U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations
|This report is an archived publication and may contain dated technical, contact, and link information|
Publication Number: FHWA-HRT-04-142
Date: December 2005
Enhanced Night Visibility Series, Volume XI: Phase II—Cost-Benefit Analysis
PDF Version (443 KB)
PDF files can be viewed with the Acrobat® Reader®
CHAPTER 4—BENEFIT ESTIMATION
The field tests conducted on the Smart Road measured the effect of 12 different VESs on drivers’ ability to detect pavement markings and drivers’ ability to detect and recognize a given object. Four studies took place in four different meteorological conditions: clear, rain, snow, and fog. Thirty drivers, ten in each of three age groups—18 to 25 years, 40 to 50 years, and 65 years and over—participated in each field test (except for snow, where the oldest age group was omitted out of concern for the risk of slips while moving between vehicles). The tests included nine different objects, including pedestrians, bicycles, and tires.
In the case of the pavement markings, the study was conducted only in clear weather conditions. Thirty drivers, in the same three age groups, participated. Each of three different marking materials (i.e., one nonfluorescent and two fluorescent) was placed on the Smart Road as a yellow centerline and a white edgeline. The detection distances of the beginning and end of each marking type were recorded.
The National Highway Traffic Safety Administration’s National Automotive Sampling System (NASS) and its General Estimates System (GES) supply historical estimates of the number of crashes each year.(14,15) The NASS database for a given year consists of a sample of all the crashes that were reported in the United States during that year.(14) The GES applies a multiplier weight to each crash in the sample to reconstruct an estimate of the total population of crashes.(15) The multiplier is based on the ratio between the total number of recorded crashes in the police jurisdiction where the crash occurred and the number of recorded crashes from that jurisdiction that are actually included in the NASS sample. The NASS database includes information such as the prevailing light and weather conditions, the severity of damage to each vehicle involved, the critical event that caused each vehicle’s involvement, the age of each driver, and the severity of injury to each person involved for every crash in its sample.(14) GES can be queried to produce an estimate that breaks down the total number of crashes into very fine categories according to the conditions prevailing at the time of the crash, the proximate cause of the crash, and the age of the driver.(15) Likewise, GES can be queried to produce estimates of the number of personal injuries that occurred in each of these crash categories, and the number of vehicles that were damaged in each of these crash categories.
The NASS database classifies personal injuries according to the KABCO scale: “K” designates a fatal injury, “A” an incapacitating injury, “B” an evident injury, “C” a possible injury, and “O” no injury (property damage only).(14) FHWA Technical Advisory T7570.2 gives an estimate of the average cost of an injury of each degree of severity in the KABCO scale in 1994 dollars.(16) To project the average cost in any subsequent year, the advisory prescribes that the 1994 cost be divided by the Gross Domestic Product (GDP) implicit price deflator for 1994 and multiplied by the implicit price deflator for the subsequent year. This conversion, using the implicit price deflators for 1994 and 2003, is used to obtain an estimate of the average cost of each personal injury of each type in 2003 dollars.
The NASS database classifies vehicle damage into four categories: no damage, minor damage, functional/moderate damage, or disabling/severe damage.(14) The estimate of the average cost of a damaged vehicle in each category of severity is based on the average cost of a property-damage-only crash(16)—that is, category “O” in the NASS database.(14) Minor damage is assigned an average cost equal to one-half the average for all property damage only (PDO) crashes, moderate damage is assigned an average cost equal to the average for all PDO crashes, and severe damage is assigned an average cost equal to twice the average for all PDO crashes.(16)
It is postulated that the motoring public would realize the benefits of enhanced night visibility in the form of reduced crash costs. To forecast the potential benefits of a VES in a given future year under this postulate, it is necessary to forecast the level of crash costs that would be incurred if the system were not adopted and the level of crash costs that would be incurred if the system were adopted. The crash data are grouped with the assumptions (1) that only a crash that occurs at night on an unlighted road might be averted by the ultraviolet or infrared VESs, and (2) that only a crash that begins with a roadway departure that occurs at night on an unlighted road might be averted by the fluorescent pavement marking technology. One might conjecture alternatively that a crash that occurs at night or at dusk or dawn may be affected. This alternative assumption expands the definition and the number of crashes that a night vision system may affect, and thus, magnifies the system’s potential effect on crash costs.
The percentage of night crashes on unlighted roads that might be affected by a VES may be supposed to be equal to the percentage of vehicles in which the system is installed. The percentage of such crashes that may be affected by a fluorescent pavement marking system may likewise be supposed to be equal to the percentage of unlighted highway miles on which the system is installed. Because the number of unlighted highway miles was unavailable, the percentage of rural highway miles is tabulated as a proxy. If it has no other virtue, this proxy jibes conveniently with the computation of the cost of the pavement marking technology, which proceeds on the assumption that fluorescent markings would be installed only on rural roads.
This benefit analysis sorts the recorded crashes into multiple categories according to the values in each of three fields from the NASS database: light conditions, weather conditions, and critical event (the critical event that initiated the crash-not necessarily identical to what the crash literature would call the “cause” of the crash).(14) Each of these categories matches one of the variables controlled in the Smart Road tests. Sorting on light conditions is obviously relevant because the VESs were tested only at night on an unlighted road, and they are not expected to help drivers avoid crashes that occur by day or on a lighted road. Sorting on weather conditions is relevant because of the different results that were obtained in clear weather and in simulated conditions of rain, snow, and fog. The critical event that precipitated a vehicle’s involvement in a crash identifies the subsets of crashes where earlier detection of a pavement marking, pedestrian, cyclist, animal, or object might have enabled the driver to avoid the crash. This is of interest because of different results that were obtained when the volunteer drivers were asked to spot pavement markings, pedestrians, and so forth. It would also be possible to sort on driver age and gender to take account of the different results that were obtained from men and women in the three driver age groups.
Driver age and driver gender are also available in the set of data from the Smart Road tests and in the NASS database. Basic statistical tests holding age, light, and weather constant show no significant difference between the sight distances obtained from male and female drivers; therefore, the crash database is not sorted on driver gender. Drivers of different ages did exhibit different responses to the VESs (see ENV Volumes III through VI), but crash numbers and costs sorted by age of driver are not presented here.
Light and Weather Conditions
Incident-specific fields in the NASS Accident File identify the light and weather conditions at the time of each crash.(14) First, a crash is grouped with all other crashes that occurred under the same light and weather conditions. These two main categories create 15 groups.
The relevant sorting by light condition distinguishes three groups: crashes where the light condition field has the value “dark,” crashes where the field has either of the values “dawn” or “dusk,” and crashes where the field has any other value. This sorting is based on the assumption that night visibility enhancements can affect only the group of crashes that occur at night on unlighted roads, which is the light condition called “dark” in the GES database,(15) or the alternative assumption that the enhancements affect also the group of crashes for which the light field has the values “dawn” or “dusk.”
The NASS database classifies the weather prevailing at the time of each crash in one of eight categories: “clear,” “rain,” “sleet,” “snow,” “fog,” “rain with fog,” “sleet with fog,” and “other.”(14) Because the Smart Road tests simulated only four weather conditions—clear, rain, snow, and fog—it is necessary to guess at the performance of each VES in the conditions “sleet,” “rain with fog,” “sleet with fog,” and “other” from its performance in the simulated conditions. The sorting in the weather field supposes that a system’s performance under conditions of sleet, rain with fog, and sleet with fog equals its performance in rain. The sorting further supposes that a system’s performance under “other” conditions equals its performance in clear weather. Therefore, the sorting by weather condition distinguishes five groups: crashes where the weather field has either the value “clear” or “other,” crashes where the weather field has either the value “rain” or “sleet,” crashes where the weather field has the value “snow,” crashes where the field has the value “fog,” and crashes where the field has either the value “rain with fog” or “sleet with fog.”
Figure 7 breaks down the estimated number of crashes in each year from 1992 to 2001 into three categories: those that occurred in daylight or on lighted roads, those that occurred in the dark, and those that occurred at dawn or dusk.(14) Some 12 percent of crashes occurred in dark conditions where enhanced night visibility would be expected to have an effect, while another 3 percent or so occurred in dawn or dusk conditions where it might be expected to have an effect.
Figure 7. Bar graph. Number of crashes, 1992 through 2001, by light condition.(14)
Figure 8 breaks down the estimated number of crashes in each year from 1992 to 2001 into five categories: those that occurred in conditions identified as “clear” or “other,” those that occurred in rain or sleet, those that occurred in snow, those that occurred in fog, and those that occurred in rain or sleet with fog.(14)
Figure 8. Bar graph. Number of crashes, 1992 through 2001, by weather condition.(14)
A vehicle-specific field called “Critical Event” in the NASS Vehicle File identifies the triggering event that involved each vehicle in each recorded crash.(14) The critical events on which enhanced night visibility would have an effect were identified. Enhanced visibility of pavement markings is postulated to have an effect on crashes triggered by a lane or roadway departure (field values 012–014); this excludes roadway departures that occur secondarily to some other precipitating event such as a blowout or the approach of another car. Enhanced visibility of pedestrians, cyclists, and inanimate objects are postulated to have an effect on crashes triggered by vehicle interaction with a pedestrian (field values 080–082), with a cyclist (values 083–085), and with an animal or an inanimate object (values 087–092).
Typically, if the Vehicle File records the involvement of more than one vehicle in a single crash, some primary cause (including those of interest to this study) is listed as the critical event for one of the vehicles, and some variety of secondary interaction between vehicles is listed as the critical event for each of the others.(14) A multicar crash, a critical event involving a lane departure, or an interaction with a nonmotorist (i.e., one of the critical events of interest) could very likely lead to a subsequent critical event of some other type (e.g., an interaction between two vehicles), but it is also possible that a critical event involving a lane departure or an interaction with a nonmotorist could be caused by a critical event of some other type. The sorting here postulates that if one of the critical event values that are relevant to this study is attributed to any vehicle in a multicar crash, then that critical event is the initial event of the multicar crash, and it is the criterion by which that crash will be categorized. Accordingly, it is possible to group the crashes by the presumed first critical event: any crash that includes a roadway departure is grouped with all other crashes that include a roadway departure; any crash that includes an interaction with a pedestrian is grouped with all other crashes that include an interaction with a pedestrian, and likewise for interactions with a cyclist, an animal, or an object.
The many other values of the critical event field represent critical events that are related in no obvious way to the sight-distance field tests(14)—that is, critical events on which enhanced night visibility would probably have no effect. A single large group of crashes remains in which none of the above critical events is listed. The largest group of critical event values within this category represents various interactions between two or more vehicles. Other critical events in this category include mechanical failure and loss of traction. The VESs studied in this project would seem unlikely to affect the night visibility of a car traveling with its headlights on, but the reader may make his or her own suppositions about the crashes in this category.
A negligible number of crashes, only a few dozen in tens of thousands of recorded crashes, involved more than one of the critical events of interest.(14) In these very few cases, the crash costs (explained below) are divided between two critical event categories.
GES estimates of the number of persons injured in each of the five degrees of severity (on the KABCO scale) and the number of vehicles damaged in each of the four degrees of severity. Using the NASS data of the years 1999, 2000, and 2001,(14) three separate estimates were generated.
GES estimates of the total number of crashes were generated for each of the 10 years 1992 to 2001. The simple log linear equation shown in figure 3 fitted to this 10-year series suggests how the number of crashes may be expected to grow over time in the absence of ENV technology.
Figure 9 compares the GES-estimated number of crashes in each year with the number implied by the fitted equation. The fitted equation provides a forecast of the number of crashes that would occur in any future year if no new VESs were introduced.
The ratio between the average number of personal injuries in each of the separate crash categories from the year 1999 through 2001 and the total number of crashes in the year 2000, multiplied by the total number of crashes forecast for any future year and described in the preceding paragraph, affords a forecast of the number of personal injuries in each crash category that would occur in that future year (if no new VESs were introduced).
The procedure for forecasting the number of vehicles damaged in any crash category for a given future year is completely analogous to the procedure for forecasting the number of personal injuries.
Figure 10 breaks down the number of persons involved in crashes from 1999 to 2001 into five categories, according to the critical event deemed to be the primary cause of the crash involved, and into seven crosscategories, according to the severity of their injuries (if any).
Figure 10. Bar graph. Estimated number of people involved in crashes, 1999 through 2001, by critical event and severity of injury.(13)
Figure 11 breaks down only the number of persons identified as injured in crashes from 1999 to 2001 into the same five cause categories and into five injury categories.
Figure 11. Bar graph. Estimated number of people injured in crashes, 1999 through 2001, by critical event and severity of injury.(13)
Figure 12 breaks down the number of vehicles involved in crashes from 1999 to 2001 into the same five cause categories and five crosscategories according to the severity of the damage (if any) they sustained.
Figure 12. Bar graph. Estimated number of vehicles involved in crashes, 1999 through 2001, by critical event and severity of damage.(13)
Following the guidelines provided in the 1994 FHWA technical advisory mentioned above,(16) a dollar value is attributed to each personal injury and damaged vehicle in the GES estimates for 1999, 2000, and 2001.(13) For example, each personal injury of type “K” (fatal) is assigned the average value, in dollars, of such injuries expressed in year 2003. The result is a tally of crash costs that can be grouped according to the crash conditions identified with the variables in the Smart Road tests: light conditions (day or lighted; dark; or dusk/dawn), weather conditions (clear; rain/sleet; snow; or fog), and critical events (pedestrian interaction, cyclist interaction, animal or object interaction, roadway departure, or other) identified with the types of objects (pedestrian, cyclist, inanimate object, or pavement stripe) used in the tests.
Figure 13 breaks down the estimated cost of the crash casualties from 1999 to 2001 into five categories, according to the critical event in which each crash is recorded.
Figure 13. Bar graph. Estimated annual crash costs, 1999 through 2001, by critical event.(13)
Because the four critical event categories that represent interaction with creatures or objects that might or might not be seen at night account for 36 percent of the total costs, it is evident that a little over a third of crash costs arise from critical events in which enhanced visibility might be expected to have an effect.