Jle Identification

Identification Strategy (jle-identification)

When to trigger

• A causal claim about a legal rule rests on OLS + controls, or TWFE on staggered statute adoption
• A court/judge design's random-assignment defense, monotonicity, or exclusion is not pinned down
• A regulatory threshold's density, bandwidth, or bunching response is unaddressed
• An antitrust/enforcement event study has contaminated windows or no clean control market
• You are unsure the design clears JLE's law-and-economics credibility bar

The JLE identification bar

JLE judges identification through a law-and-economics lens: the mapping from a legal/regulatory source of variation to the causal estimand must be explicit, defended, and consistent with how the rule actually operates. Referees here are economists who also know the institution — they will ask whether the statute really bound when you say, whether enforcement varied, whether the "control" jurisdiction had its own contemporaneous reform. State the estimand, name the identifying assumption, show the diagnostic that could have failed, and keep the claim inside what the legal design supports. Inference must match the design (cluster at the level of legal variation — usually the jurisdiction; correct for few jurisdictions).

Design paths

Path A: Law change / staggered statute adoption (DiD / event study)

• With staggered adoption across jurisdictions, move beyond TWFE — Callaway–Sant'Anna, Sun–Abraham, de Chaisemartin–D'Haultfœuille, or Borusyak–Jaravel–Spiess imputation; a Goodman-Bacon decomposition to show which 2×2s drive the estimate.
• Clean event-study leads for pre-trends; date treatment to when the rule binds (effective date, not signing date), and handle anticipation and phase-ins.
• Rule out contemporaneous legal change in treated jurisdictions (other reforms riding the same bill); defend parallel trends across jurisdictions, ideally with Rambachan–Roth honest-DID.

Path B: Court / judge / case-assignment design (IV or as-good-as-random)

• Establish random (or as-good-as-random) assignment of cases to judges/panels; test for balance on case characteristics; document the assignment rule and exceptions (recusal, specialization).
• Judge leniency/stringency as an instrument: report first-stage strength, defend exclusion (the judge affects the outcome only through the ruling), and monotonicity (no defiers).
• State the LATE (compliers near the assignment margin); resist reading it as the population effect.

Path C: Regulatory threshold / eligibility cutoff (RD / bunching)

• Density test (McCrary / Cattaneo–Jansson–Ma) for manipulation around the regulatory cutoff; covariate smoothness; local-linear with data-driven bandwidth and bias-corrected robust CIs (rdrobust); donut and bandwidth sensitivity.
• Where firms/agents sort to avoid a rule, treat bunching as evidence, not noise (size-based regulation, tax/penalty kinks); estimate the response from the missing mass.
• State whether the estimand is the local effect at the threshold and resist extrapolation.

Path D: Antitrust / enforcement event (merger, decree, entry, ruling)

• Define the event window tightly and rule out confounding announcements; for prices/output, use clean control markets unaffected by the action.
• Get market definition right (the antitrust object); show the treated and control markets were comparable pre-event.
• For merger price effects, consider difference-in-differences across affected vs. unaffected products/regions; for enforcement deterrence, distinguish the targeted firm from spillover deterrence on others.

Execution bridge (StatsPAI / Stata MCP)

Estimate and audit the design, don't only describe it. Full map: execution-with-mcp. JLE is empirical law-and-economics — DiD around legal/regulatory change is central.

• detect_design → recommend → fit with as_handle=true → audit_result.
• Observational causal claims: staggered DiD (callaway_santanna / sun_abraham + bacon_decomposition + honest_did_from_result); IV (effective_f_test + anderson_rubin_ci); RDD (rdrobust + mccrary_test).
• Experiments: randomization-based inference + romano_wolf for many-outcome control.
• Sensitivity: oster_delta / sensemakr for observational claims.

Report the magnitude in interpretable units; route the full battery to the appendix. A run end-to-end (synthetic data, real returns) is in the JF execution walkthrough.

Checklist

• Design chosen; the legal-variation-to-estimand mapping stated in one sentence
• Estimand named (ATT / LATE / local-at-threshold / event effect) and matched to the design
• Treatment dated to when the rule binds; anticipation/phase-in handled
• Design-appropriate diagnostic shown (event-study leads + Bacon / assignment balance + first-stage / density + bandwidth / clean control market)
• Modern estimator where TWFE or 2SLS would bias; contemporaneous legal change ruled out
• Inference clustered at the jurisdiction/assignment level; few-cluster correction (wild-cluster bootstrap)
• The claim never exceeds what the legal design identifies (no extrapolation past the local/LATE object)