Friday, January 12, 2001

Part XIII

Department of Transportation National Highway Traffic Safety Administration

49 CFR Part 575 Consumer Information Regulations; Federal Motor Vehicle Safety Standards; Rollover Resistance; Final Rule

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00001 Fmt 4717 Sfmt 4717 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3388 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

DEPARTMENT OF TRANSPORTATION reached by phone at (202) 366–5559 or note vehicles that are equipped with by facsimile at (202) 493–2739. For ‘‘electronic stability control’’ National Highway Traffic Safety public comments and other information technology, which may reduce the risk Administration related to previous notices on this of a vehicle getting into an incipient subject, please refer to: rollover situation. 49 CFR Part 575 DOT Docket No. NHTSA–2000–6859, The agency requested comments on Docket Management, Room PL–401, 400 its tentative decision to implement such [Docket No. NHTSA±2000±8298] Seventh Street, SW, Washington, D.C. a program on June 1, 2000.1 The closing Consumer Information Regulations; 20590 (hours 10:00 a.m. to 5:00 p.m. date for comments was August 30, 2000. Federal Motor Vehicle Safety Monday through Friday) or on the Twenty-five commenters responded. Standards; Rollover Resistance internet at www.dms.gov/search, and This notice addresses the major issues Docket No. 91–68; Notice 3, NHTSA presented by the commenters, our AGENCY: National Highway Traffic Docket, Room 5111, 400 Seventh Street, response to those comments, and the Safety Administration (NHTSA), DOT. SW, Washington, DC 20590. NHTSA procedures and protocol we will use to ACTION: Response to Comments, Notice Docket hours are from 9:30 am to 4:00 implement a rollover consumer of Final Decision. pm Monday through Friday. information program based on SSF. For SUPPLEMENTARY INFORMATION: complete background and rationale for SUMMARY: The agency has concluded I. Introduction the program, please see the June 1, 2000 that consumer information on the II. Background notice. III. Discussion of Commenters’ Issues rollover risk of passenger cars and light II. Background multipurpose passenger vehicles and A. SSF as a Measure of Rollover Risk B. NHTSA’s Statistical Analysis Linking trucks will reduce the number of Rollover crashes are complex events SSF to Rollover Rates that reflect the interaction of driver, rollover crashes and the number of C. Comments on Practical Problems with injuries and fatalities from rollover SSF Ratings road, vehicle, and environmental crashes. This information will enable D. Consumer’s Ability to Understand SSF factors. We can describe the relationship prospective purchasers to make choices as a Measure of Rollover Risk in the between these factors and the risk of about new vehicles based on differences Event of a Single-vehicle Crash rollover using information from the in rollover risk and serve as a market E. The Question of Electronic Stability agency’s crash data programs. We limit Control our discussion here to light vehicles, incentive to manufacturers in striving to F. Alternative Programs Suggested by design their vehicles with greater which consist of (1) passenger cars and Commenters (2) multipurpose passenger vehicles and rollover resistance. The consumer G. Commenters’ Desire for a Minimum information program will also inform Standard Based on a Dynamic Test trucks under 4,536 kilograms (10,000 drivers, especially those who choose IV. Rollover Information Dissemination using pounds) gross vehicle weight rating 2 vehicles with poorer rollover resistance, SSF in NCAP (collectively, ‘‘light trucks’’). that their risk of harm can be greatly V. Rulemaking Analyses and Notices According to the 1999 Fatality Analysis Reporting System (FARS), reduced with seat belt use to avoid Appendix I Statistical Analysis in Response to Comments 10,142 people were killed as occupants ejection. Appendix II Proposed List of Test Vehicles in light vehicle rollovers, including The agency has decided to use the for MY2001 8,345 killed in single-vehicle rollovers. Static Stability Factor to indicate Eighty percent of the people who died rollover risk in single-vehicle crashes I. Introduction in single-vehicle rollovers were not and to incorporate the new rating into This notice outlines the plan the using a seat belt, and 64 percent were NHTSA’s New Car Assessment Program National Highway Traffic Safety ejected from the vehicle (including 53 (NCAP). As part of these ratings, the Administration (NHTSA) will use to percent who were completely ejected). agency also has decided to note vehicles incorporate a new rollover rating of new FARS shows that 55 percent of light that are equipped with ‘‘electronic cars and light trucks into its existing vehicle occupant fatalities in single- stability control’’ technology, which New Car Assessment Program (NCAP). vehicle crashes involved rollover. The may reduce the risk of a vehicle getting NCAP currently gives consumers proportion differs greatly by vehicle into an incipient rollover situation. This crashworthiness ratings for new light type: 46 percent of passenger car notice summarizes the comments vehicles in frontal and side crashes. The occupant fatalities in single-vehicle received in response to the agency’s ratings are based on vehicle crashes involved rollover, compared to June 1, 2000 Request for Comment performance with respect to occupant 63 percent for pickup trucks, 60 percent regarding the addition of rollover ratings injury criteria gathered in crash tests for vans, and 78 percent for sport utility based on SSF to NCAP, our response to and are presented using one to five vehicles (SUVs). those comments, and the procedures stars, one star for the highest risk and Using data from the 1995–1999 and protocol we will use to implement five for the lowest. We intend to use the National Automotive Sampling System a new rollover consumer information same star rating system to present the (NASS) we estimate that 253,000 light program. risk of rollover in the event of a single- vehicles were towed from a rollover FOR FURTHER INFORMATION CONTACT: For vehicle crash. One star would represent crash each year (on average), and that the most up to date vehicle star ratings a Static Stability Factor (SSF) 27,000 occupants of these vehicles were call the Auto Safety Hotline at 888–327– corresponding to a 40 percent or greater seriously injured (defined as an 4236 or refer to NHTSA’s website at risk of a single-vehicle crash resulting in Abbreviated Injury Scale (AIS) rating of www.nhtsa.dot.gov. For technical rollover, while five stars would at least 3).3 This includes 205,000 questions you may contact Gayle represent an SSF corresponding to a risk Dalrymple, NPS–23, Office of Safety of less than 10 percent. Static Stability 1 65 FR 34999 (June 1, 2000). Performance Standards, National Factor is one-half the track width of a 2 Light trucks include vans, minivans, sport utility vehicles (SUVs), and pickup trucks under Highway Traffic Safety Administration, vehicle divided by the height of its 4,536 kilograms (10, 000 pounds) gross vehicle 400 Seventh Street, SW, Washington, center of gravity. As part of the rating weight rating. DC 20590. Ms. Dalrymple can be based on SSF, the agency also has to 3 A broken hip is an example of an AIS 3 injury.

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00002 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3389 single-vehicle tow-away rollovers with rollover (trial attorney, consumer basic parameters of [sic] influencing 19,000 serious injuries. Sixty-five groups). The commenters raised issues resistance.’’ percent of those people who suffered a in four areas: In 1973, all of the manufacturers serious injury in single-vehicle tow- The suitability of SSF as a measure of opposed NHTSA’s plans for a standard away rollovers were not using a seat rollover risk, regarding rollover prevention in extreme belt, and 50 percent were ejected • Whether NHTSA’s statistical accident avoidance maneuvers because (including 41 percent who were analysis linking SSF to single-vehicle of their expectation of negligible completely ejected). Estimates from rollover rates was correct, benefits, concern about banning vehicle NASS are that 81 percent of tow-away • Whether consumers are capable of types, degradation of vehicle rollovers occurred in single-vehicle understanding the concept of single- capabilities including braking traction crashes, and 87 percent (178,000) of the vehicle crash as exposure to rollover, and handling performance, and single-vehicle rollover crashes occurred and unresolved problems with maneuver • after the vehicle left the roadway. The need for a minimum standard, testing. General Motors presented a very Based on the 1995–1999 General or consumer information, for rollover detailed set of comments that remain Estimates System (GES) data we based on a dynamic test. relevant today. For example, its estimate that 241,000 light vehicles Alternative consumer information observations on the effect of restraint rolled over each year (on average) in programs for rollover prevention were use on rollover fatality rates and on the police-reported crashes, and that 57,000 also offered by some commenters. Those breakdown of the rollover problem occupants in rollover crashes received four issues and the alternative programs between multi-vehicle and single- injuries rated as K or A on the police are discussed in this section. vehicle crashes and on-road and off- injury scale. (The police KABCO scale A. SSF as a Measure of Rollover Risk road incidences are largely supported by calls these injuries ‘‘incapacitating,’’ but present data. Likewise, its discussion of their actual severity depends on local Many respondents to the RFC believe the problems of maintaining consistent practice. ‘‘Incapacitating’’ injury may that SSF is not a good measure of pavement surface and tire traction mean that the injury was visible to the rollover risk for various reasons. properties, the use of automatic controls reporting officer or that the officer Comments and the parties that made and outriggers, the types of maneuvers them were the following: called for medical assistance.) This • and their relationship to real crashes is includes 205,000 single-vehicle NHTSA has exaggerated the still meaningful. We also think its rollovers with 46,000 K or A injuries. importance of SSF in rollover crashes. comments regarding SSF (which it Fifty-four percent of those with K or A Vehicles have little to do with rollover; called geometric stability measurement) injury in single-vehicle rollovers were the driver and road conditions bear so are still accurate. General Motors said: not using a seat belt, and 20 percent much of the blame that the vehicles should not be rated for rollover.—The Resistance to rollover is mainly influenced were ejected from the vehicle (including by the following factors: 18 percent who were completely Alliance of Automobile Manufacturers 1. Height of the center of gravity. ejected). Estimates from GES are that 16 (Alliance), Association of Import 2. Horizontal distance from center of Automobile Manufacturers (AIAM) gravity to wheel track. percent of light vehicles in police- • reported single-vehicle crashes rolled Isuzu SSF is too simplistic. SSF 3. Capability for generating large forces in over. The estimated risk of rollover ignores tire properties, suspension the lateral direction of the tire contacts due differs by vehicle type: 13 percent of compliance, handling characteristics, to high tire friction. Lateral forces sufficient for rollover can cars and 14 percent of vans in police- antilock brakes, electronic stability control, vehicle shape and structure result from severe maneuvers under high tire- reported single-vehicle crashes rolled road friction conditions; from collisions with over, compared to 24 percent of pickup (post-impact rollover), and tripping other vehicles, curbs, or road furniture (signs, trucks and 32 percent of SUVs. factors (tires).—Alliance, University of lamp posts, guard rails), and from maneuvers The data presented above demonstrate Michigan Transportation Research in roadside soil capable of sustaining high that rollover crashes create a serious Institute, JCW Consulting, SiSan, lateral forces. safety problem and that a reduction in Automotive Testing Inc., Toyota, Isuzu, General Motors qualified the the number of rollovers can make a Honda discussion as pertaining to relatively significant contribution to motor vehicle 1. Origin of Static Stability Factor simple maneuvers, but cautioned safety. Static Stability Factor is not a measure against the use of ‘‘special’’ braking and III. Discussion of Commenters’ Issues of rollover resistance invented by the steering inputs for rollover maneuver tests as unrepresentative of vehicle The Request for Comment (RFC) was agency. It was introduced to the agency in 1973 by vehicle manufacturers as a operation. It also discussed the relative published June 1, 2000. The comment importance of secondary vehicle period closed August 30, 2000. Twenty- scientifically valid potential substitute for the dynamic maneuver tests the characteristics other than those above five commenters replied. The which are the components of SSF. respondents were vehicle manufacturers agency wanted to develop regarding and their associations, testing untripped on-road rollover.4 The Motor It was noted in a previous section that the laboratories, independent researchers, Vehicle Manufacturers Association dominant factors in flat road rollover (which has evolved into the present resistance are the center of gravity height, consumer safety groups, an insurance track width, and the ability of the tire-road association, a trial attorney, and two Alliance of Automobile Manufacturers) interface to generate high levels of lateral consumers. Two commenters agreed stated the following about SSF, force. Suspension geometry, component with the inclusion of rollover rating in ‘‘Although this method does not stiffness factors, allowable ride travel, and NCAP as it was presented in the RFC. embrace all vehicle factors relating to tire stiffness factors also exert a measurable The other commenters were divided rollover resistance, it does involve the influence on rollover performance. But, these among those who opposed the plan latter factors are considered to be of secondary importance. It should be noted (manufacturers, dealers, testing labs) 4 In 1973, NHTSA published an Advance Notice of Proposed Rulemaking on Rollover Prevention (38 that in many cases, very careful laboratory and those who thought it did not go far FR 9598, April 18, 1973). The comments cited here tests are required to establish the influence enough that a minimum standard, can be found in NHTSA Docket No. 73–10; Notice of suspension modifications on rollover based on a dynamic test, is needed for 1, comments 11 (MVMA) and 14 (GM). resistance.

VerDate 112000 23:26 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00003 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm03 PsN: 12JAR10 3390 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

In its conclusions, General Motors tests do not represent tripped rollover. incidence of rollovers occurring in maintained that there was no safety Once the vehicle is in a tripping single-vehicle crashes. The Alliance need for the on-road rollover resistance situation (e.g., has left the road), tire responded in August 2000 with the standard the agency intended to propose traction is largely irrelevant to tripped position that vehicle characteristics are and that, if the agency decided to act at rollover. Center of gravity height and now deemed largely irrelevant to the all, it should pursue consumer track width (and to a much lesser extent occurrence of rollover crashes and information based on SSF. roll moment of inertia) are the only consumer information on vehicle If any regulation is required, some benefit vehicle properties with general rollover resistance is inherently may be derived at minimal cost by better applicability to tripped rollover misleading. The Alliance provided a informing the customer of relative product situations. So, in 95 percent of rollovers, statistical study purporting to rollover performance, so he can assess this these vehicle properties would be the demonstrate that the influence of SSF vehicle performance factor in making his most relevant vehicle influences on the was limited to three to eight percent of selection in a free market. This information likelihood of rollover. In the five the variability between vehicles in could be based on geometric stability percent of the problem involving rollover crashes. measurements for the full range of highway While the laws of physics prove vehicles. untripped rollover, a choice exists between using static measurements and beyond question that vehicles with low This comment was made before the performance in maneuver tests. To get SSF roll over at lower lateral NCAP program was established to data to make an informed choice accelerations than vehicles with high provide consumer information on safety between the two, NHTSA conducted a SSF, the effect of SSF must be shown to performance and before the consumer maneuver test program using 12 have a significant influence on the was faced with such a large range of vehicles in 1998. That testing confirmed outcome of actual crashes (rollover vs. geometric stability (SSF) in non- General Motors’ opinion of 25 years no rollover) to be worth using for commercial passenger vehicles. Also, earlier that the static measurements consumer information. It is a fact that most of the practical difficulties in correspond well to dynamic maneuver types of vehicles with SSFs lower than seeking objective, relevant and tests.7 It also confirmed that the passenger cars, as a group, have greater repeatable driving maneuver tests problems with maneuver testing numbers of rollover crashes than discussed by General Motors in 1973 identified by GM in 1973 are still largely passenger cars, either as a percentage of remain unsolved. Note that GM unresolved today. Accordingly, we all crashes (passenger cars, 1.6 percent; suggested the static laboratory concluded in our June 2000 notice that vans, 2.0 percent; pickup trucks, 3.7 measurement as a substitute for there were no practical improvements in percent; SUVs, 5.1 percent) or as a maneuver tests when only on-road rating overall rollover resistance to be percentage of single-vehicle crashes untripped rollover was under gained at this time by using something (passenger cars, 13 percent; vans, 14 consideration. This is an even stronger other than static measurements. percent, pickup trucks, 24 percent; endorsement of static measurements SUVs, 32 percent). The Alliance than that represented by NHTSA’s 2. The Importance of the Effect of SSF attributes these differences primarily to reasons for using SSF for consumer on Rollover Rate differences in the driver and road information on all single-vehicle When the agency first sought public conditions associated with the various rollovers, tripped and untripped.5 comment on rollover issues in 1973, the vehicle types, rather than to the We view the rollover safety problem industry’s position was that the characteristics of the vehicles. For as 95 percent a problem of tripped frequency of untripped on-road example, if young males using alcohol rollover and five percent a problem of rollovers was too low to justify and driving on rural roads with high 6 on-road untripped rollover. Maneuver significant vehicle modifications and speed limits are over-represented as constraints on future vehicle design. drivers of four-wheel drive pickup 5 Untripped rollover is a rollover induced by tire trucks in crashes, could these road-use friction with the driving surface alone, resulting The vehicle manufacturers questioned from a driving maneuver and usually occurring on the benefit/cost relationship and variables outweigh the vehicle property the roadway. Tripped rollovers usually occur when practicability of a minimum standard on to the point of insignificance? a vehicle runs off the roadway and the tires and rollover resistance, but they did not According to the current industry view, wheels contact a tripping mechanism (curb, soft deny the relationship between SSF and the correlation between the SSF of a soil, pavement drop off) which causes the vehicle vehicle and its ability to attract risky to roll. A much smaller number of tripped rollovers rollover crashes. The agency’s June 2000 occur on the road as a result of the wheel rim plan for consumer information on drivers who operate vehicles under digging into the pavement during an extreme rollover resistance expressed adverse road conditions is the maneuver. Whether or not a vehicle rolls when it considerable agreement with the 1973 fundamental reason vehicles with low encounters a tripping mechanism is highly SSF are involved in a higher proportion dependent on the geometric properties represented industry position on rollover and by SSF. In an untripped rollover, SSF is still very offered a statistical study of modern of rollover crashes. important, but other factors come into play (such crash data in order to quantify the The agency agrees that driver as tire properties). Therefore, GM’s suggestion to relationship between SSF and the behavior and road conditions are use SSF to characterize a vehicle’s tendency for significant factors in understanding why untripped rollover was a very strong endorsement single-vehicle crashes of any type occur, of the relationship between SSF and vehicle Sampling System team did its own audit of the rollover. 1992–96 rollover data and concluded that some and that they have a strong influence on 6 In 1998, the agency was performing research on tripped rollovers were miscoded as untripped whether single-vehicle crashes result in driving maneuvers to see if we could develop a way rollovers (typically these were onroad rollovers in rollover. However, we think that the to ameliorate the incidence of onroad, untripped which the vehicle was sliding sideways and tripped rollover resistance of the vehicle rollover, which we estimated at the time to be less on its own wheel rim). Using corrected 1992–96 than 10 percent of rollover crashes. The American data, our National Center for Statistics and Analysis represented by SSF also exerts a strong Automobile Manufacturers Association (one of the estimated that 3.7 percent of rollovers are untripped influence on whether single-vehicle predecessors of the Alliance) contracted with and 3.5 percent are both untripped and onroad, crashes result in rollover. The statistical Calspan Corporation to review all the cases in while 4.4 percent of single-vehicle rollovers are study in our previous notice attempted NHTSA’s Crashworthiness Data System coded as untripped. (Research Note, ‘‘Passenger Vehicles in untripped to try to demonstrate that we were Untripped Rollovers,’’ September 1999.) to address the important question of misplacing our research funds on a very small 7 See the June 1, 2000 Request for Comments for whether road-use differences between problem. Consequently our National Automotive a summary of that research. vehicles relegate their difference in

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00004 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3391 rollover resistance to insignificance in make/model, and thus can consider statistic for SSF is greater than for any actual crash experiences. We analyzed interactions among variables. It is a of the other variables in the logistic state accident reports in six states popular statistical tool in the health model using individual variables. In the (1994–1997) on 184,726 single-vehicle sciences. The Alliance also introduced logistic model using scenario risk to crashes with 36,575 rollovers involving the concept of scenario risk in its combine all the variables except SSF, 100 vehicle make/models. The road-use logistic regression model. In this the ‘‘Chi-square’’ statistic for SSF is variables available in all six states technique, each combination of road-use greater than that of the scenario risk identified male drivers, young drivers, variables (with some states providing as variable in three of the eight states. This alcohol involvement, darkness, wet or many as 14 variables) is a scenario. result also contradicts the Alliance’s icy surface, speed limit 55 mph or Scenario risk becomes a continuous assertion that SSF is relegated to greater, storm, hill, and curve. We used variable. insignificance by the importance of multiple linear regression because its Appendix I of this notice presents a road-use variables on the rollover ‘‘R-squared statistic’’ provided an new statistical study which adds experience of vehicles in use. intuitive method of comparing the another year of state crash data to the The Alliance’s assertion that the effect explanatory power of individual database of our previous notice and of SSF on rollover is negligible was not variables and because we could control contrasts analyses of the crash data a consequence of the possible the effect of the large differences in the using logistic regression of individual superiority of logistic regression over number of crash samples for the various variables and risk scenarios to the linear linear, nor of the use of scenario risk vehicles. Each vehicle was represented regression method used in the previous rather than individual variables. by its SSF and the average of each road- notice. We found that it made very little Instead, the Alliance assertion depends use variable over the number of crashes difference to the logistic regression upon a subtle change in the definition in each state. Systematic differences models whether the road-use variables of the variables which serve as between states in rollover rate due to were used as individual variables or alternatives to SSF in explaining factors such as accident reporting combined to form risk scenarios, but rollovers. thresholds were accommodated by the that the curve estimating rollovers per NHTSA used the number of police- inclusion of a dummy variable for each single-vehicle crash produced by the reported single-vehicle crashes as a state. The ‘‘R-squared statistic’’ for the logistic regression was slightly different measure of each make/model’s exposure complete model was 0.88, indicating from that previously reported for linear to rollover risk. We did not include that the model explained 88 percent of regression. collisions with pedestrians or animals the observed differences in rollover rate The estimated risk of rollovers per in the roadway in our definition of per single-vehicle crash between the single-vehicle crash is six times as high single-vehicle crashes because, while vehicle make/models. for a vehicle with an SSF of 1.00 as for those crashes generate a police report, The linear regression that used only a vehicle with an SSF of 1.53 (the range the collision itself poses no risk of the SSF and the state dummy variables of the observed data) based on the linear rollover of the vehicle. Our sample size as predictor variables had an ‘‘R- regression model. The average slope of was large enough that we did not need squared’’ of 0.73, which means that the rollover risk versus SSF curve for to further investigate pedestrian and almost three-quarters of the variability the linear regression model (Figure 1) in animal crashes for relevance. We did in rollover risk between vehicle models the range of observed data was ¥0.713. include collisions with parked vehicles is explained by the SSF plus the The slope of the corresponding curve of because they represented a type of adjustments for state-to-state differences the logistic models is ¥0.598 or roadway departure and a collision with in crash reporting. This is greater than ¥0.580, depending on whether we use a fixed object, although these collisions the ‘‘R-squared’’ for the best model that the individual variables or the scenario- offer the least exposure to typical used only the road-use variables plus risk variable. Both the linear and logistic tripping mechanisms. the state dummy variables (0.58). Thus, approaches produced models that fit the Our analysis examined the effects of the SSF appears to have greater data well, and both estimated a road-use variables because their explanatory value than the combination coefficient for the SSF term that was correlations with SSF were the basis of of the road-use variables. We conclude very important (in terms of statistical an alternative theory of rollover that the SSF is not relegated to significance and the magnitude of the causation. It is plausible that the greater insignificance by the road-use variables effect). rate of rollover of vehicles with low SSF in describing rollover risk. The linear regression is judged by the is not caused by low SSF but rather by The Alliance comment criticized the ‘‘R-squared’’, a measure of fit that is characteristics of drivers and roads agency’s use of linear regression because familiar to many people. The logistic which happen to be correlated with low it operates on averages of road-use regression is less well known, but it also SSF vehicles. The example of young variables and cannot consider the has a standard measure of fit, the males being the predominant driver possible interaction among variables. association of predicted probabilities population of particularly low SSF For example, the linear regression and observed responses. The percentage pickup trucks shows that this model would consider that the crashes of concordant pairs for our logistic alternative has plausibility. of a particular make/model may involve models was very high (for example, it However, the Alliance departed from 30 percent young drivers, 20 percent was 71.4 percent for the six-state the road-use variables as alternative with alcohol involvement and 15 combined model). causes of rollover. The Alliance analysis percent on curves, but it cannot We can also measure the ‘‘Chi-square’’ was not an explanation of alternative distinguish crashes in which all of the value for the coefficient of the SSF term theories of rollover causation but rather factors were present simultaneously. in each model to describe the an attempt to show that there is little, The Alliance used logistic regression significance of that term. Logistic if any, effect of SSF on rollover rather than linear regression in its regression models were calculated for causation. To do this, the Alliance analysis. Logistic regression operates on the original six states, plus Ohio and created a category of ‘‘non-vehicle’’ every individual crash circumstance New Mexico, which report rollover only variables. This category allowed the sampled, rather than on averages of the if it is the first harmful event. In seven addition of one variable whose effect road-use variables for crashes of each of the eight states, the ‘‘Chi-square’’ overwhelmed the effects of all other

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00005 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3392 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations variables. That variable was ‘‘first could fly, there would be far fewer as an example. The commenter harmful event, collision with a traffic rollover crashes, based on the suggested using logistic regression to unit.’’ It separated crashes which were experience of actual aircraft. Since resolve the problem of variable collisions with pedestrians, animals or vehicles can not fly, and run-off-the- interaction. parked vehicles from other single- road crashes can not be changed into As introduced in the previous section, vehicle crashes. In essence, the extra different types of crashes, positing these Appendix I of this notice presents a new variable separates crashes with impossibilities as a means of analyzing, statistical study which adds another minimum exposure to tripping or addressing, the real world problem of year of state crash data to the database mechanisms from all other single- more than 10,000 Americans dying each relied on in our previous notice and vehicle crashes. This would seem to be year in rollover crashes does not seem contrasts analyses of the crash data a meaningless addition because there is either helpful or insightful. using logistic regression of individual NHTSA seeks ways to address real no reason to expect a significant variables and risk scenarios to the linear world safety problems constructively. In correlation between SSF and collisions regression method used in the previous the real world, driver and roadway with pedestrians, animals and parked notice. The model curves estimating vehicles. However, it sets up what the factors are certainly important factors in rollovers per single-vehicle crash using Alliance calls its ‘‘low risk scenario’’ all crashes, including rollovers. That is logistic regression were nearly identical which serves as a basis for comparison why NHTSA spends so much effort to regardless of whether the road-use of rollover risk factors. increase belt use, reduce speeding, variables were entered individually or The Alliance then compared the effect eliminate impaired driving, and so as combinations in risk scenarios. on rollover risk of increased SSF to the forth. However, the vehicle is also a However, logistic regression does effect on rollover risk made by moving significant factor in crash safety. If we produce a slightly different curve from the scenarios of actual crashes to take the driver and roadway conditions estimating rollovers per single-vehicle the ‘‘low risk scenario’’. The effect on as givens (for example, a young male crash from that previously reported for rollover risk of moving actual crash driver in a rural area), the physical linear regression. scenarios to the ‘‘low risk scenario’’ is attributes of different vehicles essentially the effect on rollover risk of determine different outcomes when, for Figure 1 shows the comparison eliminating tripping mechanisms. The example, the vehicle drops two wheels between the updated linear regression effect is huge. In simplified terms, the off the road, and the driver responds analysis of the summarized data and the Alliance has argued that the effect on incorrectly. Some vehicles will roll over two logistic models (the six-state models tripped rollover gained by an increase of much more often than others in these using either the individual variables or SSF is minimal compared to the effect situations. Such vehicle differences the scenario-risk variable). The linear on tripped rollover of removing tripping have been shown to strongly correlate regression curve of the previous notice mechanisms. The statistical study in with rollover resistance expressed by was essentially unchanged by the Appendix I includes a discussion of SSF. We believe the American public addition of another year of state crash how this type of analysis, in which should have this information available data (for a total of 226,117 single-vehicle characteristics of the crash itself are to consider when making purchase crashes with 45,574 rollovers). The used to define the risk scenarios, is decisions. logistic models are very similar to each equally useful for ‘‘demonstrating’’ that other, and all the models indicate that seat belts have negligible safety benefit. B. NHTSA’s Statistical Analysis Linking the SSF is very important in We do not find the Alliance analysis SSF to Rollover Rates understanding rollover risk. As noted persuasive. It may well be true that The Alliance commented that the previously, the average slope of the changing a single-vehicle run-off-the- method NHTSA used to analyze the rollover risk vs. SSF curve estimated by road crash (where there is a high risk of statistical relationship between state the linear regression model in the range rollover) into a crash in which the crash data and SSF used in the RFC of observed data was ¥0.713, and the vehicle, for example, hits an animal in failed to take into account possible average slope of the corresponding the road (where there is no risk of interactions between the various non- curve of the logistic models is ¥0.598 rollover) virtually eliminates the risk of vehicle variables, and therefore or ¥0.580, depending on whether we rollover, and may do far more to underestimated the role of the non- use the individual variables or the minimize rollover risk than changing vehicle factors in rollover risk. The scenario-risk variable. Also, logistic any single vehicle or driver factor. possible interaction between alcohol regression estimates a greater risk of However, the point of this is unclear. involvement and the crash occurring on rollover than does linear regression for One could also show that if vehicles a curve in a particular crash was given vehicles with SSFs higher than 1.10.

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00006 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3393

BILLING CODE 4910±59±C regression curve proposed in our June 1, regression curve was that reasonable The logistic regression and linear 2000 notice. one-star and five-star SSF boundaries regression separate the effects of vehicle The proposed rating system was based occurred at predicted levels of rollover rollover resistance and those of road-use on equal intervals of risk and positioned risk of 10 percent and 40 percent, variables by different processes, and the the five-star level at a value of SSF permitting three equal intervals of risk logistic regression predicts a curve with achievable by favorably designed family between them divisible by ten for the a lower average slope. The Alliance sedans. It also positioned the one-star two-star, three-star and four-star commented that logistic regression range where it captured some popular boundaries. Having the star rating considers the potential effect of SUVs and pickup trucks of the recent intervals bounded at 10, 20, 30 and 40 variables in combination which may past. The manufacturers of the one-star percent rollover risk levels would make intensify or dilute their individual vehicles generally have improved the the meaning of the ratings easier to effects, but that linear regression would current versions of the equivalent explain to consumers. Figure 2 presents neglect combination effects. With this vehicles to the two-star level, but we the proposed rating system in graphical possibility in mind, we considered believe the one-star rating ceiling would form. The updated linear regression whether the use of the curve be stringent enough to discourage curve in Figure 1 is nearly identical to corresponding to logistic regression on companies from returning to old design the linear regression curve in Figure 2, individual variables would serve as a practices or from importing less except that it would set the one star better basis of rollover risk for the advanced vehicles. A fortuitous feature boundary for 40 percent rollover risk at vehicle star ratings than the linear of the ratings based on the linear 1.03 instead of 1.04.

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00007 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3394 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

BILLING CODE 4910±59±C demonstrated very good performance in and five-star boundaries to reflect the We considered the merits of the resisting rollover. We believe that difference in average slope between the various ways in which the rollover risk maintaining the 10, 20, 30 and 40 linear regression curve and the logistic versus SSF curve produced by logistic percent star boundaries with the logistic regression curve. A five-star boundary of regression (Figure 1, Individual regression curve would have the 1.46 corresponds to a rollover risk of Variables) could be used to replace that practical effect of replacing the five-star less than 12 percent on the logistic produced by linear regression (also in rating system with a three-star rating regression curve. (The previous Figure 1) as the basis for defining system. At the low end of the SSF scale, boundary of 1.45 would require a rollover risk in the rating system. If the the distinction between some statement of risk of 12.1 percent which proposed rating intervals in terms of historically poor performing vehicles would not be desirable for consumer SSF (1.04, 1.12, 1.24, 1.45) were and their improved replacements would information). Similarly, a one-star maintained, they would no longer be lost. At the higher end of the SSF boundary of 1.05 would correspond to satisfy their rationale of representing scale, the distinction between some very a rollover risk greater than 36 percent. equal increments of rollover risk in a good performing mid-sized and large These one-star and five-star boundaries single-vehicle crash. Conversely, if the sedans and some clearly poorer would allow for equal risk intervals of risk intervals at 10, 20, 30 and 40 performing sub-compacts would be lost. eight percentage points between the percent are maintained, the one-star SSF It would appear that the best way to other star boundaries. A change from 10 level would become 1.01 and the five- incorporate the rollover risk levels percent risk intervals to eight percent star level would become 1.51. A one-star estimated by logistic regression while risk intervals would be proportional to level of 1.01 is so low that we know of maintaining the usefulness of the rating the difference in average slope between only one vehicle (not in current system to the consumer is to maintain the linear regression curve and the production) that it would describe. the proposed one-star and five-star logistic regression curve. Figure 3 Similarly, a five-star level of 1.51 boundaries as closely as possible. This illustrates this idea for using the logistic appears to be out of reach for even the approach would require adjustment of curve in a revised rating system in a most stable family sedans which have the equal risk intervals between the one- graphical form comparable to Figure 2.

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00008 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3395

However, this idea also has serious the difference in risk analysis methods percent is associated with SSF 1.25 to drawbacks. It would move the three star is great enough to compel a change in 1.44. level from 1.13 SSF to 1.17 and the four the proposed star rating levels to the Five Stars ★★★★★: Risk of Rollover star level from 1.25 to 1.29 because the detriment of manufacturers who are less than 10 percent is associated with logistic regression shows less of the trying to achieve them and to the SSF 1.45 or more. asymptotic shape of the raw data (Figure detriment of consumers who we believe 1 of Appendix 1) than does the linear will find the proposed rating system C. Comments on Practical Problems regression (of the log of SSF) curve simpler. We also note that the linear with SSF Ratings previously proposed. This is troubling regression curve presents a more 1. Difficulty of Improving Vehicles for two reasons. The shape of the conservative estimate of rollover risk for original linear regression curve vehicles with SSF greater than 1.10, and The Alliance and the import conforms better than does the logistic we anticipate vehicles with SSF lower manufacturers’ organization, AIAM, regression curve to the expectation that than 1.10 becoming rare in light of asserted that improvements in a a given increase in SSF produces a manufacturers’ reported efforts at vehicle’s SSF are not practicable since substantially greater benefit for a vehicle improving rollover resistance. SSF is largely determined by its vehicle with a low SSF than for one with a high The rating system that NHTSA will type. That is, the track widths and c.g. SSF. Also, NHTSA believes that the use to define rollover risk and assign heights of pickups, SUVs, vans, and proposed star rating levels may have star rating is based on the updated passenger cars are more or less fixed become design goals for manufacturers linear regression curve in Figure 1 of within certain limits. Significant seeking to improve rollover resistance. this section. It would be described changes to those measurements would A change in star rating levels at this verbally as follows: simply eliminate the vehicle attributes time may have the counterproductive which are common to the category and One Star ★: Risk of Rollover 40 effect of denying manufacturers which are presumably desirable to percent or greater in a single-vehicle recognition for substantial consumers. These comments noted, for crash is associated with SSF 1.03 or improvements in rollover resistance of example, that significantly lowering the less. c.g. (thus raising the SSF) of an SUV vehicle designs. ★★ While we do not deny the theoretical Two Stars : Risk of Rollover 30 could be accomplished by decreasing advantages of logistic regression cited percent or greater but less than 40 ground clearance, but doing so might by the Alliance regarding interactions percent is associated with SSF 1.04 to make it unappealing compared to other between road use variables, the 1.12. vehicles in the SUV category. similarity in curves describing rollover Three Stars ★★★: Risk of Rollover 20 Conversely, the comments contended risk as a function of SSF in the linear percent or greater but less than 30 that marginal changes to track width and logistic regression approaches percent is associated with SSF 1.13 to and c.g. height small enough to suggests that such interactions do not 1.24. maintain attributes in a vehicle category exert a great influence on the effect of Four Stars ★★★★: Risk of Rollover 10 would not improve rollover risk. They SSF. Therefore, we do not believe that percent or greater but less than 20 conclude that SSF is not a useful design

VerDate 112000 23:26 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00009 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm03 PsN: 12JAR10 3396 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations criterion, and it lacks the potential to vehicle. As a contrasting example, Furthermore, SSF is relevant to stability reduce rollover rates if current vehicle buyers who desire passenger and cargo under virtually any circumstance, types are to be preserved. capacity may choose a van or minivan whether it be a run-off-the-road crash, We disagree that significant over a conventional station wagon after an obstacle avoidance scenario, or even improvement in SSF will necessarily deciding that their priorities outweigh collisions with objects or other vehicles, eliminate desirable attributes within a the increase in rollover risk associated though it is obviously more significant class of vehicles. We are aware of a with that choice. in some of those events, i.e., single- recent redesign of a production SUV 8 in We believe that vehicle modifications vehicle crashes, than in others, i.e., the U.S. that achieved a decrease in c.g. to improve rollover resistance ratings collisions, where impact forces can height of approximately 2.0 inches are both achievable and beneficial. Press overwhelm other factors. (along with a significant increase in accounts suggest that manufacturers are, It was suggested in the comments that track width) while actually increasing in fact, making such modifications as vehicle characteristics which an SSF- the ground clearance. We estimate those they redesign their light trucks. based rating ignores, like body shape changes represent an improvement in However, the ratings do not force and tire profile, influence rollover rate SSF equivalent to at least one star rating manufacturers to modify vehicles, nor because they determine how a vehicle interval, and we would expect a do they force consumers to accept only interacts with roadside objects and significant decrease in rollover risk in certain vehicle alternatives. The ratings terrain during a crash event. As an single-vehicle crashes. will have a positive effect on the light example, Honda suggested that lowering We also would note that passenger vehicle rollover problem by making a vehicle’s c.g., thus improving its SSF, car-based SUV’s with significantly consumers more aware of trade offs in by equipping it with low-profile tires better SSFs than traditional, truck-based rollover stability, allowing consumers to could increase the risk of tripped SUVs have been gaining popularity in make more informed purchase rollover by making sideward wheel the absence of any consumer decisions, and influencing their contact with tripping mechanisms more information program for rollover. The awareness of the need to wear seat belts likely. This is speculative and not range of SSF among ten SUVs in our to prevent ejection in rollover crashes. persuasive. Each single-vehicle crash is, 1998 SSF measurements of a group of 32 This improvement will accrue even if more or less, a unique event, because of then-new vehicles was equivalent to a the manufacturers make no changes to the variety and complexity of rollover risk reduction of approximately vehicles whatsoever in response to the circumstances involved. Although we 14 percent using the predictive curve program. agree that tripping usually initiates through interaction of a vehicle’s wheels from the linear regression analysis 2. Possible Consequences of Improving (i.e., tires and/or rims) with the roadway explained in this notice. So-called SSF ‘‘crossover’’ vehicles promise even environment, generalizations about the greater improvement. While these Honda and the Alliance also influence of low-profile tires, or vehicles may offer less of some suggested that, with a design criterion differences in body shape, on tripping attributes of traditional SUVs, like like a rollover rating based on SSF, frequency are extremely difficult to overall ride height, the increasing manufacturers may be inclined to substantiate, given the limitless popularity of crossover vehicles ‘‘design for the test.’’ The manufacturer combinations of terrain, pavement indicates that those attributes may be of a vehicle whose score falls just below condition, shoulder design, barriers, less important to consumers than the a rating cutoff point might be able to soil, vegetation, etc. A vehicle feature ones which they maintain in common make design adjustments that shift the like taller, more flexible tire sidewalls with traditional SUVs, such as cargo vehicle’s score into the next higher may help avoid tripping in a few room and traction on snowy roads. category. We believe there is no reason crashes, but is likely to be ineffective in Thus, the suggestion that no changes to to discourage manufacturers from taking the vast majority of others, and may be current vehicle designs are possible such actions because an improvement in counterproductive in some cases. Even without significant customer resistance SSF will result in a corresponding if it were possible for a manufacturer to appears to be an assertion unsupported improvement in rollover risk. In fact, we identify tires and rims that were by what has happened recently in the believe that a major advantage of SSF, supposedly more resistant to tripping, market. one that distinguishes it from other safe handling and road holding On the other hand, one of the models measures of rollover resistance, is that it considerations should certainly weigh that scored highest among the ten SUVs ‘‘does no harm.’’ Since SSF is a more heavily in tire and rim selection. in the 1998 measurements was a more fundamental measure of inherent A notable exception to this involves the problem of tire debeading. Clearly, or less traditional design, i.e., it was not vehicle stability, there is no realistic risk a wheel rim that becomes exposed when passenger car-based.9 This gives that increasing SSF will degrade actual a tire debeads either as a precursor to a evidence that more stable light truck rollover rate or have other unintended, single-vehicle crash or in the course of design is not incompatible with negative consequences. In contrast, one, can become a primary tripping traditional design attributes. improvement in other metrics can result The fact that SUVs are seldom used in trade-offs that compromise overall mechanism. We believe that tire and rim combinations that are more resistant to off-road indicates that not all SUV safety. For example, maximizing a debeading may indeed lessen the risk of buyers really want off-road capability. vehicle’s Tilt Table Ratio can be rollover in a single-vehicle crash. The Buyers who are aware of the tradeoff in accomplished by trading off some agency is already planning to improve risk of rollover that such off-road vehicle directional control (oversteer/ debeading requirements in FMVSS No. capability usually entails, may decide understeer) characteristics. As another 109. they can obtain the attributes they want example, it is apparent that the Stability A further difficulty in identifying or need in a more rollover-resistant Margin metric can be improved by reducing tire grip, which could decrease vehicle features that might improve 10 tripping resistance is that crash data is 8 Mitsubishi Montero redesign from model year driver control of the vehicle. (MY) 1991–99 design to MY 2000 version of the limited. The minute level of detail same nameplate. 10 These metrics are explained in detail in the required to thoroughly analyze the 9 Isuzu Rodeo. June 1, 2000 notice. interaction of a vehicle’s wheels,

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00010 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3397 undercarriage, body components, etc., documents,12, 13, 14 describing the of the VIMF. The calibration fixture can with the roadway environment in a run- design of the VIMF give the same value be set up in either a light or heavy off-the-road event is generally for ‘‘repeatability’’ or ‘‘two standard configuration, the latter achieved by unavailable in state or national crash deviation error’’ for c.g. height adding weight in precise locations to databases. NHTSA’s NASS–CDS measurements. increase the c.g. height by a known database does contain a high level of Basically, ‘‘repeatability,’’ as used in amount. In the light configuration, the detail, but it focuses on a relatively the referenced documents in regard to fixture is representative of the mass and small sampling of crashes. In contrast, the VIMF, is not separate from the c.g. height of a mid-size passenger car. the SSF of vehicles in crashes can be ‘‘accuracy’’ of the system. It is incorrect In the heavy configuration, it is determined as long as the data contain to assume that the total VIMF system representative of a light truck. a few details about the vehicle, like error in c.g. height measurements is the In calibration tests using this fixture, make and model. Availability of sum of the 0.5 percent repeatability and the VIMF consistently measures the c.g. extensive crash data is important for 0.5 percent accuracy, for a total system location to within 0.5 percent of the analyses like NHTSA’s statistical error of one percent in c.g. height known value. Tables 6 and 7 of the 1995 analysis of crashes in six U.S. states as measurements. The total system error of Heydinger paper cited here indicate that reported in the RFC and in Appendix I the VIMF for c.g. height measurement is the VIMF was able to measure the c.g. here. 0.5 percent or less, as explained below. height of the fixture to within 0.46 When the VIMF was under percent (2.6 mm in 561.2 mm) and 0.32 Honda also suggested that development, an error analysis was percent (2.6 mm in 809.2 mm) of its problematic suspension behavior such conducted based on experience with theoretically known values in the light as ‘‘suspension-jacking’’ can lead to a NHTSA’s Inertial Parameter and heavy configurations, respectively. higher risk of rollover regardless of SSF, Measurement Device (IPMD), a Those results correspond well with the and that this exemplifies why SSF alone precursor to the VIMF. Over the course VIMF error analysis which predicts that is not an adequate indicator of rollover of several years, the IPMD underwent the degree of accuracy should be resistance. Although vehicles with successive updates and improvements, somewhat higher when measuring particular suspensions, most notably culminating in a fifth and final version heavier, higher c.g. vehicles. That is, the ‘‘swing-axle’’ designs, historically may of the machine that ultimately served as measurement accuracy for vehicles have been associated with rollover, we a model for the VIMF. The error analysis which are likely to fall into the lower believe those represent relatively few accounted for all the known sources of SSF categories is significantly better cases out of a very large population of error arising from each system than 0.5 percent. rollover crashes and that such examples component, for example, platform While we believe the NHTSA of suspension design are uncommon in deflection and vehicle restraint rigidity, measurements will be sufficiently current vehicles. Furthermore, as experience with the IPMD had accurate, no degree of measurement suspension behavior is less important indicated. By mathematical modeling, accuracy can prevent borderline cases. than SSF once a vehicle has left the the contribution of each component to There is always a possibility of a vehicle roadway, where factors like shoulder the whole system error was determined. score falling so close to a cutoff point condition and terrain interact with the The final design specifications for the between star ranges that applying even basic stability characteristics of the VIMF were set by that analysis. Each a small amount of measurement vehicle to determine crash outcome. component was selected or fabricated so uncertainty to the score results in 3. SSF Measurement Accuracy as to limit the combined error from all ambiguity about the category to which the known contributions to 0.5 percent the vehicle belongs. This situation is Honda stated in response to the RFC of the measured value for c.g. height. characteristic of any rating scheme and that the Vehicle Inertia Measurement The details of the error analysis are is no different from what currently Facility(VIMF) that NHTSA will use to discussed in the referenced documents. exists in the NHTSA frontal and side ascertain SSF is not accurate enough to Since it was designed and NCAP. We plan to use conventional repeatably give useful vehicle ratings. constructed, the accuracy of the VIMF rounding methodology to determine the Honda suggested that for c.g. height has been evaluated using a custom-built SSF of each test vehicle to two decimal measurement the measurement error is calibration fixture with a known c.g. places and assign stars based on that the sum of 0.5 percent ‘‘repeatability’’ location. This fixture is a heavy result. error and 0.5 percent ‘‘accuracy’’ error, weldment made from stock steel plates If a manufacturer determined that one giving a total measurement error of ±1.0 and box section beams whose of its models was on the border between percent of the measured value. Honda individual c.g. locations are easily star levels, the manufacturer could, if it believes an error of that magnitude is determined by geometry. Because it is a wished, make changes to the vehicle to significant, compared to the small very rigid body and is fabricated from improve its SSF to the point where it differences between vehicles being such geometrically simple components, falls comfortably in the higher category. compared, and that a vehicle could be the calibration fixture’s c.g. location, as If the vehicle was indeed on the border, assigned an incorrect number of stars well as its mass moments of inertia, are the changes necessary would probably due to measurement error. known theoretically, and it is thus a be very minor, and it would be benchmark for reckoning the accuracy Honda appears to have misinterpreted voluntary, not mandatory. the published reports available on the 12 Heydinger, G.J., et al; ‘‘The Design of Vehicle D. Consumers’ Ability to Understand VIMF. The document cited in Footnote Inertia Measurement Facility’’; SAE Paper 950309; SSF as a Measure of Rollover Risk in the 19 of the RFC does indicate, in Table 1, February 1995. Event of a Single-vehicle Crash ‘‘error bounds’’ for c.g. height of ±0.5 13 Bixel, R.A., et al; ‘‘Developments in Vehicle percent of the measured value.11 Other Center of Gravity and Inertial Parameter Estimation Some commenters had misgivings and Measurement’’; SAE Paper 950356; February about consumers’ abilities to understand 1995. and use the new rollover rating 11 14 Heydinger, G.J., et al; ‘‘Measured Vehicle Heydinger, G.J., et al; ‘‘An Overview of a information in three areas. They believe: Inertial Parameters—NHTSA’s Data through Vehicle Inertia Measurement Facility’’; Intl. • November 1998’’; Society of Automotive Engineers Symposium on Automotive Technology; Paper Consumers are not capable of 1999–01–1336; March, 1999. 94SF034; October 1994. understanding that the star rating

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00011 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3398 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations describes the risk of rollover in the object causing it to roll over. These are The more ‘‘top-heavy’’ the vehicle, the event that the vehicle is involved in a called single-vehicle crashes because more likely it is to roll over. Based on single-vehicle crash. the crash did not involve a crash with a study of 185,000 single-vehicle • Consumers will not find the another vehicle. The Rollover Rating is crashes, this measurement has been information useful in making a vehicle an estimate of your risk of rolling over shown to relate very closely to the real- choice. if you have a single-vehicle crash. The world rollover experience of vehicles. • Even if consumers use the Rollover Rating essentially measures • NHTSA’s Front and Side Crash Test information, the new program will not how ‘‘top-heavy’’ a vehicle is. The more Ratings predict a vehicle occupant’s lead to a decrease in rollover crashes. ‘‘top-heavy’’ the vehicle, the more likely chance of serious injury if the vehicle is Each of these areas are discussed and it is to roll over. The lowest rated involved in that type of crash. The responded to below. vehicles (1-star) are at least 4 times more Rollover Rating predicts the risk of a 1. Are Consumers Capable of likely to roll over than the highest rated rollover if your vehicle is involved in a vehicles (5-stars). single-vehicle crash. (It does not, Understanding That the Star Rating • Describes the Risk of Rollover in the Here are the Rollover Ratings: however, predict the likelihood of that crash.) Event That the Vehicle Is Involved in a In A Single-vehicle Crash, a vehicle with • Single-vehicle Crash? While the Rollover Rating does not a rating of: directly predict the risk of injury or Auto manufacturers and the National Five Stars ★★★★★ death, keep in mind that rollovers have Automobile Dealers’ Association Has a risk of rollover of less than 10% a higher fatality rate than other kinds of (NADA) believe that consumers are not Four Stars ★★★★ crashes. Even the highest rated vehicle capable of understanding that the star Has a risk of rollover greater than 10% can roll over, but you can reduce your rating describes the risk of rollover in but less than 20% chance of being killed in a rollover by ★★★ the event that the vehicle is involved in Three Stars about 75% just by wearing your seat a single-vehicle crash. The following is Has a risk of rollover greater than 20% belt. a list of comments and the commenters but less than 30% • Here are the Rollover Ratings: who made them: Two Stars ★★ • Consumers will be confused Has a risk of rollover greater than 30% In A Single-vehicle Crash, a vehicle with because the rollover ratings are not in but less than 40% a rating of: ★ terms of injury risk like other NCAP One Star Five Stars ★★★★★ ratings—Alliance Has a risk of rollover greater than 40% Has a risk of rollover of less than 10% • Consumers will not understand that We also asked the focus groups to Four Stars ★★★★ the rollover ratings do not include evaluate the following longer version: Has a risk of rollover greater than 10% crashworthiness attributes—AIAM but less than 20% Description of Rollover Resistance • Consumers will think the rollover Three Stars ★★★ Rating risk is the life-time rollover risk from Has a risk of rollover greater than 20% driving the vehicle or the risk of rollover • Thousands of crashes occur each but less than 30% each time they drive the vehicle— year when a driver loses control of his/ Two Stars ★★ Alliance, Suzuki, Toyota, Honda her vehicle and runs off the road. These Has a risk of rollover greater than 30% • Consumers will think risk is the are called single-vehicle crashes because but less than 40% same for all drivers in all conditions and the crash did not involve a collision One Star ★ have the false impression that the with another vehicle. Once the vehicle Has a risk of rollover greater than 40% vehicle design is the principal cause of leaves the road it can hit an object (pole, The focus group testing pointed out rollover—Suzuki, NADA tree, guardrail, etc.), or the wheels can areas of difficulty in comprehension The language that will be used in contact a ditch, soft soil, a curb or other that were addressed in writing the final consumer information products object, tripping the vehicle and causing language. concerning this rollover rating (see it to roll over. Single-vehicle rollovers Focus group participants felt that Section IV) was developed using the can also occur on the road, but most while the shorter explanation was too outcome of focus group testing. As rollover crashes occur when a vehicle short to fully comprehend the rating, the discussed in the June 2000 notice, in runs off the road, usually sliding longer version was overwhelming and April 1999 NHTSA conducted a series sideways. included unnecessary information. of six focus groups to examine ways of • The National Highway Traffic Based on the focus group inputs, we presenting comparative rollover Safety Administration (NHTSA) has have developed the following language: information. As a result of the provided consumers with frontal and Description of Rollover Resistance comments to our June 2000 notice, side impact crash test ratings for several Rating NHTSA conducted another series of years. Because more than 10,000 people focus groups in November 2000. Two die each year in rollover crashes, • Most rollover crashes occur when a versions of explanatory language were NHTSA has added a Rollover Rating to vehicle runs off the road and is tripped presented to a total of 12 groups of nine provide consumers with better overall by a ditch, curb, soft soil, or other object consumers each in two different cities. safety information on new vehicles. causing it to roll over. These crashes are NHTSA asked the focus groups to • The Rollover Rating is an estimate usually caused by driver behavior such evaluate a short version of rollover of your risk of rolling over if you have as speeding or inattention. These are rating explanatory language that read as a single-vehicle crash. If that happens, called single-vehicle crashes because follows: the risk of rollover for the highest rated the crash did not involve a collision vehicles (5-star) is less than 10%, but with another vehicle. More than 10,000 Description of Rollover Resistance that risk factor increases by a factor of people die each year in all rollover Rating 3 to 4 for the lowest rated vehicles (1- crashes. Most rollover crashes occur when a star). • The Rollover Resistance Rating is vehicle runs off the road and is tripped • The Rollover Rating essentially an estimate of your risk of rolling over by a ditch, soft soil, a curb or other measures how ‘‘top-heavy’’ a vehicle is. if you have a single-vehicle crash. It

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00012 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3399 does not predict the likelihood of that are aware that rollover is heavily vehicle crash by up to 24% in some crash. The Rollover Resistance Rating influenced by driver and road cases. essentially measures vehicle characteristics. In almost all groups the In addition, another safety benefit of characteristics of center of gravity and first cause of rollover cited by the NCAP program is the general track width to determine how ‘‘top- participants was speed. Participants also improvements manufacturers have heavy’’ a vehicle is. The more ‘‘top- mentioned road conditions and driver made to vehicles as the result of heavy’’ the vehicle, the more likely it is behavior and/or experience as factors. publishing such ratings. These to roll over. The lowest rated vehicles However, the participants also seemed improvements benefit all consumers (1-star) are at least 4 times more likely to understand that the vehicle can also regardless of their choice of vehicle. to roll over than the highest rated play a part in determining whether or Over the years, manufacturers have vehicles (5-stars). not a rollover occurs, and that this responded to the frontal NCAP program • The Rollover Resistance Ratings of rating was only a measure of that factor. and as a result the number of models vehicles were compared to 220,000 NHTSA notes that the explanatory achieving a five-star rating today is 2.7 actual single-vehicle crashes, and the language will be used in the Buying a times what it was when the program ratings were found to relate very closely Safer Car brochure, and other places started in 1979. As for the criticism that to the real-world rollover experience of that present the star ratings. This star ratings do not indicate the tiny vehicles. brochure’s primary focus is how a difference among vehicles near the • While the Rollover Resistance person can purchase a safer vehicle. It dividing lines, this is also true for the Rating does not directly predict the risk does not include extensive discussion of frontal and side NCAP ratings. Just as of injury or death, keep in mind that driver behaviors that can increase with these ratings, the actual scores for rollovers have a higher fatality rate than safety, as those types of issues tend to the vehicles will be available on the other kinds of crashes. Remember: Even be addressed by other agency programs. NCAP website to anyone who is the highest rated vehicle can roll over, NHTSA will include additional interested. but you can reduce your chance of being information about rollover in the form Finally, with regard to comments that killed in a rollover by about 75% just by of Q&A’s on the agency’s website, and options can cause wide difference in the wearing your seat belt. is considering developing additional rating for a specific model, over the • Here are the Rollover Resistance rollover consumer information, both of years that we have been researching Ratings: which would be more appropriate vehicle inertial parameters, four-wheel places for discussion of other factors drive is the only equipment option for In A Single-Vehicle Crash, a vehicle that can reduce the risk of rollover. with a rating of: which we have observed a large 2. Will Consumers Find the Information potential effect on SSF. NHTSA intends Five Stars ★★★★★ Useful in Making a Vehicle Choice? to test the most common versions of all Has a risk of rollover of less than 10% Four Stars ★★★★★ The commenters listed below believe vehicles. Where two- and four-wheel Has a risk of rollover between 10% that even if consumers do understand drive versions of the same vehicle are and 20% the risk represented by the stars, this available, we will test them both and Three Stars ★★★★★ information will not be useful to them report them as separate models. We will Has a risk of rollover between 20% in choosing a vehicle. They assert the accurately describe the actual test and 30% following: vehicle in the literature reporting the Two Stars ★★ • Consumers pick a vehicle class rating. Has a risk of rollover between 30% before they select a particular model. Manufacturers who believe there are and 40% There are not enough differences in star significant differences in SSF for One Star ★ ratings among vehicles in the same class different vehicle configurations may Has a risk of rollover greater than 40% to make the information useful to fund an optional NCAP measurement, The length of the final version is consumers. The stars reflect only tiny just as they may fund optional frontal or midway between the two versions differences on each side of the dividing side NCAP tests. Then if the difference tested. It adds information not included line.—Alliance, Ford, BMW, CU in equipment or configuration makes a in the tested short version that • The difference in SSF made by difference in the SSF, that difference participants felt was particularly options and configurations available on will be available to the public. important in understanding the a single vehicle are too great to allow 3. Even If Consumers Use the information and/or particularly meaningful ratings—Alliance Information, Will the New Program compelling to cause them to pay While it is true that many consumers Lead to a Decrease in Rollover Crashes? attention to the information. It deletes limit their vehicle choices early in the information in the tested long version purchase-decision process (e.g., must be Some commenters believe that even if that participants felt was unnecessary an SUV), many others are also consumers do use the new ratings, the and/or confusing. In addition, the considering vehicles in more than one outcome of that use will be other than explanation of the star ratings was class (e.g., a van or an SUV). As the what we desire. The following are simplified because the original format availability of rollover resistance rating comments and who made them. seemed to cause some confusion about information becomes more widely • Rollover ratings will encourage whether more stars or less stars was a known, consumers will begin to know consumers to purchase cars instead of better rating. Finally, NHTSA has that certain types of vehicles have better trucks and cars are less safe than chosen to use the term ‘‘Rollover ratings than others. In addition, while trucks.—Alliance Resistance Rating’’ rather than ‘‘Rollover we cannot predict the final spread of • A system based on RO/SVC may Rating’’ as this seemed to help ratings for the 2001 models that will be cause the choice of a less-safe vehicle participants understand the rating. tested, in our research there was usually because it doesn’t take the make/ The potential confusions cited by the a two- to three-star rating range for each model’s risk of becoming involved in a commenters did not occur in the focus class. Thus, by his or her vehicle choice crash into account.—Suzuki, Tenneco groups. From the discussions during the alone, a consumer could reduce his or • Consumers will think that if they focus groups, it is clear that participants her chance of a rollover in a single- drive a vehicle with a high SSF they

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00013 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3400 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations will be immune to rollover and this will of these types of analysis will be across all crash types indicates that lead them to drive unsafely—Alliance possible for side impact and rollover large cars have a lower fatality rate than • There is no demonstrated safety NCAP after an adequate number of years SUV’s and small pickup trucks, and the benefit of rollover rating.—Alliance, of experience. There is no evidence that same as the rate for standard pickups. BMW consumers have responded to vehicles Medium cars have a rate about the same The best indicator of the potential with high frontal NCAP scores or other as SUV’s and lower than the rate for benefits of any new ratings program is safety features by riskier driving small pickup trucks. Small cars and the frontal NCAP program. As discussed behavior, and no reason to believe that small pickup trucks have about the previously, there are now many more they will respond differently to rollover same rate. See Figure 4. If we narrow the five-star vehicles than when the frontal ratings. Similarly, there is no indication picture to rollover crashes, as in Figure NCAP program started. Research also that consumers believe they are immune 5, we see that SUV’s and small pickups indicates that a five-star rating correlates to injury by driving a vehicle with a have the highest rates, at least 75 to enhanced real-world safety. five-star frontal or side NCAP rating or Therefore, all consumers benefit from with additional safety features. percent higher than the rate for small these improvements in the vehicle fleet NHTSA disagrees that cars are less cars. The rates for medium and large even if they don’t make purchase safe than light trucks. Occupant fatality cars are below any of the light truck decisions based on the star ratings. Both rates (average 1991–98, FARS data) types.

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00014 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3401

BILLING CODE 4910±59±C side impact NCAP ratings, and we plan rating presentation after there is some However, NHTSA is aware that as we to do for rollover crashes. In addition, evidence of its effectiveness. BMW, expand the areas in which we provide NHTSA has been considering possible Toyota, Isuzu, Tenneco and the Alliance consumer information ratings, it is ways to provide consumers with a offered similar comments. All expressed becoming more and more important to single summary rating of a vehicle’s confidence that the technology would provide consumers with guidance on safety. reduce the number of on-road loss-of- how to weigh ratings in different control situations that often result in off- categories. For example, it is quite E. The Question of Electronic Stability road tripped rollovers. The Alliance common for SUVs to receive five-star Control suggested that ESC may also reduce the ratings in side impact NCAP, but our Continental Teves objected to the use risk of untripped rollover, and research indicates that these vehicles of SSF to rate rollover resistance Continental believes that it may help will have rollover ratings in the one- to because the ratings would not reward drivers regain control after they leave three-star range. NHTSA can help manufacturers for equipping vehicles the roadway. Many commented that consumers understand these differences with Electronic Stability Control (ESC). ratings based on SSF would stifle and by providing them with information on It was also dissatisfied with language in undercut advanced vehicle technology. the frequency of various crash types, as the notice promising consumer The notice specifically asked we have been doing with the front and information about ESC as part of the commenters to share any data they may

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00015 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3402 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations have on the effectiveness of stability effectiveness of ESC. Based on the ‘‘trips’’ it. The Rollover Resistance Ratings control technologies in preventing relative data available, it would not be estimate the risk of rollover in event of a single-vehicle crashes, but none did so. appropriate to abandon SSF. We single vehicle crash, usually when the The NCAP program rates the risk of encourage manufacturers to assist us in vehicle runs off the road. Electronic Stability rollover in the event of a single-vehicle Control (which is offered under various trade determining the effectiveness of ESC by names) is designed to assist drivers in crash. Most of these single-vehicle identifying optional ESC systems in VIN maintaining control of their vehicles during crashes involve hitting a curb or codes and sharing available data. We extreme steering maneuvers. It senses when running off the road accidently and will continually monitor data on the a vehicle is starting to spin out (oversteer) or encountering soft soil, a ditch or real-world effectiveness of ESC and plow out (understeer), and it turns the something that trips the vehicle. To make appropriate changes based on that vehicle to the appropriate heading by repeat, 95 percent of rollovers are data. We do not expect that automatically applying the brake at one or tripped. Once a vehicle is in this manufacturers will abandon ESC, since more wheels. Some systems also situation and strikes a tripping they express so much confidence in its automatically slow the vehicle with further mechanism, its chances of rolling over brake and throttle intervention. What makes ultimate effectiveness. Electronic Stability Control promising is the depend heavily on its SSF. NHTSA wants to encourage The promise of ESC is not that it can possibility that with its aid many drivers will technological applications that enhance avoid running off the road and having a change what happens when a vehicle vehicle stability, provide drivers with single vehicle crash in first place. However, hits a tripping mechanism but that it more control of their vehicle, and help ESC cannot keep a vehicle on the road if its may help the driver to avoid going off prevent rollover and other crashes. For speed is simply too great for the available the roadway in the first place. ESC can ESC in particular, it is reasonable to traction and the maneuver the driver is apply one or more brakes automatically assume that it will help some drivers attempting or if road departure is a result of to keep the yaw rate of the vehicle use the available traction to stay on the driver inattention. In these cases, a single proportional to its speed and lateral road in circumstances that would vehicle crash will happen, and the Rollover Resistance Rating will apply as it does to all acceleration. Essentially, it corrects for otherwise result in panic-driven errors vehicle understeer or oversteer, and vehicles in the event of a single vehicle and roadway departure. We have asked crash. some systems may override a driver’s the National Academy of Sciences to failure to brake when in fear of losing recommend ways of combining the A similar discussion will accompany control. This benefit could minimize the effect of ESC on exposure to single- the rollover resistance ratings in the driver’s chances of compounding his or vehicle crashes, with the effect of SSF April issue of Buying a Safer Car. her driving errors in a panic situation. on rollover resistance in a single-vehicle F. Alternative Programs for Rollover However, it cannot keep a vehicle from crash, as part of its Congressionally- Consumer Information Suggested by leaving the roadway if the vehicle is mandated study of rollover consumer Commenters going too fast for the maneuver the information. We do not expect that a Three commenters to the RFC driver is attempting. recommendation can be implemented Like frontal and side NCAP ratings, presented ideas for consumer without some determination of ESC’s the Rollover Resistance Rating is information programs to be used in real-world effectiveness, but in the concerned with vehicle attributes that place of the agency’s proposal to use meantime we will identify in our affect the outcome of a crash. None of SSF to rate vehicles. The Alliance had Buying a Safer Car brochure the vehicles the present ratings attempt to describe four suggestions: for which ESC is available and provide the probability of a vehicle’s • Cause drivers to obey the speed an explanation of these systems. The involvement in a crash. For example, limits, be alert and unimpaired, and use identification of vehicles with ESC will the frontal crashworthiness star rating proper restraints, and provide driver start in the December 2000 issue of does not reward manufacturers who training in off-road recovery and crash Buying a Safer Car. The April 2001 issue equip vehicles with advanced braking avoidance maneuvering. systems. Also, the agency cannot rely on of Buying a Safer Car will also present • Improve the roadways with paved skid pad demonstrations to determine Rollover Resistance Ratings. shoulders to eliminate road edge drop- The first presentation of Rollover the effectiveness of a safety device in offs and provide road edge rumble strips Resistance Ratings will be on the the hands of the public. Anti-lock to help alert drivers. NHTSA website. The website will also brakes were once considered likely to • Promote Electronic Stability present Questions and Answers reduce rollover crashes because they Control. regarding rollover crashes including one had the potential to reduce the number • Promote crashworthiness of vehicles exiting the road sideways as discussing the effect of ESC and its improvements including active restraint a result of rear brake lock-up. This relationship to the Rollover Resistance systems, tubular and side curtain air expectation has not been realized in Ratings. Until the Rollover Resistance bags, new belt reminder systems, passenger cars according to years of Ratings are integrated into Buying a structural crashworthiness crash statistics. There has actually been Safer Car, the NHTSA website will improvements, FMVSS 201 interior an increase in the rollover rate of provide a chart of rated vehicles which protection, new locks and latches and passenger cars equipped with anti-lock will include a column indicating the alternative glazings. brakes that researchers have not yet availability of ESC. The heading of that Ford and Suzuki commented that SSF been able to explain. column will provide a link to the Q&A should be used only to rate vehicle The commenters suggest that NHTSA about ESC. classes and should not be used to show The Q&A section will include the should abandon SSF as a basis for distinctions between make/models in following discussion: rollover rating because it does not the same class. These commenters also reward ESC in the star rating and that Question: How does Electronic Stability believe that the program should not without such a reward the use of the Control affect rollover, and what is its relationship to the Rollover Resistance present the risk of rollover technology would be in doubt. The Ratings? quantitatively. importance of SSF to rollover resistance Answer: Most rollovers occur when a The NADA recommended that is supported by abundant real-world vehicle runs off the road and strikes a curb, NHTSA put more emphasis on the seat evidence, while there is no data on the soft shoulder, guard rail or other object that belt message in the context of rollover,

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00016 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3403 including child safety restraints and passenger cars.16 NHTSA also possible influence on SSF will be suggested that manufacturers include in previously decided not to set a vehicle included in the vehicle description their vehicles’ owners manuals material rollover standard at a level that would when the rating is reported. Two-wheel about crash avoidance driving practices. effectively force a redesign even of drive and four-wheel drive versions of The manufacturers’ association, the certain vehicle types like small pickups a vehicle are treated as separate models, Alliance, on the other hand, wanted to and small SUVs 17 because it would not because a four-wheel drive option can see seat belt information only in a be appropriate to prohibit the have a significant effect on SSF. As general sense, not specifically referring manufacture and sale of those vehicles provided for in the frontal and side to rollover. without some predictable benefit NCAP, manufacturers can, at their The major flaw with all of these commensurate with the cost of that option, pay for tests of vehicles, models, suggestions is that they do not deliver action. However, we can still provide or configurations not included in what the consumer wants—definitive, accurate and meaningful information NHTSA’s test plan, if they wish to comparative, information about the about rollover resistance to allow the inform consumers about those vehicles relative risk of rollover in specific public to make fully informed choices through the program.19 The SSF will be vehicles. We have shown, in the when selecting a new vehicle. converted to a star rating according to previous sections of this notice and the the curve presented in Section III and notices that have preceded it, that we IV. Rollover Information Dissemination using SSF in NCAP Appendix I at the intervals specified in can link rollover risk to the SSF of Section III. The rollover rating specific make/models. Any rollover- The agency has decided to go forward information will be available on the specific consumer information product with a pilot consumer information agency’s website, and will be included that NHTSA develops in the future will program on vehicle rollover resistance, in all NHTSA publications and press mention driving habits that contribute using the SSF as a basis for the rating releases which use NCAP data. The to rollover prevention and emphasize system. This program will be part of brochures and the website presentation the importance of seat belt use. NCAP, which currently gives consumers will explain the basis of the ratings, However, the focus of the present action information on frontal and side-impact present the SSF measurements, and is on allowing consumers to make an crashworthiness. Today we are discuss the magnitude of rollover harm informed choice about the safety of the announcing the 2001 model year prevention provided by seat belt use. vehicles they purchase, both by class vehicles to be tested and how the As part of the presentation on rollover and by model. information will be disseminated to the the following explanatory text will be public. used: G. Commenters Preference for a There are two activities ongoing in Minimum Standard Based on a NHTSA that may change this pilot Description of Rollover Resistance Dynamic Test program: the study by the National Rating Tab Turner, a plantiff’s attorney, and Academy of Science mandated by • Most rollover crashes occur when a Insurance Institute for Highway Safety, Congress in the Department’s Fiscal vehicle runs off the road and is tripped Consumers Union, and Advocates for Year 2001 appropriations bill 18 and the by a ditch, curb, soft soil, or other object Highway and Auto Safety, stated in Congressional requirement contained in causing it to roll over. These crashes are their comments that, while they had no the TREAD Act that the agency develop usually caused by driver behavior such objection to using SSF to provide a dynamic test for consumer as speeding or inattention. These are consumer information, an information information on rollover, conduct the called single-vehicle crashes because program was not sufficient to address tests, and determine how best to the crash did not involve a collision the rollover problem. They believe a disseminate the test results to the public with another vehicle. More than 10,000 federal motor vehicle safety standard, by November 1, 2002. Changes or people die each year in all rollover based on a dynamic track test of additions to this program will be crashes. vehicles, is needed. developed if necessary to conform to the • Notwithstanding the recent requirements of these two statutes. The Rollover Resistance Rating is Transportation Recall Enhancement, The rollover information program will an estimate of your risk of rolling over Accountability, and Documentation Act operate just as the current frontal and if you have a single-vehicle crash. It (TREAD) 15 which requires the agency to side NCAP does. New models are does not predict the likelihood of that issue ratings based on a dynamic test selected for testing before the beginning crash. The Rollover Resistance Rating within two years, we believe that of the model year. Selection is based essentially measures vehicle consumer information based on SSF is primarily on production levels characteristics of center of gravity and an appropriate way to proceed at this predicted by the manufacturers and track width to determine how ‘‘top- time to address rollover. Two issues are submitted to the agency confidentially. heavy’’ a vehicle is. The more ‘‘top- involved here: the issue of a minimum Consideration is given also to vehicles heavy’’ the vehicle, the more likely it is standard versus consumer information, scheduled for major changes, or new to roll over. The lowest rated vehicles and the issue of dynamic testing versus models with specific features that may (1-star) are at least 3 times more likely a static metric. Both of these issues were affect their SSF’s. The vehicles chosen to roll over than the highest rated for NCAP testing will be obtained and vehicles (5-stars). addressed at length in the RFC. • We agree that it would be desirable to measured by NHTSA, as the vehicles The Rollover Resistance Ratings of have a standard to address a safety issue become available. Vehicles are obtained vehicles were compared to 220,000 as significant as rollover resistance. with popular equipment, typical of a actual single-vehicle crashes, and the However, as explained in the RFC, rental fleet, and the equipment with ratings were found to relate very closely NHTSA previously decided not to set a to the real-world rollover experience of vehicle rollover standard at a level that 16 Denial of the Wirth petition, 52 FR 49033 vehicles. would effectively force nearly all light (December 29, 1987). 17 Termination to establish a minimum vehicle 19 The manufacturer pays for the vehicle and the trucks to be redesigned to be more like standard for rollover resistance based on TTR or test, however, actual vehicle leasing and testing is CSV, 59 FR 33254 (June 28, 1994.) done by a testing laboratory under contract to 15 P.L. 106–414, November 1, 2000. 18 P.L. 106–346, October 23, 2000. NHTSA.

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00017 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3404 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

• While the Rollover Resistance Issued on: January 8, 2001. questioned the use of linear models of Rating does not directly predict the risk Stephen R. Kratzke, summarized data, and recommended logistic models of individual crash events as an of injury or death, keep in mind that Associate Administrator for Safety improvement (their goals 1 and 2). In Performance Standards. rollovers have a higher fatality rate than response, we have made the suggested other kinds of crashes. Appendix I: Statistical Analysis in corrections, used updated VIN-decoded data, Remember: Even the highest rated vehicle Response to Comments added a year of data (the 1998 calendar year can roll over, but you can reduce your chance data are now available for all six states used of being killed in a rollover by about 75% Response to Comments of the Alliance of in our original analysis), and refit the model. just by wearing your seat belt. Automotive Manufacturers based on a Study Details on the data definitions are included by Exponent Failure Analysis Associates, Inc. below in ‘‘Available Data,’’ and the results of • Here are the Rollover Resistance titled: The Relative Importance of Factors are described in ‘‘Refitting the Linear Ratings: Related to the Risk of Rollover Among Model.’’ We have also used our data to fit Passenger Vehicles logistic regression models, and these results In A Single-Vehicle Crash, a vehicle are described in ‘‘Fitting Logistic Models.’’ A Background with a rating of: comparison of the two approaches is The agency has proposed expanding the provided in ‘‘Comparing the Models.’’ Five Stars ★★★★★ New Car Assessment Program (NCAP), which Our logistic models produced results that tests vehicle performance in front and side were similar to those produced by our linear Has a risk of rollover of less than 10% crashes, to include information on rollover model of summarized data and to the logistic Four Stars ★★★★ resistance. We proposed a rollover metric for models described in the Exponent report Has a risk of rollover between 10% and consumer information based on the Static (which were based on a slightly different Stability Factor (SSF) and described the group of states, calendar years, and 20% approach in a Request for Comments, Notice explanatory variables). That is, the choice of Three Stars ★★★ for Rollover NCAP (‘‘the Notice,’’ docket model form and data source do not affect our NHTSA 2000–6859, item 1, June 1, 2000). essential conclusion: the SSF is strongly Has a risk of rollover between 20% and The Appendix to the Notice described a related (both in terms of statistical 30% statistical analysis of four years of data (1994 significance and magnitude of effect) to Two Stars ★★ to 1997) from six states (Florida, Maryland, rollover risk. However, there are some Missouri, North Carolina, Pennsylvania, and differences among the models in the Has a risk of rollover between 30% and Utah), and we provided more details of the estimated sensitivity of rollover risk to 40% analysis (definitions, programming changes in the SSF. One Star ★ statements, and computer output) in another Where we disagree most with the Exponent submission to the Rollover NCAP docket report is in the interpretation of the results. Has a risk of rollover greater than 40% (item 4). The Alliance of Automobile The authors of the Exponent report argue that As part of these ratings, the agency also Manufacturers (‘‘the Alliance’’) reviewed the the SSF plays a smaller role in rollover has decided to note vehicles that are Notice and the supplemental material and causation than do driver and other road-use submitted their comments to that docket factors (their goals 2 and 4). Goal 2 (gauging equipped with ‘‘electronic stability (item 25). the relative strength of the SSF and non- control’’ technology, which may reduce Appendix 4 of their comments is a paper vehicle factors) is so important to the authors the risk of a vehicle getting into an prepared for the Alliance by Exponent that they used it as the title of their report. incipient rollover situation. Failure Analysis Associates, Inc. (‘‘the We believe that our analysis indicates that Appendix II contains a preliminary Exponent report’’) on The Relative the SSF is very important in describing Importance of Factors Related to the Risk of rollover risk, as measured by the fit of each list of vehicles we will measure and for Rollover Among Passenger Vehicles (Alan C. model, the significance of the coefficient of which we will report SSF and star Donelson, Farshid Forouhar, and Rose M. the SSF term, and the magnitude of the ratings. The vehicles will be tested as Ray, in a paper dated August 30, 2000). The coefficient of the SSF term. We do recognize they become available to the test Exponent report critiqued our linear that driver and other road-use variables are facility. As of today 24 vehicles have regression analysis of the summarized crash also important. Federal, state, and local been tested; the results are available data and suggested an alternative approach education and enforcement programs are all from the Auto Safety Hotline (888–327– based on logistic regression analysis of aimed at the vulnerability of road users to 4236) or on the NHTSA website at individual crash events. This paper is a human error, and we recognize that the comparison of the two approaches (the linear driver plays a large role in causing or www.nhtsa.dot.gov. The remainder of model from summarized data and the logistic avoiding crashes. However, what we set out the test results and star ratings for the model of individual crash events) in response to address in the Notice is whether the SSF 2001 model year will be available by to those comments. provides information that is useful to April 30, 2001. consumers—information they can use in Overview selecting a vehicle, deciding whether to use V. Rulemaking Analyses and Notices The Exponent report listed four goals for seat belts and child seats, and adapting their Executive Order 12866 their study (page 4 of that report), and we driving style to a new vehicle. We describe will address their conclusions in our this point in more detail below, in This notice was not reviewed under response. The four goals were as follows: ‘‘Interpreting the Analytical Results,’’ using Executive Order 12866 (Regulatory (1) ‘‘To evaluate the statistical study an example based on the relationship Planning and Review). NHTSA has offered by NHTSA as a basis for comparative between crash severity, belt use, and injury ’ratings’ [emphasis in original] of rollover severity. analyzed the impact of this decision and risk,’’ In summary, we believe that our statistical determined that it is not a ‘‘significant (2) ‘‘To gauge the strength of SSF as a models (both the linear model of summarized regulatory action’’ within the meaning predictor of rollover relative to the influence data and the logistic models of individual of Executive Order 12866. The agency of non-vehicular factors,’’ crash events) and the statistical models anticipates that providing information (3) ‘‘To quantify the relationship between offered in the Exponent report support our on rollover risk under NHTSA’s New SSF and risk of rollover after adjusting for the conclusion that the SSF is a useful measure Car Assessment Program would impose influence of non-vehicular factors,’’ and of rollover risk that will help the consumer no regulatory costs on the industry. (4) ‘‘To estimate the magnitude and choose a new vehicle and use it wisely. reliability of apparent changes in rollover Authority: 49 U.S.C. 322, 30117, and risk with changes in SSF.’’ Available Data 32302; delegation of authority is at 49 CFR The Exponent report offered three The analysis described in the Notice was 1.50 and 49 CFR 501.8. corrections to our vehicle group definitions, based on single-vehicle crashes, which we

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00018 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3405 defined to exclude crashes with another first harmful event in the crash. We have and DRINK. Speed limit is not reported in motor vehicle in transport or with a made some use of all eight states in this New Mexico, so we defined FAST based on nonmotorist (such as a pedestrian or updated analysis, but most of the analysis is the roadway function class after reviewing pedalcyclist), animal, or train. We eliminated based on the six states with the best rollover the relationship between these two variables vehicles without a driver and all vehicles reporting. These are the six states that were among New Mexico cases in the 1994–1998 that were parked, pulling a trailer, designed the basis for the analysis described in the Fatality Analysis Reporting System (FARS) for certain special or emergency uses Notice. data. We assumed, based on our review of the (ambulance, fire, police, or military), or on an For this analysis, we used the SDS data FARS data, that (1) interstate and rural emergency run at the time of the crash. Our and the VIN-decoded data available on arterial roads had a speed limit of at least 55 only criterion for including a vehicle model NHTSA’s Research and Development Local mph, (2) local roads and urban arterial roads, in the analysis was a reliable measure of the Area Network ( LAN). The National Center collectors, and ramps had a speed limit of no SSF. The 100 vehicle groups we identified for Statistics and Analysis (NCSA, an office more than 50 mph, and (3) the speed limit were described in the Notice, and the in R&D) recently rebuilt the 1997 VIN files was unknown for all other roads. RURAL was definitions for these groups were included in for Maryland and Missouri, and the numbers unavailable for two states (Maryland and another submission to the same docket (item of relevant cases differ slightly from those Missouri), BADROAD was unavailable for 4). reported in the Notice. The major changes two states (Missouri and Pennsylvania), Exponent reviewed this information and were a slightly more-conservative approach NOINSURE was unavailable for three states pointed out three errors in the specifications to dealing with mistakes in VIN transcription (Maryland, North Carolina, and Utah), and of the vehicle groups (page 37). First, vehicle and some additional vehicle-make codes. We NUMOCC was unavailable for Missouri group 65 should have been defined as model also expanded somewhat our definition of (where uninjured passengers need not be years 1990–1995 (not 1988–1996). Second, ‘‘rollover’’ in North Carolina (adding reported). vehicle group 66 should have been defined information from the four impact-type as model years 1996–1998 (not 1997–1998). variables), which increased the number of Refitting the Linear Model And third, vehicle group 91 should have rollovers in that state over what was reported We refit the linear model using the included model code ‘‘SKI’’ (not ‘‘SCI’’), as in the Notice. The number of relevant approach described in the Notice. There were defined by the output from The Polk vehicles identified for each state and 241,036 single-vehicle crashes available for Company’s PC VINA software (PC VINA calendar year are shown as Table 6. Note that this analysis (that is, involving a vehicle in for Windows User’s Manual, October 20, Ohio reported a relatively small percentage of one of the 100 vehicle groups, occurring 1998). We also found a typographical error in VINs in 1998 (about 29 percent of vehicles between 1994 and 1998, and occurring in the the specification of vehicle group 79: the had a VIN on the electronic file), so case six states we studied in preparing the Notice number of drive wheels should have been counts for the vehicles relevant to this study (Florida, Maryland, Missouri, North Carolina, specified as ‘‘not equal to 4’’ (rather than are low. Our analysis is not too sensitive to Pennsylvania, and Utah), and 48,996 of these ‘‘equal to 4’’). We corrected these mistakes in missing VIN information because it is based (20.33 percent) involved rollover. We the list and computer programs, and the on internal comparisons of the crash data eliminated the 1998 Pennsylvania data corrected list of vehicles is included here as (specifically, on rollovers per single-vehicle because CURVE and GRADE are not available Tables 1 through 4. crash); this would not be the case if we were on the electronic file, and this left 227,194 Our understanding of some important basing our analysis on comparisons with an single-vehicle crashes, of which 45,880 differences in state crash reporting are external source, such as rollovers per (20.19 percent) involved rollover. included in Table 5. The Notice described registered vehicle. We summarized the data for each vehicle our criteria for including a state in the We added a calendar year of data (1998) for group in each state, which produced 599 analysis, which were as follows: the six states used in the analysis described summary records (there were no reported (1) Data availability (the state must in the Notice. However, Pennsylvania no single-vehicle crashes involving vehicle participate in the agency’s State Data System longer includes on the electronic file some group 54 in Utah). As with the earlier (SDS) and have provided the 1997 data), environmental variables that we need for this analysis, we eliminated any summary record (2) VIN reporting (the vehicle identification analysis (specifically, CURVE and GRADE), that was based on fewer than 25 cases number (VIN) must be coded on the so we could not use the 1998 Pennsylvania because we thought estimates based on electronic file), and data in the analysis. The variables available smaller samples were too unreliable. This left (3) Rollover identification (we must be able for this analysis are shown as Table 7. We us with 518 summary records, representing to determine whether a rollover occurred, calculated the SSF to two decimal places the experiences of 226,117 single-vehicle regardless of whether it was a first or (with observed values between 1.00 and crashes, including 45,574 (20.16 percent) subsequent event in the crash). 1.53), we defined NUMOCC as the count of rollovers. Figure 1 shows the rollover rate Six states (Florida, Maryland, Missouri, occupants in each vehicle, and we defined all (rollovers per single-vehicle crash) as a North Carolina, Pennsylvania, and Utah) met the other road-use factors as dichotomous function of the SSF plotted for each of the all three criteria. Two states (New Mexico variables (with ‘‘0’’ coded for ‘‘no,’’ and ‘‘1’’ 100 vehicle groups. These data have not been and Ohio) met two of the three criteria; these coded for ‘‘yes’’). adjusted for differences in vehicle use or states participate in the SDS and the VIN is All eight states reported the following data: state reporting practices, but they do show a available on the electronic file, but rollovers ROLL, SSF, DARK, STORM, FAST, HILL, strong tendency for lower rollover rates with are identified only if they are reported as the CURVE, BADSURF, MALE, YOUNG, OLD, higher values of the SSF.

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00019 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3406 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

BILLING CODE 4910±59±C for all crashes in Florida, Missouri, North driving situation (DARK, FAST, and CURVE), We used the 1994–1998 General Estimates Carolina, Pennsylvania, and Utah). These are three variables that describe the driver System (GES) for a comparison with the six- the fourteen variables we used in the earlier (MALE, YOUNG, and DRINK), and all five state rollover rate for the study vehicles as a analysis (described in the Notice), plus the state dummy variables. The F-statistic for the group. The five years of GES data include variable OLD. model as a whole was 311, and the 9,910 sampled vehicles that we identified as We ran the stepwise linear regression probability of a value this high by chance being in one of the 100 vehicle groups (based analysis against these 518 summary records alone is less than 0.0001. The model had an on decoding the VIN with the PC VINA to describe the natural logarithm of rollovers R2 of 0.88 and the coefficient of the SSF term software for those states that include the VIN per single-vehicle crash, which we call (¥3.3760) was highly significant; more on their police reports) involved in a single- LOGROLL, as a function of a linear details on the fit of the model are included vehicle crash, and 2,377 of these rolled over. combination of the explanatory variables. (To as Table 8b. Note that adding the road-use Weighting the GES data to reflect the sample avoid losing information on vehicle models variables increased both the model R2 (from scheme (but not adjusting for missing VIN with a low risk of rollover, we set ROLL to 0.73 to 0.88) and the absolute value of the data) produces estimates of 1,185,474 single- 0.0001 if there were no rollovers represented coefficient of the SSF term (from ¥2.8634 to vehicle crashes per year, of which 236,335 by the summary record.) We used the option ¥3.3760). That is, the effect of the SSF on (19.94 percent) involved rollover. That is, the that gives more weight to data points that are rollover risk is estimated to be even greater six states in our study have a rollover rate for based on more observations, so vehicle after adjusting for differences in road use. police-reported crashes that is essentially the groups with more crashes count for more in We used the results of the model to adjust same as the national estimate produced from the analysis. Each data point was weighted the observed number of rollovers per single- the GES data (with the qualification that the by the number of single-vehicle crashes it vehicle crash to account for differences GES estimate is based on data from just those represented, but the weighting was capped at among vehicle groups in their road-use states that include the VIN on the police 250. That is, data points based on more than characteristics in single-vehicle crashes. For report). 250 observations were weighted by 250. Our each of the 518 summary records, we used We defined the dependent variable ROLL rationale was that we wanted the model to the regression results and the typical road as the fraction of single-vehicle crashes that fit well across the full range of SSF values, use to estimate what LOGROLL would have involved rollover. The independent so we did not want to over-weight the data been if road use for that vehicle group had (explanatory) variables in the six-state for the most-common models on the road. been the typical road use observed for all the combined model were those available in all We ran a preliminary model using the SSF vehicles in the study. The approach is the six states. They were expressed as the and the five state dummies to estimate one used in the Notice. We used an fraction of single-vehicle crashes that LOGROLL. The model had an R2 of 0.73, and intermediate step to account for differences involved each of the following ten situations: the coefficient of the SSF term (¥2.8634) was in road use and adjust the data towards the DARK, STORM, FAST, HILL, CURVE, highly significant (the t-statistic indicates average experience for the study vehicles: BADSURF, MALE, YOUNG, OLD, and that the probability that the coefficient is ADJlLOGROLLi DRINK. We also defined dummy variables for really zero is less than 0.0001); the details are =LOGROLLi l l five states (DUMMY FL, DUMMY MD, included as Table 8a. Thus, it appears that BETAlDARK × (DARKi ¥ DUMMYlNC, DUMMYlPA, and the SSF is very useful in understanding MEANlDARK) l DUMMY UT, with Missouri used as the rollover risk. We then performed a stepwise ¥BETAlFAST × (FASTi ¥ baseline case) to capture state-to-state linear regression (using forward variable MEANlFAST) differences in reporting thresholds and selection and a significance level of 0.15 for ¥BETAlCURVE × (CURVEi ¥ definitions. These variables have the value entry and removal from the model) on the MEANlCURVE) ‘‘1’’ if the crash occurred in that state (for six-state data; this is the same approach we ¥BETAlMALE × (MALEi ¥ example DUMMYlMD = 1 for all Maryland used for the analysis described in the Notice. MEANlMALE) crashes), and they have the value ‘‘0’’ The stepwise regression procedure with the ¥BETAlYOUNG × (YOUNGi ¥ otherwise (for example, DUMMYlMD = 0 SSF chose three variables that describe the MEANlYOUNG)

VerDate 112000 23:32 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00020 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm03 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3407

¥BETAlDRINK × (DRINKi ¥ DUMMYlFLi through DUMMYlUTi are The adjusted rollover rate for each vehicle MEANlDRINK) the values of the state dummy variables group is then estimated by: ¥BETAlDUMMY—FL × DUMMYlFL for each summary record (with no more ( ) i ADJlROLL=e ADJlLOGROLL . ¥BETAlDUMMYlMD × DUMMYlMD than one of these equal to ‘‘1,’’ and all i This is our estimate of what the rollover rate ¥BETAlDUMMYlNC × DUMMYlNCi the rest equal to ‘‘0’’), would have been if all vehicle groups were ¥BETAlDUMMYlPA × DUMMYlPAi MEANlDARK through MEANlDRINK ¥BETAlDUMMYlUT × DUMMYlUTi are the average values of the road-use used in the same way, and it reflects the + MEANlDUMMIES, variables observed in the study data average use patterns of all vehicles in the l where: (with MEAN DARK=0.4314, study. The adjusted rollover rates are shown l MEANlFAST=0.4807, in Figure 2. ADJ LOGROLLi is the estimate of what l l LOGROLL would have been for each MEAN CURVE=0.3315, MEAN MALE The average adjusted number of rollovers summary record if all vehicles were used = 0.6276, per single-vehicle crash for all the study MEANlYOUNG=0.3987, and the same way, l vehicles in the six states is 0.1982, which is LOGROLL is the value of LOGROLL MEAN DRINK=0.1509), and i MEANlDUMMIES is the average state essentially the same as the rollover rate in the observed for each summary record, original study data (0.2016) and the rollover BETAlDARK through BETAlDRINK are adjustment in the study data. l rate estimated from the GES data (0.1994) for the coefficients (Beta-values) of the road- MEAN DUMMIES was calculated for these 100 vehicle groups. A linear model fit use variables, DARK through DRINK, these 226,117 single-vehicle crashes that were produced by the model (as from the coefficient of the state dummy through the adjusted data is described by the shown in Table 8b), variables and the number of cases in equation: BETAlDUMMYlFL through each state as follows: LOGROLL = 2.5861—3.3760 × SSF. l l × BETA DUMMY UT are the (1.2253 number of Florida cases The model has an R2 of 0.85, and the coefficients of the state dummy +0.6933 × number of Maryland cases coefficient of the SSF term was highly variables, DUMMYlFL through +0.0000 × number of Missouri DUMMYlUT, that were produced by +0.6969 × number of North Carolina cases significant. Details on the fit of the the model, +1.2449 × number of Pennsylvania cases model through the adjusted rollover rates DARKi through DRINKi are the values of +0.8622 × number of Utah cases) are included as Table 8c. the road-use variables observed for each /Total number of cases summary record, =0.8019,

BILLING CODE 4910±59±C The estimated rollover rates for the SSF associated with halving the number of Exponentiating both sides of the equation values between 0.95 and 1.55 are shown in rollovers per single-vehicle crash is estimated produces an estimate that the number of Table 19 in the column labeled ‘‘Model 1,’’ as 0.21. For example, the number of rollovers rollovers per single-vehicle crash is and the estimates for the observed range (SSF per single-vehicle crash under average approximated by the curve: values from 1.00 to 1.53) are shown as Figure conditions is estimated as: ¥ 2. This model form has very useful ROLL = 13.28 × e( 3.3760 × SSF). 0.44 for a SSF of 1.01 properties. The increase in the SSF that is 0.23 for a SSF of 1.22, and

VerDate 112000 23:26 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00021 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm03 PsN: 12JAR10 3408 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

0.11 for a SSF of 1.43. model of state data, and the results are T = LOGODDSALL ¥ (BETASSF × Thus, rollover risk drops by a half when the included in Relationships between Vehicle MEANSSF). SSF increases from 1.01 to 1.22, and it drops Size and Fatality Risk in Model Year 1985– The rollover rate for all the vehicles included in half again when the SSF increases from 93 Passenger Cars and Light Trucks (Charles in the Florida study was 0.2044 and their 1.22 to 1.43. J. Kahane, Evaluation Division, Office of average SSF was 1.2894, which means that: Plans and Policy, National Highway Traffic The SSF is both highly significant in the T = log (0.2044/0.7956) ¥ (¥4.3908 × Safety Administration, DOT HS 808 570, e model and very important in describing 1.2894) and January 1997). The results of the six-state rollover risk (the estimated rollover risk T = 4.3025 at the average rollover odds and combined model are shown as Table 17a. The increases by a factor of 6.0 over the observed SSF values. range of the data, from a SSF of 1.00 to 1.53). model fits the data well, and the SSF is We call this specific value of the function T, This means that changes in the SSF (or highly significant in the model (with a chi- ‘‘T0.’’ Then, after controlling for other changes in how vehicles with low SSF values square value of 7,230). The coefficient of the SSF term in the factors, LOGODDSSSF is estimated as: are used) has the potential for large × reductions in rollover risk. logistic model for each state and for the six- LOGODDSSSF = T0 + (BETASSF SSF), state combined model describes the and at SSF=1.00 in Florida, this is calculated Fitting Logistic Models relationship between the rollover rates for as: any two values of the SSF, and we can use The Exponent report questioned the LOGODDS1.00 = 4.3025 ¥ (4.3908 × 1.00), validity of using a linear regression analysis this relationship to estimate the rollover rate under average road-use conditions for each so of summarized data, though they noted the ¥ advantages of this approach for describing value of the SSF. We used the method that LOGODDS1.00 = ( 0.0883). the data. They suggested using a logistic Ellen Hertz described in her study of the ROLL1.00 is estimated from the LOGODDS1.00 regression analysis with the SSF and road- safety effects of vehicle weight. She as: use variables, and they also suggested (as a estimated injury risk based on a logistic ex/(1 + e x), model of state data, and the results are way of dealing with potential cross- where x is the LOGODDS , so the rollover included in A Collection of Recent Analyses 1.00 correlations) an approach that uses crash-risk rate at a SSF value of 1.00 is estimated as of Vehicle Weight and Safety (T.M. Klein, E. scenarios in place of the road-use variables. 0.4778 rollovers per single-vehicle crash. The Hertz, and S. Borener, Mathematical Analysis They provided results from the states they rollover rate for all other values of the SSF Division, National Center for Statistics and used in their analysis, and we did a similar can be estimated using: analysis of the eight states available to us. Analysis, Research and Development, ((SSF ¥ 1.00)×BETASSF) × National Highway Traffic Safety ODDSSSF = e ROLL1.00/ The data for two states, New Mexico and ¥ Ohio, were not combined with the data from Administration, DOT HS 807 677, May 1991). (1 ROLL1.00) the other six states because a rollover is We defined: and reported in New Mexico or Ohio only if it is BETASSF = the coefficient of the SSF term ROLLSSF = ODDSSSF/(1 + ODDSSSF). considered to have been the first harmful in the logistic model for a state, We used this approach for each state and event in the crash. However, we did look ROLLSSF = the rollover rate at a specific value for the six-state combined model. The briefly at these data because we were curious of the SSF, and average rollover rate and SSF for each state about how the rollover definition affects the ODDSSSF = the odds of rollover at a specific and for the six-state combined data are analysis. We wanted to see how the risk of value of the SSF. shown in Table 18, along with the estimated a rollover occurring as the first harmful event We choose a SSF of 1.00 as the basis for the rollover rates for a SSF of 1.00. For example, in a single-vehicle crash varies as a function calculations. The relationship between the rollover risk for the six-states combined of the SSF as reported in these two states. ROLL1.00 and any other ROLLSSF can be is estimated as 0.4031 at an SSF of 1.00, and We ran a logistic regression analysis for calculated for each state as follows: it is shown in the column for the results of each state to model rollover as a function of the models based on ‘‘individual variables.’’ ROLLSSF = ODDSSSF / (1 + ODDSSSF) the SSF and the road-use variables. For each (The results of the models based on ‘‘crash where state, we used the explanatory variables scenarios’’ are described below.) The results ((SSF ¥ 1.00)×BETASSF) × available for the linear regression analysis ODDSSSF = e ROLL1.00 for each value of the SSF are shown in the ¥ plus other variables that were available in / (1 ROLL1.00). column labeled ‘‘Model 2’’ in Table 19. each state, as described in Table 7. The fits The results of the logistic analysis of the As a check of the six-state combined of the models are summarized in Tables 9a Florida data are shown in Table 9a, including model, we calculated the average rollover through 16a. Each model seems to fit the data an estimate that: risk for each value of the SSF based on the well. The coefficient of the SSF term varies BETASSF = (¥4.3908), individual state models. For example, we from (¥3.0800) in North Carolina to calculated the average rollover rate for a ¥ so all we need for rollover rate estimates ( 4.3908) in Florida. The values for New across the range of the SSF is an estimate of vehicle with a SSF of 1.00 by taking the Mexico (¥3.0809) and Ohio (¥4.3642) fall in average of the estimates for these six states ROLL1.00 in Florida. We estimated ROLL1.00 this range, which suggests that the choice using the following approach. For each state, (that is, Florida, Maryland, Missouri, North between ‘‘all rollovers’’ and ‘‘first harmful we defined: Carolina, Pennsylvania, and Utah), weighted event rollovers’’ may not be critical for a by the size of each state (as measured by the basic understanding of the sensitivity of ODDSALL = odds of rollover for the study number of single-vehicle crashes involving rollover risk to the SSF (though the choice is vehicles as a group, any study vehicle in each state). The result important in determining the absolute level LOGODDSALL = the natural logarithm of is an estimated risk of 0.4101 rollovers per of rollover risk). In all cases, the coefficient ODDSALL, and single-vehicle crash for an SSF of 1.00, and of the SSF term was highly significant; the MEANSSF = the average SSF for the study the same procedure was applied to each vehicles. probability of a chi-square this large by value of the SSF from 0.95 to 1.55. The chance alone (the smallest chi-square values The model says that: results are shown as the column labeled were 209 for New Mexico and 416 for Utah) LOGODDS = T + (BETASSF × SSF), ‘‘Model 3’’ in Table 19. was estimated as less than 0.0001. where The Exponent report also suggested using We then combined the data from the six T = a linear function of the explanatory an approach they called a ‘‘crash scenario states that have the best rollover reporting variables, analysis’’ to address possible interactions (that is, data that were not limited to first- among the explanatory variables. This idea is harmful-event rollovers) and used them and we solved for the ‘‘average’’ value of T interesting and conceptually simple. The together in a logistic model, using the such that: single-vehicle crashes from each state are × explanatory variables they have in common. LOGODDSALL = T + (BETASSF MEANSSF). categorized into cells defined by the possible We used the approach Charles Kahane That is, we assumed that the results of the combinations of the road-use variables. For described in his study of the safety effects of logistic model apply to the average rollover example, the Florida logistic analysis used 14 vehicle size. He used dummy variables to rate and SSF value for the vehicles as a road-use variables: DARK, STORM, RURAL, capture reporting differences in a logistic group, and this means that: FAST, HILL, CURVE, BADROAD, BADSURF,

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00022 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3409

MALE, YOUNG, OLD, NOINSURE, DRINK, probability of a greater chi-square value was state combined data to model rollover risk as and NUMOCC. NUMOCC is the count of less than 0.20. We reasoned that using a large a function of two variables: the SSF and occupants in each vehicle, and the other 13 number of variables to define the crash SCENRISK. The fits of the models are variables take on the value ‘‘0’’ or ‘‘1’’ scenarios would tend to produce many cells summarized in Tables 9b through 17b. Each (indicating ‘‘no’’ or ‘‘yes’’). This produces a with small sample sizes, and that the table shows the number of crash scenarios large number of possible combinations of the variables with smaller chi-square values with at least 25 observations and the total variable values: would be missed less. A review of Tables 9a number of crashes in these more-frequent 213 × the number of levels of NUMOCC. through 16a shows that this eliminated only scenarios. Each model seems to fit the data one variable in Florida (DARK), but it Converting NUMOCC into a dichotomous well. The coefficient of the SSF term in the eliminated five variables in Utah (STORM, variable (for example, one that identifies crash-scenario logistic model for each state vehicles with at least three occupants) yields HILL, MALE, YOUNG, and OLD). Second, we converted NUMOCC into MANYOCC (with describes the relationship between any two 14 dichotomous variables, which means 214 values of the SSF. We applied the approach combinations of these variables, or 16,384 value ‘‘1’’ meaning three or more occupants, and ‘‘0’’ meaning one or two occupants). we used for the individual-variable logistic cells for the various crash scenarios. In Again, the purpose of this was to reduce the model to estimate the rollover risk for each practice, not all combinations will occur number of cells with small sample counts, value of the SSF and to combine the values (there were 2,034 non-zero cells in the while retaining the essential information. across states. The rollover rates at a SSF of Florida data), and some non-zero cells have Third, we tabulated the number of single- 1.00 are shown in Table 18, and the very low counts (there were 267 cells in the vehicle crashes (SVACCS) and the number of estimated rollover rates as a function of the Florida data with at least 25 observations). rollovers (ROLLACCS) for each combination SSF are shown in Table 19. The column The rollover rate for each cell can be of DARK, STORM, RURAL, FAST, HILL, labeled ‘‘Model 4’’ shows the results for the calculated from these data, and this is a CURVE, BADROAD, BADSURF, MALE, measure of the risk associated with that six-state model, and the column labeled YOUNG, OLD, NOINSURE, DRINK, and scenario. This rate can be used in place of all ‘‘Model 5’’ shows the average of the MANYOCC that had been selected for the road-use explanatory variables (for individual models for the six states. Note that inclusion in each state. We eliminated any example, in place of the 14 original road-use the individual-variable and the crash- combination (that is, any crash scenario) with variables in the Florida analysis). The scenario approaches produce very similar fewer than 25 observations. The results are Exponent report recommends a refinement to numbers. This is consistent with the results summary data describing the experience of this calculation so that the scenario-risk reported in the Exponent report (and all vehicles in each crash scenario. Fourth, variable for each specific vehicle reflects the summarized in Table 20, using the we merged the crash-scenario summary data rollover rate for all other vehicles in its cell. standardized estimates of the coefficients). for each state back onto the original data (that For example, in a cell with 100 vehicles and is, the data for each individual single-vehicle 20 rollovers, the scenario-risk variable Comparing the Models crash), so that each crash was linked to a (SCENRISK) will be calculated as: The rollover rates estimated across the count of the total number of single-vehicle ¥ range of SSF values for the six states 20/(100 1) for each nonrollover vehicle crashes and the total number of rollovers that and as occurred in its crash scenario (its cell). We combined are shown in Table 19 for all five ¥ ¥ defined the scenario-risk variable, statistical models (the linear model of (20 1)/(100 1) for each rollover vehicle. summarized data and the four versions of the Using a crash-scenario variable is an SCENRISK, as the rollover rate for all other vehicles in that crash scenario in that state. logistic model), and the estimates for the interesting idea, even though the analytical observed values of the SSF are plotted in results in the Exponent report seem to show The calculation was as follows: ¥ Figure 3. The five models are as follows: that the individual-variable and crash- SCENRISK = (ROLLACCS ROLL)/ ¥ scenario logistic models produced very (SVAACCS 1). Model 1: Linear model of the summarized similar results. The standardized estimates Recall that ROLL is coded as ‘‘1’’ if the data, for the coefficients of the SSF term produced vehicle rolled over and ‘‘0’’ if it did not, so Model 2: Logistic model of the six-state by the two approaches (and our own results) this equation produces an estimate of the combined data, based on individual are shown in Table 20. We attempted to rollover rate for all vehicles in the crash variables, duplicate the crash-scenario analysis based scenario except for the one case under study; Model 3: Average of the logistic models for on the description provided in the Exponent this was the method recommended by the the six states, based on individual report. The concept seems clear and logical, Exponent report. This scenario-specific variables, and we made the following decisions in rollover rate is calculated for each vehicle on Model 4: Logistic model of the six-state implementing it for this analysis. First, we the file and is then available as an combined data, based on crash scenarios, reviewed the output from the logistic explanatory variable for a logistic model. and regression on individual variables for each We ran a logistic regression analysis Model 5: Average of the logistic models for state and selected those for which the against the data for each state and for the six- the six states, based on crash scenarios.

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00023 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3410 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

BILLING CODE 4910±59±C Model 5 = 0.3929 (average of the logistic more negative in the presence of wheelbase, There are important similarities between the models for the six states with crash and wheelbase itself had a negative estimates produced by the two approaches: scenarios), coefficient in each model in which it was both the linear model of summarized data The results of the four logistic models are included. and the logistic models suggest a strong almost indistinguishable in Figure 3: the Adding wheelbase seemed to produce relationship (in terms of statistical crash-scenario approach produces results results closer to those in the Exponent report. significance and in terms of the magnitude of that are only slightly different from the That report does not include the estimates of the effect) between the SSF and rollover risk. individual-variable approach (the former are the variable coefficients, but it does include The average slope across the range of the a little lower at a SSF of 1.00 and little higher the standardized coefficients. These are observed SSF values (from 1.00 to 1.53) at an SSF of 1.53), and the average of the shown in our Table 20, along with the shown in Figure 3 is ¥0.713 for the linear logistic models for the six states produces corresponding values from our analysis. For model; the logistic models produce estimates results that are only slightly different from example, when we ran the logistic regression of a slightly smaller effect, with average the logistic model of the six-state combined analysis on the Florida data and used slopes between ¥0.598 and ¥0.555. Both data (the former are a little lower at a SSF wheelbase as one of the explanatory ¥ types of models agree in estimating a large of 1.00 and little higher at an SSF of 1.53). variables, we obtained values of ( 0.392) ¥ increased risk for vehicles with a low SSF. The results of our logistic analyses seem to and ( 0.374) for the standardized The four logistic models produce very similar differ only slightly from those described in coefficients from the individual-variable and results, and each suggests that rollover risk the Exponent report, and much of the crash-scenario models, respectively. These difference may be the result of our decision are higher than the values we obtained is very sensitive to the SSF (only slightly less ¥ ¥ so than estimated from the results of the to omit wheelbase from the models. We did without wheelbase, ( 0.349) and ( 0.327), and they are very close to the values in the linear model of the summarized data). not include wheelbase as an explanatory Exponent report, (¥0.383) and (¥0.381). Figure 3 shows that the greatest absolute variable because we could not identify any Adding wheelbase to our models produced differences in the rollover rate estimates are physical reason for an effect on rollover risk. higher estimates of the coefficient for the SSF at the lowest values of the SSF. The values However, we reran each analysis with the addition of wheelbase to test the sensitivity term and higher estimated rollover rates for of the rollover rate estimated for a SSF of of the results to this decision. In every case, vehicles with lower SSF values. For example, 1.00 were as follows: adding wheelbase to the model produced a the six-state models that included wheelbase Model 1 = 0.4551 (linear model of the higher estimate of the effect of the SSF on produced estimates that the coefficients of summarized data), rollover risk and a higher estimate of rollover the SSF term are (¥3.9525) and (¥3.7918) Model 2 = 0.4101 (logistic model of the six- risk for the lowest values of the SSF. This and the estimated rollover rates for a SSF of state combined data with individual occurred for all 18 models (those estimated 1.00 are 0.4338 and 0.4228 for the individual- variables), using both the individual-variable and crash- variable and crash-scenario approaches, Model 3 = 0.4031 (average of the logistic scenario approaches for each of the eight respectively. models for the six states with individual states and for the six-state combined data), There is also one important difference variables), despite a negative value for the coefficient of between the linear analysis of summary data Model 4 = 0.3999 (logistic model of the six- the wheelbase term in each model. That is, and the logistic analysis of individual state combined data with crash scenarios), the coefficient of the SSF term was negative crashes. We limited the summary data to and in each of the original models, it became those based on at least 25 observations and

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00024 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3411 we capped the weighting at 250 to avoid estimate of 171,284 drivers involved each carefully and when fully sober), but others over-emphasizing the more-popular vehicles. year. An estimated 11,569 of these were are beyond the control of the driver (roads However, the logistic regression analysis on seriously injured (that is, they died or are curved, through rural areas, and with individual crashes uses all observations received an injury rated as three or higher on speed limits of 55 mph so traffic can move equally. When we removed the two the Abbreviated Injury Scale). The overall efficiently through all parts of the country). thresholds from the linear analysis, we risk of serious injury was 6.75 percent, but Young drivers gain experience through obtained slightly lower estimates of the effect the risk varied greatly as a function of the driving, and they eventually become mature of the SSF on rollover risk, and the change in vehicle velocity during the impact drivers; in the meantime, they also benefit relationship between the adjusted rollover (that is, the delta V). For delta V less than 10 from more-stable vehicles. It is difficult to see rates and the SSF is described by: mph, the risk of serious injury was 0.76 how Exponent’s the low-risk scenario could (¥ 3.2356 × SSF) ROLL = 10.99e percent. If all 171,284 drivers in these towaway be used as an alternative to the SSF as the This model produces an estimate of 0.4323 basis for a rollover safety program. rollovers per single-vehicle crash at an SSF crashes had been injured at the same rate as those in the lowest delta V range, we would The approach described in the Exponent of 1.00, which is closer to the estimates from report (comparing the risk associated with our logistic models (and essentially the same have seen: × the SSF to all the risks associated with road- as the estimates from the logistic models that 0.0076 171,284 = 1,302 serious injuries use factors) would suggest, in our example include wheelbase as an explanatory among unbelted drivers in frontal crashes. variable). Half of these (601 serious injuries) could based on NASS data, that reducing delta V have been prevented if the drivers had used should be a higher safety priority than Interpreting the Analytical Results a lap-and-shoulder belt. Thus, we have the increasing belt use. (To use an extreme Many of the comments in the Exponent following: example to make a point, using the approach report reflect an interest in evaluating the 171,284 serious injuries among unbelted described in the Exponent report for a study relative strength of the driver and vehicle drivers, of which 1,202 would have of air crashes would suggest that preventing contributions to rollover risk. We agree that occurred if delta V was low, of which gravity is more important than regular this is an interesting question, but it is not 601 would have occurred if belts were maintenance of the airplane.) However, belt the one we set out to address. Our used. use programs have been successful because perspective is that of a person choosing a the remedy is simple and cost-effective and new vehicle who wants to know how his According to the logic proposed by Exponent, we would interpret the results as because the importance of delta V does not choice of vehicle will affect his risk of being reduce the importance of belt use in involved in a rollover. We are interested in follows: 99.30 percent of serious injuries are preventing injury. We believe a similar eliminating the confounding effects of road argument can be made for focusing on the use so we can isolate the effect of the vehicle attributable to high crash speeds, and SSF, while agreeing that driver and other on rollover risk. The importance of road-use 0.35 percent are attributable to road-use variables may be the basis for other factors does not preclude a role for vehicle- neglecting to use belts. specific information. Clearly this is nonsense. Belt use will safety improvements. Also, a factor can be important without prevent serious injury even among those in Conclusion suggesting an easy remedy. Consider two higher-speed crashes (half of the 11,569 The Exponent report acknowledged the factors that increase the risk of rollover given serious injuries that did occur among a single-vehicle crash: driver age unbelted drivers at any crash speed could potential advantages of multiple linear (specifically, the effect of young, have been prevented by belt use, for a analysis, and their recommendation is inexperienced drivers) and curved roads. We reduction of 5,784 serious injuries from belt relevant here: do have some influence over their effect on use). More importantly, belts offer a practical Multiple regression analysis can have some rollover risk: better driver training and better solution, while there is no practical way to value as an explanatory tool for describing road design can help reduce rollovers even reduce all crash speeds to less than 10 mph. factors related to vehicle rollover. Linear among young drivers on curved roads. Note that this is comparable to the regression analysis, however, must only be However, some additional risk is a given for approach that the Exponent report used in used in this heuristic way and only when people who are still gaining on-road arguing that the value of the SSF in prior research has demonstrated that linear experience, and curved roads are a necessity understanding rollover risk was in the range regression produced essentially the same in many places. So, while driver and other of 3–8 percent. They estimated the relative results as did a rigorous and valid road-use factors are important to risk of the lowest-risk scenario, estimated statistical analysis. [page 28] understanding rollover risk, this is not the how many rollovers could be prevented if all Table 19, Figure 3, and the sensitivity same as saying that all rollovers can be single-vehicle crashes occurred with the risk analyses described above suggest that the prevented by driver and other road-use of the lowest-risk scenario, and relegated the remedies. Vehicle design plays an important importance of the SSF to a fraction of the linear and logistic regression approaches role in understanding and mitigating rollover small amount of risk that remained. The produce essentially the same results. The risk even among young drivers on curved lowest-risk scenario that they use as their Exponent report recommended a logistic roads by making vehicles more-forgiving of standard appears to be (based on the table on approach and concluded that the linear driver and road limitations, and our analysis page 31 of their report) crashes that did not approach based on summarized data describes the magnitude of that effect. involve a vehicle defect and that did involve overstated the value of the SSF in Another comparison may help clarify why a mature driver who had not been drinking understanding rollover risk. This does not we believe that the SSF can be useful even or engaged in risky driving, on a straight, seem to be the case. The linear approach though driver and other road-use factors are urban road with a speed limit of 50 mph or produces estimates of rollover risk that are a such valuable predictors of rollover risk. less, and for which the first harmful event little more conservative (in the sense that Using the same approach Exponent used for was a collision with a traffic unit in a single- they are lower) than those from the logistic SSF and other factors involved in rollover, vehicle crash; the bulk of these crashes may models for most observed values of the SSF one can statistically demonstrate that seat be collisions with pedestrians and and for most vehicles on the road today. The belt use is insignificant in preventing injuries pedalcyclists, which would tend to be Exponent report included much lower from a crash. The 1998–1999 National reported because of the injuries to the non- estimates for rollover risk across the range of Automotive Sampling System (NASS) data motorists. SSF values, but this was not a result of the include 7,631 investigated unbelted drivers These are crashes with almost no chance logistic approach. Rather, it was the result of of light passenger vehicles that were towed of rollover, and so they are essentially tying the estimates to the low-risk scenario from a frontal nonrollover crash (Table 22), irrelevant to a rollover-prevention program. (where rollover is unlikely). and weighting these data to reflect the Also note that some of these factors can be sampling plan produces an annual average addressed by the driver (driving more BILLING CODE 4910±59±P

VerDate 112000 23:26 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00025 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm03 PsN: 12JAR10 3412 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00026 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3413

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00027 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3414 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00028 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3415

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00029 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3416 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00030 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3417

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00031 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3418 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00032 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3419

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00033 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3420 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00034 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3421

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00035 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3422 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00036 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3423

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00037 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3424 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00038 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3425

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00039 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3426 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00040 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3427

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00041 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3428 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00042 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3429

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00043 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3430 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00044 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3431

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00045 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3432 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00046 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3433

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00047 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3434 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

BILLING CODE 4910±49±C model, where applicable. In no case will a determined by the test laboratory and will two-wheel-drive measurement be applied to Appendix II: List of Test Vehicles for depend primarily on model availability. a four-or all-wheel-drive variant, or vice The following class abbreviations are used: MY2001 Rollover Resistance Ratings versa. The agency may need to make LPC = light passenger car NHTSA expects to measure the Static substitutions for some of the models listed depending on availability. The list is CPC = compact passenger car Stability Factor and provide rollover arranged largely alphabetically within each MPC = medium passenger car resistance ratings for each of the following vehicle category, and passenger cars are HPC = heavy passenger car model year 2001 vehicles. For pickups and sorted by class according to the SUV = sport utility vehicle SUVs, the agency plans to measure and classifications used in the NHTSA NCAP LT = light truck report separately on both two-wheel-drive frontal and side crash test programs. The and four-or all-wheel-drive variants of each order in which vehicles will be tested will be BILLING CODE 4910±59±C

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00048 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3435

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00049 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 3436 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00050 Fmt 4701 Sfmt 4725 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10 Federal Register / Vol. 66, No. 9 / Friday, January 12, 2001 / Rules and Regulations 3437

BILLING CODE 4910±59±C [FR Doc. 01–973 Filed 1–9–01; 2:33 pm] BILLING CODE 4910±59±P

VerDate 112000 22:28 Jan 11, 2001 Jkt 194001 PO 00000 Frm 00051 Fmt 4701 Sfmt 4700 E:\FR\FM\12JAR10.SGM pfrm02 PsN: 12JAR10