From NHTSA, September 2006
The Current Approach
The current estimate is constructed from a combination of two studies, the 2004 Death Certificate Study’s collection of backover deaths from 1998, and the FARS backover deaths within the same time period. In order to produce a set of FARS cases, we must first establish a definition for a FARS backover crash, specifically eliminating cases that include driverless vehicles (crashes that involved a child interfering with the controls and crashes where a driver left their vehicle and was subsequently killed due to the car’s movement).
Our new definition for these crashes will be examined in greater detail here, in addition to being attached as an appendix. After loading and merging the Accident and Person files, the people in the file are divided into Drivers and Nonoccupants (Pedestrians, Bicyclists, Other Cyclists, Other Pedestrians, and Unknown Non-motorist) that died. The nonoccupants were then associated with the vehicle that struck them, by the variable N_MOT_NO. Next, all of the vehicles in the Vehicle File were merged with their corresponding drivers.
It is in this step that driverless vehicles are accounted for. If no specific person file of type “driver” is associated with a vehicle in FARS, this means that the driver was unknown or not present. This solution avoids cases of a driver exiting their vehicle and dying as a result of the vehicle moving backwards (due to technical or driver error). The official FARS coding for such a case states that the victim will be coded as a pedestrian after successfully exiting the vehicle after it has come to rest. Suppose, for instance, that in such a case, the victim was somehow coded as a driver. As such, the algorithm does not catalogue driver deaths, and the case is not counted. Next, suppose that the victim was coded as a non-occupant. This non-occupant would be related to the vehicle that struck them, but the vehicle would have no corresponding identified driver (there exists no such person in the Person file that is coded to be a driver as we have already assumed that by hypothesis. Additionally, the FARS coding manual specifically instructs the user to only code an individual as a driver if they were driving the vehicle, not merely if a child was sitting in the driver’s seat. A handful of young drivers caused backover fatalities according to our definition over 1995-2004, but the youngest was 14 years old, which is a legal driving age in some states.
After this collection of identified vehicles with identified drivers is collected, it is merged with the non-occupant fatalities. Specifically, these cases are examined for those that fit the characteristics of a backup crash. That is, among these non-occupant fatalities (caused by identified vehicles with identified drivers), we only will examine the cases where a number of factors coincide. The first harmful event must be an impact with a non-occupant, and the vehicle must be performing a backover maneuver. Here the vehicle maneuver variable VEH_MAN was used, but with special attention to parking and exiting parking. If the vehicle maneuver was related to parking, an additional check was made with the initial impact. Because the initial impact was with a nonoccupant, the initial impact variable will describe where that non-occupant was struck by the vehicle. Therefore, if the initial impact was a non-occupant collision from 4 to 8 o’clock on the vehicle, then that vehicle would be associated with a backover maneuver.
The description above, with corresponding SAS code serves as a definition of a “backover crash” with respect to FARS. So far, no other definition has been documented to our knowledge regarding these cases. Alternatives exist amongst the choices of variables, especially the “first” impact and “first” harmful event being replaced with “principle” impact and “most” harmful event. By using the variables that describe the first impact, cases where pedestrian impacts are a result of some other action are less likely to be included.
According to FARS, there were 83 backover deaths in 1998 across the nation, and 58 backover deaths within the set of the 35 states and the District of Columbia over which the Death Certificate Study was done. By matching “backover” cases based on age, sex, state, and date of death, 14 matches were found. That is, it is highly likely that FARS and the Death Certificate Study have 14 cases in common. So, if we sum the distinct cases across both studies, there exist (91-14) = 77 cases just in the Death Certificate Study, (58-14) = 44 cases just in FARS, and 14 cases in both, for a sum of (91-14) + (58-14) + 14 = 135 cases. The Death Certificate study used a linear projection based on the number of death certificates they were granted access to by their respective states, and the number of death certificates overall that fit a preliminary search of standard coding. We then use the same ratio used in the death certificates of (123/91) to expand the 135 cases to a national total of 183 deaths.1 Thus, according to our national estimate of 183 deaths, 69 were children under 5 and 76 were children under 15. Also, 76 percent of the fatalities were caused by passenger vehicles, 24 percent were caused by medium/heavy trucks or other vehicles.
We justify using the 1998 data for a current estimate after examining the FARS data across several years. From 1991 to 2004, there is no definite trend in the number of fatalities, and it seems to hover in a range of 60-80 cases a year (averaging 76 cases per year in 1991 to 2004), with a slight downward trend (averaging 74 cases per year in the last 10 years from 1995 to 2004). One would expect with increasing vehicle registrations and vehicle miles traveled, there might be an increase over time, but FARS data do not support such a hypothesis in recent years.
Because the current approach uses a combination of death certificate cases and FARS data, only one year of information (1998) is available. Despite this limitation, distributions over several years worth of FARS can provide insight to the problem. Table 2 examines the distribution of the age of the victim by the striking vehicle.
