Jpam Research Design

Research Design & Identification (jpam-research-design)

Credible identification is JPAM's core bar. The journal evaluates the effects of real policies and programs, so the design must connect the theory of change (jpam-theory-building) to evidence a policymaker can trust. State the estimand, the assumptions that license a causal reading, and how each is defended — then rule out the single strongest rival explanation. Selection-on- observables alone rarely clears the bar.

When to trigger

• Specifying or defending identification for a policy evaluation
• A reviewer questioned causal claims, parallel trends, the instrument, the discontinuity, or confounding
• Choosing among RCT / DiD / RD / IV / synthetic control for a given policy variation
• Preparing a pre-analysis plan for a prospective program evaluation

Design menu (match to the policy variation)

• RCT / field experiment. The gold standard where feasible. Report randomization unit, balance, power/MDE, take-up, attrition, and ITT vs. TOT/LATE. Pre-register primary outcomes and subgroups.
• Difference-in-differences / event study. For staggered policy adoption use heterogeneity-robust estimators (Callaway–Sant'Anna, Sun–Abraham, de Chaisemartin–D'Haultfœuille, BJS imputation) — not naive TWFE. Show pre-trends as an event study; test, don't assert, parallel trends.
• Regression discontinuity / kink. For eligibility thresholds and benefit formulas. Report bandwidth selection, local-polynomial robustness, density/manipulation tests (McCrary/rddensity), covariate continuity, and the local nature of the estimand.
• Instrumental variables. For policy-induced variation. Defend the exclusion restriction substantively, report first-stage strength (effective F / weak-IV-robust inference), and interpret the LATE — whose behavior the instrument shifts.
• Synthetic control. For a single treated unit (a state/country policy). Report donor pool, pre- period fit, placebo/permutation inference, and leave-one-out robustness.

Inference & policy-evaluation standards

• Cluster at the level of treatment assignment; with few clusters use wild-cluster bootstrap.
• Adjust for multiple outcomes/subgroups (and say which test is primary).
• Distinguish ITT vs. treatment-on-the-treated; report take-up for any offer-based program.
• Specify the estimand and target population — JPAM cares which population the policy decision is about.

The adjudication test (JPAM-specific)

For the single strongest rival explanation (selection, anticipation, concurrent policy, mean reversion), write one sentence: "If the rival were driving the result, the data would look like ___; instead they look like ___." If you cannot, the design does not yet identify the policy effect.

Execution bridge (StatsPAI / Stata MCP)

Estimate and audit the design, don't only describe it. Full map: execution-with-mcp. JPAM is policy analysis — program evaluation is the core; DiD/IV/RDD and the policy-relevant magnitude are decisive.

• 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.