Synthesis of Safety for Traffic Operations - Transports Canada

Introduction
adequate size control group, otherwise random variations in the control group
data could jeopardize the validity of the results.
• Cross-sectional study: This study methodology involves the study of two different
groups of sites that vary only by the feature of interest. For instance, the crash
records of two-lane rural arterials with an 80 km/h speed limit would be
compared to the crash record of two-lane rural arterials with a 60 km/h speed
limit, to determine the safety impacts of a reduction in the speed limit from 80
km/h to 60 km/h. The most obvious shortcoming of this study methodology is the
ability to match the two groups of sites on all other features that will impact on
crash occurrence and severity.
• Empirical Bayes (EB) Techniques: The EB technique is actually an observational
before-after study that uses advanced mathematics to minimize biases introduced
by unrelated effects, and regression-to-the-mean. The procedure is based on the
premise that the true safety record of a site is some combination of the actual
crash frequency at the site itself, tempered by the mean crash record for a
collection of sites with similar characteristics. These two measures of site safety
are combined by considering the amount of site data available, and the reliability
of the group mean. The EB technique most often takes the form of using SPFs (as
discussed previously), combined with site-specific crash data in the before and
after periods to determine the safety impacts. Appendix D provides a more detail
on how this method is used.
The foregoing is a very brief description of the more popular study methodologies found
in the literature and is not exhaustive or complete. For more information and discussion
on each of these techniques the reader is referred to Hauer (1997), and Hamilton
Associates (1997).
Caveats and Cautions
Caution should always be exercised when attempting to apply the results of a study
undertaken in one location to another analogous location. This is particularly true when
attempting to transfer results from one jurisdiction to another. The most common pitfalls
are as follows:
Reporting methods vary between jurisdictions
The definition of an “intersection crash” is a prime example of varying methods. In some
jurisdictions/research, intersection crashes are defined as all crashes that occur with 30
metres of the intersection, and all crashes that are recorded as intersection-related. In
other research the limit is extended within 250 metres of the intersection.
Page 8