Jeem Identification

Identification Strategy (jeem-identification)

When to trigger

• A regulation/permit-market effect rests on TWFE with staggered adoption, or on OLS + controls
• A hedonic or travel-cost estimate could be confounded by spatial sorting or omitted amenities
• A stated-preference WTP could be contaminated by hypothetical bias, scope insensitivity, or yea-saying
• A weather/pollution IV's exclusion restriction or its adaptation interpretation is challenged
• You are unsure the design recovers a welfare-relevant parameter, not just a reduced-form effect

The JEEM identification bar

JEEM identification is judged on two axes at once: the causal/preference-recovery argument must be credible, and the recovered object must be welfare-relevant — a marginal damage, a WTP, a pass-through, an abatement cost. Environmental data carry field-specific threats (spatial dependence, sorting, monitoring-station selection, weather endogeneity through adaptation) that generic applied-micro referees miss but JEEM referees will not. Pick the branch and make the mapping from data to the welfare object explicit.

Branch A — Environmental-policy causal design

• Regulation / standards DiD: with staggered rollout, abandon plain TWFE for heterogeneity-robust estimators (Callaway–Sant'Anna, Sun–Abraham, de Chaisemartin–D'Haultfœuille); show a clean event study with pre-period leads; report a Goodman-Bacon decomposition. Argue the regulation timing is not driven by prior pollution trends.
• Cap-and-trade / permit markets: the price and the cap are equilibrium objects — instrument or bound the endogeneity; watch for leakage to uncovered sources and reshuffling, which change the welfare sign.
• RD in standards / eligibility: sharp thresholds (an attainment cutoff, a plant-size or emissions threshold) — McCrary/Cattaneo–Jansson–Ma density test, covariate smoothness, bias-corrected CIs.
• Inference: cluster at the regulatory/assignment level; use spatial (Conley) SEs when units are geographic and shocks are correlated across space.

Branch B — Climate / weather IV and damages

• Argue weather realizations are as-good-as-random conditional on location and time fixed effects; defend exclusion (weather affects the outcome only through the modeled channel).
• Separate weather (short-run shock) from climate (long-run expectation) — the adaptation margin is the contribution; a panel weather coefficient is not a long-run climate-damage estimate without an adaptation argument.
• Handle spatial and serial correlation in errors; report the estimand (marginal damage at current vs. future climate).

Branch C — Revealed-preference valuation (hedonics, travel cost)

• Hedonics: the amenity coefficient is biased by sorting (Tiebout) and omitted correlated amenities. Use a quasi-experimental shock to the amenity (a plant opening/closing, a Superfund listing, a regulation), boundary discontinuities, or a sorting model; do not present a cross-sectional hedonic as causal WTP.
• Travel cost: address endogenous trip cost, multi-purpose trips, and the recreation-demand censoring (count models, Kuhn–Tucker demand systems).
• Be explicit about what welfare measure the capitalization or demand estimate recovers (marginal WTP vs. total amenity value) and its partial- vs. general-equilibrium scope.

Branch D — Stated-preference valuation (CV, discrete-choice experiments)

• Survey design is the identification: incentive compatibility, a credible payment vehicle, a consequential decision, and a scope test (WTP rises with the size of the good).
• Address hypothetical bias (cheap-talk, certainty calibration, inferred valuation), protest responses, and status-quo/yea-saying effects.
• Pre-specify the choice model (RUM / mixed logit / latent class) and report welfare (compensating variation) with its uncertainty, not just utility coefficients.

Execution bridge (StatsPAI / Stata MCP)

Estimate and audit the design, don't only describe it. Full map: execution-with-mcp. JEEM is environmental economics — policy/regulation designs and non-market valuation; the causal chain serves its program-evaluation lane.

• 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.