Crash cost data & calculation audit

November 3, 2026 · v1.5 · national unit costs in 2024 USD

Published cost per crash

FHWA-SA-25-021, October 2025, Table 1 supplies the following national values. Each crash is classified by its highest injury severity; average injuries and vehicles involved are already incorporated. Comprehensive values include economic losses plus quality-of-life valuation. Select one column; do not add them or apply another occupant/vehicle multiplier.

These retain the national price basis; no Michigan income adjustment is applied. The FHWA workbook supports additional adjustments. Economic loss is broader than insurance payouts or cash spending. Comprehensive loss includes nonmonetary harm.

Michigan severity counts used for the preset

Michigan Traffic Crash Facts 2024, Crash Experience by Crash Type, page 5 supplies crash counts by maximum severity across all Michigan roads. The arrays below pool the specified rows without mixing counts of injured people into crash totals.

Rear-end pools ordinary rear-end, rear-end–left-turn and rear-end–right-turn. Opposing direct-left uses head-on–left-turn. Angle uses the broader angle row. Sideswipe pools same- and opposite-direction rows. The publication does not isolate fixed-object and overturn severity within its single-vehicle category; neither that category nor its Other category substitutes for our fifth group.

Pooling is itself an approximation: statewide subtype weights need not match either layout's historical subtype weights. The same statewide conditional severity distribution is copied independently to each layout initially. Editing Standard never alters Michigan. The preset does not encode a measured MUT-specific severity reduction.

Calculated group averages

For each known group, divide each severity count by its group total, multiply by the selected FHWA crash-unit value, then sum. Full-precision source ratios drive the calculation; six-decimal percentages are display values. These averages are calculations from the two sources, not published layout-specific cost estimates.

Annualization and two rate sources

For each seed, annual external entrants = annual intended-left maneuvers × simulated external entrants / simulated intended-left entrants. Each complete Michigan-left trip enters once. For layout L and group t, expected annual cost = annual external entrants / 1,000,000 × type-specific rate(L,t) × sum over severity s of probability(s | L,t) × unit cost(s). Subtract Standard from Michigan within each seed, then average.

The Michigan severity counts supply probabilities conditional on crash type; they do not supply type frequencies or crash occurrence rates. In applied replay base-rate mode, type rates come from the traffic run's applied total rates and normalized detailed type weights, aggregated into the five groups. The historical reference assumptions are 1.644 Standard and 1.388 Michigan crashes per million entrants, with source and boundary limitations documented in crash evidence. Their system boundaries are unverified. Replay multiplier and incident on/off state do not affect this base-rate sensitivity calculation. Changing unapplied controls does not silently alter the retained run.

In custom rate mode, users enter group-specific rates per million unique external entrants and assert compatible main-junction/crossover coverage, crash definitions, periods and exposure units. That assertion cannot fix incompatible data. In custom cost mode, user-entered cost per group replaces the severity-derived average and applies to both layouts. Saved custom values remain intact when switching to or from the source preset.

Incomplete costs stay incomplete

A row with all five severity fields blank is unknown. Partial or invalid mixtures block calculation; known rows must total 100% within 0.0001 percentage point. To complete the fifth group, enter defensible severity shares separately for each layout or use a sourced custom group cost. Do not enter zero merely to make an unknown disappear.

The priced-group difference includes only groups with costs available in both layouts. No type rates are renormalized to exclude the unknown group. Priced coverage and excluded expected crashes remain visible. A partial difference is not a lower or upper bound and cannot establish a full safety ranking. The 200M/550M/1.1B exposure cases scale the same priced groups; they are not confidence bounds.

No combined operating-plus-crash total is produced for replay-rate mode. A custom scenario can combine only when all groups are priced, matching exposure is asserted, all-traffic scope is selected, and operating/crash valuations share a dollar year. Operating results use actual matched crash-free runs so incident delay and fuel are not added twice. Consistent cost definitions still require review. Infrastructure and other unpriced effects remain excluded.

Paired-seed intervals capture exposure/traffic sampling only, not crash-rate, severity, valuation or model uncertainty. Rare-event replay draws do not determine expected annual costs. Exports retain applied rates, both severity mixtures, sources, unit costs, expected severity counts, missing values and dollar year.

View annual crash scenario · Annual operating methods · Crash-rate evidence