METHODS PROGRAM · VERSION 2.0 · 2026-07-20

Two tracks.
One evidence boundary.

这不是寻找一个更显著系数的流程，而是一套阻止错误确定性的研究模板。

Track A stress-tests observational associations. Track B permits causal language only after a policy or natural-experiment design passes its identification gates.

CURRENT VERDICTThe present country panel does not identify the causal direction or magnitude of PM2.5's effect on period life expectancy.This is evidence of design sensitivity—not evidence that PM2.5 is harmless.

01 — THE DUAL TRACK

Association is examined.
Causation is earned.

Both tracks begin from one registered protocol, preserve contradictory results and end with a rule-governed conclusion rather than a preferred coefficient.

AOBSERVATIONAL

Stress-test the association

Two-way fixed effects, country trends, change models, alternative exposure windows, nonlinear response and future-exposure placebo tests.

Status · Complete

BCAUSAL IDENTIFICATION

Qualify an external change

Policy timing, comparison-group overlap, exposure first stage, event-study pre-trends, negative controls and interference checks.

Status · Engine validated; empirical policy data required

02 — TRACK A RESULT

The estimate changes
with the design.

Effects are life-expectancy years per 10 µg/m³ higher PM2.5. Direction and interval conclusions change across plausible specifications, while the future-exposure placebo remains positive.

SpecificationEstimate95% interval

A1 · Primary TWFE+0.401+0.150 to +0.651

A2 · Country trends−0.038−0.183 to +0.107

A3 · One-year change−0.006−0.086 to +0.074

A8 · Future placebo+0.364+0.121 to +0.607

EVIDENCE GRADEA1 · Inconclusive association

Current data do not support a stable independent association. The result remains useful because it locates the design assumptions that drive the answer.

03 — TRACK B GATE

A blank estimate is the
correct scientific result.

The engine recovers known simulated effects, but simulation validates software—not the PM2.5 claim. No empirical estimate is produced until a versioned intervention panel is registered.

REQUIRED INPUTPolicy name, legal source, timing, geographic scope and assignment mechanism

FIRST STAGEThe intervention must measurably alter pollution exposure

DESIGN CHECKSPre-trends, anticipation, overlap, negative controls and spillovers

OUTCOME UPGRADEAnnual age-standardized all-cause and cardiopulmonary mortality

04 — REUSABLE WORKFLOW

A method that travels
beyond air pollution.

The transferable product is the protocol and decision system, not a PM2.5-specific coefficient. It can be adapted to climate, water, food, occupational exposure, policy and health-system questions.

Define

Map roles

Separate confounders from mediators, descendants and descriptive stratifiers.

Register data

Record sources, versions, measurement type, licensing and uncertainty.

Run Track A

Compare fixed effects, trends, changes, exposure windows and placebos.

Qualify Track B

Require an external assignment mechanism and pass identification checks.

Reconcile

Report support, counter-evidence, limitations and the next evidence gate.

05 — REPRODUCIBILITY

Inspect the protocol,
results and decision.

Release 1.0 remains frozen as the historical baseline. Version 2 adds the reusable dual-track layer without overwriting the original study.

Methodology report · PDFSix-page reviewed report and reusable research templateDownload ↓Registered protocol · JSONQuestion, variable roles, estimands and decision rulesDownload ↓Track A results · JSONAll observational specifications and uncertainty estimatesDownload ↓Track B validation · JSONSimulation-only engine checks; not empirical PM2.5 evidenceDownload ↓Dual-track reconciliation · JSONUnified evidence boundary and next-data requirementDownload ↓

Two tracks.One evidence boundary.

Association is examined.Causation is earned.