Recovering from Selection Bias in Causal and Statistical Inference

By Elias Bareinboim, Jin Tian, Judea Pearl. 2014

199 citations recorded by the source corpus.

Selection bias is caused by preferential exclusion of units from the samples and represents a major obstacle to valid causal and statistical inferences; it cannot be removed by randomized experiments and can rarely be detected in either experimental or observational studies. In this paper, we provide complete graphical and algorithmic conditions for recovering conditional probabilities from selection biased data. We also provide graphical conditions for recoverability when unbiased data is available over a subset of the variables. Finally, we provide a graphical condition that generalizes the backdoor criterion and serves to recover causal effects when the data is collected under preferential selection.

Subfields: Artificial Intelligence, General Decision Sciences, 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 Artificial Intelligence

In General Decision Sciences

In Statistics and Probability

Citations in the corpus

Multi-Source Domain Adaptation: A Causal View (2015, methodology): “They, for instance, help characterize transportability of experimental ﬁndings (Pearl and Bareinboim 2011) or recoverability from selection bias (Bareinboim, Tian, and Pearl 2014).”
A Complete Generalized Adjustment Criterion (2015, methodology): “Future work might also explore under which circumstances our restriction to not allow for latent selection variables might be relaxed, o r whether our criterion could be combined with methods to recover from selection bias (Bareinboim et al., 2014).”
Causal and Counterfactual Inference (2018, background): “Another problem that falls under the Data Fusion umbrella is that of “Selection Bias” (Bareinboim et al., 2014), which requires a generalization from a subpopulation selected for a study to the population at large, the target of the intended policy.”
TRYGVE HAAVELMO AND THE EMERGENCE OF CAUSAL CALCULUS (2014, unknown): “A complete theory can be found in Bareinboim and Pearl (2012) and Bareinboim et al. (2014).”
Generalized Adjustment Under Confounding and Selection Biases (2018, background): “Formally, the task of estimating a probabilistic quantity from a selection-biased distribution is known as recovering from selection bias (Bareinboim, Tian, and Pearl 2014).”
The Causal Foundations of Structural Equation Modeling (2012, background): “The analysis of (Bareinboim et al., 2014) identifies conditions under which such non-representative selection of units can be neutralized.”
Fides: Towards a Platform for Responsible Data Science (2017, background): “For example, if the school-age youth are disproportionately male, one might infer that the sampling was biased [6], and the results of downstream analyses could be automatically flagged.”
Robust Text Classification in the Presence of Confounding Bias (2016, background): “In the machine learning community, selection bias has received some attention (Zadrozny 2004; Sugiyama, Kraule-dat, and M¨uller 2007; Bareinboim, Tian, and Pearl 2014).”
An examination of multivariable Mendelian randomization in the single-sample and two-sample summary data settings (2018, result): “MVMR, by contrast, will then provide a consistent estimator of the direct effect of the exposure on the outcome; the consistency of IV analysis under a range of scenarios that include those discussed here has been proved elsewhere.(3,4,30) These simulation results also highlight that MVMR does not i”
Recovering Causal Effects from Selection Bias (2015, background): “Further, a complete condition for recovering conditional probabilities from selection biased data has been given in (Bareinboim, Tian, and Pearl 2014) and read as follows: Theorem 1 The conditional distribution P(y|t) is recoverable (as P(y|t, S = 1)) if and only if (Y ⊥ S |T ).”
Causality matters in medical imaging (2019, background): “If selection is purely image-based ( X → S ), we may exploit the conditional independence S ⊥⊥ Y | X , which implies that the predictive relation is directly recoverable 29 , i.e.”
Generalizing Experimental Findings (2015, background): “3 (b) of (Pearl and Bareinboim, 2014 ) where a disparity in the average reading ability of two cities requires two different treatments, depending on what causes the disparity.”
Causal inference and the data-fusion problem (2016, methodology): “Definition 4 [Selection backdoor criterion (38)]: Let a set Z of variables be partitioned into Z+ ∪Z− such that Z+ contains all nondescendants of X and Z− the descendants of X, and letGs stand for the graph that includes the sampling mechanism S.”
Multiple Causes: A Causal Graphical View (2019, methodology): “It aligns with common conditions required by recovery under selection bias (e.g., Theorem 2 of Bareinboim et al. (2014)).”