Optimally combining propensity score subclasses

Kara Rudolph, K. Ellicott Colson, Elizabeth A. Stuart, Jennifer Ahern

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

Propensity score methods, such as subclassification, are a common approach to control for confounding when estimating causal effects in non-randomized studies. Propensity score subclassification groups individuals into subclasses based on their propensity score values. Effect estimates are obtained within each subclass and then combined by weighting by the proportion of observations in each subclass. Combining subclass-specific estimates by weighting by the inverse variance is a promising alternative approach; a similar strategy is used in meta-analysis for its efficiency. We use simulation to compare performance of each of the two methods while varying (i) the number of subclasses, (ii) extent of propensity score overlap between the treatment and control groups (i.e., positivity), (iii) incorporation of survey weighting, and (iv) presence of heterogeneous treatment effects across subclasses. Both methods perform well in the absence of positivity violations and with a constant treatment effect with weighting by the inverse variance performing slightly better. Weighting by the proportion in subclass performs better in the presence of heterogeneous treatment effects across subclasses. We apply these methods to an illustrative example estimating the effect of living in a disadvantaged neighborhood on risk of past-year anxiety and depressive disorders among U.S. urban adolescents. This example entails practical positivity violations but no evidence of treatment effect heterogeneity. In this case, weighting by the inverse variance when combining across propensity score subclasses results in more efficient estimates that ultimately change inference.

Original languageEnglish (US)
Pages (from-to)4937-4947
Number of pages11
JournalStatistics in Medicine
Volume35
Issue number27
DOIs
StatePublished - Nov 30 2016
Externally publishedYes

Fingerprint

Propensity Score
Weighting
Treatment Effects
Positivity
Proportion
Therapeutics
Estimate
Vulnerable Populations
Depressive Disorder
Causal Effect
Anxiety Disorders
Anxiety
Confounding
Meta-Analysis
Disorder
Overlap
Control Groups
Alternatives
Simulation

Keywords

  • observational studies
  • propensity score
  • stratification
  • subclassification

ASJC Scopus subject areas

  • Epidemiology
  • Statistics and Probability

Cite this

Rudolph, K., Colson, K. E., Stuart, E. A., & Ahern, J. (2016). Optimally combining propensity score subclasses. Statistics in Medicine, 35(27), 4937-4947. https://doi.org/10.1002/sim.7046

Optimally combining propensity score subclasses. / Rudolph, Kara; Colson, K. Ellicott; Stuart, Elizabeth A.; Ahern, Jennifer.

In: Statistics in Medicine, Vol. 35, No. 27, 30.11.2016, p. 4937-4947.

Research output: Contribution to journalArticle

Rudolph, K, Colson, KE, Stuart, EA & Ahern, J 2016, 'Optimally combining propensity score subclasses', Statistics in Medicine, vol. 35, no. 27, pp. 4937-4947. https://doi.org/10.1002/sim.7046
Rudolph, Kara ; Colson, K. Ellicott ; Stuart, Elizabeth A. ; Ahern, Jennifer. / Optimally combining propensity score subclasses. In: Statistics in Medicine. 2016 ; Vol. 35, No. 27. pp. 4937-4947.
@article{fe2ba091c3fa46f9b74ce363171d353a,
title = "Optimally combining propensity score subclasses",
abstract = "Propensity score methods, such as subclassification, are a common approach to control for confounding when estimating causal effects in non-randomized studies. Propensity score subclassification groups individuals into subclasses based on their propensity score values. Effect estimates are obtained within each subclass and then combined by weighting by the proportion of observations in each subclass. Combining subclass-specific estimates by weighting by the inverse variance is a promising alternative approach; a similar strategy is used in meta-analysis for its efficiency. We use simulation to compare performance of each of the two methods while varying (i) the number of subclasses, (ii) extent of propensity score overlap between the treatment and control groups (i.e., positivity), (iii) incorporation of survey weighting, and (iv) presence of heterogeneous treatment effects across subclasses. Both methods perform well in the absence of positivity violations and with a constant treatment effect with weighting by the inverse variance performing slightly better. Weighting by the proportion in subclass performs better in the presence of heterogeneous treatment effects across subclasses. We apply these methods to an illustrative example estimating the effect of living in a disadvantaged neighborhood on risk of past-year anxiety and depressive disorders among U.S. urban adolescents. This example entails practical positivity violations but no evidence of treatment effect heterogeneity. In this case, weighting by the inverse variance when combining across propensity score subclasses results in more efficient estimates that ultimately change inference.",
keywords = "observational studies, propensity score, stratification, subclassification",
author = "Kara Rudolph and Colson, {K. Ellicott} and Stuart, {Elizabeth A.} and Jennifer Ahern",
year = "2016",
month = "11",
day = "30",
doi = "10.1002/sim.7046",
language = "English (US)",
volume = "35",
pages = "4937--4947",
journal = "Statistics in Medicine",
issn = "0277-6715",
publisher = "John Wiley and Sons Ltd",
number = "27",

}

TY - JOUR

T1 - Optimally combining propensity score subclasses

AU - Rudolph, Kara

AU - Colson, K. Ellicott

AU - Stuart, Elizabeth A.

AU - Ahern, Jennifer

PY - 2016/11/30

Y1 - 2016/11/30

N2 - Propensity score methods, such as subclassification, are a common approach to control for confounding when estimating causal effects in non-randomized studies. Propensity score subclassification groups individuals into subclasses based on their propensity score values. Effect estimates are obtained within each subclass and then combined by weighting by the proportion of observations in each subclass. Combining subclass-specific estimates by weighting by the inverse variance is a promising alternative approach; a similar strategy is used in meta-analysis for its efficiency. We use simulation to compare performance of each of the two methods while varying (i) the number of subclasses, (ii) extent of propensity score overlap between the treatment and control groups (i.e., positivity), (iii) incorporation of survey weighting, and (iv) presence of heterogeneous treatment effects across subclasses. Both methods perform well in the absence of positivity violations and with a constant treatment effect with weighting by the inverse variance performing slightly better. Weighting by the proportion in subclass performs better in the presence of heterogeneous treatment effects across subclasses. We apply these methods to an illustrative example estimating the effect of living in a disadvantaged neighborhood on risk of past-year anxiety and depressive disorders among U.S. urban adolescents. This example entails practical positivity violations but no evidence of treatment effect heterogeneity. In this case, weighting by the inverse variance when combining across propensity score subclasses results in more efficient estimates that ultimately change inference.

AB - Propensity score methods, such as subclassification, are a common approach to control for confounding when estimating causal effects in non-randomized studies. Propensity score subclassification groups individuals into subclasses based on their propensity score values. Effect estimates are obtained within each subclass and then combined by weighting by the proportion of observations in each subclass. Combining subclass-specific estimates by weighting by the inverse variance is a promising alternative approach; a similar strategy is used in meta-analysis for its efficiency. We use simulation to compare performance of each of the two methods while varying (i) the number of subclasses, (ii) extent of propensity score overlap between the treatment and control groups (i.e., positivity), (iii) incorporation of survey weighting, and (iv) presence of heterogeneous treatment effects across subclasses. Both methods perform well in the absence of positivity violations and with a constant treatment effect with weighting by the inverse variance performing slightly better. Weighting by the proportion in subclass performs better in the presence of heterogeneous treatment effects across subclasses. We apply these methods to an illustrative example estimating the effect of living in a disadvantaged neighborhood on risk of past-year anxiety and depressive disorders among U.S. urban adolescents. This example entails practical positivity violations but no evidence of treatment effect heterogeneity. In this case, weighting by the inverse variance when combining across propensity score subclasses results in more efficient estimates that ultimately change inference.

KW - observational studies

KW - propensity score

KW - stratification

KW - subclassification

UR - http://www.scopus.com/inward/record.url?scp=84993982925&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84993982925&partnerID=8YFLogxK

U2 - 10.1002/sim.7046

DO - 10.1002/sim.7046

M3 - Article

C2 - 27426623

AN - SCOPUS:84993982925

VL - 35

SP - 4937

EP - 4947

JO - Statistics in Medicine

JF - Statistics in Medicine

SN - 0277-6715

IS - 27

ER -