceptable. Whenever a weighted average does not estimate an effect (e.g. important effect modification is ignored), or the weights are senseless - the models should not be fit.

Original languageEnglish (US)
Pages (from-to)214-222
Number of pages9
JournalJournal of Evaluation in Clinical Practice
Volume19
Issue number1
DOIs
StatePublished - Feb 2013

Fingerprint

Structural Models
Weights and Measures

Keywords

  • causal diagrams
  • derived variables
  • inverse probability of treatment weighting
  • marginal structural models
  • standardization
  • thought bias

ASJC Scopus subject areas

  • Public Health, Environmental and Occupational Health
  • Health Policy

Cite this

Marginal structural models : Much ado about (almost) nothing. / Shahar, Eyal; Shahar, Doron J.

In: Journal of Evaluation in Clinical Practice, Vol. 19, No. 1, 02.2013, p. 214-222.

Research output: Contribution to journalArticle

@article{f5706ebc95df4173bfb87d85b8ec532f,
title = "Marginal structural models: Much ado about (almost) nothing",
abstract = "Marginal structural models were developed to account for a so-called time-dependent confounder and to estimate the presumed effect of 'treatment regime' (treatment over time). We present a set of causal axioms, according to which the problem of time-dependent confounding does not exist, and 'treatment regime' affects nothing. Per our axiomatization, marginal structural models do not introduce a new idea of deconfounding, but simply estimate a weighted average of effects. Whenever a weighted average and the weighting scheme can both be rationalized, the models are acceptable. Whenever a weighted average does not estimate an effect (e.g. important effect modification is ignored), or the weights are senseless - the models should not be fit.",
keywords = "causal diagrams, derived variables, inverse probability of treatment weighting, marginal structural models, standardization, thought bias",
author = "Eyal Shahar and Shahar, {Doron J.}",
year = "2013",
month = "2",
doi = "10.1111/j.1365-2753.2011.01757.x",
language = "English (US)",
volume = "19",
pages = "214--222",
journal = "Journal of Evaluation in Clinical Practice",
issn = "1356-1294",
publisher = "Wiley-Blackwell",
number = "1",

}

TY - JOUR

T1 - Marginal structural models

T2 - Much ado about (almost) nothing

AU - Shahar, Eyal

AU - Shahar, Doron J.

PY - 2013/2

Y1 - 2013/2

N2 - Marginal structural models were developed to account for a so-called time-dependent confounder and to estimate the presumed effect of 'treatment regime' (treatment over time). We present a set of causal axioms, according to which the problem of time-dependent confounding does not exist, and 'treatment regime' affects nothing. Per our axiomatization, marginal structural models do not introduce a new idea of deconfounding, but simply estimate a weighted average of effects. Whenever a weighted average and the weighting scheme can both be rationalized, the models are acceptable. Whenever a weighted average does not estimate an effect (e.g. important effect modification is ignored), or the weights are senseless - the models should not be fit.

AB - Marginal structural models were developed to account for a so-called time-dependent confounder and to estimate the presumed effect of 'treatment regime' (treatment over time). We present a set of causal axioms, according to which the problem of time-dependent confounding does not exist, and 'treatment regime' affects nothing. Per our axiomatization, marginal structural models do not introduce a new idea of deconfounding, but simply estimate a weighted average of effects. Whenever a weighted average and the weighting scheme can both be rationalized, the models are acceptable. Whenever a weighted average does not estimate an effect (e.g. important effect modification is ignored), or the weights are senseless - the models should not be fit.

KW - causal diagrams

KW - derived variables

KW - inverse probability of treatment weighting

KW - marginal structural models

KW - standardization

KW - thought bias

UR - http://www.scopus.com/inward/record.url?scp=84872940976&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84872940976&partnerID=8YFLogxK

U2 - 10.1111/j.1365-2753.2011.01757.x

DO - 10.1111/j.1365-2753.2011.01757.x

M3 - Article

VL - 19

SP - 214

EP - 222

JO - Journal of Evaluation in Clinical Practice

JF - Journal of Evaluation in Clinical Practice

SN - 1356-1294

IS - 1

ER -