Efficient Estimation of Average Treatment Effects Using the Estimated Propensity Score 1

By Keisuke Hirano, Guido W. Imbens. 2002

Subfields: Statistics and Probability, Causal Inference & Discovery

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Statistics and Probability

In Causal Inference & Discovery

Citations in the corpus

Doubly Robust Off-policy Value Evaluation for Reinforcement Learning (2015, background): “In medical and public-policy domains (Murphy et al., 2001; Hirano et al., 2003), it is often hard to run a controlled experiment to estimate the treatment effect, and off-policy There are roughly two classes of approaches to off-policy value evaluation.”
Efficiently Breaking the Curse of Horizon in Off-Policy Evaluation with Double Reinforcement Learning (2019, background): “In finite-horizon problems (i.e., when the estimand is ρπ e T ), when the behavior policy is known, IS is unbiased and consistent but its variance tends to be large and it is inefficient (Hirano et al., 2003).”
Breaking the Curse of Horizon: Infinite-Horizon Off-Policy Estimation (2018, background): “This problem is of great practical importance in many application domains where deploying a new policy can be costly or risky, such as medical treatments [26], econometrics [13], recommender systems [19], education [23], Web search [18], advertising and marketing [4, 5, 38, 40].”
More Robust Doubly Robust Off-policy Evaluation (2018, background): “2015), healthcare (e.g., Murphy et al. 2001; Hirano et al. 2003), and education (e.”
Optimal and Adaptive Off-policy Evaluation in Contextual Bandits (2016, background): “, 2015), but it becomes (asymptotically) optimal when the true propensity scores are replaced by suitable estimates (Hirano et al., 2003).”
Double Reinforcement Learning for Efficient Off-Policy Evaluation in Markov Decision Processes (2019, background): “In particular, under M 1 , IS with ˆ ν t = ν t is known to be ineﬃcient (Hirano et al., 2003), which implies it must be ineﬃcient under M 2 as well.”
Towards Optimal Off-Policy Evaluation for Reinforcement Learning with Marginalized Importance Sampling (2019, methodology): “IS-based off-policy evaluation methods have also seen lots of interest recently especially for short-horizon problems, including contextual bandits [Murphy et al., 2001; Hirano et al., 2003; Dudík et al., 2011; Wang et al., 2017].”
Efficient Policy Learning (2017, background): “(2016) show that the double machine learning estimator (9) achieves the semiparametrically efficient rate for estimating Q(π) (Hirano et al., 2003), √ n ( Q̂DML(π)−Q(π) ) ⇒ N (0, V (π)) , (11) V (π) = Var [π(X)τ(X)] + E [ Var [ Y (−1) ∣∣X = Xi] 1− e(Xi) + Var [ Y (+1) ∣∣X = Xi] e(Xi) ] , (12) provid”
Importance Sampling Policy Evaluation with an Estimated Behavior Policy (2018, background): “, 2007; Delyon & Portier, 2016) and causal inference (Hirano et al., 2003; Rosenbaum, 1987).”
Incentives to teach badly: After-school tutoring in developing countries (2014, background): “I show that the empirical results are robust to dropping observations at the tails of the propensity-score distribution and with propensity-score weighting (Hirano et al., 2003).”
Policy Learning With Observational Data (2017, methodology): “The doubly robust form used here is not the only way to construct eﬃcient estimators for the value of a single policy π —for example, Hirano, Imbens, and Ridder (2003) show that inverse-propensity weighting with non-parametrically estimated propensity scores may also be eﬃcient—but it plays a key…”
Double/debiased machine learning for treatment and structural parameters (2017, unknown): “See, for example, the work by Levit (1975), Ibragimov and Hasminskii (1981), Bickel (1982), Robinson (1988), Newey (1990), van der Vaart (1991), Andrews (1994a), Newey (1994), Newey et al.”
Unconditional Quantile Regressions (2007, unknown): “In the first application, we revisit the analysis of birthweight of Koenker and Hallock (2001), where quantile regressions are used to show that there are differential impacts of being a boy or having a black mother, for example, at different quantiles of the conditional birthweight distribution.”
Identification and information in monotone binary models (2007, unknown): “Hirano et al. (2003) report similar results in the context of treatment models where the nuisance parameter is the inverse of a propensity score, in a set of moment restrictions.”
SHARP BOUNDS ON THE DISTRIBUTION OF TREATMENT EFFECTS AND THEIR STATISTICAL INFERENCE (2009, unknown): “…with selection-on-observables, the following two assumptions are often used to evaluate the effect of a treatment or program; see e.g., Rosenbaum and Rubin (1983a,1983b), Hahn (1998), Heckman et al. (1998), Dehejia and Wahba (1999), and Hirano, Imbens, and Ridder (2003), to name only a few.”
A Distributional Approach for Causal Inference Using Propensity Scores (2006, methodology): “We turn to the case where our knowledge about the propensity score is parametric and expressed as π(·;γ ) or model S. Hirano, Imbens, and Ridder (2003) considered nonparametric estimation of the propensity score and showed that the IPW estimator achieves the semiparametric variance bound under…”