Crash Rates and Countermeasures

A busy intersection with many crashes can be safer than a quiet one with few.

Two intersections need a safety upgrade, and the budget covers one. Over the last three years, Main and 5th, downtown, logged 24 crashes. The rural crossroads at Route 9 and Mill Road logged 9. On a crash map, Main and 5th glows red. But 40,000 vehicles a day enter Main and 5th, and only 6,000 enter the crossroads. Per vehicle that drives through, the quiet crossroads is two and a half times as dangerous. By the end of this Lesson you'll be able to show that in two lines, then estimate how many crashes a fix would prevent and what that's worth.

Counting chances to crash

A crash count mixes two things: how dangerous a place is, and how much traffic passes through it. Send twice the traffic through a perfectly ordinary intersection and you'd expect about twice the crashes, though nothing about the intersection got worse. To see the danger on its own, divide the count by the number of chances to crash, which is called the exposure. A crash rate is crashes per unit of exposure, for the same reason a batting average is hits per at-bat and not just hits.

What counts as one chance depends on the kind of site.

  • At an intersection, the risk is in passing through the place where paths cross. Every vehicle that enters takes that chance once, whatever road it came in on and wherever it's headed, so the exposure is the number of vehicles entering.
  • On a road segment, a vehicle is at risk for every mile it drives, so a car that covers 4 miles of the segment has four times the chances of one that covers 1. The exposure is vehicle miles: vehicles times the distance each one travels.

Traffic comes as an ADT, the average daily traffic in vehicles per day. Crashes come as a count over a study period. Turning the first into an exposure that matches the second is most of the arithmetic in this Lesson.

Checkpoint

A county is ranking its sites for safety funding. Which of these comparisons are fair?

A.
Two segments of different lengths, by crashes per million vehicle miles
B.
Two intersections with three years of data each, by crash counts
C.
Two intersections, by crashes per million vehicles entering on the major road
D.
Two intersections, by crashes per million vehicles entering from all approaches
E.
Two segments of different lengths, by crashes per million vehicles that use them

Crash rates at intersections

The Handbook's Traffic Safety Equations give the crash rate per million entering vehicles as

RMEV=A×1,000,000VV=ADT×365\text{RMEV} = \frac{A \times 1{,}000{,}000}{V} \qquad V = \text{ADT} \times 365

where AA is the number of crashes and ADT is the average daily traffic entering the intersection, summed over every approach. The 1,000,000 only keeps the numbers readable, so a rate reads 0.7 instead of 0.0000007. The 365 turns a daily volume into a yearly one, and that's the catch: the Handbook defines AA as the crashes in a single year. Crash studies usually pool several years, since one year holds too few crashes to judge a site by. With several years of crashes on top, the exposure has to cover the same years:

RMEV=A×1,000,000ADT×365×years\text{RMEV} = \frac{A \times 1{,}000{,}000}{\text{ADT} \times 365 \times \text{years}}

That step, a daily volume times 365 days times the number of years, expands traffic to the total over the study period, and it's the one most often dropped. Run the Hook's two intersections through it, with 3 years of crashes at each:

Main and 5th:24×1,000,00040,000×365×3=2443.8=0.55Route 9 and Mill:9×1,000,0006,000×365×3=96.57=1.37\begin{aligned} \text{Main and 5th:}&\quad \frac{24 \times 1{,}000{,}000}{40{,}000 \times 365 \times 3} = \frac{24}{43.8} = 0.55 \\ \text{Route 9 and Mill:}&\quad \frac{9 \times 1{,}000{,}000}{6{,}000 \times 365 \times 3} = \frac{9}{6.57} = 1.37 \end{aligned}
Crashes in 3 years 24 Main & 5th 9 Rt 9 & Mill Crashes per million entering vehicles 0.55 Main & 5th 1.37 Rt 9 & Mill
Crash counts and crash rates at the Hook's two intersections
Main and 5th has the most crashes, but 40,000 vehicles a day enter it against 6,000 at Route 9 and Mill Road. Per million entering vehicles, the crossroads' rate is 2.5 times higher.

Counts don't always arrive as entering volumes. A count station on one leg of an intersection usually counts that leg's traffic in both directions, the vehicles heading in and the vehicles heading out. Every vehicle uses one leg to enter and another to leave, so it crosses two count stations, and adding the two-way counts on every leg counts each vehicle twice:

entering ADT=12∑(two-way ADT on each leg)\text{entering ADT} = \tfrac{1}{2}\sum \left(\text{two-way ADT on each leg}\right)
North leg South leg East leg West leg two-way count station 1 counted entering 2 counted leaving
One vehicle, counted on two legs

So entering volumes by approach get added as they are, and two-way volumes by leg get added and halved.

Worked Example A four-leg intersection over three years
1 of 2

A four-leg intersection recorded 18 crashes over 3 years. The average daily volumes entering on its approaches are 5,200 vehicles northbound, 4,800 southbound, 7,100 eastbound, and 6,900 westbound. Find its crash rate per million entering vehicles.

Crash rates on road segments

For a segment, the Handbook gives the rate per million vehicle miles:

RMVM=A×1,000,000VMTVMT=ADT×(days in study period)×(length of road)\begin{aligned} \text{RMVM} &= \frac{A \times 1{,}000{,}000}{\text{VMT}} \\ \text{VMT} &= \text{ADT} \times (\text{days in study period}) \times (\text{length of road}) \end{aligned}

VMT is the vehicle miles traveled, the segment's exposure. ADT is the daily traffic on the segment, both directions together, since every one of those vehicles drives its length, and the length is in miles. Unlike the intersection formula, this one already spans the whole study period: AA is the crashes over the period, and the days are the period's days, 365 times the years.

Many agencies, and some exam questions, report segment rates per 100 million vehicle miles instead. The Handbook prints only the per-million form. For the other, put 100,000,000 where the 1,000,000 was. Nothing else changes, so the per-100-million rate is 100 times the per-million rate: 2.3 crashes per million vehicle miles is 230 per 100 million. Read which one the question asks for.

Checkpoint
1 of 2

A 0.8-mile urban arterial carries an ADT of 22,000 and had 30 crashes over 2 years. What is its crash rate per million vehicle miles?

per million vehicle miles

Ranking sites, and running a rate backward

With rates in hand, ranking is sorting: compute each site's rate in the same units, and the highest rate marks the most dangerous site. Rank intersections against intersections and segments against segments, because a rate per entering vehicle and a rate per vehicle mile divide by different kinds of exposure.

A rate also runs backward. If a site keeps its rate, the crashes to expect over a coming period are the rate times that period's exposure, in millions:

A=RMEV×V1,000,000orA=RMVM×VMT1,000,000A = \frac{\text{RMEV} \times V}{1{,}000{,}000} \qquad\text{or}\qquad A = \frac{\text{RMVM} \times \text{VMT}}{1{,}000{,}000}
Checkpoint
1 of 2

Three road segments each have 3 years of crash data. Which has the highest crash rate?

Three segments' lengths, traffic, and crashes
SegmentLength (mi)ADT (veh/day)Crashes
A2.015,00024
B0.89,00010
C5.04,00018
A.
Segment A
B.
Segment B
C.
Segment C

Countermeasures act in sequence

A countermeasure's crash reduction factor, CR, is the fraction of crashes it's expected to prevent. A CR of 0.30 means 30% fewer crashes. The Handbook gives the formulas but no CR values, so a question supplies them. You'll also see the same effect stated the other way around, as a crash modification factor, CMF, the fraction of crashes that remain:

CMF=1−CR\text{CMF} = 1 - \text{CR}

A CR of 0.30 and a CMF of 0.70 describe the same countermeasure. The Handbook uses only CR, while the CMF is the Highway Safety Manual's form. The CMF multiplies directly: 0.70 times the crashes before is the crashes after.

Now stack three countermeasures at one intersection, say a protected left-turn phase with CR = 0.40, better lighting with CR = 0.30, and high-friction surfacing with CR = 0.40. Adding the factors promises 110%, more crashes prevented than there were. The adding goes wrong because no countermeasure can prevent a crash that an earlier one already prevented. Each works only on the crashes left over. Start from 100 crashes. The turn phase prevents 40 and leaves 60. The lighting prevents 30% of those 60, or 18, and leaves 42. The surfacing prevents 40% of 42, or 16.8, and leaves 25.2. All three together prevent 74.8 crashes of the 100.

0 25 50 75 100 Crashes 100 40 prevented 60 18 prevented 42 16.8 prevented 25.2 Before After CR1 = 0.40 After CR2 = 0.30 After CR3 = 0.40
Three countermeasures applied to 100 crashes, one after another

The Handbook's overall crash reduction factor "for multiple mutually exclusive improvements at a single site" writes out that same sequence:

CR=CR1+(1−CR1) CR2+(1−CR1)(1−CR2) CR3+⋯\text{CR} = \text{CR}_1 + (1 - \text{CR}_1)\,\text{CR}_2 + (1 - \text{CR}_1)(1 - \text{CR}_2)\,\text{CR}_3 + \cdots

Each term is one countermeasure's share of what the earlier ones left. For the three above, the terms are the figure's 40, 18, and 16.8 crashes, as fractions of the original 100:

CR=0.40+(0.60)(0.30)+(0.60)(0.70)(0.40)=0.40+0.18+0.168=0.748\text{CR} = 0.40 + (0.60)(0.30) + (0.60)(0.70)(0.40) = 0.40 + 0.18 + 0.168 = 0.748

The fraction of crashes that survives all of them is the product of the (1−CR)(1 - \text{CR}) terms, the CMFs, which gives a shortcut to the same number:

CR=1−(1−CR1)(1−CR2)(1−CR3)=1−(0.60)(0.70)(0.60)=0.748\text{CR} = 1 - (1 - \text{CR}_1)(1 - \text{CR}_2)(1 - \text{CR}_3) = 1 - (0.60)(0.70)(0.60) = 0.748

A product doesn't care about order, so it doesn't matter which countermeasure you list first. And because each (1−CR)(1 - \text{CR}) is above zero, the combined reduction stays below 100% unless a single countermeasure removes every crash by itself. The crashes expected afterward are N(1−CR)N(1 - \text{CR}), here 100(0.252) = 25.2.

Checkpoint
1 of 3

A site averages 22 crashes per year. Three countermeasures with crash reduction factors of 0.40, 0.20, and 0.15 are planned. With traffic unchanged, how many crashes per year should you expect afterward?

crashes/yr

Crashes prevented, and what they're worth

The combined CR is the fraction of crashes a project prevents. The Handbook's Crash Reduction equation turns it into a count:

crashes prevented=N×CR×ADT after improvementADT before improvement\text{crashes prevented} = N \times \text{CR} \times \frac{\text{ADT after improvement}}{\text{ADT before improvement}}

where NN is the expected number of crashes if the countermeasure isn't built and the traffic volume stays the same, usually the crashes per year the site has been having. The ADT ratio is there because traffic rarely holds still. If the ADT rises 20% after the project, the site would have had about 20% more crashes without the fix, and the fix prevents its share of that larger number. So the ratio is after over before, and growing traffic raises the crashes prevented. If you're ever unsure which way up the ratio goes, that's the check. The crashes that still happen are the larger number less the prevented ones:

crashes remaining=N×ADTafterADTbefore×(1−CR)\text{crashes remaining} = N \times \frac{\text{ADT}_{\text{after}}}{\text{ADT}_{\text{before}}} \times (1 - \text{CR})

A crash count still isn't what an agency budgets with, because crashes differ enormously in cost. A property-damage-only (PDO) crash costs a repair bill, while an injury crash adds medical care and lost work. To put a dollar value on a project, find the crashes prevented at each severity and multiply by that severity's average cost per crash:

annual savings=∑severities(crashes prevented per year)×(cost per crash)\text{annual savings} = \sum_{\text{severities}} (\text{crashes prevented per year}) \times (\text{cost per crash})

The Handbook gives no crash costs, so a question states them. Those annual savings are the benefit side of a benefit-cost ratio, the subject of Benefit-Cost Analysis.

Worked Example A new signal with growing traffic
1 of 3

An intersection averages 10 crashes per year, 2 with injuries and 8 PDO, at an ADT of 18,000. A new traffic signal with CR = 0.35 is planned, and the ADT is expected to reach 21,000 once it's running. Injury crashes cost $180,000 each and PDO crashes $12,000. Find the crashes prevented per year and the annual savings.

Recap

  • A crash count mixes danger with traffic. A crash rate divides by exposure, the chances to crash: vehicles entering for an intersection, vehicle miles for a segment. Rank sites of the same kind by rate, and multiply a rate by the exposure in millions to forecast crashes.
  • At an intersection, RMEV=A×106/(ADT×365×years)\text{RMEV} = A \times 10^6 / (\text{ADT} \times 365 \times \text{years}), with ADT entering from every approach, or half the sum of two-way leg counts.
  • On a segment, RMVM=A×106/(ADT×days×length)\text{RMVM} = A \times 10^6 / (\text{ADT} \times \text{days} \times \text{length}). A rate per 100 million vehicle miles uses 10810^8 and is 100 times larger.
  • Countermeasures act in sequence: CR=CR1+(1−CR1)CR2+⋯\text{CR} = \text{CR}_1 + (1 - \text{CR}_1)\text{CR}_2 + \cdots, which is one minus the product of the CMFs, with CMF=1−CR\text{CMF} = 1 - \text{CR}.
  • Crashes prevented = N×CR×ADTafter/ADTbeforeN \times \text{CR} \times \text{ADT}_{\text{after}} / \text{ADT}_{\text{before}}. Annual savings are the prevented crashes at each severity times its cost per crash.