Index: Dahabreh 2020 - Extending Inferences from Trial to Target Population

Dahabreh 2020 - Extending Inferences from a Randomized Trial to a New Target Population

Routing Summary

This folder fully captures Dahabreh, Robertson, Steingrimsson, Stuart & Hernán (2020), Extending inferences from a randomized trial to a new target population (Statistics in Medicine 39(14); arXiv:1805.00550v3 tutorial): generalizability vs transportability, five identification conditions, outcome-model / inverse-odds-weighting / doubly robust estimators, the 36-scenario simulation (Tables 2-3), the CASS application (Tables 4-5), and practical diagnostics. The Appendix cited in the paper is absent from the PDF and is not captured.

• Need the paper on one page or the section-to-note map? → Dahabreh 2020 - Extending Inferences from a Randomized Trial - Overview
• Need generalizability vs transportability, nested vs non-nested designs, Table 1 data structure? → Dahabreh 2020 - Trial Participation Process, Designs and Estimands
• Need the five identification conditions or the transport formula for $E[Y^a\mid S=0]$? → Dahabreh 2020 - Identifiability Conditions and Identification
• Need formulas for OM, IOW1/IOW2, DR1-DR3, double robustness, inference? → Dahabreh 2020 - Outcome Model, Inverse Odds Weighting and Doubly Robust Estimators
• Need the simulation design and bias/variance tables, or a Python replication? → Dahabreh 2020 - Simulation Study of Estimators
• Need the CASS covariate and results tables? → Dahabreh 2020 - CASS Coronary Surgery Application
• Need sensitivity analysis, covariate selection, positivity diagnostic, or the paper’s recommendations? → Dahabreh 2020 - Practical Issues, Diagnostics and Discussion

Concept Map

Concept	Note	Type	Depends On	Key Result
Trial participation process	Dahabreh 2020 - Trial Participation Process, Designs and Estimands	definition	none	eligible → invited → participant
Generalizability vs transportability	Dahabreh 2020 - Trial Participation Process, Designs and Estimands	definition	participation process	target within vs beyond trial-eligible population
Nested / non-nested design	Dahabreh 2020 - Trial Participation Process, Designs and Estimands	definition	none	cohort-wide ATE not identified in non-nested
Table 1 data structure	Dahabreh 2020 - Trial Participation Process, Designs and Estimands	example	designs	outcome missing by design for $S=0$
Identifiability conditions (1)-(5)	Dahabreh 2020 - Identifiability Conditions and Identification	definition	estimands	exchangeability over $S$ untestable
Identification theorem	Dahabreh 2020 - Identifiability Conditions and Identification	theorem	conditions	$E[Y^a\mid S=0]=E\{E[Y\mid X,S=1,A=a]\mid S=0\}$
OM estimator	Dahabreh 2020 - Outcome Model, Inverse Odds Weighting and Doubly Robust Estimators	definition	identification	average outcome-model predictions over target X
IO weights and IOW1/IOW2	Dahabreh 2020 - Outcome Model, Inverse Odds Weighting and Doubly Robust Estimators	definition	identification	weights $(1-p)/(p{e}_a)$; normalize
DR1-DR3 and double robustness	Dahabreh 2020 - Outcome Model, Inverse Odds Weighting and Doubly Robust Estimators	theorem	OM, IOW	consistent if either model correct
Simulation Tables 2-3	Dahabreh 2020 - Simulation Study of Estimators	example	estimators	trial-only bias 0.66-0.89; OM lowest variance
CASS Tables 4-5	Dahabreh 2020 - CASS Coronary Surgery Application	example	estimators	RD -1.3% to -1.9% vs trial -3.0%
Sensitivity, covariate selection	Dahabreh 2020 - Practical Issues, Diagnostics and Discussion	concept	conditions	effect modifiers suffice for ATE
Positivity diagnostic	Dahabreh 2020 - Practical Issues, Diagnostics and Discussion	definition	weights	summed inverse odds $\approx n_{obs}$
Recommendation	Dahabreh 2020 - Practical Issues, Diagnostics and Discussion	concept	simulation	prefer DR with normalized weights

Notes

• Dahabreh 2020 - Extending Inferences from a Randomized Trial - Overview — CONTAINS: research question, contributions, headline numbers, full Source Map, marketing relevance.
• Dahabreh 2020 - Trial Participation Process, Designs and Estimands — CONTAINS: Sec. 1-3; participation process; generalizability vs transportability; nested vs non-nested designs; Table 1; target estimand.
• Dahabreh 2020 - Identifiability Conditions and Identification — CONTAINS: Sec. 4; conditions (1)-(5); identification theorem with proof sketch; weaker identification in measure.
• Dahabreh 2020 - Outcome Model, Inverse Odds Weighting and Doubly Robust Estimators — CONTAINS: Sec. 5; eqs. 1-6; why IOW is unbiased; double robustness; special cases; inference; runnable Python.
• Dahabreh 2020 - Simulation Study of Estimators — CONTAINS: Sec. 6; data generation; Tables 2 and 3 reproduced; interpretation; independent Python replication.
• Dahabreh 2020 - CASS Coronary Surgery Application — CONTAINS: Sec. 7; models; Table 4; Table 5; Figure 1 diagnostics; interpretation.
• Dahabreh 2020 - Practical Issues, Diagnostics and Discussion — CONTAINS: Sec. 8-9; sensitivity analysis; covariate selection; flexible models and cross-fitting; positivity diagnostic; recommendations; lift-study checklist.

Sources

• Dahabreh 2020 - Extending Inferences from a Randomized Trial to a New Target Population — Dahabreh IJ, Robertson SE, Steingrimsson JA, Stuart EA, Hernán MA (2020), Statistics in Medicine 39(14); arXiv:1805.00550v3, 21 pp. (main text and references; Appendix not included).

See Also

• Allcott 2015 - Site Selection Bias in Program Evaluation - Overview — empirical test of the same exchangeability condition at site level.
• Bayesian Inverse Probability Weighting, Doubly-Robust Estimands for ATT(g,t) — weighting and DR machinery elsewhere in the vault.
• Geo-Experiment Methodology - Overview, Conversion Lift Studies on Ad Platforms — marketing experiments to which transportability applies.