A General Algorithm for Deciding Transportability of Experimental Results

By Elias Bareinboim, Judea Pearl. 2013

224 citations recorded by the source corpus.

Abstract Generalizing empirical findings to new environments, settings, or populations is essential in most scientific explorations. This article treats a particular problem of generalizability, called “transportability”, defined as a license to transfer information learned in experimental studies to a different population, on which only observational studies can be conducted. Given a set of assumptions concerning commonalities and differences between the two populations, Pearl and Bareinboim [1] derived sufficient conditions that permit such transfer to take place. This article summarizes their findings and supplements them with an effective procedure for deciding when and how transportability is feasible. It establishes a necessary and sufficient condition for deciding when causal effects in the target population are estimable from both the statistical information available and the causal information transferred from the experiments. The article further provides a complete algorithm for computing the transport formula, that is, a way of combining observational and experimental information to synthesize bias-free estimate of the desired causal relation. Finally, the article examines the differences between transportability and other variants of generalizability.

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

Citations in the corpus

Domain Adaptation by Using Causal Inference to Predict Invariant Conditional Distributions (2017, methodology): “ standard causal discovery techniques (designed for random variables) jointly on system and intervention variables. 5Our graphical representation also bears some similarities with selection diagrams [Bareinboim and Pearl, 2013], but one crucial difference is that we are modeling the joint distributi”
Transportability of Trial Results Using Inverse Odds of Sampling Weights. (2017, background): “More recently, Bareinboim and Pearl (3) highlighted several key distinctions between generalizability and transportability and introduced a method for deriving a transport formula, which relies on a detailed understanding of the causal relationships among all relevant variables.”
Do-calculus when the True Graph Is Unknown (2015, background): “In the future, we hope our approach can aid the identiﬁcation of causal effects from multiple data sets of non-identical populations (Bareinboim and Pearl, 2013).”
Recovering Causal Effects from Selection Bias (2015, background): “This paper complements recent work on transportability (Pearl and Bareinboim 2014; Bareinboim and Pearl 2013), which deals with transferring causal information across disparate, heterogeneous environments.”
Joint Causal Inference from Multiple Contexts (2016, methodology): “…in (Spirtes et al., 2000), “force variables” in Pearl (1993), “decision variables” in influence diagrams (Dawid, 2002), “regime indicators” in Didelez et al. (2006), “selection variables” in selection diagrams (Bareinboim and Pearl, 2013), and the “environment variable” in Peters et al. (2016).”
Towards multi-modal causability with Graph Neural Networks enabling information fusion for explainable AI (2021, background): “A particular problem of generalizability has been described by (Bareinboim & Pearl, 2013) [20], which is called transportability and can be seen as a ‘‘data fusion framework’’ for the external validation of intervention models and counterfactual queries.”
Generalizing Experimental Findings (2015, background): “The analysis reported in (Pearl and Bareinboim, 2014) has resulted in an algorithmic criterion for deciding whether transportability is feasible and, when confirmed, the algorithm produces an estimand for the desired effects (Bareinboim and Pearl, 2013).”
Causal Transfer Learning (2017, methodology): “(5)Our graphical representation also bears some similarities with selection diagrams [Bareinboim and Pearl, 2013], but one crucial difference is that we are modeling the joint distribution on the intervention and system variables, whereas a selection diagram represents the conditional distribution o”
How Generalizable Is Your Experiment? An Index for Comparing Experimental Samples and Populations (2014, methodology): “Similarly, Bareinboim and Pearl (2013) investigated the assumptions necessary for generalizability—what they call ‘‘transportability’’—using a graph theoretic approach.”
Support and Invertibility in Domain-Invariant Representations (2019, methodology): “We summarize this section in a statement inspired by Lemma 2 in Bareinboim & Pearl (2013).”
Causal inference and the data-fusion problem (2016, unknown): “45, 52, and 54.”
Transportability from Multiple Environments with Limited Experiments: Completeness Results (2014, background): “[6] introduced a formal language for encoding differences and commonalities between domains accompanied with necessary or sufficient conditions under which transportability of empirical findings is feasible between two domains, a source and a target; then, these conditions were extended for a comple”