jole-identification-strategy

Identification Strategy (jole-identification-strategy)

When to trigger

• The empirical core is OLS + controls with an undefended causal claim
• A DID uses two-way fixed effects (TWFE) on staggered policy timing without modern estimators
• A shift-share / Bartik IV's exclusion or exposure-share exogeneity is unargued
• An RDD at an eligibility threshold lacks a density test or bandwidth robustness
• AKM firm–worker fixed effects may carry limited-mobility bias

The JOLE identification bar

JOLE publishes empirical labor papers only when the causal claim is credible and the data are replicable. Labor referees apply the standard credibility ladder, tuned to labor settings (strong → weaker):

1. RCT / field experiment (e.g., training, hiring, job-search interventions) with balance and pre-registration
2. RDD at a clean eligibility threshold (program cutoffs, age/score rules)
3. DID / event study off a credibly exogenous labor reform (minimum wage, UI, mandate) with modern estimators
4. IV, including shift-share / Bartik designs, with a strong first stage and a defended exclusion restriction
5. AKM two-way (firm + worker) fixed effects for wage decompositions, with limited-mobility-bias correction
6. Matching / selection-on-observables — a complement, rarely the spine

A novel or newly linked labor dataset (matched employer–employee registers, administrative earnings) answering a first-order question can carry a paper even when the design is more descriptive — but only with disciplined measurement and a clear labor lesson.

Branch paths

Branch A: DID / event study (policy reforms)

• Staggered adoption? Move beyond TWFE: Callaway–Sant'Anna, Sun–Abraham, or de Chaisemartin–D'Haultfœuille; report a Goodman-Bacon decomposition to expose invalid negative-weight comparisons.
• Pre-trends: clean event-study plot with leads near zero.
• Inference: cluster at the treatment-assignment level (often state); wild-cluster bootstrap with few clusters.
• Placebo timing and placebo outcomes that should not move.

Branch B: IV (incl. shift-share / Bartik)

• First-stage F strong; with weak instruments use Anderson–Rubin / weak-IV-robust sets.
• For shift-share: argue exogeneity of the shares (or of the shocks, per the design) and run share-level balance / falsification.
• Exclusion argued in three registers: labor theory, institutional detail, and falsification where the channel is absent.
• Report reduced form and OLS alongside IV; state the LATE/complier interpretation.

Branch C: RDD (eligibility thresholds)

• Density / manipulation test at the cutoff; covariate smoothness and placebo cutoffs.
• Optimal bandwidth (Calonico–Cattaneo–Titiunik) plus bandwidth-robustness; bias-corrected CIs.
• Fuzzy RDD: report the first stage in the discontinuity.

Branch D: RCT / field experiment

• Pre-analysis plan referenced; deviations reported.
• Balance table; attrition analysis (Lee bounds if differential).
• Multiple-hypothesis adjustment across outcomes/subgroups.
• External validity: what does this labor population teach beyond itself?

Branch E: AKM firm–worker FE

• Identify off the connected set of firms linked by job movers.
• Address limited-mobility (incidental-parameter) bias — leave-out (KSS) or bias-corrected estimators — before interpreting firm-effect variance.
• Be explicit about what the firm and worker effects do and do not identify (exogenous-mobility assumption).

Execution bridge (StatsPAI / Stata MCP)

Estimate and audit the design, don't only describe it. Full map:

[execution-with-mcp](../../../shared-resources/empirical-methods/execution-with-mcp.md). JOLE is labor economics — the home of clean identification; DiD/IV/RDD and selection corrections are the binding constraint.

• 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](../../../Journal-of-Finance-Skills/resources/worked-examples/02-execution-walkthrough.md).

Checklist

• [ ] Identifying variation named in one sentence and defended as exogenous
• [ ] Design-appropriate diagnostics (pre-trends / density / first-stage F / balance / connected set)
• [ ] Modern estimator where TWFE would be biased (staggered policy DID)
• [ ] Shift-share exogeneity (shares or shocks) argued and tested
• [ ] AKM limited-mobility bias corrected before variance decomposition
• [ ] Inference matched to assignment level; few-cluster issues addressed
• [ ] LATE / ATT / external-validity interpretation stated
• [ ] The claim never exceeds what the design supports

Anti-patterns

• TWFE on staggered minimum-wage / reform timing with no heterogeneity-bias discussion
• A shift-share IV with no argument for share (or shock) exogeneity
• Interpreting AKM firm-effect dispersion without correcting limited-mobility bias
• RDD reporting one bandwidth and hiding sensitivity
• A clean local labor estimate oversold as a universal structural parameter

Output format

【Design】RCT / RDD / DID / IV (incl. shift-share) / AKM / descriptive
【Identifying variation】one sentence
【Diagnostics done】[pre-trends, density, first-stage F, balance, connected set, ...]
【Diagnostics missing】[...]
【Inference】clustering level + few-cluster handling
【Interpretation】LATE / ATT / firm-effect caveat / external validity
【Next step】jole-data-analysis