On two-stage estimation of structural instrumental variable models

Byeongyeob Choi, Jason P. Fine, M. Alan Brookhart

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

Two-stage least squares estimation is popular for structural equation models with unmeasured confounders. In such models, both the outcome and the exposure are assumed to follow linear models conditional on the measured confounders and instrumental variable, which is related to the outcome only via its relation with the exposure. We consider data where both the outcome and the exposure may be incompletely observed, with particular attention to the case where both are censored event times.A general class of two-stage minimum distance estimators is proposed that separately fits linear models for the outcome and exposure and then uses a minimum distance criterion based on the reduced-form model for the outcome to estimate the regression parameters of interest. An optimal minimum distance estimator is identified which may be superior to the usual two-stage least squares estimator with fully observed data. Simulation studies demonstrate that the proposed methods perform well with realistic sample sizes. Their practical utility is illustrated in a study of the comparative effectiveness of colon cancer treatments, where the effect of chemotherapy on censored survival times may be confounded with patient status.

Original languageEnglish (US)
Pages (from-to)881-899
Number of pages19
JournalBiometrika
Volume104
Issue number4
DOIs
StatePublished - Dec 1 2017

Fingerprint

Two-stage Estimation
Instrumental Variables
Least-Squares Analysis
Linear Models
least squares
Minimum Distance Estimator
Two-stage Least Squares
Structural Models
linear models
Sample Size
Colonic Neoplasms
Linear Model
colorectal neoplasms
Drug Therapy
drug therapy
Survival
Structural Equation Model
Oncology
Model
Chemotherapy

Keywords

  • Censored data
  • Endogeneity
  • Instrumental variable
  • Resampling
  • Unmeasured confounder.

ASJC Scopus subject areas

  • Statistics and Probability
  • Mathematics(all)
  • Agricultural and Biological Sciences (miscellaneous)
  • Agricultural and Biological Sciences(all)
  • Statistics, Probability and Uncertainty
  • Applied Mathematics

Cite this

On two-stage estimation of structural instrumental variable models. / Choi, Byeongyeob; Fine, Jason P.; Brookhart, M. Alan.

In: Biometrika, Vol. 104, No. 4, 01.12.2017, p. 881-899.

Research output: Contribution to journalArticle

Choi, Byeongyeob ; Fine, Jason P. ; Brookhart, M. Alan. / On two-stage estimation of structural instrumental variable models. In: Biometrika. 2017 ; Vol. 104, No. 4. pp. 881-899.
@article{705ee4f4e82148118d648bd9bf5dff4e,
title = "On two-stage estimation of structural instrumental variable models",
abstract = "Two-stage least squares estimation is popular for structural equation models with unmeasured confounders. In such models, both the outcome and the exposure are assumed to follow linear models conditional on the measured confounders and instrumental variable, which is related to the outcome only via its relation with the exposure. We consider data where both the outcome and the exposure may be incompletely observed, with particular attention to the case where both are censored event times.A general class of two-stage minimum distance estimators is proposed that separately fits linear models for the outcome and exposure and then uses a minimum distance criterion based on the reduced-form model for the outcome to estimate the regression parameters of interest. An optimal minimum distance estimator is identified which may be superior to the usual two-stage least squares estimator with fully observed data. Simulation studies demonstrate that the proposed methods perform well with realistic sample sizes. Their practical utility is illustrated in a study of the comparative effectiveness of colon cancer treatments, where the effect of chemotherapy on censored survival times may be confounded with patient status.",
keywords = "Censored data, Endogeneity, Instrumental variable, Resampling, Unmeasured confounder.",
author = "Byeongyeob Choi and Fine, {Jason P.} and Brookhart, {M. Alan}",
year = "2017",
month = "12",
day = "1",
doi = "10.1093/biomet/asx056",
language = "English (US)",
volume = "104",
pages = "881--899",
journal = "Biometrika",
issn = "0006-3444",
publisher = "Oxford University Press",
number = "4",

}

TY - JOUR

T1 - On two-stage estimation of structural instrumental variable models

AU - Choi, Byeongyeob

AU - Fine, Jason P.

AU - Brookhart, M. Alan

PY - 2017/12/1

Y1 - 2017/12/1

N2 - Two-stage least squares estimation is popular for structural equation models with unmeasured confounders. In such models, both the outcome and the exposure are assumed to follow linear models conditional on the measured confounders and instrumental variable, which is related to the outcome only via its relation with the exposure. We consider data where both the outcome and the exposure may be incompletely observed, with particular attention to the case where both are censored event times.A general class of two-stage minimum distance estimators is proposed that separately fits linear models for the outcome and exposure and then uses a minimum distance criterion based on the reduced-form model for the outcome to estimate the regression parameters of interest. An optimal minimum distance estimator is identified which may be superior to the usual two-stage least squares estimator with fully observed data. Simulation studies demonstrate that the proposed methods perform well with realistic sample sizes. Their practical utility is illustrated in a study of the comparative effectiveness of colon cancer treatments, where the effect of chemotherapy on censored survival times may be confounded with patient status.

AB - Two-stage least squares estimation is popular for structural equation models with unmeasured confounders. In such models, both the outcome and the exposure are assumed to follow linear models conditional on the measured confounders and instrumental variable, which is related to the outcome only via its relation with the exposure. We consider data where both the outcome and the exposure may be incompletely observed, with particular attention to the case where both are censored event times.A general class of two-stage minimum distance estimators is proposed that separately fits linear models for the outcome and exposure and then uses a minimum distance criterion based on the reduced-form model for the outcome to estimate the regression parameters of interest. An optimal minimum distance estimator is identified which may be superior to the usual two-stage least squares estimator with fully observed data. Simulation studies demonstrate that the proposed methods perform well with realistic sample sizes. Their practical utility is illustrated in a study of the comparative effectiveness of colon cancer treatments, where the effect of chemotherapy on censored survival times may be confounded with patient status.

KW - Censored data

KW - Endogeneity

KW - Instrumental variable

KW - Resampling

KW - Unmeasured confounder.

UR - http://www.scopus.com/inward/record.url?scp=85039150036&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85039150036&partnerID=8YFLogxK

U2 - 10.1093/biomet/asx056

DO - 10.1093/biomet/asx056

M3 - Article

AN - SCOPUS:85039150036

VL - 104

SP - 881

EP - 899

JO - Biometrika

JF - Biometrika

SN - 0006-3444

IS - 4

ER -