Estimating treatment effect heterogeneity in randomized program evaluation

By Kosuke Imai, Marc Ratkovic. 2013

534 citations recorded by the source corpus.

When evaluating the efficacy of social programs and medical treatments using randomized experiments, the estimated overall average causal effect alone is often of limited value and the researchers must investigate when the treatments do and do not work. Indeed, the estimation of treatment effect heterogeneity plays an essential role in (1) selecting the most effective treatment from a large number of available treatments, (2) ascertaining subpopulations for which a treatment is effective or harmful, (3) designing individualized optimal treatment regimes, (4) testing for the existence or lack of heterogeneous treatment effects, and (5) generalizing causal effect estimates obtained from an experimental sample to a target population. In this paper, we formulate the estimation of heterogeneous treatment effects as a variable selection problem. We propose a method that adapts the Support Vector Machine classifier by placing separate sparsity constraints over the pre-treatment parameters and causal heterogeneity parameters of interest. The proposed method is motivated by and applied to two well-known randomized evaluation studies in the social sciences. Our method selects the most effective voter mobilization strategies from a large number of alternative strategies, and it also identifies the characteristics of workers who greatly benefit from (or are negatively affected by) a job training program. In our simulation studies, we find that the proposed method often outperforms some commonly used alternatives.

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

The Econometrics of Randomized Experiments (2016, background): “Athey, Eckles and Imbens (2015) also consider richer hypotheses, such as hypotheses about what types of link definitions correspond to meaningful peer effects; they propose a test of the hypothesis that a sparser definition of the network is sufficient to capture relationships for which treating a f”
The State of Applied Econometrics - Causality and Policy Evaluation (2016, positive): “In addition, it is important to recognize that regularized regression models reward parsimony: if there are several correlated variables, LASSO will prefer to put more weight on one and drop the others.”
Efficient Policy Learning (2017, methodology): “the treatment effect is weaker than the main effect, this re-parametrization can interact with the lasso penalty in a way that improves the resulting fit (Imai and Ratkovic, 2013).”
Quasi-oracle estimation of heterogeneous treatment effects (2017, methodology): “For example, with the lasso, Imai and Ratkovic (2013) advocate replacing (6) with a single lasso as follows, ˆ b, ˆ δ = argmin b,δ n i =1 where then ˆ τ ( x ) = x (cid:62) ˆ δ .”
A Survey of Learning Causality with Data: Problems and Methods (2018, background): “Different from many other studies, in the study of the efficacy for social programs and medical treatments by Imai and Ratkovic [65], there are many different strategies (treatments) available for each instance.”
Research Dilemmas with Behavioral Big Data (2017, unknown): “(e.g., Imai and Ratkovic 43 and Yahav et al. 31 ).”
Decomposing Treatment Effect Variation (2016, background): “…2017), biostatistics (Huang, Gilbert, and Janes 2012;Matsouaka, Li, andCai 2014), education (Raudenbush and Bloom 2015), economics (Heckman, Smith, and Clements 1997; Crump et al. 2008; and Djebbari and Smith 2008), political science (Green and Kern 2012; Imai and Ratkovic 2013), and other areas.”
From Local to Global: External Validity in a Fertility Natural Experiment (2015, unknown): “Crump et al. (2008), Green and Kern (2012), and Imai and Ratkovic (2013) developed nonparametric methods for characterizing effect heterogeneity, including sieve estimators, Bayesian additive regression trees, and support vector machines, respectively.”
Estimating Treatment Effects with Causal Forests: An Application (2019, unknown): “1 Methodology and Motivation There has been considerable recent interest in methods for heterogeneous treatment effect estimation in observational studies (Athey and Imbens, 2016; Athey, Tibshirani, and Wager, 2019; Ding, Feller, and Miratrix, 2016; Dorie, Hill, Shalit, Scott, and Cervone, 2017; Hah”
Discovery of Treatments from Text Corpora (2016, methodology): “Interaction effects could be modeled, but it would require us to make much stronger parametric assumptions using a method for heterogeneous treatments such as (Imai and Ratkovic, 2013).”
A Nonparametric Bayesian Analysis of Heterogenous Treatment Effects in Digital Experimentation (2014, background): “The framework implied by (18) is related to a semiparametric literature (e.g., Hill 2011; Green and Kern 2012; Grimmer, Messing, and Westwood 2013; Imai and Ratkovic 2013) that studies the difference between ﬂexible regression functions in each of the treatment groups.”
Data ex Machina: Introduction to Big Data (2017, background): “Others are developing algorithms that seek to inductively identify groups for whom effects are especially large (Athey & Imbens 2016, Green & Kern 2012, Imai & Ratkovic 2013, Taddy et al. 2016).”