Identification Strategy (jeea-identification)
When to trigger
- An empirical causal claim rests on OLS + controls, or TWFE on staggered timing
- A structural model's parameters are estimated but it is unclear what in the data identifies them
- A theory paper's result depends on assumptions whose role is not transparent
- You are unsure the identification clears JEEA's general-interest theory-and-empirics bar
The JEEA identification bar
JEEA spans theory and empirics, so "identification" means different things by branch — but in every case the mapping from assumptions/data to the object of interest must be explicit and defended, and credible enough for a general-interest readership and a co-editor who is not a subfield specialist. JEEA's house norms reinforce this: report standard errors and confidence sets (no significance asterisks/boldface for significance) and make the empirical strategy reproducible for the JEEA Data Editor's pre-acceptance replication check (DCAS). Pick the branch and make the argument legible.
Branch paths
Branch A: Structural / quantitative identification
- Name what identifies each parameter. Tie parameters to specific data features / moments; argue identification from the model's structure, not "the estimator converged."
- Targeted vs. untargeted moments: report fit to targeted moments and untargeted-moment validation as out-of-sample discipline.
- Sensitivity / informativeness: report parameter sensitivity to moments (sensitivity matrix) so readers see which data move which parameters.
- Estimation regularity: state the objective (MLE / GMM / MSM / indirect inference), starting values, tolerances, multi-start; report Monte Carlo evidence recovering known parameters.
- Counterfactual validity: argue the estimated parameters are policy-invariant enough for the counterfactual (Lucas critique).
Branch B: Empirical causal design (applied micro / development / finance)
- DID / event study: with staggered adoption move beyond TWFE (Callaway–Sant'Anna, Sun–Abraham, de Chaisemartin–D'Haultfœuille); show clean event-study leads; report a Goodman–Bacon decomposition.
- IV: strong first stage; with weak instruments use Anderson–Rubin / weak-IV-robust sets; defend the exclusion restriction in theory, institutions, and falsification.
- RDD: Cattaneo–Jansson–Ma density test; optimal bandwidth + robustness; covariate smoothness; bias-corrected CIs.
- Inference clustered at the assignment level; address few-cluster issues (wild-cluster bootstrap).
Branch C: Theory / mechanism identification
- What assumptions do the work. Identify the minimal assumptions driving the headline result; show which can be relaxed and which are essential.
- Comparative statics as identification: make clear which primitive moves which prediction, so the model's empirical content is testable.
- Source of the result: distinguish a genuinely new mechanism from a re-parameterization; route to
jeea-theory-model for generality and proof discipline.
Branch D: Experimental / own-data
- Pre-registration in a recognized registry; report deviations and the explicit estimand.
- Randomization balance; attrition (Lee bounds if differential); multiple-hypothesis adjustment; external-validity discussion.
Checklist
Anti-patterns
- "The estimator converged" presented as if it were identification (structural)
- TWFE on staggered treatment with no heterogeneity-bias discussion (empirical)
- A theory result whose driving assumption is hidden in notation, so its empirical content is unclear
- Calibrating parameters and running a counterfactual without arguing policy-invariance
- Reporting significance with asterisks instead of standard errors / confidence sets
Referee pushback mapped to the identification fix
- "This is OLS with controls dressed up as causal." → Provide a design (DID/IV/RDD) or a credible selection-on-observables defense with sensitivity (Oster) and falsification.
- "Staggered TWFE here is biased." → Re-estimate with Callaway–Sant'Anna or Sun–Abraham; show flat event-study leads.
- "Your structural estimates are calibration in disguise." → Show the sensitivity matrix and which moment moves which parameter; report untargeted fit.
- "The model's headline result is an artifact of one assumption." → Name the assumption, relax it, and show the result survives (or scope it honestly).
Output format
【Branch】structural / empirical / theory / experimental
【Assumption-or-data-to-object mapping】one sentence
【Identification evidence】[moments+sensitivity / pre-trends+density+first-stage / minimal-assumptions / balance]
【Estimation/inference】objective + SEs/confidence sets (no asterisks); clustering if any
【What it does NOT identify】[...]
【Next step】jeea-theory-model or jeea-robustness