Crash Rate Calculation

A crash rate normalizes crash frequency by exposure so a busy arterial and a quiet collector can be compared. Two standard forms: segments per million vehicle-miles traveled (MVMT) and intersections per million entering vehicles (MEV). This note is the math behind the rate bullets in How to Summarize 5-Year Crash Data.

Segment rate (per MVMT)

R = (C × 1,000,000) / (365 × N × AADT × L)
 
C    = crashes on the segment in the study period
N    = number of years
AADT = annual average daily traffic (two-way, veh/day)
L    = segment length in miles

Many agencies (and the summary note in this vault) report segment rates per 100 MVMT — same formula with 100,000,000 in the numerator, i.e., the per-MVMT rate × 100. State which basis you used; a 100× labeling mix-up is a classic reviewer catch.

Intersection rate (per MEV)

R = (C × 1,000,000) / (365 × N × V)
 
C = crashes at the intersection in the study period
N = number of years
V = total daily entering volume (veh/day)

V is entering vehicles, not the sum of two-way leg AADTs: if all you have is two-way AADT per leg, entering volume ≈ half the sum of the leg AADTs (assumes balanced directional split); turning movement counts expanded to daily volumes are better where available.

Worked example

Intersection (the practice question in How to Summarize 5-Year Crash Data): 12 crashes over 5 years, 25,000 entering ADT. R = (12 × 1,000,000) / (365 × 5 × 25,000) = 12,000,000 / 45,625,000 ≈ 0.26 crashes/MEV

Segment: 30 crashes over 5 years on a 2.0-mile segment with 20,000 AADT. R = (30 × 1,000,000) / (365 × 5 × 20,000 × 2.0) = 30,000,000 / 73,000,000 ≈ 0.41 crashes/MVMT (= 41 per 100 MVMT)

Exposure data sources

• Segment AADT: Florida Traffic Online (FTO) — pull per How to Pull AADT and K-D-T Factors from Florida Traffic Online; note the count-station year and whether the value is an estimate.
• Off-system/local roads: County & MPO Count Programs (Tampa Bay) or project counts.
• Use the AADT that matches the crash years — a 5-year crash period paired with a single-year AADT is common practice but say so; averaging available years across the period is cleaner.
• Crash counts: How to Pull Crashes from Signal Four Analytics — the geography definition (segment limits, intersection buffer) must match between the C and the exposure, or the rate is fiction.

Rate vs frequency vs EB-expected

Three tiers, increasing rigor:

1. Frequency (crashes/yr): fine for ranking sites with similar volumes; severity-weight it before it means anything (see KABCO Injury Scale).
2. Rate (this note): controls for exposure, but assumes crashes scale linearly with volume — they don’t.
3. EB-expected (SPF + observed, Empirical Bayes): corrects both the nonlinearity and regression-to-the-mean; the Highway Safety Manual (HSM) Part B method and the right answer for serious screening or before/after work.

When rates mislead

• Low-volume sites: a tiny denominator inflates the rate — 2 crashes on a 500-AADT local street posts a huge rate from what may be pure chance. Never rank a mixed network by raw rate; the top of the list fills with low-volume sites.
• Nonlinearity: because SPFs are nonlinear in AADT, low-volume facilities systematically show higher per-exposure rates than high-volume ones even when performing “normally” for their class.
• Regression to the mean: a site picked because it spiked will tend to drop with no treatment; rates don’t fix this — EB does.
• Comparator abuse: “above average” requires a stated comparator (district/statewide average for the same facility type and area type) #status/verify current FDOT district values — never quote a comparator average from memory.

Critical rate

The statistically honest version of “above average”: compare the site rate to a critical rate — the average rate for similar facilities plus an allowance for random variation at the site’s own exposure (HSM network screening, Crit. Rate method):

Rc = Ra + P × sqrt(Ra / M) + 1 / (2 × M)
 
Ra = average rate for the reference population (same facility type)
P  = z-value for the chosen confidence level
M  = site exposure in MEV (intersections) or MVMT (segments)

Only sites with R > Rc are flagged — high-exposure sites face a tighter threshold, which is exactly the low-volume correction the raw rate lacks.

Warning

A crash rate with an unstated denominator basis (MEV vs MVMT vs 100 MVMT), an unmatched geography, or no comparator is a number, not a finding. Document C, N, AADT/V source and year, and length in the methodology — same discipline as the pull log in How to Pull Crashes from Signal Four Analytics.

See also: How to Summarize 5-Year Crash Data · KABCO Injury Scale · Highway Safety Manual (HSM) · Florida Traffic Online (FTO)