Causal Transportability
Routing Summary
This topic covers external validity: when and how a causal effect measured in one population (a trial, test market or campaign) carries over to another. It has two halves. The graphical theory covers selection diagrams, do-calculus transport formulas and data fusion. The statistical estimation covers participation weights, outcome models, doubly robust estimators and empirical evidence of site-selection bias. There are 5 sub-topics and 36 notes.
- Need the formal definition of transportability, selection diagrams and S-admissibility? → Pearl & Bareinboim 2014
- Need one framework for confounding, selection bias and transportability (data fusion)? → Bareinboim & Pearl 2016
- Need a statistical review (PATE vs SATE, assumptions, IPPW, outcome regression, doubly robust, sensitivity)? → Degtiar & Rose 2023
- Need trial-to-target estimators with a simulation and an application? → Dahabreh et al. 2020
- Need evidence that early test sites overstate effects (Opower, 111 sites)? → Allcott 2015
- Transporting a geo test or lift study to new markets? → Pearl 2014 - Marketing Application Transporting Geo and Lift Results, Degtiar 2023 - Marketing Application Transporting Experiment Results, Allcott 2015 - Conclusions, Remedies and Marketing Implications
Sub-topics
- Pearl & Bareinboim 2014 - External Validity (8 notes) — COVERS: threats vs licenses and SCM basics; selection diagrams and the formal definition of transportability (Defs. 1–8); motivating examples and transport formulas; S-admissibility and Theorems 1–3; appendix derivations; a marketing application; limits.
- Bareinboim & Pearl 2016 - Data Fusion (4 notes) — COVERS: SCM principles, backdoor, front-door and z-identification; sample selection bias and recoverability (selection-backdoor criterion); transportability and data-fusion criteria.
- Degtiar & Rose 2023 - Review (10 notes) — COVERS: populations, estimands and the decomposition of external-validity bias; identifying assumptions; assessing study–target dissimilarity; weighting and matching; outcome regression; doubly robust methods; heterogeneity tests; comparison tables; marketing application.
- Dahabreh et al. 2020 - Trial to Target (7 notes) — COVERS: the participation process, designs and estimands; identifiability conditions; outcome-model, inverse-odds-weighting and doubly robust estimators; a 36-scenario simulation; the CASS application; diagnostics.
- Allcott 2015 - Site Selection Bias (7 notes) — COVERS: a model of external validity; the Opower program and data; microdata extrapolation overpredicting by 0.40–0.64 pp; metadata analysis of selection mechanisms; microfinance and clinical-trial evidence; remedies and marketing implications.
Cross-Cutting Concepts
- Reweighting to the target population: the graphical transport formula in Pearl 2014 - S-Admissibility and the Graphical Criterion and its estimators in Dahabreh 2020 - Outcome Model, Inverse Odds Weighting and Doubly Robust Estimators and Degtiar 2023 - Weighting and Matching Methods.
- Selection of study sites: in theory, selection nodes (Pearl 2014 - Selection Diagrams and Definition of Transportability); in practice, Opower (Allcott 2015 - Metadata Analysis of Site Selection Mechanisms).
- Pooling vs transporting: hierarchical pooling across sites in Meager 2019 - Pooling Factors and Measuring External Validity versus covariate-based transport.
Gaps
- Cole & Stuart 2010 (ACTG 320) and Stuart et al. 2011 (propensity scores for generalizability) were planned but not ingested: their only free copies (PMC) are behind a bot check. Their methods are covered by Dahabreh 2020 - Outcome Model, Inverse Odds Weighting and Doubly Robust Estimators and Degtiar 2023 - Assessing Dissimilarity Between Study and Target.
Sources
- Pearl 2014 - External Validity From Do-Calculus to Transportability
- Bareinboim 2016 - Causal Inference and the Data-Fusion Problem
- Degtiar 2023 - A Review of Generalizability and Transportability
- Dahabreh 2020 - Extending Inferences from a Randomized Trial to a New Target Population
- Allcott 2015 - Site Selection Bias in Program Evaluation
See Also
- Meta-Analysis — pooling evidence across studies
- Do-Calculus in Summary Causal DAGs — the do-calculus the transport theorems rely on
- Geo-Experiment Methodology - Overview — the marketing experiments whose results need transporting