Skip to main content
Studies · CA Air Quality · Investigation 08

Does the Health Model Change the Optimal Policy?

No — but it changes the burden estimate by 4×. Both Di et al. 2017 (Medicare ≥65) and Krewski et al. 2009 (ACS ≥30) pick T2 accelerated electrification as the optimal transport portfolio. They disagree on how much benefit T2 delivers, not on whether to deploy it. The age-threshold gap leaves 94,490 contested annual deaths across 14.7 million Californians aged 30–64.

T2
Both Pick T2
4.0×
Burden Ratio
94,490
Contested Deaths/yr
$0.05B
Residual EVSI
The Two Models

Di vs Krewski: Two Cohorts, Different Age Coverage

The concentration–response function (CRF) maps PM2.5 exposure to mortality risk. Two studies dominate the field. They differ in slope, but the sharper difference is who they include:

Di et al. (2017) — 61 million Medicare enrollees, ages ≥65 only. HR = 1.073 per 10 µg/m³ (β = 0.00705). Higher slope, but the cohort cannot speak to anyone under 65.

Krewski et al. (2009) — ACS CPS-II, 500K participants, ages ≥30. HR = 1.056 per 10 µg/m³ (β = 0.00545). Lower slope, but covers 14.7 million additional Californians in the 30–64 bracket where Di is silent.

Di → T2 Accelerated
$5.8B
Net benefit. 630 deaths avoided (ages ≥65 only). T2 is optimal under Di.
Krewski → T2 Accelerated
$22.5B
Net benefit. 2,001 deaths avoided (ages ≥30). T2 is also optimal under Krewski.

Both CRFs are peer-reviewed, both are used by EPA in regulatory impact analyses, and under each one independently the optimal transport portfolio is the same: T2 accelerated electrification. The difference between them is not the ranking of policies — it is the absolute burden each implies, driven primarily by who the cohort covers.

The Policy Ranking

Same Ranking, Different Magnitudes

Under both CRFs independently, T2 accelerated electrification tops the ranking. T4 equity-targeted comes second. T3 delayed phase-out is consistently last. The CRF choice is not a structural lever over the policy decision — it is a multiplier on the estimated benefit.

Policy Di Net Benefit ($B) Krewski Net Benefit ($B) Di Optimal? Krewski Optimal?
T1 Baseline $5.62 $17.77
T2 Accelerated $5.81 $22.52 Yes Yes
T3 Delayed $2.85 $7.96
T4 Equity $5.78 $22.51
T5 Heavy-Duty $4.58 $16.73

Under Di (ages ≥65 only), T2 wins by a narrow $0.19B over T1 baseline, with T4 a close second. Under Krewski (ages ≥30), T2 wins by a wider $4.75B over T1, with T4 again close behind. The ranking of all five portfolios is preserved across the two CRFs — the policy decision is structurally robust to the CRF choice.

The Real Divergence

Burden Magnitude: 4× Driven by Age-Threshold

The CRFs do not disagree about the policy. They disagree about how big the problem is to begin with. Under Di (Medicare ≥65 only) the CA-wide annual PM2.5-attributable baseline is 11,689 deaths. Under Krewski (ACS ≥30) it is 17,660 — a 1.5× increase in attributable burden. The underlying cohort baseline mortality pool over which the CRF applies differs by 4× (31,407 vs 125,897 annual all-cause deaths), tracking the cohort age floor, not the CRF slope.

Di Coverage (≥65)
4.8M
Californians covered. 31,407 baseline annual cohort mortality — the all-cause death pool over which the Di CRF applies.
Krewski Coverage (≥30)
19.5M
Californians covered. 125,897 baseline annual cohort mortality. 14.7M additional people aged 30–64 where Di is silent.
Contested Burden
4.0×
Ratio of Krewski to Di cohort mortality pool. 94,490 contested annual baseline deaths — 75.1% of the total — sit in the age-30–64 bracket outside Di’s cohort. PM2.5-attributable deaths differ by 1.5× (17,660 vs 11,689).

Both CRFs are scientifically credible. Di covers Medicare ≥65; Krewski covers ACS ≥30. They answer different questions on different cohorts — neither replaces the other. The cohort you report against sets the burden; it does not set the policy. Rankings hold; the scale of benefits does not.

Building Policy

Building Ranking Is Also Preserved

The building electrification decision is stable across both CRFs in the same way. B1 baseline is preferred over B2 aggressive electrification under both Di and Krewski. The CRF changes the magnitude of the benefit, but not the ranking. B2 does not justify its incremental cost under either cohort, because fossil space- and water-heating already contribute a small share of ambient PM2.5 compared with transport and wildfire.

No policy under study reorders under the Di ↔ Krewski CRF switch. The CRF is an uncertainty on the scale of benefit, not on which portfolio to deploy.

Finding
The Di vs Krewski CRF choice does not flip the optimal transport portfolio — both independently select T2 accelerated electrification, with T4 equity a consistent second. What the CRF does change is the burden framing: the Krewski ACS ≥30 cohort covers 14.7 million additional Californians aged 30–64, giving a 4× larger baseline cohort mortality pool (125,897 vs 31,407 annual all-cause deaths over which the CRF applies) and a 1.5× larger PM2.5-attributable burden (17,660 vs 11,689 annual deaths). Residual CRF uncertainty after hierarchical-Bayes pooling is quantified by Inv 21 at $0.05B EVSI; a $0.5M meta-analysis resolves it at 100:1 ROI.

What this changes about the decision. Because the ranking holds under both CRFs, California does not need to pick one before acting — T2 accelerated is optimal either way. The CRF choice is a burden-reporting decision, not a policy-selection one. An age-stratified meta-analysis ($0.5M, 6 months) is still worth running to narrow the $5.8B ↔ $22.5B reported-benefit range before CARB files GFO-25-304.

10,000-draw Monte Carlo · Di (Medicare ≥65, β = 0.00705) vs Krewski (ACS ≥30, β = 0.00545) run independently · Sobol global sensitivity analysis · ISRM source–receptor matrix · 5 transport + 4 building policies · Burden ratio from age-decomposition of ambient CA population against each cohort’s age floor · Residual EVSI anchored on the Inv 21 hierarchical posterior over the CA-pooled CRF