Risk AssessmentAuthor: Dr. Jean AndreyNOTE: Contents phased out but kept online for reference.1. Transportation SafetyRisk is a combination of uncertainty and negative consequences. The
development of the field of risk assessment was fostered by a series
of unrelated but significant health studies and was based very much
on the dual development of probability theory and scientific method.
Risk assessment involves three steps : hazard identification, risk estimation
risk evaluation. A fourth aspect of risk studies follows from the
risk assessment process and deals with ameliorating the risk; it has
been labeled risk response, risk control or, most commonly, risk management.
Transportation safety is a term that is intended to convey something
about the avoidance of risk from bodily harm and property damage associated
with the operation of transport systems. However, there is a paradox
in the conventional usage of the term in that most safety statements
instead discuss the amount of risk present. Indeed, safety typically
refers to the characteristic or long-term average risk of an entity.
Low levels of risk are thus equated with safe conditions/systems.
Government agencies maintain databases that document the frequency
and nature of unsafe incidents for the various transportation modes.
There are different indicators for the different modes, e.g. air crashes,
pipeline spills, capsized boats, train derailments and road collisions.
Not all modes contribute equally to the overall transportation risk
experienced by a society. In fact, in most nations of the world, including
Canada, more than 90 percent of casualties (fatalities and injured persons
combined) are victims of road collisions. 2. Road SafetyMotor vehicle accidents are not a simple stimulus-response process;
rather, many factors interact to create collisions events. Sometimes
the phrase 'cause' is used to identify the factor that assigns legal
responsibility for the accident (e.g. driving while alcohol-impaired;
exceeding the speed limit). However, in most accident research, the
concept of cause has been replaced with the concept of contributing
factor. The term contributing factor refers to those conditions/ whose
variation will alter the risk of collision. The first formal articulation of the factors contributing to motor
vehicle collisions was developed by the public health community in the
1940's and is referred to as the Haddon Matrix.
The Haddon Matrix is a useful conceptual framework for organizing
both risk factors and accident-countermeasures strategies. Because there are so many potential hazards affecting the safe operation
of motor vehicles, risk analysis tend to focus mainly on risk estimation,
i.e. quantifying and comparing collision-involvement rates for different
combinations of drivers, vehicles and environments. 3. Collision Involvement RatesCollision-involvement rates are calculated as as follows :
NUMBER OF COLLISIONS EXPOSURE
Depending on the focus of the study, the collision data may pertain
to all reported collisions or to specific subsets, for example according
to severity (e.g. property damage only (PDO), injury or fatal), number
of motor vehicles involved (e.g. one, two, three or more), or pre-collision
maneuver (e.g. turning maneuver, ran off road). It is also common to focus on particular subsets of drivers (e.g.
by age or sex), vehicles (e.g. passenger vs. freight), roads (e.g. rural
vs. urban; two-lane vs. multi-lane), and driving conditions (e.g. according
to time of day, weather condition or political jurisdiction). In these
cases, it is important that the collision and exposure data are compatible.4. Exposure to Risk and the Calculation of Accident RatesThe concept of exposure has been an integral part of road safety
analysis since WWII. Although defined in various way, the idea is fairly
straightforward : one would expect accident frequency to increase as
travel increases, both at the individual and societal level. As a result,
exposure is used to normalize accident data to allow for comparison.
For example, if Driver A has 10 accidents while driving one million
kilometers and Driver B has 5 accidents while driving 200,000 kilometers,
then one would conclude that Driver A's accident risk is lower because,
per kilometer, Driver A is less likely to be involved in a collision
(.00001 involvements/km for Driver A vs. .000025 involvements/km for
Driver B). This difference in risk could be due to differences in driver
characteristics (e.g. skill, experience, risk taking), vehicle characteristics
(e.g. size, state of repair) or the quality of exposure (e.g. proportion
of driving at night, during inclement weather, in high traffic, in urban
areas, on undivided highways).Because exposure can be interpreted very broadly, a wide variety
of exposure measures have been used in the field of road safety; ultimately,
the choice of exposure measure is dependent upon the elements under
investigation. Byun et al (1979) conceptualize that there are four denominations
of exposure measures : time-, event-, activity-, and population-based
measures. Chipman et al (1991) suggest that risk exposure can be measured
in terms of population, driver-distance, or driver-time. This latter
classification seems the most appropriate because travel, and thus exposure,
occurs within these three dimensions. The following table lists examples
for each class of measure. Throughout the literature, distance travelled
and time travelled are the most widely used measures of driver exposure
Average Daily Traffic (ADT)
Number of licensed drivers
Number of registered vehicles
Number of trips
Number of passengers
Another way of comparing accident rates is to calculate risk ratios.5. Measuring ExposureTo be of use in risk analysis, exposure data need to be assembled
from reliable sources. There are three widely accepted methods of collecting
exposure data : direct observation, survey techniques, and induced measurement.
A less popular method is to use fuel sales to estimate total distance
travelled. The advantages and disadvantages of each are discussed below.
Direct methods include personal observation and mechanical counting.
In the health sciences, direct measurements can be obtained through
laboratory experiments. The driving environment, however, cannot be
as easily modeled in a lab; however, personal observations about driving
behavior are possible. Researchers can observe driver behavior as a
passenger or as a roadside observer (Wolfe, 1982). These method allows
the researcher to collect a wide variety of exposure measures in a diverse
range of situations. Direct Observation is, obviously, very time consuming
and not very efficient for studies of large groups of drivers or vehicles.
A less costly method of direct observation uses mechanical counting
devices to record the number of vehicle axles that pass by (Wolfe, 1982).
This method is useful in local site evaluations, but again is not amenable
to studies of large driver or vehicle groups, and provides no information
about the quality of trips. Overall, direct methods can provide very
reliable data, but these methods tell us little about driver or trip
characteristics. Surveys are the most widely used method to gather exposure information
about drivers in the road safety community (e.g., see Wolfe, 1982; Cerrelli,
1989; Joly et al., 1991; Chipman et al., 1991; Stutts and Martell, 1992).
Two examples of exposure databases compiled from survey data include
the National Personal Transportation Survey (NPTS) in the United States
and the 1988 and 1994 Exposure Surveys for Ontario funded by the Ontario
Ministry of Transportation. Typically, a random sample is drawn from
the entire driver population and trip data are collected for a period
of one or more days. Exposure information is then extrapolated to make
inferences about various driver sub-populations. There are several methods of conducting surveys. Researchers may
use personal interviews, self-administered questionnaires, or mail-back
surveys. Personal interviews may consist of roadside surveys or telephone
interviews. The presence of an interviewer is likely to reduce the number
of incomplete records (Wolfe, 1982). The presence of a researcher may
or may not be required for self-administered questionnaires. Mail-back
questionnaires typically provide less complete data than the personal
interviews. Problems may arise when drivers are asked to recall past
trips. Also, the season in which data are collected may introduce bias.
Respondents are usually asked to provide trip details for a period of
one or more days. The lure of the mail-back questionnaires is, however,
the opportunity to gather data for selected variables for a larger sample.
Studies of driver populations at the regional, provincial, or national
level are very expensive and for this reason, empirical data on travel
behaviour are not collected annually. In 1964, Thorpe (1964) introduced a technique that can be used to
by-pass the need for traditionally gathered empirical estimates of driver
exposure. The technique is known as the induced exposure method, and
is based on several assumptions. The technique has been applied in modified
form in several recent studies (e.g. Stamatiadis and Deacon, 1997).
Theoretically, the induced-exposure method allows risk comparisons to
be made for a wide range of situations. Another clever way to bypass the need to collect exposure data is
to use the matched-pair (also known as the matched-sample) approach.
This approach is commonly used in studies of driving-related weather
6. GIS Applications in Road Safety ResearchAn early application of GIS was to determine optimal routes for emergency
vehicles (Bridgehouse, 1993). Addresses obtained from incoming 911 calls
are matched to a road network and the quickest route can be resolved.
The cost of implementing such a system is enormous, primarily because
of the time needed to create an accurate province- or state-wide address
database. Like the emergency vehicle application, the transport of hazardous
materials is another area of safety research where GIS technology has
been successfully employed (Lepofsky and Abkowitz, 1993; Brainard et
al., 1996). These studies evaluate the variations in risk along any
given road and then determine the safest route of transport. The topological
structure of a vector-based GIS is well suited to route assignment based
on origin-destination data. Overlay and buffer operations are particularly
helpful when examining different hazard scenarios. The application of GIS to issues in accident analysis has been
minimal, although the number of publications has increased recently.
Most studies using GIS have been limited primarily to black spot
analyses. Black spots occur on a map when several accidents occur at the
same location. The use of GIS permits analysts to visually inspect the
spatial distribution of accidents and interactively query individual
sites. It is also possible to use GIS to evaluate the safety benefits
of various local countermeasure activities by examining the frequency
and spatial distribution of accidents (e.g., Affum and Taylor, 1997).
As a consequence, most of the studies that focus on the accident ignore
the social factors that give rise to the risk of an accident. Hence, there is an opportunity to use GIS to move beyond the accident
and focus on the travel that gives rise to traffic accidents. Many researchers
in the road safety community have stated that poor exposure data compromise
their ability to estimate accident rates, especially when the data are
disaggregated by driver and driving situation. Two recent studies illustrate
the use of GIS to estimate risk exposure. Although the above studies found the use of GIS beneficial, there
are some limitations of GIS technology that have restricted the amount
of research on travel. The data structure in a GIS over-simplifies the
transportation environment. This has serious implications for modeling
accuracy in large scale urban transportation modeling. Furthermore,
there are two fundamental problems when using the network assignment
procedures of a GIS to predict travel or travel demand. First, most
travel models require traffic flow or congestion information. These
values vary with every trip and most GISs are not able to account for
any iterative variations in traffic flow in a user friendly manner.
Second, the topological data model used by vector GISs cannot adequately
account for the "virtual" relationships between unconnected map features
essential to calculating and predicting urban travel demand. Shaw (1993)
surmises that a GIS is not presently suitable for any of the four steps
in the urban travel demand model. He concludes that most GISs are insufficient
for present traffic demand modeling requirements because the strengths
of a GIS lie in its spatial analytic capabilities rather than its spatial
modeling capabilities. Researchers have therefore questioned the validity
of GIS model outputs (Hartgen et al., 1993). Many of the studies cited
above had to transfer data back and forth between the GIS and other
statistical software complicating the use of GIS for transport and safety-related
studies. Many software developers have recognized these limitations
and have begun to produce GIS packages oriented toward the transportation
modeller (e.g., TransCAD, GIS-Trans).