Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
A Dissent on Safety Issues: Fuel Economy and Highway Safety David L. Greene and Maryann Keller The relationship between fuel economy and highway safety is complex, ambiguous, poorly understood, and not measurable by any known means at the present time. Im- proving fuel economy could be marginally harmful, benefi- cial, or have no impact on highway safety. The conclusions of the majority of the committee stated in Chapters 2 and 4 are overly simplistic and at least partially incorrect. We make a point of saying fuel economy and safety rather than weight or size and safety, because fuel economy is the subject at hand. While reducing vehicle weight, all else equal, is clearly one means to increasing fuel economy, so is reducing engine power, all else equal. To the extent that con- sumers value power and weight, manufacturers will be re- luctant to reduce either to improve mpg. Indeed, Chapter 3 of this report, which addresses the likely means for improv- ing passenger car and light-truck fuel economy, sees very little role for weight or horsepower reduction in comparison with technological improvements. However, we will spend most of this appendix discussing the relationships between vehicle weight and safety, because the more important tech- nological means to improving fuel economy appear to be neutral or beneficial to safety. In analyzing the relationships between weight and safety it is all too easy to fall into one of two logical fallacies. The first results from the very intuitive, thoroughly documented (e.g., Evans, 1991, chapter 4, and many others), and theo- retically predictable fact that in a collision between two ve- hicles of unequal weight, the occupants of the lighter vehicle are at greater risk. The fallacy lies in reasoning that, there- fore, reducing the mass of all vehicles will increase risks in collisions between vehicles. This is a fallacy because it is the relative weight of the vehicles rather than their absolute weight that, in theory, leads to the adverse risk consequences for the occupants of the lighter vehicle. In fact, there is some evidence that proportionately reducing the mass of all ve- hicles would have a beneficial safety effect in vehicle colli- sions (Kahane, 1997, tables 6-7 and 6-8; Joksch et al., 1998, p. ES-2. 117 The second fallacy arises from failing to adequately ac- count for confounding factors and consequently drawing conclusions from spurious correlations. In analyzing real crashes, it is generally very difficult to sort out "vehicle" effects from driver behavior and environmental conditions. Because the driver is generally a far more important deter- minant of crash occurrences than the vehicle and a signifi- cant factor in the outcomes, even small confounding errors can lead to seriously erroneous results. Evans (1991, pp.92- 93), for example, cites research indicating that the road user is identified as a major factor in 95 percent of traffic crashes in the United Kingdom and 94 percent in the United States. The road environment is identified as a major factor in 28 percent and 34 percent of U.K. and U.S. crashes, respec- tively, while the comparable numbers for the vehicle are 8 and 12 percent. Of the driver, environment, and vehicle, the vehicle is the least important factor in highway fatalities. Moreover, there are complex relations among these factors: Younger drivers tend to drive smaller cars, smaller cars are more common in urban areas, older drivers are more likely to be killed in crashes of the same severity, and so on. To isolate the effects of a less important factor from the effects of more important yet related factors is often not possible. In the case of vehicle weight and overall societal highway safety, it appears that there are not adequate measures of exposure with which to control for confounding factors so as to isolate the effects of weight alone. THE PROBLEM IS COMPLEX Part of the difficulty of estimating the true relationships between vehicle weight and highway safety is empirical: real- ity presents us with poorly designed experiments and incom- plete data. For example, driver age is linearly related to ve- h~cle weight (Joksch, personal communication, June 19, 2001), and vehicle weight, size, and engine power are all strongly correlated. This makes it difficult to disentangle driver effects from vehicle effects. As another example, pe- . . . . . . ,
118 EFFECTIVENESS AND IMPACT OF CORPORATE AVERAGE FUEL ECONOMY (CAFE) STANDARDS destrian fatalities are most concentrated in dense urban areas, where smaller vehicles predominate. In Washington, D.C.,42 percent of traffic fatalities are pedestrians; in Wyoming only 3 percent are pedestrians (Evans, 1991, p. 4~. Failing to accu- rately account for where vehicles are driven could lead one to conclude that smaller vehicles are more likely to hit pedestri- ans than larger vehicles. Measures of vehicle exposure with which to control for confounding influences of drivers, envi- ronment, and other vehicle characteristics are almost always inadequate. Under such circumstances it is all too easy for confounding effects to result in biased inferences. Another part of the problem is the systematic nature of the relationships. To fully analyze the effect of weight on safety, one must consider its impacts on both the probability of a crash (crash involvement) and the consequences of a crash (crashworthiness or occupant protection). Crashes among all types of highway users must be considered not just crashes between passenger cars, or even all light-duty vehicles, but also crashes between light vehicles and heavy trucks, pedes- trians, and cyclists, as well as single-vehicle crashes. Only one study has attempted to fully address all of these factors. That is the seminal study done by C. Kahane of the National Highway Traffic Safety Administration (Kahane, 1997~. No other study includes pedestrian and cyclist fatali- ties. No other study also explicitly addresses crash involve- ment and occupant protection. Kahane's study stands alone as a comprehensive, scientific analysis of the vehicle weight and safety issue. It makes the most important contribution to our understanding of this issue that has been made to date. But even Kahane's study has important limitations. As the author himself noted, he was unable to statistically sepa- rate the effects of vehicle size from those of vehicle weight. This would have important implications if material substitu- tion becomes the predominant strategy for reducing vehicle weight, since material substitution allows weight to be re- duced without reducing the size of vehicles. Both the steel and aluminum industries have demonstrated how material substitution can produce much lighter vehicles without re- ducing vehicle dimensions (e.g., see, NRC, 2000, pp. 46- 49~. Not only prototype but also production vehicles have confirmed the industries' claims that weight reductions of 10 to 30 percent are achievable without reducing vehicle size. Kahane's analysis (1997) is thorough and careful. It de- tails at length the approximations and assumptions necessi- tated by data limitations. These have also been enumerated in two critical reviews of the work by the NRC (North, 1996) and industry consultants (Pendelton and Hocking, 1997~. We will not belabor them here. It is important, however, to re- peat the first finding and conclusion of the panel of eight experts who reviewed the Kahane (1997) study, because it is identical to our view of this issue. We quote the panel' s No. 1 finding and conclusion in full. 1. The NHTSA analysts' most recent estimates of vehicle weight-safety relationships address many of the deficiencies of earlier research. Large uncertainties in the estimates re- main, however, that make it impossible to use this analysis to predict with a reasonable degree of precision the societal risk of vehicle downsizing or downweighting. These uncer- tainties are elaborated below. (North, 1996, p. 4.) Despite these limitations, Kahane's analysis is far and away the most comprehensive and thorough analysis of this subject. We will return to it below for insights on several issues. THE LAWS OF PHYSICS There is no fundamental scientific reason why decreasing the mass of all highway vehicles must result in more injuries and fatalities. In debates about CAFE and safety, it has fre- quently been claimed that the laws of physics dictate that smaller, lighter vehicles must be less safe. This assertion is quite true from the perspective of a single private individual considering his or her own best interests and ignoring the interests of others, but it is false from a societal perspective. Therefore, the safety issues surrounding a general down- weighting or downsizing of highway vehicles are concerned with the details of how vehicle designs may change, differ- ences in the performance of lighter weight materials, the pre- cise distribution of changes in mass and size across the fleet, and interactions with other highway users. The One Point on Which Everyone Agrees There is no dispute, to the best of our knowledge, that if a collision between two vehicles of different mass occurs, the occupants of the heavier vehicle will generally fare better than the occupants of the lighter vehicle. The evidence on this point is massive and conclusive, in our opinion. This conclusion is founded in the physical laws governing the changes in velocity when two objects of differing mass col- lide. In a direct head-on collision, the changes in velocity (/\v) experienced by two objects of differing mass are in- versely proportional to the ratio of their masses (Joksch et al., 1998, p. 11), as shown in the following equations: Avl = (V1 + V2) m1 + m2 . /\V2 m1 (1) Because the human body is not designed to tolerate large, sudden changes in velocity, /\v, correlates extremely well with injuries and fatalities. Empirically, fatality risk in- creases with the fourth power of /\v (Joksch et al., 1998~. The implications are extreme. If vehicle 2 weighs twice as much as vehicle 1, the fatality risks to occupants of vehicle 1 will be approximately 24 = 16 times greater than those to the occupants of vehicle 2 in a head-on collision. Lighter ve- hicles will generally experience greater /\v's than heavier vehicles, and their occupants will suffer greater injuries as a
APPENDIX A result. Evans (1991, p. 95) has summarized this relationship in two laws: When a crash occurs, other factors being equal, 1. The lighter the vehicle, the less risk posed to other road users. 2. The heavier the vehicle, the less risk posed to its occu- pants. Evans' two laws make it clear that there are winners and losers in the mass equation. In free markets, this relationship causes a kind of market failure called an externality, which leads to oversized and overweighed vehicles. This market failure, combined with the aggressive designs of many heavier vehicles, is very likely a much more important soci- etal safety concern than improving fuel economy. It is well known that in collisions with sport utility vehicles, pick-up trucks, and vans, car drivers are at a serious safety disadvan- tage, not only because of the disparity in vehicle weights but because of the aggressivity of light-truck designs (Joksch, 2000). The simple relationship expressed by equation (1) tells us two important things. First, suppose that the masses of both vehicle 1 and vehicle 2 are reduced by 10 percent. This is equivalent to multiplying both m2 and me by 0.9. The result is a canceling of effects and no change in the /\v's. Thus, this simple application of the laws of physics would predict that a proportionate downweighting of all light vehicles would result in no increase in fatalities or injuries in two-car crashes. We emphasize this point because it is entirely con- sistent with the findings of Kahane's seminal study (1997) of the effects of downsizing and downweighting on traffic fatalities. Second, the distribution of vehicle weights is im- portant. Because the probability of fatalities increases at an increasing rate with /\v, a vehicle population with widely disparate weights is likely to be less safe than one with more uniform weight, at any overall average weight. IN COLLISIONS BETWEEN VEHICLES OF THE SAME WEIGHT, IS LIGHTER OR HEAVIER BETTER? Kahane's results (1997) suggest that in car-to-car or light truck-to-light truck collisions, if both vehicles are lighter, fatalities are reduced. The signs of the two coefficients quan- tifying these effects are consistent for the two vehicle types, but neither is statistically significant. Focusing on the crash- worthiness and aggressivity of passenger cars and light trucks in collisions with each other, Joksch et al. (1998) stud- ied fatal accidents from 1991 to 1994 and found stronger confirmation for the concept that more weight was, in fact, harmful to safety. In their analysis of the effects of weight and size in pas- senger car and light-truck collisions, Joksch et al. (1998) paid special attention to controlling for the age of the occu- pants and recognizing nonlinear relationships between key variables. Their analysis led them to two potentially very 119 important conclusions: (1) increased weight of all cars was not necessarily a good thing for overall safety and (2) greater variability of weights in the vehicle fleet was harmful. Among cars, weight is the critical factor. Heavier cars im- pose a higher fatality risk on the drivers of other cars than lighter cars. A complement to this effect is that the driver fatality risk in the heavier car is lower. However, the reduc- tion in the fatality risk for the driver of the heavier car is less than the increase of the fatality risk for the driver of the lighter car. Thus, the variation of weight among cars results in a net increase of fatalities in collisions. (Joksch et al., 1998, p. 62.) Studies like those of Kahane (1997) and Joksch et al. (1998) that take greater pains to account for confounding factors appear to be less likely to find that reducing weight is detrimental to highway safety in vehicle-to-vehicle crashes than studies that make little or no attempt to control for con- founding factors. This suggests to us that confounding fac- tors are present and capable of changing the direction of a study's conclusions. FINALLY, ALL THINGS CONSIDERED (KAHANE) The most comprehensive assessment of the impacts of vehicle weight and size on traffic safety was undertaken by the National Highway Traffic Safety Administration, partly in response to a request from the 1992 NRC study Automo- tive Fuel Economy. The NRC study pointed out that a soci- etal perspective that included all types of crashes and all highway users was needed, and that crash involvement as well as crashworthiness needed to be considered. The 1992 NRC study also noted the possibility that the downweighting of vehicles could increase or decrease fatalities, depending on the resulting weight distribution. Kahane's study (1997) attempted to address all of these issues. Because of its thor- oughness, technical merit, and comprehensiveness, it stands as the most substantial contribution to this issue to date. Based on traffic fatality records for model years 1985 to 1993, Kahane (1997) estimated the change in fatalities at- tributable to 100-lb reductions in the average weight of pas- senger cars and light trucks. The author carefully and prominently notes that the data and model did not allow a distinction between weight and other size parameters such as track width or wheelbase. This implies that the 100-lb downweighting includes the effects of whatever down- sizing is correlated with it in the fleet under study. The report also meticulously documents a number of data prob- lems and limitations, and the procedures used to circum- vent them. While these problems have important implica- tions for interpreting the study's results (more will be said on this subject below), in our opinion, the seriousness and professionalism with which Kahane tried to address them cannot be questioned. Perhaps the most interesting implication of Kahane's
120 EFFECTIVENESS AND IMPACT OF CORPORATE AVERAGE FUEL ECONOMY (CAFE) STANDARDS TABLE A-1 Estimated Effects of a 10 Percent Reduction in the Weights of Passenger Cars and Light Trucks Cars Light Trucks Type of Fatalities in Effect Changein Fatalities in Effect Changein Crash 1993 Crashes (%) Fatalities 1993 Crashes (%) Fatalities Single-vehicle Rollover 1,754 4.58 272 1,860 0.81 67 Object 7,456 1.12 283 3,263 1.44 208 Subtotal 555 275 Crashes with others Pedestrian 4,206 -0.46 -66 2,217 -2.03 -199 Big truck 2,648 1.40 126 1,111 2.63 129 Car 5,025 -0.62 -105 5,751 -1.39 -354 Light truck 5,751 2.63 512 1,110 -0.54 -27 Subtotal 467 - 51 Subtotal single-vehicle crashes: 555 + 275 = 830 Subtotal crashes with others: 467 - 451 = 16 Total 846 + 147 NOTE: Weight reductions of 10 percent in MY 2000 vehicles are assumed to be 338.6 lb (0.1 x 3,386) for passenger cars and 443.2 lb (0.1 x 4,432) for light trucks. SOURCE: Based on Kahane (1997), tables 6-7 and 6-8. (1997) study has received little attention. Following the logic of the simple laws of physics of equation (1), one would predict that if the weight of all light-duty vehicles were re- duced by an equal proportion, there would be no change in fatalities in crashes among these vehicles. Calculating a 10 percent change in weight for model year 2000 passenger cars and light trucks, and applying Kahane's estimates of the per- cent changes in fatalities per 100 lb of weight, one sees that Kahane's model also predicts little or no change in fatalities. The calculations are shown in Table A-1. Adding the change in fatalities in car-to-car, car-to-light truck, light truck-to- light truck, and light truck-to-car collisions produces a net change of +26, a result that is not close to being statistically different from zero.) Of course, the simple laws of physics say nothing about crash avoidance, which Kahane's study partially addresses. Nonetheless, these results provide em- pirical evidence that, from a societal perspective, an appeal cannot be made to the laws of physics as a rational for the beneficial effects of weight in highway crashes. Of course, if cars are downweighted and downsized more than light trucks, the increased disparity in weights would increase fa- talities. Conversely, if trucks are downsized and down- weighted more than cars, the greater uniformity would re- duce fatalities. These results are also entirely consistent with the conclusions of Joksch et al. (1998~. Other studies have predicted substantial increases in fatali- iThe authors confirmed with Dr. Kahane that the calculations shown in Table A-1 were consistent with the proper interpretation of his model. ties for just such vehicle-to-vehicle collisions (e.g., Partyka, 1989; Lund et al., 2000~. But these studies make much more modest attempts to correct for confounding factors. The Kahane (1997) and Joksch et al. (1998) studies suggest that the more thoroughly and carefully one controls for confound- ing effects, the weaker the apparent relationships between ve- hicle weight and highway fatalities become. This is evidence, albeit inconclusive, that adequately correcting for confound- ing effects might reduce or even eliminate the correlations between weight and overall highway safety. Kahane's study also found that downweighting and downsizing cars and light trucks would benefit smaller, lighter highway users (pedestrians and cyclists). But the ben- efits to pedestrians are approximately canceled by the harm- ful effects to light-duty-vehicle occupants in collisions with larger, heavier highway vehicles (i.e., trucks and buses). Kahane's finding that downweighting and downsizing are beneficial to pedestrians and cyclists is important because no other study includes impacts on pedestrians, a fact that biases the other studies toward finding negative safety im- pacts. An important result is that in downsizing and down- weighting light-duty vehicles, there will be winners and losers. Including pedestrians, cyclists, and heavy truck colli- sions leads to an even smaller net change for collisions among all highway users of +16, again not close to being statistically different from zero. The bottom line is that if the weights of passenger cars and light trucks are reduced proportionally, Kahane's study predicts that the net effect on highway fatalities in collisions
APPENDIX A among all highway users is approximately zero. Given the history of the debate on this subject, this is a startling result. The story for single-vehicle accidents, however, is not good. Kahane's model predicts that fatalities in rollovers would increase by over 300 and fatalities in fixed object crashes by almost 500, for a total of 800 more annual fatali- ties. These numbers are statistically significant, according to the model. Thus, the predicted increase in fatalities due to downweighting and downsizing comes entirely in single- vehicle accidents. This is puzzling because there appears to be no fundamental principle that underlies it. Rollover pro- pensity and crashworthiness in collisions with fixed objects should, with the exception of crashes with breakable or de- formable objects, be a matter of vehicle design rather than mass. This issue will be taken up next. WHY QUESTION THE RESULTS FOR SINGLE-VEHICLE ACCIDENTS? Kahane's results (1997) for single-vehicle accidents are suspect, though not necessarily wrong, because other objec- tive measures of rollover stability and crashworthiness in head-on collisions with fixed objects are not correlated with vehicle weight. An advantage of crash test data and engi- neering measurements is that they are controlled experiments that completely separate vehicle effects from driver and en- vironmental effects. A disadvantage is that they may over- simplify real-world conditions and may measure only part of what is critical to real-world performance. Nonetheless, the fact that such objective measures of vehicle crashworthiness and rollover potential do not correlate with vehicle weight is cause for skepticism. In Figures A-1 and A-2 we show the National Highway Traffic Safety Administration's five-star frontal crash test results for MY 2001 passenger cars plotted against vehicle weight (NHTSA, 2001~. What is clear to the naked eye is confirmed by regression analysis. There is no statistically significant relationship between mass and either driver-side or passenger-side crash test performance. A plausible explanation for this result may lie in the fact 6- 5 - 4 - o Q A 3 - 2 - 1 - 1000 2000 3000 4000 Curb Weight (Ibs.) 5000 6000 FIGURE A-1 NHTSA passenger-side crash ratings for MY 2001 passenger cars. 121 6 - ~ 5 - cn 4 o 3 - 2 - 1000 2000 3000 4000 Curb Weight (Ibs.) 5000 6000 FIGURE A-2 NHTSA driver-side crash ratings for MY 2001 pas- senger cars. that as mass is reduced, the amount of kinetic energy that a vehicle must absorb in a crash is proportionately reduced. Of course, the material available to absorb this energy must also be reduced and, other things equal, so would the distance over which the energy can be dissipated (the crush space). However, vehicle dimensions tend to decrease less than pro- portionately with vehicle mass. Wheelbase, for example, decreases with approximately the one-fourth power of mass. That is, a 10 percent decrease in mass is associated with roughly a 2.5 percent decrease in vehicle wheelbase. With mass decreasing much faster than the length of structure available to absorb kinetic energy, it may be possible to maintain fixed-object crash performance as mass is reduced. The NHTSA crash test data suggest that this has, in fact, been done. So we are left with the question, If lighter vehicles fare as well in fixed-barrier crash tests as heavier vehicles, why should Kahane's results (1997) indicate this as one of the two key sources of increased fatalities? There are several possibilities. First, despite Kahane's best efforts, confound- ing of driver, environment, and vehicle factors is very likely. Second, the crash tests could be an inadequate reflection of real-world, single-vehicle crash performance. Third, the dif- ference could, in part, be due to the ability of vehicles with greater mass to break away or deform objects. The issue of collisions with breakable objects was inves- tigated by Partyka (1995), who found that there was indeed a relationship between mass and the likelihood of damaging a tree or pole in a single-vehicle crash. Partyka (1995) exam- ined 7,452 vehicle-to-object crashes in the National Acci- dent Sampling System. Light-duty vehicles were grouped by 500-lb increments and the relationships between weight and the probability of damaging a tree or pole estimated for the 3,852 records in which a tree or pole was contacted. Partyka concluded, It appears that about half of vehicle-to-object crashes in- volved trees and poles, and about a third of these trees or poles were damaged by the impact. Damage to the tree or
122 EFFECTIVENESS AND IMPACT OF CORPORATE AVERAGE FUEL ECONOMY (CAFE) STANDARDS pole appears more likely for heavier vehicles than for lighter vehicles in front impacts, but not in side impacts. When front and side impacts are combined, the result is an uneven relationship between mass and the probability of dam- aging a tree or pole, but one which generally indicates increas- ing probability of damage to the object with increasing ve- hicle mass. Figure A-3 shows the percent of time a tree or pole will be damaged by collision with passenger cars given a col- lision with a fixed object. Frontal and side impacts have been combined based on their relative frequency (Partyka, 1995, table 2~. Roughly, the data suggest that the chances of break- ing away an object may increase by 5 percent over a greater than 2,000-lb change in weight, an increase in breakaway probability of 0.25 percent per 100 lb. If one assumes that a life would be saved every time a pole or a tree was damaged owing to a marginal increase in vehicle weight (in what other- wise would have been a fatal accident), then the breakaway effect could account for about 100 fatalities per year per 10 percent decrease in light-duty vehicle weight. This is about 1 percent of annual fatalities in single-vehicle crashes with fixed objects but still represents a large number of fatalities and a potentially important concern for downweighted vehicles. While the assumption that one life would always be saved if a tree or pole broke away is probably extreme, on the other hand objects other than trees and poles can be moved or deformed. But Partyka's study (1995) is also incomplete in that it does not address crash avoidance. To the extent that smaller and lighter vehicles may be better able to avoid fixed objects or postpone collision until their speed is reduced, there could conceivably be crash avoidance benefits to offset the reduced ability to break away or deform fixed objects. The net result is not known. ARE LIGHTER CARS MORE LIKELY TO ROLL OVER? The other large source of single-vehicle fatalities based on Kahane's analysis (1997) is rollover crashes. Others have 25.0% - c' ~ 20.0%- 8 . _ 15.0%- i A' 10.0%- in c' 5.0% 0.0% / 1500 2000 2500 3000 3500 4000 4500 Vehicle Weight Class FIGURE A-3 Estimated frequency of damage to a tree or pole given a single-vehicle crash with a fixed object. also found that rollover propensity is empirically related to vehicle mass (e.g., GAO, 1994; Farmer and Lund, 2000~. It is tempting to attribute these empirical results to an inherent stability conferred by mass. But there is good reason to doubt such an inference. The stability of vehicles depends on their dimensions, especially track width, and the height of their center of grav- ity. If downweighting and downsizing imply a reduction in track width or a raising of a vehicle' s center of gravity, the result would be greater instability. Data provided by NHTSA on its measurements of the Static Stability Factor (SSF)2 of MY 2001 passenger cars and light trucks indicates that there is no relationship between SSF and vehicle weight within the car and truck classes (Figure A-4) (data supplied by G.J. Soodoo, Vehicle Dynamics Division, NHTSA,2001~. How- ever, combining passenger cars and light trucks, one sees stability decreasing as vehicle weight increases. This is en- tirely due to the lesser stability of light trucks as a class. One clear inference is that a vehicle's rollover stability based solely on the SSF is a matter of design and not inher- ent in its weight. Certainly, real-world performance may be far more complicated than can be captured even by theoreti- cally valid and empirically verified measures of stability. On the other hand, the difficulty of sorting out confounding in- fluences may also be biasing the results of statistical analy- ses based on real-world crashes. What raises doubts is the fact that a theoretically valid measure of vehicle stability shows no relation (or a negative relation if trucks are in- cluded) to vehicle weight. Given this, it is reasonable to sur- mise that some other, uncontrolled factors may account for the apparent correlation between vehicle weight and rollover fatalities. THE BIG PICTURE (TIME SERIES DATA) From a cursory examination of overall trends in fatality rates and light-duty vehicle fuel economy, it appears that the two move in opposite directions: fatality rates have been going down; fuel economy has been going up (Figure Am. But the trend of declining fatality rates antedates fuel economy standards and can be observed in nearly every country in the world. Can anything be learned by statistical analysis of aggregate national fatality and fuel economy trends? Probably not. The question is relevant because one of the earliest and most widely cited estimates of the effects of CAFE standards on traffic fatalities comes from a study by Crandall and Gra- ham (1989) in which they regressed highway fatality rates against the average weight of cars on the road and other vari- ables. The data covered the period 1947 to 1981. CAFE stan- dards were in effect for only the last 4 years of this period, but 2NHTSA's SSF is defined as a vehicle's average track width divided by twice the height of its center of gravity. It is measured with a driver in the vehicle.
APPENDIX A 1.6 - 1.4 - ~ <5; ~ `,.~. 1.2 - ~~ He 0.8 - 0.6 - 0.4 - 0.2 - . O- 2000 3000 4000 Pounds 5000 6000 FIGURE A-4 NHTSA static stability factor vs. total weight for MY 2001 vehicles. had been known for the last 6. Statistically significant effects of weight were found in 3 of 4 regressions presented, but other variables one might have expected to be statistically signifi- cant including income, fraction of drivers aged 15-25, con- sumption of alcohol per person of drinking age, and measures of speed were generally not significant. Time-series regressions including variables with clear time trends, such as the declining trend of highway fatalities, are notorious for producing spurious correlations. One tech- nique for removing such spurious correlations is to carry out the regressions on the first differences of the data. First differencing removes linear trends but retains the informa- tion produced when variables deviate from trendlines. A1- though it typically produces much lower correlation coeffi- cient (R2) values, it is generally regarded as producing more robust estimates. We regressed total U.S. highway fatalities directly against light-duty-vehicle fuel economy and several other variables using first differences and first differences of the logarithms of the variables. The data covered the period 1966 to 1999 (data and sources available from the authors on request). No 6.0 ~ 5.0 > ~ 4.0 o . _ . _ ~ 3.0 lo A ~ 2.0 .~ `~ 1.0 IL 0.0 1966 1976 1986 1996 T25.0 nn - 15.0 Cal - 10.0 - On _._ 1 1 1 1 1 - 0.0 FIGURE A-5 Traffic fatality rates and on-road light-duty miles per gallon, 1996-2000. 123 statistically significant relationship between the on-road fuel economy of passenger cars and light trucks and highway fa- talities was found in any of the many model formulations we tried. Other variables tested included real GDP, total vehicle miles of highway travel, total population, the price of motor fuel, the product of the shares of light-truck and car travel, and the years in which the 55-mph speed limit was in effect. Only GDP and the 55-mph speed limit were statistically sig- nificant. The speed dummy variable assumes a value of 1 in 1975, when the 55-mph speed limit was implemented, and drops thereafter to 0.5 in 1987, when it was lifted for rural interstates, and then to 0 in 1995 and all other years. This variable may also be picking up the effects of gasoline short- ages in 1974. Most often, miles per gallon appeared with a negative sign (suggesting that as fuel economy increases, fatalities decline), but always with a decidedly insignificant coefficient, as shown in typical results illustrated by the fol- lowing equation: loge(F)logoff ~) = - 0.0458 + 1.33[1Oge(GDP`)loge(GDP~ it] - 0.0875(D~ - Do ~) - 0. 112[1Ogetmpg`) - logetmpg~ id] Adj. R2 = 0.57 All variables except mpg are significant at the 0.01 level based on a two-tailed l-test. The P-value for mpg is 0.69, indicating that the odds of obtaining such a result if the true relationship is zero are better than two in three. The constant suggests that fatalities would decline at 4.6 percent per year if GDP (i.e., the size of the economy) were not growing. We present these results here only because they demon- strate that the aggregated national data covering the entire time in which fuel economy standards have been in effect and a decade before show not the slightest hint of a statisti- cally significant relationship between light-duty-vehicle fuel economy and highway traffic fatalities. The idea that a clear and robust relationship can be inferred from aggregated na- tional data is not supportable. SUMMARY The relationships between vehicle weight and safety are complex and not measurable with any reasonable degree of certainty at present. The relationship of fuel economy to safety is even more tenuous. But this does not mean there is no reason for concern. Significant fuel economy improve- ments will require major changes in vehicle design. Safety is always an issue whenever vehicles must be redesigned. In addition, the distribution of vehicle weights is an im- portant safety issue. Safety benefits should be possible if the weight distribution of light-duty vehicles could be made more uniform, and economic gains might result from even partly correcting the negative externality that encourages individuals to transfer safety risks to others by buying ever larger and heavier vehicles. Finally, it appears that in certain kinds of accidents, re- ducing weight will increase safety risk, while in others it
124 EFFECTIVENESS AND IMPACT OF CORPORATE AVERAGE FUEL ECONOMY (CAFE) STANDARDS may reduce it. Reducing the weights of light-duty vehicles will neither benefit nor harm all highway users; there will be winners and losers. All of these factors argue for caution in formulating policies, vigilance in testing vehicles and moni- tonng safety trends, and continued efforts to increase under- standing of highway safety issues. In conclusion, we again quote from the conclusions of the eight-member NRC panel convened to evaluate Kahane's analysis (1997) of the weight and safety issue. Nonetheless, the committee finds itself unable to endorse the quantitative conclusions in the reports about projected high- way fatalities and injures because of the large uncertainties associated with the results uncertainties related both to the estimates and to the choice of the analytical model used to make the estimates. Plausible arguments exist that the total predicted fatalities and injuries could be substantially less, or possibly greater, than those predicted in the report. Moreover, possible model misspecification increases the range of uncer- ta~nty around the estimates. Although confidence intervals could be estimated and sensitivity analyses conducted to pro- vide a better handle on the robustness of the results, the com- plexity of the procedures used in the analysis, the ad hoc ad- justments to overcome data limitations, and model-related uncertainties are likely to preclude a precise quantitative as- sessment of the range of uncertainty. (North, 1996, p. 7.) Although Kahane (1997) did estimate confidence inter- vals and did partially address some of the other issues raised by the NRC committee, it was not possible to overcome the inherent limitations of the data that real-world experience presented. The NRC committee' s fundamental observations remain as valid today as they were in 1996. REFERENCES AND BIBLIOGRAPHY Crandall, Robert W., and John D. Graham. (1989). "The Effect of Fuel Economy Standards on Automobile Safety." Journal of Law & Eco- nomics XXXII (April). Evans, Leonard. 1991. Traffic Safety and the Driver. New York, N.Y.: Van Nostrand Reinhold. Evans, Leonard. 1994. "Small Cars, Big Cars: What Is the Safety Differ- ence?" Chance 7 (3): 39-16. Farmer, Charles M., and Adrian K. Lund. 2000. Characteristics of Crashes Involving Motor Vehicle Rollover, September. Arlington, Va.: Insur- ance Institute for Highway Safety. GAO (General Accounting Office). 1991. Highway Safety: Have Automo- bile Weight Reductions Increased Highway Fatalities? GAO/PEMD- 92-1, October. Washington, D.C.: GAO. GAO. 1994. Highway Safety: Factors Affecting Involvement in Vehicle Crashes. GAO/PEMD-95-3, October. Washington, D.C.: GAO. GAO.2000. Automobile Fuel Economy: Potential Effects of Increasing the Corporate Average Fuel Economy Standards. GAO/RCED-00-194, August. Washington, D.C.: GAO. Joksch, Hans C. 1985. Small Car Accident Involvement Study. Draft final report prepared for DOT, October. Washington, D.C.: NHTSA. Joksch, Hans C. 2000. Vehicle Design Versus Aggressivity. DOT HS 809 194. April. Washington, D.C.: NHTSA. Joksch, Hans, Dawn Massie, and Robert Pichler. 1998. Vehicle Aggressi- vity: Fleet Characterization Using Traffic Collision Data. DOT HS 808 679, February. Washington, D.C.: NHTSA. Kahane, Charles J. 1997. Relationships Between Vehicle Size and Fatality Risk in Model Year 1985-93 Passenger Cars and Light Trucks. DOT HS 808 570, January. Washington, D.C.: NHTSA. Khazzoom, J. Daniel. (1994). "Fuel Efficiency and Automobile Safety: Single-Vehicle Highway Fatalities for Passenger Cars." The Energy Journal 15(4): 49-101. Lund, Adrian K., Brian O'Neill, Joseph M. Nolan, and Janella F. Chapline. 2000. Crash Compatibility Issue in Perspective. SAE Technical Paper Series, 2000-01-1378, SAE International, March. Warrendale, Pa.: SAE. NHTSA (National Highway Traffic Safety Administration). 1995. Impacts with Yielding Fixed Objects by Vehicle Weight, NHTSA Technical Report, DOT HS 808 574, June. Washington, D.C.: NHTSA. - NHTSA. 1996. Effect of Vehicle Weight on Crash-Level Driver Injury Rates, NHTSA Technical Report, DOT HS 808 571, December. Wash- ington, D.C.: NHTSA. NHTSA. 1997. Relationships between Vehicle Size and Fatality Risk in Model Year 1985-1993 Passenger Cars and Light Trucks. NHTSA Technical Report, DOT HS 808 570, January. Washington, D.C.: NHTSA. NHTSA. 2001. New Car Assessment Program. U.S. Department of Trans- portation (DOT). Available online at <www.nhtsa.dot.gov/NCAP>. North, D. Warner. 1996. National Research Council, Transportation Re- search Board, Letter to Ricardo Martinez, National Highway Traffic Safety Administration, Department of Transportation, June 12. NRC (National Research Council). 1992. Automotive Fuel Economy. Washington, D.C.: National Academy Press. NRC. 2000. Review of the Research Program of the Partnership for a New Generation of Vehicles. Sixth Report. Washington, D.C.: National Academy Press. Partyka, Susan C. 1989. Registration-based Fatality Rates by Car Size from 1978 through 1987, Papers on Car Size Safety and Trends, NHTSA Technical Report, DOT HS 807 444, June. Washington, D.C.: NHTSA. Partyka, Susan C. 1995. Impacts with Yielding Fixed Objects by Vehicle Weight, NHTSA Technical Report, DOT HS 808-574, June. Washing- ton, D.C.: NHTSA. Pendleton, Olga, and Ronald R. Hocking. 1997. A Review and Assessment of NHTSA's Vehicle Size and Weight Safety Studies, October. Ishpeming, Mich.: Pen-Hock Statistical Consultants.