Nonparametric Causal Effects Based on Incremental Propensity Score Interventions

Kennedy, Edward H.

doi:10.1080/01621459.2017.1422737

Cited by 103 publications

(140 citation statements)

References 38 publications

Supporting

Mentioning

137

Contrasting

Order By: Relevance

“…Alternatively, a tilted intervention distribution may be defined as

g_{δ} false(a false| w false) = \frac{\exp (δ a) g (a | w)}{\int \exp (δ a) g (a | w) d κ (a)},

for

δ \in R

, letting the hypothetical post‐intervention exposure A δ be a random draw from g δ , conditional on the natural value of the observed covariates W . For binary A , Kennedy () proposed to evaluate the total effect of a binary exposure A in terms of incremental propensity score interventions that replace the propensity score g (1| w ) with a shifted version based on multiplying the odds of exposure by a user‐given parameter δ ′. In particular, the post‐intervention propensity score is given by

g_{δ^{'}} false(1 false| w false) = \frac{δ^{'} g false(1 false| w false)}{δ^{'} g false(1 false| w false) + 1 - g false(1 false| w false)},

for

0 < δ^{'} < \infty

.…”

Section: Mediation Analysis For Population Intervention Effectsmentioning

confidence: 99%

“…In particular, the post‐intervention propensity score is given by

g_{δ^{'}} false(1 false| w false) = \frac{δ^{'} g false(1 false| w false)}{δ^{'} g false(1 false| w false) + 1 - g false(1 false| w false)},

for

0 < δ^{'} < \infty

. The proposal of Kennedy () is thus a case of exponential tilting (3) under the parameterization

δ^{'} = \exp false(δ false)

. This choice of parameterization is motivated by the fact that

δ^{'}

can be interpreted as an odds ratio indicating how the intervention changes the odds of exposure.…”

Section: Mediation Analysis For Population Intervention Effectsmentioning

confidence: 99%

“…Estimation of total effects of stochastic interventions was first considered by Stock () and has been the subject of recent study (Robins et al ., ; Didelez et al ., ; Tian, ; Pearl, ; Taubman et al ., ; Stitelman et al ., ; Díaz and van der Laan, ; Dudík et al ., ; Haneuse and Rotnitzky, ; Young et al ., ). Particularly relevant to this work are the methods of Díaz and van der Laan (), Haneuse and Rotnitzky (), who defined total effects for modified treatment policies, and Kennedy (), who studied identification and estimation of the total effect of propensity score interventions that shift a binary exposure distribution. These works do not address decomposition of the effects of stochastic interventions on the exposure into direct and indirect effects, which is the central theme of the present work.…”

Section: Introductionmentioning

confidence: 99%

“…Díaz and van der Laan () studied the effect of a (hypothetical) policy that enforces pollution levels below a certain cut‐off point. Kennedy () showed that stochastic interventions can also be used in longitudinal studies to define and estimate total effects without reliance on the positivity assumption.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Causal Mediation Analysis for Stochastic Interventions

Díaz

Hejazi

2020

Journal of the Royal Statistical Society Series B: Statistical Methodology

View full text Add to dashboard Cite

Mediation analysis in causal inference has traditionally focused on binary exposures and deterministic interventions, and a decomposition of the average treatment effect in terms of direct and indirect effects. In this paper we present an analogous decomposition of the population intervention effect, defined through stochastic interventions on the exposure. Population intervention effects provide a generalized framework in which a variety of interesting causal contrasts can be defined, including effects for continuous and categorical exposures. We show that identification of direct and indirect effects for the population intervention effect requires weaker assumptions than its average treatment effect counterpart, under the assumption of no mediator-outcome confounders affected by exposure. In particular, identification of direct effects is guaranteed in experiments that randomize the exposure and the mediator. We discuss various estimators of the direct and indirect effects, including substitution, re-weighted, and efficient estimators based on flexible regression techniques, allowing for multivariate mediators. Our efficient estimator is asymptotically linear under a condition requiring n 1/4 -consistency of certain regression functions. We perform a simulation study in which we assess the finite-sample properties of our proposed estimators. We present the results of an illustrative study where we assess the effect of participation in a sports team on BMI among children, using mediators such as exercise habits, daily consumption of snacks, and overweight status.

show abstract

“…Alternatively, a tilted intervention distribution may be defined as

g_{δ} false(a false| w false) = \frac{\exp (δ a) g (a | w)}{\int \exp (δ a) g (a | w) d κ (a)},

for

δ \in R

g_{δ^{'}} false(1 false| w false) = \frac{δ^{'} g false(1 false| w false)}{δ^{'} g false(1 false| w false) + 1 - g false(1 false| w false)},

for

0 < δ^{'} < \infty

.…”

Section: Mediation Analysis For Population Intervention Effectsmentioning

confidence: 99%

“…In particular, the post‐intervention propensity score is given by

g_{δ^{'}} false(1 false| w false) = \frac{δ^{'} g false(1 false| w false)}{δ^{'} g false(1 false| w false) + 1 - g false(1 false| w false)},

for

0 < δ^{'} < \infty

. The proposal of Kennedy () is thus a case of exponential tilting (3) under the parameterization

δ^{'} = \exp false(δ false)

. This choice of parameterization is motivated by the fact that

δ^{'}

can be interpreted as an odds ratio indicating how the intervention changes the odds of exposure.…”

Section: Mediation Analysis For Population Intervention Effectsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Causal Mediation Analysis for Stochastic Interventions

Díaz

Hejazi

2020

Journal of the Royal Statistical Society Series B: Statistical Methodology

View full text Add to dashboard Cite

show abstract

“…However, misspecification of regression models can lead to extreme Cheng Ju and David Benkeser contributed equally to this manuscript. bias in estimates of the ATE (Kang and Schafer, 2007), which has led to growing interest in the use of nonparametric methods (Hubbard et al, 2000;Robins and van der Vaart, 2006); locally efficient estimation of the treatment-specific survival distribution with right censored data and covariates in observational studies (Farrell, 2015;Kennedy, 2018); and adaptive regression techniques to estimate the ATE (van der Laan and Rubin, 2006;Lee et al, 2010;Karim and Platt, 2017;Karim et al, 2018;Wyss et al, 2018). In particular, the field of targeted learning has emerged as a paradigm for deriving formal statistical inference about estimated treatment effects when machine learning techniques are used to fit regressions (van der Rose, 2011, 2018).…”

Section: Introductionmentioning

confidence: 99%

Robust inference on the average treatment effect using the outcome highly adaptive lasso

Benkeser

Laan

2019

Biometrics

View full text Add to dashboard Cite

Many estimators of the average effect of a treatment on an outcome require estimation of the propensity score, the outcome regression, or both. It is often beneficial to utilize flexible techniques, such as semiparametric regression or machine learning, to estimate these quantities. However, optimal estimation of these regressions does not necessarily lead to optimal estimation of the average treatment effect, particularly in settings with strong instrumental variables. A recent proposal addressed these issues via the outcome‐adaptive lasso, a penalized regression technique for estimating the propensity score that seeks to minimize the impact of instrumental variables on treatment effect estimators. However, a notable limitation of this approach is that its application is restricted to parametric models. We propose a more flexible alternative that we call the outcome highly adaptive lasso. We discuss the large sample theory for this estimator and propose closed‐form confidence intervals based on the proposed estimator. We show via simulation that our method offers benefits over several popular approaches.

show abstract

Machine Learning and Causal Inference

Zivich

Breskin

Kennedy

2022

Wiley StatsRef: Statistics Reference Online

View full text Add to dashboard Cite

In this article, we review the motivation behind the use of machine learning or data‐adaptive algorithms for the estimation of causal effects. To guide our discussion, we focus on the estimation of the causal mean as a didactic example. Estimation of the causal mean involves estimation of at least one nuisance model for which machine learning is considered. Two key challenges to reliable application of machine learning for causal inference are reviewed: slower than parametric convergence, and complexity of the nuisance model and its estimator. Current approaches to address each of these challenges are then reviewed. We then summarize the extensions to other estimands, identification strategies, and estimators. We conclude with practical advice for application.

show abstract

Nonparametric Causal Effects Based on Incremental Propensity Score Interventions

Cited by 103 publications

References 38 publications

Causal Mediation Analysis for Stochastic Interventions

Causal Mediation Analysis for Stochastic Interventions

Robust inference on the average treatment effect using the outcome highly adaptive lasso

Machine Learning and Causal Inference

Contact Info

Product

Resources

About