Highway

Safety II

Crash Pa2erns

Two commonly used techniques to determine crash pa5erns are:

(1) Expected value analysis and

(2) Cluster analysis.

A suitable summary of crash data also can be used to determine pa5erns.

Expected value analysis is a mathemaCcal method used to idenCfy locaCons with abnormal
crash characterisCcs. It should be used only to compare sites with similar characterisCcs (for
example, geometrics, volume, or traffic control), since the analysis does not consider exposure
levels. The analysis is carried out by determining the average number of a specific type of crash
occurring at several locaCons with similar geometric and traffic characterisCcs. This average,
adjusted for a given level of confidence, indicates the “expected” value for the specific type of
crash. LocaCons with values higher than the expected value are considered as over-represenCng
that specific type of crash. The expected value can be obtained from the expression:

Methods of Summarizing Crash Data

A summary of crashes can be used to idenCfy safety problems that may exist at a parCcular site.
It can also be used to idenCfy the crash pa5ern at a site from which possible causes may be
idenCfied, leading to the idenCficaCon of possible remedial acCons (countermeasures). There
are five different ways in which a crash at a site can be summarized:
• Type

• Severity

• ContribuCng circumstances

• Environmental condiCons

• Time periods

Summary by Type

This method of summarizing crashes involves the idenCficaCon of the pa5ern of crashes at a
site, based on the specific types of crashes. The types of crashes commonly used are:

• Rear-end

• Right-angle

• LeR-turn

• Fixed object

• Sideswipes

• Pedestrian related

• Run off road

• Head-on

• Parked vehicle

• Bicycle related

Summary by Severity.

This method involves lisCng each crash occurring at a site under one of three severity classes:
fatal (F), personal injury (PI), and property damage only (PDO). Fatal crashes are those that
result in at least one death. Crashes that result in injuries, but no deaths, are classified as
personal injury. Crashes that result in neither death nor injuries but involve damage to property
are classified as property damage only. This method of summarizing crashes is commonly used
to make comparisons at different locaCons by assigning a weighted scale to each crash based on
its severity. Several weighCng scales have been used, but a typical one is given as:

Fatality =12

Personal injury = 3

Property damage only =1

For example, if one fatal crash, three personal injury crashes, and five property damage crashes
occurred during a year at a parCcular site, the severity number of the site is obtained as follows.

Severity number = (12 x1) + (3 x 3) + (1 x 5) = 26

The disadvantage in using the severity number is the large difference between the severity
scales for fatal and property damage crashes. This may overemphasize the seriousness of
crashes resulCng in fataliCes over those resulCng in property damage. For example, a site with
only one fatal crash will be considered much more dangerous than a site with nine property
damage crashes. This effect can be reduced by using a lower weighCng, for example, 8 for fatal
crashes, especially at locaCons where fatal crashes are very rare in comparison with other
crashes.

Summary by Contribu@ng Circumstances.

In this method, each crash occurring at a site is listed under one of three contribuCng factors:

(1) Human factors,

(2) Environmental factors, and

(3) vehicle-related factors. The necessary informaCon is usually obtained from accident reports.

Summary by Environmental Condi@ons. This method categorizes crashes based on the

(1) LighCng condiCon (i.e., daylight, dusk, dawn, or dark) and

This method of summarizing crashes facilitates the idenCficaCon of possible causes of crashes
and safety deficiencies that may exist at a parCcular locaCon. The expected value method may
be used to ascertain whether crash rates under a parCcular environmental condiCon are
significantly greater at one site than at other similar sites.

Summary by Time Period.

This method categorizes all crashes under different Cme periods to idenCfy whether crash rates
are significantly higher during any specific Cme periods. Three different Cme periods can be
used:

(1) Hour

(2) Day, and

(3) Month.

This method of summarizing data also facilitates the use of the expected value method to
idenCfy Cme periods during which crash occurrences take place.

Iden@fying and Priori@zing Hazardous Loca@ons and Elements

Hazardous locaCons are sites where crash frequencies, calculated on the basis of the same
exposure data, are higher than the expected value for other similar locaCons or condiCons.
Several methods have been used to idenCfy and prioriCze hazardous locaCons. Any of the crash
rates or summaries described earlier may be used to idenCfy hazardous locaCons. A common
method of analysis involves the determinaCon of crash rates based on the same exposure data
for the study site with apparent high rates and several other sites with similar traffic and
geometric characterisCcs. An appropriate staCsCcal test, such as the expected value analysis is
then performed to determine whether the apparent high crash rate at the study site is
staCsCcally significant. If the staCsCcal test shows that the apparent high crash rate is
significantly higher, an abnormal rate of crashes at the test locaCon is likely and that site is to be
considered a hazardous locaCon. All hazardous locaCons are then prioriCzed with that having
the highest crash rate assigned the highest priority.

A technique that is used to idenCfy possible hazardous locaCons is known as the criCcal CRF
method. Since traffic crashes are random occurrences and can be considered as “rare events,” it
is not possible to idenCfy hazardous locaCons simply on the basis of the number of crashes.
Rather, the criCcal rate method incorporates the traffic volume to determine if the crash rate at
a parCcular locaCon is significantly higher than the average for the type of facility. StaCsCcs are
typically maintained by facility type, which is determined by factors such as traffic volume,
traffic control, number of lanes, land-use density, and funcConal classificaCon (see Chapter 15).
The criCcal crash rate method involves the following expression.

Where CR = criCcal crash rate, per 100 million vehicle-miles per million entering
vehicles

AVR = Average crash rate for the facility type

TF= test factor, the number of standard deviaCons given at a confidence level

TB = Traffic base, per 100 million vehicle-miles or million entering vehicles

The crash raCo of actual crash occurrence for the segment being studied with respect to the
criCcal rate is determined. LocaCons with crash raCos greater than 1 are considered hazardous.
All hazardous locaCons can then be prioriCzed by assigning the locaCon with the highest crash
raCo the highest priority, that with the next highest crash raCo the next highest priority, and so
on. All of the methods discussed so far have the inherent flaw of using only number of crashes
or crash rates to make conclusions on safety.
Example: IdenCfying a Hazardous LocaCon Using the CriCcal Crash Rate Procedure

An urban arterial street segment 0.2 mile long has an average annual daily traffic (AADT) of
15,400 veh/day. In a three-year period, there have been eight crashes resulCng in death and/or
injuries and 15 involving property damage only. The statewide average crash experience for
similar types of roadway is 375 per 100 mvm for a three-year period of which 120 involved
death and/or injury and 255 caused property damage only. Is the 0.2 mile long street segment
hazardous? In idenCfying hazardous locaCons, consider that a single death/injury crash is
equivalent to three property damage crashes. Use a 95% confidence level.

SoluCon:

Step 1. Calculate the traffic base, TB.

Years x A ADT x segment length x 365 days per year
TB =
100 million

3 x 15400 x 0.20 x 365
TB =
100 x106

TB = 0.0337/100mvm

Step 2. Calculate the 3-year average crash rate for this type of facility.

AVR = 3 x 120 + 255 = 615 equivalent crashes per 100 mvm per year

Step 3. Select a test factor based on confidence level. Since a confidence level of 95% is
specified, the test factor is 1.96.

Step 4. Compute the criCcal rate.

0.5 AVR
CR = AVR + + TF
TB TB

0.5 615
CR = 615 + + 1.96
0.0337 0.0337
 = 895 crashes per 100 mvm

Step 5. Determine the raCo of actual crash occurrence for the segment with respect to the
criCcal rate.

3 x 8 + 15
segment crash histor y =
0.0337

= 1157 equivalent crashes per 100 mvm
segment crash histor y
crash ratio =
statewide crash histor y

1157
= = 1.29
895

Since the raCo exceeds 1, a safety problem is likely to exit. Specific crash records for the
segment should be reviewed so that appropriate measures can be recommended.