Key Transportation Indicators: Summary of a Workshop (2002)

This chapter begins with a brief review of existing measures of transportation safety and their possible use. It then summarizes the subgroup’s discussion of issues associated with existing measures and ends with consideration of possible new indicators.

Three main groupings illustrate the range of current safety indicators considered by the subgroup. They are counts of fatalities, injuries, and accidents; rates (counts per exposure units); and countermeasure-related measures (alcohol-related incidents, seat belt use, truck-related incidents, etc.). Measures in each of these groupings convey very different information and serve different purposes. There may be other types of measures not covered in these groups that have been overlooked.

The number of injured persons, often tabulated by the severity of the injury, is taken as the fundamental measure of transportation safety. Although other losses or disbenefits may be of interest, such as property damage or congestion delay, they are generally not considered safety measures.

Several possible uses have been suggested for existing safety indicators. First, they can describe the current state of transportation safety relative to

the past. This is the most straightforward use of a key indicator and usually implies a comparison over time, typically year-to-year changes. Second, safety indicators can indicate the status of countermeasure programs. In this case, the understanding is derived from the research underlying the development of the countermeasure in the first place. However, the meaning is likely to be limited to the context of the countermeasure. Such measures can be a useful report card for a particular program, but they may not have use beyond the program. Finally, safety indicators can be used to support policy decisions regarding existing and new countermeasures.

The final use, supporting policy decisions, may be problematic. Countermeasure development usually involves some detailed understanding of the relationship of the countermeasure to some safety objective, such as reducing injuries or accidents. Out of that understanding can come some key indicators. But time trends in key indicators characterizing the overall state of transportation safety are not likely to include any information that is relevant to decisions on new countermeasures or programs. In general, key indicators can tell in which directions things are headed, but they usually do not contain information on why the trend is up or down. Thinking that they do appears to be a common misconception with any key indicator. This theme of appropriate and inappropriate use continues through the remaining material discussing issues in the selection of key safety indicators.

Key indicators are numbers that convey information. By its nature, any indicator is necessarily an oversimplification and carries with it an inherent risk of misuse. In addition to selecting the measure, it is also necessary to document the intended meaning of the indicator. What understanding can the user expect to obtain? Four main issues were identified by the subgroup. The first issue is to identify the audience and exactly what the indicator is intended to communicate. The second issue has to do with counts versus rates. The third issue deals with data quality, and the final issue is exposure.

Based on the group’s discussions, a fundamental issue in selecting a key indicator is to identify the audience and clarify what the indicator is in-

tended to communicate. Beyond this, the geographic scope and time period need to be defined. Can national figures support disaggregation to the state or county level? Can data systems that provide annual data also provide quarterly or monthly data? What comparisons are appropriate, and what comparisons are not? Different indicators serve different purposes. Reliability of both the current point estimate and change over time is also an important consideration. What differences are significant?

Another important issue is counts versus rates. The recent public discussion over truck safety featured some disagreement over whether increasing numbers of truck-related fatalities or a declining rate of fatalities per truck miles traveled correctly indicated the current trend in truck safety. Of course, both indicators were correct and taken together show that increases in travel were greater than the decrease in the risk of fatality per miles traveled. The two serve different purposes, and both are useful. Counts are a measure of prevalence and describe the magnitude or size of a problem. Comparisons are generally appropriate within or across modes. Prevalence is often a consideration (but not the only one) in allocating resources among different problem areas. Rates provide a different kind of information and imply a comparison based on the denominator selected, for example, vehicle miles traveled or per capita. The appropriate use of rates rests on the validity of the denominator for the comparison of interest. Driver age is an example for which rates have been of primary interest. But rates require exposure data, and exposure data are often difficult and expensive to collect. Some estimates have large errors. Exposure data are the topic of the final issue.

The third important issue is data quality. This discussion is taken entirely from remarks made at the workshop by Lindsay Griffin of the Texas Transportation Institute. His judgment is that the fatal crash data that are available for developing highway safety indicators are fairly reliable, but crash and injury data are much more suspect.

The reporting threshold for traffic crashes, which defines when a crash is severe enough to be officially reported by the states, can and does change without warning. In Texas, for example, traffic crashes stood at 414,614 in

1994; in 1996 the figure was 298,143—28 percent fewer. This 28 percent reduction in reported crashes was the result of an increase in reporting threshold. After July 1, 1995, reported crashes in Texas were defined to include only those crashes in which someone is injured and/or one or more vehicles are towed from the scene (the same definition that is used in Pennsylvania).

Police-reported injury and injury severity are also of questionable reliability—and may be changing over time. In Table 2-1, five years of Texas data are presented (1975, 1980, 1985, 1990, and 1995). For each year, the number of crashes and injuries recorded in Texas is depicted.

Clearly, less serious (C-level) injuries have come to constitute a larger percentage of motor vehicle injuries in Texas in recent years. Some of this increase may result from displacements out of the more serious injury categories, but some of this increase may result from an informal lowering of the threshold for C-level injuries.

While it is fairly easy to define a motor vehicle fatality (an injured party who succumbs to his or her injuries within 30 days of the crash), it is much more difficult to define traffic crashes and injuries. These definitions

vary not only from state to state, but also across years within the same state. Despite differing definitions and the challenges of achieving consistent definitions and their consistent application, it can nonetheless be useful to track time-series data within a single jurisdiction.

The final issue identified by the subgroup is exposure. This entire discussion is taken from remarks made at the workshop by John Miller of the Virginia Transportation Research Council. Exposure is a critical element to address in order to make safety indicators broadly understood beyond the transportation community. It appears that direct measures of harm—crashes, property damage, fatalities, injuries, injury severity, environmental consequences, and even stress on the users of the facility—are readily understood. Within the transportation community, measures of exposure—vehicle miles traveled (VMT) or millions of entering vehicles (which are vehicles that enter an intersection, merge onto a roadway, or enter a traffic circle)—have facilitated comparisons of different types of highway facilities: for example, the often-repeated comparison between interstate highways and two-lane highways, in which the rate of crashes per VMT is much lower for the former than the latter. Outside the transportation community, however, the average citizen may not have a ready handle on typical measures of exposure. Certainly one can understand the definition of VMT, but it may not necessarily be apparent how to relate an average annual number of miles put on a vehicle (e.g., 15,000 miles per year) to an average trip rate (e.g., 3.2 crashes per million VMT). In short, it seems that we need to relate measures of harm to measures of exposure, or daily events, with which users are familiar. There are several possibilities to consider:

• Risk of harm per person per year, which as an aggregate measure could use crashes per VMT in conjunction with numbers of VMT traveled. The value of such a measure is that it would facilitate comparisons for different geographical areas (in the short term). In the long term, it might be helpful for comparing subareas with different trip-making characteristics (e.g., people living in the central business district versus persons living in the suburbs). Recently, an editorial in a local newspaper argued that the risk of harm from crime (for people living in the central business district) was lower than the risk of harm from crashes (for people living in the suburbs).

• Risk of harm by activity. Comparison of an hour of driving, an hour of recreational boating, an hour of hiking outdoors, and an hour of walking in an urban area may be an indicator worth considering. The disadvantage of such an indicator is that it does not take into account the number of hours the activity is done annually: for example, one may spend 500 hours annually driving yet only 30 hours annually boating.

• Risk of harm after removing deliberate and egregious user errors. The Bureau of Transportation Statistics (BTS) reports that in bicycle crashes the bicyclist alone was at fault half the time, whereas in pedestrian crashes the pedestrian was at fault about a third of the time (for pedestrians 15 years and older). The extent to which the user is at fault when a crash occurs (as opposed to the extent to which the other party is at fault) gives additional information for understanding the relative safety of these two modes, especially if appropriate exposure information, such as risk of a fatal crash per year or per trip, can be included. A complicating factor is the definition of what constitutes an “egregious” error: an intoxicated bicyclist riding against traffic or a small child stepping into the street are both at fault, but one may reflect a more conscious disregard for safety than the other.

If a measure of exposure is selected to be used with an indicator, it still will not eliminate the need for the raw data. For example, if one has 10 crashes per 12,000 VMT for road A and 100 crashes per 120,000 VMT for road B, then, although the rates are identical, some will argue that road B merits attention because an improvement therein will affect more people than would be the case for road A. Thus, as a complement to presenting crashes as a measure of rate of travel, we may want to consider an indicator that reflects the number of people affected. The activity indicator suggested above is a step in this direction: the practical value would be that when deciding where to allocate resources, one can consider how many people will be affected, in addition to the relative risk by travel amount or activity type.

The group suggested seven indicators that should not replace but could be collected in addition to those currently collected. One new safety indicator could be “risk per basket of trips.” Several workshop participants discussed the concept of developing an identifiable set of trip types that would reflect travel on a per capita basis for a metropolitan area, using the

term a “basket of trips.” This trip set would be based on trip types comparable to those used in urban travel demand models for a region. Although travel demand models are constantly under revision, the current state of practice for most urban areas is still based on three trip types: home-based work, which are trips from home to work or vice versa; home-based other, which are any other kind of trips with one end at home; and nonhome based, which are trips that do not come from or go to home. For example, a basket of trips for a region could be three home-based work trips, five home-based other trips, and four nonhome-based trips over a 24-hour period. The advantage of this indicator is that it facilitates comparison between regions, where even if crash rates are identical (e.g., number of crashes per VMT), one can understand how changes in land use and the use of transportation modes affect crash risk because the number of VMT is thus affected. The disadvantage is that its unit of exposure is based on the trip type, which is known to be imperfect—but VMT is not perfect, either. Thus, three subindicators are

• number of injuries per basket of trips,

• number of fatalities per basket of trips, and

• number of noninjury crashes per basket of trips.

Looking ahead, given the initiatives with voluntary electronic vehicle monitoring information in which individuals agree to have vehicle performance data monitored for research purposes, it may be possible to derive risk per trip type for specific regions, such as risk per home-based work trip. Right now, though, that does not seem practical.

A highly correlated indicator to risk per basket of trips is risk per household. The advantage of this indicator is that it presents regional comparison data in a form that may be easier to understand. For example, not everyone may relate to number of injuries per “set of weekday trips in the Washington, D.C., area,” but some will immediately recognize the meaning of the number of injuries per “household in the Washington, D.C., area.” Accordingly, three subindicators are

• number of injury crashes per household,

• (if resources permit) number of noninjury crashes per household, and

• (if resources permit) number of fatal crashes per household.

Another participant noted that, in the medical field, an obvious goal of emergency response is to minimize the harm that occurs following an event. A relevant indicator is thus the probability of harm once such an event has occurred. The advantage of such an indicator is that it reflects the performance of vehicle design, medical response, and roadway design (e.g., guardrail installation) in terms of mitigating the adverse impacts of a crash. Accordingly, one subindicator is:

• probability of a serious injury or fatality given that a collision occurred.

If this indicator is divided into two indicators (one for fatality and one for serious injury), it will be necessary to consider the time definition of a fatality. Unfortunately, in some cases, the linkage between emergency department data and hospital data is not always firm, thus making it difficult to know whether a serious injury at the crash scene did or did not result in a fatality.

Resources are limited. The subgroup suggested using the following criteria when selecting indicators:

• The indicators should be understood by nonspecialists. Not everyone knows how much they drive in a year, making risk per VMT somewhat harder to intuitively understand, but risk per VMT per type of household (such as an average suburban household, an average inner-city large urban household, or a small-city household), for example, may be easier to grasp.

• The indicators should have a meaningful range between low values and high values. Injury crashes occur often enough that different values in that indicator can be presented for different regions in a meaningful way. Fatalities, in contrast, are consistently quite low, which is good news but may render an indicator such as “fatalities per household” less meaningful.

• The meaning of the indicators should be clear, even when computational methods are not. It was pointed out during the workshop that not everyone understands how the Consumer Price Index is determined, but many persons in different disciplines still have a grasp for what a rise in it

means. Similarly, the number of injuries per basket of trips can be meaningful, even if a person is not a transportation demand modeler per se.

• Indicators should be developed recognizing that not everyone has the time to do the computations themselves. Theoretically a person with an interest can place the number of injuries in the numerator and the number of households in the denominator and compute the number of injuries per household, but many groups that want a quick grasp of the data may not have time to conduct a detailed study.

• If possible, the indicators should use exposure measures from other disciplines. For example, other disciplines do not use “risk per vehicle mile traveled,” but they do use “risk per capita,” which makes some of the indicators shown above more attractive. Of course, not all indicators have to meet this criterion; obviously the transportation community should continue to use risk per VMT for many situations. For a wider audience, there may be other indicators in addition to this one that provide a better picture of risk.

In some situations, error bars or confidence intervals can be included in a graphic for the reader to understand whether changes in indicators are significant or not. For example, regarding the seasonal variations in general aviation travel, bars giving a range of expected values (thus enabling one visually to inspect whether the difference between January 1999 and January 2000 fatalities is meaningful) would be appropriate, once such fatalities had been normalized with the appropriate exposure data.

For property damage only (PDO) crashes, BTS could usefully store the estimated dollar value of damage, in order to have a consistent set of PDO crash data, even if states have different reporting requirements. BTS relies on states and other organizations for data collection, and of course reporting requirements vary from state to state and change over time— both of these circumstances are currently beyond BTS’s control. For future reference, it may be fruitful to identify reporting requirements that can reliably be expected to change and to plan for these changes. For example, in terms of motor vehicle crashes, states that use a dollar threshold for noninjury crashes (e.g., in Virginia, noninjury crashes below $1,000 damage are not reported) can be expected periodically to adjust their thresholds

over time. In such an instance, by including a field for the dollar amount of property damage in BTS’s internal database, one can estimate how changes in dollar thresholds will affect the number of crashes reported after considering inflation. Thus, when a set of states change their dollar thresholds, BTS will be in a position to know whether a change in the number of PDO crashes results simply from the threshold reporting change or from some other factor.