Jpube Identification Strategy

Identification Strategy (jpube-identification-strategy)

When to trigger

• The empirical core is OLS + controls with an undefended causal claim
• A reform DID uses two-way fixed effects (TWFE) on staggered timing
• A bunching estimate lacks a defensible counterfactual density
• An IV's policy instrument has an unargued exclusion restriction
• You are unsure the design clears the JPubE public-finance bar

The JPubE identification bar

JPubE rewards credible identification of a policy-relevant parameter, evaluated by public-finance specialists under single anonymized review (a minimum of two reviewers, with author identity known to them). Because the field's payoff is usually a behavioral elasticity feeding a welfare formula, the design must pin down the response to a tax, transfer, or program rule cleanly. The credibility ranking referees implicitly apply (strong → weaker):

1. Bunching / notch designs at a known kink or eligibility threshold, recovering an elasticity from excess mass
2. RDD / RKD at a sharp policy cutoff (eligibility, benefit schedule kink)
3. DID / event study off a credibly exogenous reform rollout, with modern estimators
4. IV with a policy instrument, strong first stage, and a defended exclusion restriction
5. Selection-on-observables — acceptable only as a complement, rarely as the spine

JPubE's comparative advantage is the policy-induced discontinuity — a tax kink, a benefit cliff, a reform date — so make the identifying variation an institutional feature a reader can see.

Branch paths

Branch A: Bunching at kinks / notches

• Estimate excess mass against a smooth counterfactual fit away from the kink; report the implied elasticity.
• Round-number / focal-point diagnostics; bin-width and excluded-region robustness.
• For notches, address the dominated region and optimization frictions; bound attenuation from frictions.

Branch B: RDD / RKD

• McCrary / Cattaneo–Jansson–Ma density test for manipulation at the cutoff.
• Optimal bandwidth (Calonico–Cattaneo–Titiunik), bias-corrected CIs, bandwidth robustness.
• For RKD, identify off the slope change in the policy schedule; placebo kinks and covariate smoothness.

Branch C: DID / event study off a reform

• Staggered adoption? Move beyond TWFE — Callaway–Sant'Anna, Sun–Abraham, or de Chaisemartin–D'Haultfœuille; report a Goodman-Bacon decomposition.
• Clean pre-trends via an event-study plot; cluster at the level of treatment (often state/jurisdiction).

Branch D: IV from a policy instrument

• First-stage F strong; weak-IV-robust inference (Anderson–Rubin) where needed.
• Exclusion argued in three registers: theory, institutional detail, falsification.
• Report reduced form and OLS; state the LATE / compliers interpretation.

Execution bridge (StatsPAI / Stata MCP)

Estimate and audit the design, don't only describe it. Full map: execution-with-mcp. JPubE is public economics — tax/transfer/program designs; DiD/IV/RDD and bunching are central, magnitudes in policy units.

• 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

• Policy-induced identifying variation named in one sentence and defended as exogenous
• Design-appropriate diagnostics done (excess-mass fit / density / first-stage F / pre-trends)
• Modern estimator used where TWFE would be biased (staggered reform)
• Inference matched to assignment level (often jurisdiction); few-cluster handling
• Frictions / manipulation / placebo tests reported
• The recovered parameter maps to the welfare quantity the paper claims

Anti-patterns

• Bunching with an arbitrary excluded region chosen to maximize the estimate
• TWFE on staggered reform timing with no heterogeneity-bias discussion
• IV that is "policy shock × lagged endogenous variable" with no exclusion argument
• An elasticity identified locally but sold as a global structural parameter

Design-credibility pushback (and the pre-emptive fix)

Address these in the manuscript before a specialist referee raises them.

Worked example: a kink-bunching elasticity, stress-tested (illustrative)

Suppose excess mass at a tax kink yields a taxable-income elasticity of e = 0.25 (illustrative). The skill's bar asks three things before any table: (1) is the counterfactual a smooth density fit away from the kink, with round-number bunching handled? (2) does the estimate survive bin-width and excluded-region variation — here it stays within 0.21–0.29 (illustrative)? (3) does e map to the welfare object the paper claims — the marginal DWL of the kink rate? Only when all three hold is the design ready for jpube-data-analysis. If the elasticity swung from 0.1 to 0.5 across reasonable excluded regions, the identification is not yet credible, regardless of the headline.

Output format

【Design】bunching / RDD / RKD / DID / IV / other
【Identifying variation】one sentence (the policy discontinuity)
【Diagnostics done】[excess-mass fit, density, first-stage F, pre-trends, ...]
【Diagnostics missing】[...]
【Inference】clustering level + few-cluster handling
【Parameter → welfare】elasticity / sufficient stat mapped? [Y/N]
【Next step】jpube-data-analysis

Source: brycewang-stanford/Awesome-Journal-Skills → Journal-of-Public-Economics-Skills/skills/jpube-identification-strategy/SKILL.md