EVlibrary

[Website Under Construction]

Similar to DeclareDesign's DesignLibrary, EVlibrary provides ready-to-use templates for common external validity scenarios. While DesignLibrary focuses on research designs, EVlibrary covers the full scope of external validity inference—including theory (Model Utility), design (Scope Plausibility), and analysis (Specification Credibility). Each template includes a vignette explaining the methodology, a function for customization, and an inspector for diagnosis.

library(EVlibrary) # Example TBA

Model Utility (Theory)

Templates for building useful theoretical models that guide mechanism-based inference.

Template Description Resources
Theoretical Estimand Define target STOUT populations and the descriptive, predictive, or causal quantity of interest.
Approach 1 PATE estimation with explicit population definition and eligibility criteria.
Approach 2 TATE estimation for cross-context transportability with differing eligibility.
Mechanisms Specify mechanistic structure including core factors, moderators, and functional forms.
Approach 1 INUS mechanism specification with necessary conditions and sufficient pathways.
Approach 2 Moderator-based mechanism specification with effect modification structure.
Queries Express external validity queries and determine identification formulas or bounds.
Approach 1 Point-identified queries with full identification formulas.
Approach 2 Partially-identified queries with informative bounds.

Scope Plausibility (Design)

Templates for designing studies that align with target populations.

Template Description Resources
Target Populations Specify eligibility criteria across STOUT dimensions and model potential effect modifiers.
Approach 1 Single target population with comprehensive covariate specification.
Approach 2 Multiple target populations with comparative eligibility criteria.
Planned Samples Stratify by key modifiers to capture real treatment heterogeneity.
Approach 1 Stratified sampling by pre-specified effect modifiers.
Approach 2 Multi-site sampling with site-level heterogeneity.
Actual Samples Assess whether realized samples support credible inference and document gaps.
Approach 1 Sample-population comparison diagnostics with covariate balance.
Approach 2 Attrition and nonresponse assessment with adjusted estimands.

Specification Credibility (Analysis)

Templates for analyzing data to support population-level inference.

Template Description Resources
Population-Level Estimation Estimate population-level effects using covariate-based, model-based, or hybrid methods.
Approach 1 Inverse probability weighting for generalizability and transportability.
Approach 2 Outcome modeling with covariate adjustment for population effects.
Uncertainty Quantification Employ alternative estimands, robustness checks, and synthesis for cautious inference.
Approach 1 Sensitivity analysis for violations of transportability assumptions.
Approach 2 Partial identification bounds under relaxed assumptions.
Abductive Prediction Anchor inferences in explicit mechanisms and generate testable predictions.
Approach 1 Mechanism-based prediction for new contexts with explicit assumptions.
Approach 2 Comparative case analysis for mechanism plausibility assessment.