From jhe-skills
Guides robustness checking for Journal of Health Economics submissions: maps threats (selection, concurrent policy, staggered timing, mismeasurement) to specific checks with point-estimate stability reporting.
How this skill is triggered — by the user, by Claude, or both
Slash command
/jhe-skills:jhe-robustnessThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
- The headline estimate may be sensitive to specification, sample window, or functional form
A wall of starred alternative specifications persuades no one. JHE referees want each robustness check mapped to a specific threat to the health-economics claim, with the point estimate's stability — not its significance — as the object. Build the robustness section as a threat-response ledger: name the threat a health economist would raise, run the check that addresses it, and report whether the magnitude moves. The threats that recur at JHE are selection, concurrent policy, measurement of health/utilization, and inference with few policy clusters.
| Threat to the claim | Check that addresses it |
|---|---|
| Residual selection into insurance/treatment | bounds (Lee/Manski/Oster); selection-on-observables-to-unobservables (Oster δ) |
| Concurrent reform contaminates the policy variation | placebo on ineligible group/period; leave-one-reform-out; pre-period falsification |
| Staggered-timing bias | heterogeneity-robust estimator (CS / SA / dCDH); honest-DID sensitivity |
| Health/utilization mismeasurement (claims coding, self-report) | alternative outcome definitions; administrative vs. survey cross-check; coding-change robustness |
| Functional form / sample window | log vs. level, trimming outliers (skewed health spending!), alternative bandwidths, donut RD |
| Few-cluster inference (states) | wild-cluster bootstrap; randomization inference; correct clustering level |
| Multiple outcomes/subgroups | MHT adjustment (Romano–Wolf / sharpened q-values); pre-specify the primary outcome |
| Mechanism is one of several | horse-race the channels; show the competing story predicts a pattern you do not see |
Health spending and utilization are the journal's signature dependent variables, and they are right-skewed, zero-inflated, and heavy-tailed — the estimator choice is itself a robustness question a referee will press. Make it a deliberate, defended choice rather than a default:
A provider-payment paper shows intensity rises after a fee change; a referee suspects it is patient selection, not a true behavioral response. The JHE fix: hold the threat ledger explicit — (a) an Oster δ shows selection on unobservables would need to be 2× the observables to overturn the result; (b) a placebo on a fee-unaffected service shows no jump; (c) the spending outcome is re-run with a two-part model given 30% zeros; (d) inference is wild-cluster bootstrapped over 41 providers. The point estimate holds across all four (say 6.2%, stable within ±0.8pp, illustrative). The mechanism — behavioral response, not selection — now survives.
【Journal】Journal of Health Economics
【Skill】jhe-robustness
【Primary threat】selection / concurrent-policy / staggered-bias / measurement / inference
【Threat→check ledger】[threat: check → estimate movement]
【Spending estimator】OLS / two-part / GLM / IHS — defended? [Y/N]
【Inference】clustering level + few-cluster correction
【MHT】adjusted across outcomes/subgroups? [Y/N]
【Verdict】estimate stable / moves (explained) / fragile
【Next skill】jhe-tables-figures
This skill stress-tests an already-identified estimate; it does not fix a broken design (that is jhe-identification) or present the results (that is jhe-tables-figures). If a robustness check reveals the estimate is not actually identified — it swings with the selection bound or fails the placebo — route back to jhe-identification rather than papering over it with more specifications. When the point estimate holds across the threat ledger, hand off to jhe-tables-figures to make the stability legible.
npx claudepluginhub brycewang-stanford/awesome-journal-skills --plugin jhe-skillsOrganizes robustness checks for JEEA manuscripts around threats a general-interest referee would raise, ensuring headline results are stable to specification, sample, inference, and assumption perturbations.
Organizes robustness checks for IER papers by threat to load-bearing assumption, without running regressions. Helps structure responses to referee concerns.
Builds robustness suites for AEJ: Applied manuscripts to show headline estimates survive specification, sample, and inference choices.