An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies
By Peter C. Austin. 2011
11,173 citations recorded by the source corpus.
The propensity score is the probability of treatment assignment conditional on observed baseline characteristics. The propensity score allows one to design and analyze an observational (nonrandomized) study so that it mimics some of the particular characteristics of a randomized controlled trial. In particular, the propensity score is a balancing score: conditional on the propensity score, the distribution of observed baseline covariates will be similar between treated and untreated subjects. I describe 4 different propensity score methods: matching on the propensity score, stratification on the propensity score, inverse probability of treatment weighting using the propensity score, and covariate adjustment using the propensity score. I describe balance diagnostics for examining whether the propensity score model has been adequately specified. Furthermore, I discuss differences between regression-based methods and propensity score-based methods for the analysis of observational data. I describe different causal average treatment effects and their relationship with propensity score analyses.
Subfields: Statistics and Probability
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
- Mihaela van der Schaar
- Ahmed M. Alaa
- Susan Athey
- Stefan Wager
- Robert Tibshirani
- Jonathan E. Taylor
- Ioana Bica
- Xinkun Nie
Citations in the corpus
- Estimating Counterfactual Treatment Outcomes over Time Through Adversarially Balanced Representations (2020, methodology): “Several ways of handling the selection bias involve using propensity matching (Austin, 2011; Imai & Ratkovic, 2014; Abadie & Imbens, 2016), building representations where treated and un-treated populations had similar distributions (Johansson et al., 2016; Shalit et al., 2017; Li & Fu, 2017; Yao et…”
- Deep Counterfactual Networks with Propensity-Dropout (2017, background): “Classical works have focused on estimating “aver-age” treatment effects through variants of propensity score matching (Rubin, 2011; Austin, 2011; Abadie & Imbens, 2016; Rosenbaum & Rubin, 1983; Rubin, 1973).”
- Learning Representations for Counterfactual Inference (2016, methodology): “The methods include nearest-neighbor matching, propensity score matching, and propensity score re-weighting (Rosen-baum & Rubin, 1983; Rosenbaum, 2002; Austin, 2011).”
- The Selective Labels Problem: Evaluating Algorithmic Predictions in the Presence of Unobservables (2017, methodology): “(2) non-parametric methods such as propensity score matching, nearest-neighbor matching, which do not explicitly model the relationship between observed features, treatments, and outcomes [4, 15, 35, 36, 41].”
- The impact of college‐ and university‐run high school summer programs on students’ end of high school STEM career aspirations (2018, background): “This uncertainty can be dealt with by weighting the participant and comparison groups such that they are similar along demographic lines and across other characteristics (Austin, 2011; Linden &Adams, 2012; Rosenbaum&Rubin, 1983; Thoemmes &Ong, 2016).”
- Quantifying Engagement with Citations on Wikipedia (2020, positive): “Finding such exact matches is unrealistic in practice, so we resort to propensity score matching [4], which provides a viable solution.”
- A Survey of Learning Causality with Data (2018, methodology): “Despite its popularity, note that the validity of a linear logistic model can be suspected and nonparametric alternatives are readily available.”
- Mobile Money and the Economy: A Review of the Evidence (2018, background): “However, conditioning on the Propensity Score need not balance unmeasured covariates; and even the balance-checking between measured co-variates is problematic because the criteria for adequate balance are ill-defined (see Hill (2008) and Austin (2011)).”
- Why adaptively collected data have negative bias and how to correct for it (2017, methodology): “We also experimented with propensity matching, a commonly used method that weights each observed value of a distribution by one over the probability that this distribution is selected [Austin, 2011].”
- A Survey of Learning Causality with Data: Problems and Methods (2018, methodology): “A combination of regression adjustment and propensity score stratification can also be used to account for the difference between instances in the same stratum [12, 66, 86].”
- Imitation-Regularized Offline Learning (2019, background): “On policy imitation, a related topic is to learn the probabilities through propensity fitting [29, 1].”
- Robust Tree-based Causal Inference for Complex Ad Effectiveness Analysis (2015, methodology): “In addition, when the outcome is binary, prior research suggested that at least 8 events should be observed for every covariate in the regression model [9], which prevents its application in practice [2].”
- Towards Non-I.I.D. image classification: A dataset and baselines (2019, methodology): “Popular methods based on observational data to estimate the causal effect of a treat-ment on the outcome include propensity score matching [36,37] , markov blankets [38] and confounder balancing [39,40] and etc. [41] .”