"Problem identification serves two important functions.

• It provides you the information you need to select an appropriate countermeasure and target audience for your program….

• It may provide you with some of the baseline data you need to determine if your program meets its objectives….

"During the problem identification step, you also lay the foundation for your data collection efforts throughout the program evaluation." (NHTSA, 1999)

A highway-safety analyses will generally be performed to identify either systemic problems (i.e., those which are generally prevalent in the system, at least on one class of roadway), or local problems (those identified as existing in a limited section, area, or at a point). The analyst is generally seeking to answer a few basic questions about potential problem areas, such as:

1. Does the type of crash represent a significant part of the total crash picture (by severity)?

2. Are there factors present that seem to be over represented?

3. How is the identified problem distributed through the population (e.g., roadway segments and driver attributes)?

4. Does the nature of the problem suggest that there are feasible solutions available to consider?

A first step in most analyses is the summary and presentation of basic safety data. For crash data, the purpose is to determine which types of crashes represent the largest part of the problem, as well as to identify if there are patterns (e.g., time of day, or location), or over representations present. (e.g., elderly, specific fixed objects, etc.), which suggest strategies that may be applied to reduce or eliminate the problem. Supplementary data are needed to account for exposure and to provide evidence supporting any potential problems identified from crash data. Where crash data are not available, supplementary data of the type listed above, may have to be relied upon solely.

Analyses of crashes often involve tabulations of frequencies and rates either as univariate or multivariate tables and graphs. Graphs should be used wherever possible since they are easier to interpret and find patterns. Diagrams such as collision diagrams and fault trees are often used.

The manner in which information is displayed can have an impact on how well it is understood. Although there are many standard computer-based tools for presenting information, users need to be "visually literate." Tufte (1983) provides some useful guidelines for graphic presentation of information.

Supplementary statistical procedures are also used to assist in identifying patterns and over-representations. For example, high-hazard locations are often identified using frequency and rate information and identifying locations that have values in excess of what would be expected for similar sites, within some reasonable statistical accuracy. There are many references on identifying high-hazard locations (for example, see Ogden, 1996, Chapter 5, and Robertson, et al, 1994, pp 201- 210). Regression to the mean is one major issue that must be addressed in identifying high-hazard locations. The problem arises due to random variations that occur in crash experience. The result is that when a particular year’s data show a significantly high frequency or rate of crashes, it is likely that the following year or two will experience a lower frequency or rate, even if no action is taken. It is usually recommended that three years of data be used to help avoid this problem. Use of statistical techniques, such as Bayesian estimators, have also been suggested (See Robertson, et al, 1994, pp 201-210).

Systemic problems. The types of systemic involvement usually being tested for include:

1. Environmental – Physical (weather, sun glare, animals, trees and poles, adjacent development), Social (laws, enforcement and adjudication system)

2. Roadway - Geometric (any single or combination of horizontal, vertical or cross-sectional elements generally present; usually due to policies or traditional practices; including roadside objects)

3. Human – DUI, drivers with suspended or revoked licenses, aggressive, confused, drowsy, law breakers, age, experience, information needs (lack of, or overload)

4. Vehicle – vehicle defects, tire-pavement friction

Simple bar or pie charts of frequencies of key crash types and contributing circumstances will provide an initial means for studying the problem. Three-dimensional bar charts can sometimes be used to visualize cross-tabulations (e.g., crash type vs. contributing circumstances). When analyzing for a particular type of problem, it is often best to identify types of crashes that relate specifically to the emphasis area being considered.

Over representation can be tested, using rates and relative percentages, comparing specific locations and populations with system averages, or averages for similar types of facilities. The availability of geographic information systems helps to visualize the distribution in the system. Where a geographic information system is not available, it is possible to segment the system, using one of a variety of available algorithms, and tabulate the frequencies as a cumulative distribution to see the degree of dispersion or concentration that is present on the system.

For example, it is obvious that if the emphasis area is focused on suspended and revoked drivers, then that portion of the crash experience involving those types of drivers should be singled out to determine if:

1. They represent a significant part of the crash problem (i.e., a significant percentage of the total, and a significant frequency)

2. They are over represented in the crash problem (i.e., consider what the contribution of the type of driver is to the crash problem, relative to the proportion the driver is of the total driver population, using numbers of persons or total vehicle-miles driven as the exposure measures). Procedures can be applied to see if the differences are statistically significant.

Another example is an analysis for the emphasis area addressing run off road crashes. It is obvious that one wants to focus upon those crashes that involve a vehicle leaving the traveled way. These would be compared to all other crashes to see if they were a significant part of the total crash picture.

In addition, cross tabulations may be needed to check for factors such as:

1. The severity of the crashes

2. Presence of curves or other geometric features

3. Day vs. night conditions

4. Wet vs. dry conditions

5. Temporal variations (e.g., time of day)

6. Presence of driver factors, such as age, presence of alcohol, drowsiness, etc.

7. Presence of vehicle factors, such as vehicle size and weight

8. Pedestrian and bicyclists involved

9. Degree of dispersion of this crash type throughout the system.

10. Fixed objects involved

Each of these summaries will provide insight which leads toward not only identifying if a problem exists, but also toward identifying candidate strategies to apply to reduce or eliminate the problem.

Systemic problems may sometimes be identified without the need for analysis of crashes. Where well-designed research has shown, for example, that:

1. An additional feature is needed (e.g., addition of edge-line markings),

2. A design policy should be modified (e.g., shoulder widths on two lane highways), and

3. A legal and enforcement effort is needed to reduce driving while unlicensed, or when having a suspended or revoked license,

then a systemic program may be mounted. In any case, it will be necessary to document measures of safety, such as the appropriate type of crash history, so that effectiveness evaluations may be made after the program is implemented.

Localized Problems. Many of the comments made above apply as well to analyses of individual intersections or road segments, corridors, or small areas. The FHWA Procedural Guide (1981) suggests crash summaries by:

1. Type

2. Severity

3. Contributing circumstances

4. Environmental conditions

5. Time period

The analysis of localized problems often is a step that follows studies that identify and prioritize high-hazard locations, or some other event (e.g., a recent fatal crash) that singled out the particular local site. Since the sights being considered are often already likely candidates for action, more resources are invested in the acquisition and analysis of information about potential problems. As a result, studies of crashes are usually accompanied by, or followed with, supplementary studies of the type listed above.

In this situation, it is often feasible to synthesize data from a variety of sources into map displays. The traditional collision diagram is a prime example (see Robertson, et al, 1994, pp211-212). However, today’s computer-based systems allow collision diagrams to be generated which contain greater detail on the geometry at the site, as well as the nature of the crashes. Some include tabulation and analytical capabilities. (see http://www.pdmagic.com/im/ as an example – no product endorsement implied).

Bowman (1986) provides an accident analysis worksheet, which provides a series of activities for the analyst to follow for localized studies (See appendix A).

References

B. Bowman, Local Highway Safety Studies – User’s Guide, National Highway Institute, FHWA, July 1986.

Federal Highway Administration, Highway Safety Engineering Studies – Procedural Guide, FHWA Report No. FHWA-TS-81-220, November 1981.

National Highway Traffic Safety Administration, The Art of Appropriate Evaluation, Report No. DOT HS 808-894, May 1999.

Douglas Robertson, et al (editors), Manual of Transportation Engineering Studies, Prentic Hall, Englewood Cliffs, NJ, 1994, 514 pp.

E, R, Tufte, The Visual Display of Quantitative Information, Graphics Press, Chesire Connecticut, 1983, 197 pp.