The "Residential" effect fallacy in neighborhood and health studies: Formal definition, empirical identification, and correction

Basile Chaix, Dustin Duncan, Julie Vallée, Anne Vernez-Moudon, Tarik Benmarhnia, Yan Kestens

Research output: Contribution to journalArticle

Abstract

Background: Because of confounding from the urban/rural and socioeconomic organizations of territories and resulting correlation between residential and nonresidential exposures, classically estimated residential neighborhood-outcome associations capture nonresidential environment effects, overestimating residential intervention effects. Our study diagnosed and corrected this "residential" effect fallacy bias applicable to a large fraction of neighborhood and health studies. Methods: Our empirical application investigated the effect that hypothetical interventions raising the residential number of services would have on the probability that a trip is walked. Using global positioning systems tracking and mobility surveys over 7 days (227 participants and 7440 trips), we employed a multilevel linear probability model to estimate the trip-level association between residential number of services and walking to derive a naïve intervention effect estimate and a corrected model accounting for numbers of services at the residence, trip origin, and trip destination to determine a corrected intervention effect estimate (true effect conditional on assumptions). Results: There was a strong correlation in service densities between the residential neighborhood and nonresidential places. From the naïve model, hypothetical interventions raising the residential number of services to 200, 500, and 1000 were associated with an increase by 0.020, 0.055, and 0.109 of the probability of walking in the intervention groups. Corrected estimates were of 0.007, 0.019, and 0.039. Thus, naïve estimates were overestimated by multiplicative factors of 3.0, 2.9, and 2.8. Conclusions: Commonly estimated residential intervention-outcome associations substantially overestimate true effects. Our somewhat paradoxical conclusion is that to estimate residential effects, investigators critically need information on nonresidential places visited.

Original languageEnglish (US)
Pages (from-to)789-797
Number of pages9
JournalEpidemiology
Volume28
Issue number6
DOIs
StatePublished - Jan 1 2017

Fingerprint

Walking
Health
Geographic Information Systems
Linear Models
Research Personnel
Organizations
Surveys and Questionnaires

ASJC Scopus subject areas

  • Epidemiology

Cite this

The "Residential" effect fallacy in neighborhood and health studies : Formal definition, empirical identification, and correction. / Chaix, Basile; Duncan, Dustin; Vallée, Julie; Vernez-Moudon, Anne; Benmarhnia, Tarik; Kestens, Yan.

In: Epidemiology, Vol. 28, No. 6, 01.01.2017, p. 789-797.

Research output: Contribution to journalArticle

Chaix, Basile ; Duncan, Dustin ; Vallée, Julie ; Vernez-Moudon, Anne ; Benmarhnia, Tarik ; Kestens, Yan. / The "Residential" effect fallacy in neighborhood and health studies : Formal definition, empirical identification, and correction. In: Epidemiology. 2017 ; Vol. 28, No. 6. pp. 789-797.
@article{bc36e6710ee745548db4625c4ca93602,
title = "The {"}Residential{"} effect fallacy in neighborhood and health studies: Formal definition, empirical identification, and correction",
abstract = "Background: Because of confounding from the urban/rural and socioeconomic organizations of territories and resulting correlation between residential and nonresidential exposures, classically estimated residential neighborhood-outcome associations capture nonresidential environment effects, overestimating residential intervention effects. Our study diagnosed and corrected this {"}residential{"} effect fallacy bias applicable to a large fraction of neighborhood and health studies. Methods: Our empirical application investigated the effect that hypothetical interventions raising the residential number of services would have on the probability that a trip is walked. Using global positioning systems tracking and mobility surveys over 7 days (227 participants and 7440 trips), we employed a multilevel linear probability model to estimate the trip-level association between residential number of services and walking to derive a na{\"i}ve intervention effect estimate and a corrected model accounting for numbers of services at the residence, trip origin, and trip destination to determine a corrected intervention effect estimate (true effect conditional on assumptions). Results: There was a strong correlation in service densities between the residential neighborhood and nonresidential places. From the na{\"i}ve model, hypothetical interventions raising the residential number of services to 200, 500, and 1000 were associated with an increase by 0.020, 0.055, and 0.109 of the probability of walking in the intervention groups. Corrected estimates were of 0.007, 0.019, and 0.039. Thus, na{\"i}ve estimates were overestimated by multiplicative factors of 3.0, 2.9, and 2.8. Conclusions: Commonly estimated residential intervention-outcome associations substantially overestimate true effects. Our somewhat paradoxical conclusion is that to estimate residential effects, investigators critically need information on nonresidential places visited.",
author = "Basile Chaix and Dustin Duncan and Julie Vall{\'e}e and Anne Vernez-Moudon and Tarik Benmarhnia and Yan Kestens",
year = "2017",
month = "1",
day = "1",
doi = "10.1097/EDE.0000000000000726",
language = "English (US)",
volume = "28",
pages = "789--797",
journal = "Epidemiology",
issn = "1044-3983",
publisher = "Lippincott Williams and Wilkins",
number = "6",

}

TY - JOUR

T1 - The "Residential" effect fallacy in neighborhood and health studies

T2 - Formal definition, empirical identification, and correction

AU - Chaix, Basile

AU - Duncan, Dustin

AU - Vallée, Julie

AU - Vernez-Moudon, Anne

AU - Benmarhnia, Tarik

AU - Kestens, Yan

PY - 2017/1/1

Y1 - 2017/1/1

N2 - Background: Because of confounding from the urban/rural and socioeconomic organizations of territories and resulting correlation between residential and nonresidential exposures, classically estimated residential neighborhood-outcome associations capture nonresidential environment effects, overestimating residential intervention effects. Our study diagnosed and corrected this "residential" effect fallacy bias applicable to a large fraction of neighborhood and health studies. Methods: Our empirical application investigated the effect that hypothetical interventions raising the residential number of services would have on the probability that a trip is walked. Using global positioning systems tracking and mobility surveys over 7 days (227 participants and 7440 trips), we employed a multilevel linear probability model to estimate the trip-level association between residential number of services and walking to derive a naïve intervention effect estimate and a corrected model accounting for numbers of services at the residence, trip origin, and trip destination to determine a corrected intervention effect estimate (true effect conditional on assumptions). Results: There was a strong correlation in service densities between the residential neighborhood and nonresidential places. From the naïve model, hypothetical interventions raising the residential number of services to 200, 500, and 1000 were associated with an increase by 0.020, 0.055, and 0.109 of the probability of walking in the intervention groups. Corrected estimates were of 0.007, 0.019, and 0.039. Thus, naïve estimates were overestimated by multiplicative factors of 3.0, 2.9, and 2.8. Conclusions: Commonly estimated residential intervention-outcome associations substantially overestimate true effects. Our somewhat paradoxical conclusion is that to estimate residential effects, investigators critically need information on nonresidential places visited.

AB - Background: Because of confounding from the urban/rural and socioeconomic organizations of territories and resulting correlation between residential and nonresidential exposures, classically estimated residential neighborhood-outcome associations capture nonresidential environment effects, overestimating residential intervention effects. Our study diagnosed and corrected this "residential" effect fallacy bias applicable to a large fraction of neighborhood and health studies. Methods: Our empirical application investigated the effect that hypothetical interventions raising the residential number of services would have on the probability that a trip is walked. Using global positioning systems tracking and mobility surveys over 7 days (227 participants and 7440 trips), we employed a multilevel linear probability model to estimate the trip-level association between residential number of services and walking to derive a naïve intervention effect estimate and a corrected model accounting for numbers of services at the residence, trip origin, and trip destination to determine a corrected intervention effect estimate (true effect conditional on assumptions). Results: There was a strong correlation in service densities between the residential neighborhood and nonresidential places. From the naïve model, hypothetical interventions raising the residential number of services to 200, 500, and 1000 were associated with an increase by 0.020, 0.055, and 0.109 of the probability of walking in the intervention groups. Corrected estimates were of 0.007, 0.019, and 0.039. Thus, naïve estimates were overestimated by multiplicative factors of 3.0, 2.9, and 2.8. Conclusions: Commonly estimated residential intervention-outcome associations substantially overestimate true effects. Our somewhat paradoxical conclusion is that to estimate residential effects, investigators critically need information on nonresidential places visited.

UR - http://www.scopus.com/inward/record.url?scp=85034983645&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85034983645&partnerID=8YFLogxK

U2 - 10.1097/EDE.0000000000000726

DO - 10.1097/EDE.0000000000000726

M3 - Article

VL - 28

SP - 789

EP - 797

JO - Epidemiology

JF - Epidemiology

SN - 1044-3983

IS - 6

ER -