National Highway Traffic Safety Administration

Actions to Reduce the Adverse Effects of Air Bags
FMVSS No. 208


Office of Regulatory Analysis
Plans and Policy

February 1997


Executive Summary

I. Introduction

II. Injury Risk Curves and Protection Reference Values

III. Air Bag Aggressiveness Testing

IV. Synthesis of Data and Benefit Estimates

V. Potential Benefits of Depowering Based on the Holden Bag

VI. Cost

VII. Small Business Impacts


This Final Regulatory Evaluation analyzes the impacts of two alternatives to allow depowering of air bags. They are: 1) changing the chest g's injury criteria of Federal Motor Vehicle Safety Standard No. 208 from 60 chest g's to 80 chest g's for unbelted dummies with air bags, or 2) changing the test requirements for the unbelted condition to a generic sled test with a crash pulse of 125 msec. The intended effect of these alternatives is to allow manufacturers to depower air bags on average by 20-35 percent.

Data are available on one less aggressive depowered air bag, an air bag on a General Motors designed Holden vehicle in Australia. These data are limited to restrained occupants because almost all occupants in Australia use safety belts. The data indicate that these depowered air bags will be more effective than current air bags for restrained occupants for injuries. If the relationship in overall effectiveness of the Holden bag to the U.S. bags for belted occupants for moderate or more severe injuries (AIS 2+) is the same for fatalities, an estimated 643 lives of belted occupants could be saved annually by having depowered air bags like the Holden bag.

The agency has done a significant amount of research and testing of depowered air bags and in this evaluation has estimated the fatality, chest injury, and arm injury impacts of depowering based on this testing. The following estimates are based on research testing and math modelling. Two separate methodologies were used to determine the impact of changes in chest g's on fatalities.

The accompanying Final Rule is part of a large package of rulemakings and efforts to warn occupants of the adverse effects of air bags, reduce these adverse effects and increase safety belt usage. With increasing safety belt use, the disbenefits of depowering for unbelted adults would be reduced, and based on the Holden analysis, the benefits of depowering for belted occupants would increase.

Passenger Side


Assuming no changes are made in where children sit in vehicles, and no changes are made to the current air bags, an estimated 140 children would be killed over the lifetime of one model year's fleet if all vehicles had passenger side air bags (107 forward-facing children and 33 infants in rear-facing child restraints). Under the 80 g's alternative, an estimated 14 children's lives could be saved and under the generic sled pulse alternative, an estimated 47 children's lives could be saved by depowering air bags on the passenger side. If certain models were depowered more than 35 percent, there could be higher benefits for children, but then there would also be more disbenefits. Benefits for rear-facing infants could not be quantified, but are believed to be minimal.

Based on the Holden bag analysis, an estimated additional 172 belted passenger side occupant lives could be saved by depowering, which would partially or fully offset the estimated unbelted passenger side occupant lives that would not be saved.

The test results indicate that there would be a savings in lives for belted passengers (in the range of 1 to 7 under the 80 g's alternative and from 4 to 22 under the generic sled alternative).

If all vehicles had passenger side air bags, over the lifetime of one model year's fleet, at least 7 adult passengers would be killed by current air bags in low-to-moderate speed crashes. The agency believes depowering could eliminate almost all of these fatalities.


The agency believes there would be significant minor to moderate injury reductions with depowered air bags, but cannot quantify them.


If all vehicles had passenger side air bags, over the lifetime of one model year's fleet, an estimated 714 lives of those over 12 years of age would be saved. Under the 80 g's alternative, an estimated 11 to 87 unbelted adult passenger lives would not be saved. Under the generic sled test alternative, an estimated 34 to 280 adult passenger lives would not be saved.

Depowering could result in 28 to 89 AIS 3-5 chest injuries not being prevented under the 80 g's alternative and 90 to 289 AIS 3-5 chest injuries not being prevented under the generic sled pulse alternative compared to current air bags.

Driver Side



If all vehicles had driver side air bags, over the lifetime of one model year's fleet, at least 25 drivers would be killed by current air bags in low-to-moderate speed crashes. The agency believes depowering could reduce a large portion, but not all of these fatalities. The exact number could not be quantified.

Based on the Holden bag analysis, an estimated additional 471 belted driver's lives could be saved by depowering, which would partially or fully offset the estimated driver's lives that would not be saved.

Under the 80 g's alternative, an estimated 1,600 to 2,700 AIS 2-3 arm injuries could be prevented by depowering. Under the generic sled alternative, an estimated 5,100 to 8,800 AIS 2-3 arm injuries could be prevented by depowering.

The agency believes there would be significant minor to moderate injury reductions with depowered air bags, but can not quantify them.


If all vehicles had driver side air bags, over the lifetime of one model year's fleet, an estimated 2,302 lives would be saved. Under the 80 g's alternative, an estimated 4 to 34 belted drivers and 1 to 13 unbelted drivers, for a total of 5 to 47 driver lives would not be saved. Under the generic sled test alternative, an estimated 13 to 110 belted drivers lives and 3 to 41 unbelted driver lives, for a total of 16 to 151 drivers lives would not be saved.

Depowering could result in 53 to 173 AIS 3-5 chest injuries not being prevented under the 80 g's alternative and 171 to 553 AIS 3-5 chest injuries not being prevented under the generic sled pulse alternative compared to current air bags.



Costs are not a significant issue for this rulemaking. Depowering could be accomplished by a simple substitution of a lower powered inflator, which might reduce variable costs by a small amount. It is also possible that costs could be reduced by economies of scale if the generic sled test allows for greater commonality among air bag module components. There would be engineering, development, and testing costs to design, develop, and certify depowered air bags.


Safety Program

In 1991, Congress directed the agency to require air bags for the front outboard seating positions of all passenger cars by model year (MY) 1998 and in all trucks, multi-purpose passenger vehicles (vans, and sport-utility vehicles), and buses with a gross vehicle weight rating of 8,500 pounds or less and an unloaded vehicle weight of 5,500 pounds or less by MY 1999. (Throughout this document, these vehicles are referred to as passenger cars and light trucks).

Frontal impacts are the number one fatality and injury causing mode of crash, resulting in 64 percent of all driver and right-front passenger fatalities and 65 percent of all driver and right-front passenger AIS 2-5 injuries. (AIS 2-5 stands for the Abbreviated Injury Scale levels of moderate to critical injuries). Table I-1 shows the fatalities and moderate-to-critical injuries to driver and right front passengers with known seating positions in passenger cars and light trucks for 1994 in all impacts and in frontal impacts.(1) Safety belts are the most effective safety device known for reducing fatalities and injuries, except for child restraints.

Table I-1

1994 Fatality Estimates

All Impact Modes Drivers Right Front Passengers Total
Passenger Cars 14,610 4,883 19,493
Light Trucks 6,037 1,608 7,645
Total 20,647 6,491 27,138
Frontal Impacts Only Drivers Right Front Passengers Total
Passenger Cars 9,427 2,913 12,340
Light Trucks 4,010 901 4,911
Total 13,437 3,814 17,251


1994 AIS 2-5 Injury Estimates

(NASS-CDS towaway crashes only)

All Impact Modes Drivers Right Front Passengers Total
Passenger Cars 143,663 37,055 180,718
Light Trucks 49,890 7,510 57,400
Total 193,553 44,565 238,118
Frontal Impacts Only Drivers Right Front Passengers Total
Passenger Cars 90,909 25,166 116,525
Light Trucks 33,575 4,683 38,258
Total 124,484 30,299 154,783

Air bags are a supplemental device to safety belts. Air bags also provide benefits for unbelted occupants in potentially fatal frontal crashes.

Consumer demand has led manufacturers to install air bags in nearly every MY 1996 and 1997 passenger car and in most light trucks. As a result, the number of air bag vehicles on the road is increasing dramatically.

The agency has estimated the number of vehicles installed with air bags on the road through the year 2000. Table I-2 shows these estimates as of July 1, 1994, 1996, and 2000.

Table I-2

Projected Air Bag Installations
As of July 1 of the Calendar Year

Driver Side Air Bags 1994 1996 2000
Passenger Cars 20,426,000 36,252,000 65,534,000
Light Trucks 2,952,000 11,484,000 31,908,000
Total 23,378,000 47,736,000 97,442,000
Right Front Passenger Side Air Bags
Passenger Cars 4,507,000 18,361,000 52,225,000
Light Trucks 285,000 3,003,000 19,182,000
Total 4,792,000 21,364,000 71,407,000

The agency has just released an analysis entitled "Fatality Reduction by Air Bags, Analyses of Accident Data Through Early 1996". This analysis contains various estimates of the effectiveness of air bags in reducing fatalities. Air bags are reducing fatalities for passenger car drivers by a statistically significant 11 percent (confidence bounds: 7 to 15 percent). Air bags are reducing fatalities of light truck drivers by a statistically significant 10 percent. Air bags are reducing fatalities for right-front seat passengers (age 13 and older) by a statistically significant 13.5 percent. Preliminary estimates of limited crash data show an increase in fatality risk for children age 0-12 in frontal impacts with air bags. Air bags are reducing fatalities for unbelted drivers by 13 percent and providing supplemental protection for those wearing safety belts of 9 percent. Air bags are estimated to have saved 277 lives in 1994, 462 lives in 1995 and 1,136 lives during 1986-1995.

Recently the agency estimated the effectiveness of air bags using NASS-CDS data up through 1995 in reducing moderate and greater injuries (AIS 2+) and serious and greater injuries (AIS 3+). These effectiveness estimates are compared to an unrestrained occupant. These data indicate that air bags are reducing overall moderate and greater injuries AIS 2+ (see Table I-3), however, the large number of AIS 3 arm injuries (also see Table IV-20) are limiting the benefits of air bags. These data are mostly on the driver side. While the number of serious head and spine injuries has been reduced, the number of injuries to the arms and hands has increased. Injuries to the head and spine are more likely to be life threatening; however, arm injuries are sometimes disabling.

The estimated reductions in injury for the various restraint systems are shown in Table I-3. The effectiveness of "air bag alone" systems was not statistically significantly different from the risk of unrestrained occupants, and the effectiveness for manual lap/shoulder belts was not significantly different from the effectiveness of manual lap/shoulder belts plus air bag systems.

Table I-3

Estimated Effectiveness of Occupant Protection Systems
Compared to Unrestrained Occupants

MAIS 2+ Injuries All Damage Areas Front Damage
Air bag plus lap-shoulder belt 60% 61%
Air bag alone 18%* 6%*
Manual lap-shoulder belt 49% 56%
* Not a statistically significant difference
MAIS 3+ Injuries All Damage Areas Front Damage
Air bag plus lap-shoulder belt 59% 69%
Air bag alone 7%* -8%*
Manual lap-shoulder belt 60% 74%
* Not a statistically significant difference

The reader is also referred to Table V-2 for effectiveness estimates for head and chest body regions separately. The effectiveness of air bags on the passenger side may be different, given that there should be much fewer arm injuries on the passenger side. There are children on the passenger side, some of whom are out-of-position before the crash and some of whom become out-of-position because of pre-crash braking.

Although the benefits of air bags in reducing fatalities are well documented, there are certain situations in which air bags can have adverse side effects. As more vehicles are equipped with them, these side effects have become better known to researchers. In a November 9, 1995 (60FR56554) notice the agency requested comments on ways to reduce the adverse side effects of air bags. The agency received 59 comments to the docket (Docket No. 74-14-N97).

One way to reduce the adverse side effects of air bags is to depower the air bag. Test data indicate that the primary effect of depowering air bags is to raise the chest deceleration (chest g's) for unrestrained occupants. The agency has proposed two regulatory approaches for permitting or facilitating depowering air bags. The first is a change in the chest deceleration limits during unbelted testing from the current standard of 60 g's to 80 g's. The agency indicated in the November 1995 Notice that recent biomechanical data suggest that the human tolerance to acceleration for serious chest injury is higher for air bags than for belts, because the air bag delivers a more broadly distributed, uniform loading to the chest than does a safety belt.

On August 23, 1996, the American Automobile Manufacturers Association (AAMA) petitioned NHTSA to take two actions: (1) an immediate amendment to revise the unbelted dummy test and (2) to define, then require, out-of position occupant tests based on the ISO Technical Report. The second regulatory approach is based upon this petition and a later letter from AAMA and proposes a generic sled pulse with a 125 msec crash pulse. AAMA claims that this sled pulse would allow depowering in a much quicker time frame than 80 chest g's, because of the testing needed to recertify vehicles with 80 chest g's.


In the AAMA petition, a discussion was provided of an air bag designed in Australia to provide additional benefits for belted occupants. This air bag is analyzed in Chapter V.

In the August 6, 1996 NPRM (61FR 40784) the agency stated that eventually NHTSA expects that smart air bags will be installed in passenger cars and light trucks. Smart passenger-side air bags were broadly defined as systems that automatically prevent an air bag from injuring the two groups of children that experience has shown to be at special risk from air bags: infants in rear-facing child seats, and children who are out-of-position when the air bag deploys. The agency proposed that vehicles without smart passenger-side air bags would be required to have new, attention-getting warning labels and permitted a manual cutoff switch for the passenger side air bag. No proposals were made regarding depowering of air bags. However, there were some docket comments regarding depowering. Chrysler stated its belief that depowering will reduce child injuries/fatalities. Volvo supported raising the allowable chest accelerations for unbelted dummy testing. TRW stated that depowering efforts should be pursued as directionally correct, reducing the level of risk from deployment related injuries. While TRW supports the reduction in deployment energy, the crash pulse of today's smaller, lighter, stiffer vehicles limits the extent to which depowering can be achieved while still meeting the specified performance requirements. TRW believes a dual-staged inflator will be necessary. AAMA petitioned the agency to replace the current 30 mph unrestrained dummy barrier test with a standard 30 mph unrestrained dummy sled test. IIHS stated that the agency should allow manufacturers to reduce the energy in the current air bag systems.

Table I-4 shows the types of situations in which the agency has some information suggesting there may be a risk of serious injury to vehicle occupants from the air bag.

Table I-4
Group Affected Seating Position of Primary Risk Probable Cause of Problem
Small Statured and/or Older People (Usually Unrestrained) Driver Position Proximity to Air Bag at Time of Deployment
Infants in Rear-Facing Child Restraints Passenger Position Proximity to Air Bag at Time of Deployment
Children Unrestrained in Front Seat Passenger Position Proximity to Air Bag at Time of Deployment
Out-of-Position Occupants Driver and Passenger Position Proximity to Air Bag at Time of Deployment
Persons with Disabilities Driver Position Proximity to Air Bag at Time of Deployment; Adaptive Equipment between Air Bag and Driver; Safety Features in Vehicle Must be Modified to Accommodate Adaptive Equipment
Persons Experiencing Extremity Injuries Driver and Passenger Position Unknown; Under Study

NHTSA has been investigating other cases of air bag induced injury to reportedly correctly restrained children. For the most part, it appears from Table I-4 that the primary task is to reduce the risk to occupants who are very near the air bag at the time of deployment. Countermeasures to accomplish this task would also potentially be beneficial for correctly restrained children.

The agency has a Special Crash Investigation program that has done in-depth crash investigations of hundreds of air bag cases. As of February 15 1997, NHTSA's Special Crash Investigation program has identified 38 crashes where the deployment of the passenger-side air bag resulted in fatal injuries to a child. Nine of these deaths were to infants in rear-facing child seats. Most of the 29 other children were determined to be unrestrained or improperly restrained (e.g., wearing only the lap belt with the shoulder belt behind them) at the time of the crash. All of these cases involved pre-impact braking. The combination of no, or improper, belt use and pre-impact braking resulted in the forward movement of the children such that they were put in close proximity to the instrument panel and the air bag system at the time of the crash and the deployment of the air bag. Because of this proximity, the children sustained fatal head or neck injuries from the deploying passenger-side air bag. The agency has examined all air bag cases with children in its Fatal Analysis Reporting System (FARS) and believes these 38 cases are a census of all cases that have occurred to February 15, 1997 involving fatalities.

Over the past several years the agency has also investigated 21 minor to moderate severity special crash investigations cases of driver fatalities and 2 adult passenger fatalities that appear to be related to the air bag. This is not a census of cases.

NHTSA is extremely concerned about these deaths. As shown in Table I-2, passenger side air bags are expected to increase by a factor of over 3.3 from 1996 to 2000. If there is no change in

driver/passenger behavior or in the technology of air bags, then we can expect that as air bags increase in number in America, injuries and fatalities such as those described above will increase.


The agency examined the potential benefits of air bags, using an average estimated 11 percent effectiveness in preventing fatalities. When all passenger cars and light trucks have air bags, an estimated 2,302 drivers and 714 right front passengers, for a total of 3,016 occupant fatalities, could be saved by air bags annually. However, if no changes are made to air bags or no changes are made in the number of children or the restraint use of children in the front right seat, the agency estimates that 140 children could be killed annually by air bags (33 in rear-facing infant seats and 107 older children). This estimate is based on the 38 known child fatalities as of February 15, 1997 and the cumulative estimated number of vehicle years with passenger air bags to this time (55.8 million). The calculation is 38 x 205.8 million estimated vehicles in the year 2000/55.8 million = 140 (90 percent Poisson confidence bounds around this number are 105 to 184).

This means that if all light vehicles in the year 2000 had a passenger side air bag, there would be an estimated 140 children fatally injured by air bags. Not all vehicles in the year 2000 will be equipped with air bags, but this gives a range of the possibility of estimates if no changes are made in air bag design or riding patterns.

The number of drivers killed by air bags cannot be precisely estimated, due to incomplete information. However, for this analysis the agency estimates the number of drivers killed by air bags when a full fleet of vehicles is equipped with air bags is 25 (90 percent Poisson confidence bounds of 17 to 37). This estimate is based on the 21 known driver fatalities as of February 15, 1997 and the cumulative estimated number of vehicle years with driver air bags to this time (171 million). The calculation is 21 x 205.8 million estimated vehicles in the year 2000/171 million = 25. A mitigating factor is that manufacturers appear to have already made some changes to their air bags to reduce the possibility of driver deaths from the air bag. A few of the models the agency has tested have made some changes. Note that whatever driver deaths may have been caused or influenced by an air bag deployment are included in the analysis resulting in the 11 percent net effectiveness for drivers.

The number of adult passengers killed by air bags appears to be very small. The agency is now aware of 2 adult passengers that appear to have been killed by an air bag. Based on these two cases, the agency estimates the number of adult passengers killed by air bags when a full fleet of vehicles is equipped with air bags is 7. This estimate is based on the 2 known adult passenger fatalities as of February 15, 1997 and the cumulative estimated number of vehicle years with passenger air bags to this time (55.8 million). The calculation is 2 x 205.8 million estimated vehicles in the year 2000/55.8 million = 7.

Child fatalities were explicitly taken out of the analysis by examining the effectiveness for passengers age 13 or older. Previously, the number of children potentially killed by air bags (140) was projected, however, no estimate can be made of the number of children saved by air bags. Thus, 140 is not a net disbenefit number, but is higher than the real disbenefits. If 140 were the real net disbenefit for children, the net fatality benefits when all passenger cars and light trucks have air bags would be 2,302 drivers, and 574 (714 - 140) right front seat passengers, for a total of 2,876. [These numbers were derived: 20,647 drivers from Table I-1 + 277 lives saved by air bags (almost all drivers for 1994) = 20,924 x 0.11 effectiveness = 2,302; and 6,491 right front passengers x .11 effectiveness = 714]. Lives saved can be further divided into restrained and unrestrained as seen in Table I-5.

In order to estimate the impacts that depowering air bags could have on adult passengers an estimate of the fatalities saved by air bags is needed. The estimates in the "Total" column on the right are based on agency analyses of the effectiveness of air bags. The breakdown of those estimates in delta V cells are based on frontal impacts in NASS data for 1991 to 1995. Whenever you break limited data from a sample into a number of cells, there are bound to be some cells where the results do not appear to make sense. This was the case for belted passengers with no air bag in the 0-20 mph delta V cell, which had a fatality rate 10 times as high as drivers with no air bag in the 0-20 mph delta V cell. Adjustments were made to these estimates. Table I-5 shows the fatality rates from limited air bag data in NASS and from a larger number of no bag/belted and non-belted drivers and right front passengers. The breakdown of the air bag cases is based on judgment. The NASS data gives indications that there are benefits for belted occupants in the 30+ delta V ranges and for unbelted occupants in the 20+ delta V ranges, but the level of benefits are not reasonable from the NASS data and are based on judgment.


The agency examined a wide range of alternatives to reduce the adverse effects of air bag aggressivity and is still in the process of considering many of the alternatives. These alternatives include labels, manual cut-off switches, weight sensitive cut-off switches, raising the minimum speed ("threshold") at which an air bag deploys, "smart" air bag systems that could turn off the air bag or have it deploy at a lower intensity if an occupant were too close, tilt/telescoping steering wheels, recessed air bags, pedal extenders, depowered air bag inflators, public information campaigns, primary seat belt laws and others. All of these alternatives for vehicles, with the exception of a manual cut-off switch for some vehicles, and depowering beyond what is needed to pass FMVSS 208's unbelted test requirements, could be used now by vehicle manufacturers. There is nothing in FMVSS 208, testing with a 50th percentile dummy, that prohibits the use of most of these alternatives.

The agency has conducted and will continue to conduct public media campaigns to warn the public about the potential dangers of being too close to an air bag. The agency is also trying to get more States to pass and enforce primary seat belt use laws.

Table I-5

Estimated Fatalities in Frontal Impacts and Lives Saved by Air Bags

By Delta V Category

Assuming all Passenger Cars and Light Trucks Have Air Bags

(Not considering fatalities caused by air bags for right front passengers)

0 - 20 mph 21 - 30 mph 31 - 40 mph 41 + mph Total
Target* D
Fatalities* RFP
Restrained T
Target * D
Fatalities * RFP
Unrestrained T
Total * D
Fatalities T
Lives Saved ** D
Restrained RFP
Lives Saved ** D
Unrestrained RFP
Total Lives ** D
Saved RFP
Total ** D
Fatalities RFP
Remaining T
    * Without air bags in the fleet
    ** Lives saved by air bags
    D = Driver
    RFP = Right front passenger
    T = Total

Response to Comments on the Preliminary Regulatory Evaluation

In the "Preliminary Regulatory Evaluation, Actions to Reduce the Adverse Effects of Air Bags, FMVSS No. 208, Depowering," December 1996, the agency estimated the benefits and disbenefits of air bags as currently being produced and attempted to estimate the benefits and disbenefits of depowering air bags. There were several comments to the docket regarding the agency's estimates, but very little new data provided. The comments and the agency's response to those comments are discussed below. Changes to the analysis are made in the following chapters, where appropriate, to provide the best estimates possible for this Final Regulatory Evaluation.

Response to AAMA's docket comments, January 30, 1997, (74-14-N108-112) about the Preliminary Regulatory Evaluation (PRE).

  1. Page 4, AAMA states that only 31 percent of their vehicles could be depowered with the 80 g's alternative, as opposed to 100 percent with the sled test alternative. Further on page 6, AAMA states that if (contrary to fact) all air bags could be depowered using the 80 g's approach, there would be no appreciable difference in the safety effects of depowering by either the 80 g's or sled test alternative.

    Response: These discussions mean that the manufacturers would depower the same way under either option, only fewer vehicles would be depowered under the 80 g's option. The industry did not provide the agency with depowering plans under the two options, or with information on what vehicle air bags had already been depowered, if any. On page 8 AAMA stated that the average depowering would be in the 20 to 35 percent range and that no manufacturer has stated that depowering would reach the 40-60 percent level used in the PRE for the generic sled alternative. Under the generic sled alternative, the agency believes certain vehicles could be depowered more than 35 percent. However, based on the industry claims, the agency will estimate the effects of depowering an average of 20 to 35 percent as the most likely result of the generic sled test. Thus, 40 percent depowering, which was analyzed in the PRE, and may well be possible with the sled test alternative, will not be considered in this Final Regulatory Evaluation. As shown in the PRE, modelling estimates indicated that the disbenefits of depowering would increase significantly with depowering at 40 percent on average. AAMA claimed that the new neck criteria would keep manufacturers from depowering more than 35 percent. A test provided by AAMA with one model on the generic sled with the 143 ms pulse supports that claim, but no data were provided to support that claim with the 125 ms pulse and the few agency tests do not provide data which would support or dispute that claim.

    In analyzing the two alternatives (80 g's and the generic sled test), this FRE will utilize AAMA's claim that only 31 percent of the fleet would be depowered under the 80g's alternative, while all of the fleet would be depowered under the generic sled test, and that the amount of depowering would be the same under both alternatives.

  2. Page 7, Point #1 states that the use of NCAP data beyond the data for which it was developed (passenger car belted drivers) is not justified. AAMA cites the evaluation "the correlation found here do not necessarily extend to other types of frontal impacts, such as collisions with fixed objects or trucks,... or crashes involving drivers who do not wear safety belts."

    Response: The use of the evaluation findings in this manner was recommended by the author of the evaluation, Dr. Charles Kahane, of NHTSA. The accuracy of the estimates is not assured by extending the analysis to unbelted occupants, light trucks etc., but the agency believes that the direction and magnitude of the relationship would be relatively accurate.

  3. Page 7, Point #2 states that the evaluation could not distinguish between the real world effects of small changes in chest g's.

    Response: Using real world data, one would not be able to show a statistically significant difference caused by a 1 g change in chest g's. Vehicles must be grouped to show a statistically significant difference. However, there is an curve showing an increasing chance of injury as chest g's increase, as shown in the PRE on Pages II-13 and II-15. Again, this methodology has been cleared through Dr. Kahane. The analysis tried to take into account test variability. Given the small number of tests, there is a chance that chest g's might increase or decrease if the difference between the baseline and depowered systems were small, due solely to test variability. The agency tried to examine this issue by using computer modelling to affirm the changes in chest g's before calculating the benefit/disbenefit estimates.

  4. Page 7, Point #3 states that the use of the relationship derived from belted occupants to unbelted occupants protected by the air bag is clearly in error.

    Response: Whether you are on a belt chest g's versus injury curve or on an air bag chest g's versus injury curve, there will be an increase in injury with an increase in chest g's. The magnitude of the increase in injury risk is dependent upon the steepness of the curve and where you are on the curve. While Method 1 does not makes adjustments for this, and implicitly assumes the same steepness in the curves, Method 2 specifically addresses this issue by translating from the belt to the air bag injury curves.

  5. Page 8, Point #4 states that agency should not have applied the air bag effectiveness value of 34 percent for 12:00 crashes to the universe of all fatal crashes. In Attachment 3, Pages 6 and 7, AAMA provides a more detailed discussion and states that the agency should not have applied the increased chest g's found in rigid barrier tests to all frontal impacts. AAMA claims that the target population of all frontal impacts should be multiplied by .445 to account for only 12:00 frontal impacts and by .23 to account for barrier-type impacts, so that the results from NHTSA barrier testing would be applied to only 10.23 percent of all frontal impacts.

    Response: It is true that the agency's evaluation estimated a much higher effectiveness in 12:00 impacts than in 11:00 and 1:00 impacts (where the effectiveness was about 7 percent, not "essentially no effectiveness" as stated by AAMA). In Chapter IV, based on the belief that the barrier test best represented direct 12:00 frontal impacts, the 34 percent effectiveness for unbelted in 12:00 frontal impacts was used. If the 18 percent effectiveness in all frontal impacts were used, the disbenefits of depowering would be much greater than estimated in the PRE. For example, the increase in fatalities due to depowering in the first calculation of page IV-12 of the PRE would be 50 to 144, instead of 40 to 116. On Attachment 3 page 6, AAMA recommends that NHTSA change the target population that the results should be applied to (multiply by .445, the ratio of pure frontals to all frontals in NHTSA's evaluation), rather than changing the percent application as discussed above. In Attachment 3 Page 7, AAMA states that the barrier represents 23 percent of the frontal crashes; taking data from the PRE page III-50-51. AAMA then concludes that both the .445 and .23 should be multiplied by the target population to represent the barrier test.

    There are several flaws in this analysis by AAMA.

    1. The 23 percent representing frontal barrier tests was in all crashes, not in fatal crashes. The agency has gone back into its NASS data bases and determined that of all fatal frontal crashes, the barrier represents 34.4 percent of fatalities.

    2. Multiplying all frontals by the two factors results in a significant amount of double counting. The appropriate multiplier is the 34.4 percent.

    3. AAMA argues that there would be no loss of safety benefits for the non-barrier like crashes. AAMA offers no proof of this statement. AAMA provided no test scores or data to support this claim. The fact that the crashes that are more off-set, run at 30 mph, have lower injury scores does not prove that there would not be an increase in dummy measurements. Occupants are dying in these other impacts, 65.6 percent of all frontal fatalities. The large number of fatalities are not coming out of these low dummy measurements at 30 mph.

    The agency has no test data in offset tests comparing a baseline and depowered air bag. The implicit assumption in the PRE was that the increase in fatality rates found in the barrier impacts would transfer over to other types of frontal impacts. AAMA is correct in pointing out that the agency has no proof of this. However, AAMA offers no proof that the contrary is true. AAMA's argument that meeting the alternative sled test assures them of no disbenefit is not proof. One set of data provided by AAMA in an earlier submission (See Docket No. PRM-208-Ref.110), November 8, 1996, using the first proposed generic sled test with a 143 ms pulse, showed a significant increase in chest g's in the generic sled for the unrestrained passenger going from 26 g's to 33 g's with a 25 percent depowered air bag. In the same tests the driver chest g's increased slightly from 30 g's to 31 g's. If chest g's increase, even within the level of the standard, there is a disbenefit since the percent chance of injury increases as chest g's increase and since these increases in chest g's will become more meaningful in higher speed crashes.

    The agency has attempted to examine the depowering of air bags using a frontal offset model (see Chapter III). The agency has limited experience with modelling the off-set frontals. Certain aspects of this model have been validated with off-set testing, however, the agency is not confident enough overall to use the model to quantify the impact on injury and fatalities, particularly in the head and neck areas. The agency believes the trend in the model is accurate. The trends show increasing HIC and neck injuries and a small increase in chest g's for the unbelted passenger. Measurements for the belted driver and passenger and unbelted driver do not appear to change significantly with 20 percent depowering in the offset modelling.

    The agency will consider the disbenefits to be a range applying the analysis to 34.4 to 100 percent of all frontals. The offset modelling would determine a point somewhere in the middle of that range, but a quantitative estimate cannot be determined using the modelling at this time.

  6. Page 8, AAMA makes some claims about chest g's and cadaver data. These are addressed at the end of Chapter II.

  7. AAMA states that one set of calculations result in an estimate that a 10 chest g's increase from 48 to 58 g's could lead to more than 800 deaths annually. AAMA states that this is clearly in error since if the agency believed this, it would lower the chest g's standard from 60 to 50 g's. In fact, the agency is proposing to increase chest g's from 60 to 80 g's.

    Response: AAMA does not challenge the statistical procedure employed in Method 2. The results cannot be proved to be "unsubstantiated" by just looking at the numerical estimates presented, even if they are perceived to be inconsistent, if the methodology used to arrive at them is not challenged. Our analysis refers to an increase in chest acceleration by 10g in a crash where currently 48g chest acceleration is observed (associated with the case of unbelted driver in 30 mph barrier test). That is, depowering of the air bag in this case results in an increase of a 20.83 percent in chest acceleration (from 48g to 58g). The analysis then proceeds by postulating that a similar increase in chest acceleration will be observed across the fleet of vehicles, where the baseline chest accelerations are different. In particular, it is assumed that all air bags will be depowered in the same way, yielding similar increases in chest accelerations for occupants involved in crashes. Thus, the assumption of proportionate (rather than additive) increase in chest accelerations for all vehicles equipped with an air bag is made. A rigorous calculation then indicates an increase in fatalities of 825 for all crash severities observed on the road. It is believed that most of the fatalities occur in crashes which cause chest accelerations higher than 48 g's or 58 g's. The argument pointing out that 58 g's is below the 60 g's threshold in FMVSS 208 (p.8, l.20 - l.28) is irrelevant. If the agency believed the manufacturers could meet a 50 g's standard, with test variability, the agency might set the standard at that level. But the standards are a minimum performance standard. The proposal to increase chest g's from 60 to 80 is only when the air bag is tested without the belt. This is based on a biomechanical basis which indicated that 60 chest g's with a safety belt are equivalent to 80 chest g's or more with an air bag alone.

  8. Attachment, pages 10-11, AAMA states the information in the PRE provides additional information to support the fact that chest g is not a sufficiently good indicator of chest injury. The average chest acceleration in 30 mph barrier test is in the 45-50 g's range. The PRE frequently notes that compliance tests average 48 g's (e.g. page IV-37). If an average acceleration of 48 g's is used, page II-4 of the PRE indicates a probability of an AIS>=5 of 0.023321 for a belted occupant. Page II-16 indicates that, for an occupant with an air bag, this probability is 0.003825. Thus, the information in the PRE indicates that at the 48 g's level expected in a 30 mph barrier crash test, an air bag provides an effectiveness of 83.5% compared to a seat belt in reducing AIS>=5 injury. One would expect to see this improved effectiveness in real world crashes. In the August 1996 Report on the Fatality Reduction by Air Bags, the agency found that air bags without seat belts provide an effectiveness of 13% in reducing fatalities, compared to an unrestrained occupant. Studies of the effectiveness of seat belts in reducing fatalities have consistently found effectiveness levels of 45%. That is, the safety belt is providing significantly more effectiveness in real-world crashes than the air bag. The data used in Method 2 in the PRE lead to a directly opposite conclusion. Clearly there are many other factors which relate to the risk of injury for occupants restrained by a safety belt or an air bag. Nevertheless, the approach used in Method 2 is too simplistic and focuses on a one-dimensional relationship between injury probability and chest g's, which is not supported.

    Response: While 0.023321 and 0.003825 are indeed the AIS 5+ injury rates associated with belted and air bag restrained individuals at the chest acceleration of 48g, it cannot be claimed that these two restraint systems would result in the same levels of occupant chest acceleration under similar crash conditions and, therefore, an effectiveness calculated using injury rates from the same chest acceleration level cannot be compared to field effectiveness estimates in a meaningful manner. However, the injury curves presented on pages II-13 and II-15 do suggest that an occupant protected by an air bag meeting the current standard would be less likely to suffer chest injuries than a belted occupant at the same chest acceleration. This is consistent with the intuition that for a properly seated occupant, air bag deployment in a frontal crash will result in a distribution of the force acting on the chest over a larger area than for a belted occupant, and hence less chance of a chest injury at a given chest acceleration. It is not inconsistent with the fact that safety belts overall are more effective in preventing fatalities than air bags alone. It should be realized that Method 2 assumes that depowering of air bags is necessary to achieve a reduction in inflation hazards to children and adults too close to the air bag and attempts to assess the effects that depowering will have on the rest of the crash population. Because biomechanical tests have related chest acceleration to injury and the VRTC tests have identified increased chest acceleration as the main effect of depowering, it is appropriate to focus on chest acceleration and injuries to analyze and estimate the effects of depowering. It should also be realized that Method 2 relies only on the air bag chest acceleration /injury relationship and is not influenced by the relationship defining chest acceleration/chest injuries for belt systems.

  9. Attachment 3, Pages 10-11, AAMA also states that on Page II-16 the PRE presents a probability of AIS >=5 of 0.003825 for 48 g's, while averaging the probabilities for 21-40 mph crash severity levels yield the probability of 0.018135 for AIS >=5 injury at 30 mph. AAMA argues that the fact that these two values are so far apart further validates flaws in the approach.

    Response: It is inappropriate to average the probabilities of chest injuries for delta v in the ranges of 21-30 mph and 31-40 mph to estimate the probability of injury at 30 mph, since the relationship between delta v and probability of injury is not linear. A computation shows that if you choose the range of 26 to 35 mph, then the estimated probability of AIS 5+ injury is 0.004044, which is in quite good agreement with 0.003825.

  10. Page 9 and Attachment 3, Page 15, The Holden air bag analysis. AAMA urges the agency to rely on the real-world experience from the Holden air bag to estimate the effects of depowering. AAMA claims that the real-world Holden experience is a "huge data source" and that the 643 lives estimated by NHTSA is "extremely supportable" and the agency should rely more on these findings by adding them into the disbenefits estimated. AAMA also states they agree with the agency's analysis, "assuming the opportunity to optimize the total restraint system package including belts, air bags, and vehicle structure. AAMA provides the results of a test done of the Holden bag with an unrestrained dummy, showing that it met the 30 mph barrier test.

    Response: The agency believes it gave the Holden air bag data the correct amount of emphasis in the PRE. There is not a huge data source available to the agency on the Holden bag (148 cases, 63 with air bags and 85 without air bags; only one case of AIS 3+ chest injury, no fatalities). The agency does not have out-of-position testing with the Holden air bag to see if it would indeed provide benefits to out-of-position children. The system is not like U.S. systems, no knee bolster, higher deployment thresholds, etc. It could well be that the higher deployment threshold is the main reason that it has less injuries than U.S. air bags. The agency did not put the results of the Holden air bag test with unrestrained dummies into the analysis, even though the test results were available, since the dummies slid under the air bag because there was no knee bolster. The test did not seem representative of U.S. systems, and would give the wrong impression. AAMA agrees with the NHTSA estimate, if there is an opportunity to optimize the system. Other commenters have argued that they need a few years to optimize their depowered systems. If the agency requires smart air bags in the near future, that chance for optimization may not materialize. The agency does not believe these estimates are "extremely supportable," but presents them as having potential benefits. No manufacturer claimed they would use an air bag like the Holden air bag in the U.S.

    Several other commenters provided their views on the Holden analysis:

    1. AVS Technologies (74-14-N108-126) stated that if the main difference is occurring in AIS 2 rates, perhaps the higher deployment threshold is as significant a factor as the optimization of these bags for the belted occupant. AVS Technologies also stated that there seems to be a "jump in logic" to conclude that a less aggressive system that will obviously result in fewer AIS 2 injuries in low to moderate severity crashes will produce similar reductions in fatalities in higher severity events.

    2. Richard l. Strombotne (74-14-N108-137) argues that the number of cases in the Holden analysis is so small that the "uncertainty in the analysis renders the results useless for the intended purposes."

    3. Parents' Coalition for Air Bag Warnings (74-14-N108-143) stated that "The better results of the Holden vehicle in terms of air bag injury may well be due more to the high deployment threshold than to the lesser force of deployment in the Holden vehicle."

    4. Ford (74-14-N108-150) states that the Holden data may reflect some "optimization" of vehicle systems for belted occupants, in addition to inflation power differences. Ford indicated that it could add belt load limiters and pretensioners to improve dummy results, but "such changes could not be timed to coincide with depowering of air bags."

  11. Attachment 3, Pages 18-20, AAMA argues that the estimate of 25 adult fatalities caused by air bag inflation is too low, but does not estimate a larger number. AAMA states that the estimated 39 to 83 lives saved by depowering is reasonable.

    Response: NHTSA agrees that there could be more than 25 adults killed by air bags per year and is investigating all cases brought to its attention. NHTSA did not estimate the number of adults that could be saved by depowering, but stated that it believed it would be a large proportion, but not all, of the 25. AAMA estimated 25.

    Both AAMA and IIHS commented on NHTSA's projection of driver fatalities caused by air bag inflation and the associated benefits of depowered air bags. AAMA indicated that the actual count may be much higher and IIHS provided analyses of specific fatality cases contained in the NASS files where the air bag may have contributed to the fatal injuries. However, without performing detailed investigations such as those performed within NHTSA's Special Crash Investigation (SCI) program it is not possible to be certain as to the role the air bag played in contributing to any injuries. Without these type of investigations, it is very difficult to determine whether someone died from the air bag or would have died without the air bag. Cases can be found where people have died in similar situations with and without an air bag deploying.

    As pointed out in this evaluation, the number of drivers killed by air bags cannot be precisely estimated, due to incomplete information. The estimates provided are based on cases contained in the SCI program.

    Along these lines, in order to obtain more detailed information on possible fatalities caused by an air bag, NHTSA requested that the manufacturers provide the agency with any information they have on air bag related fatalities. The agency has evaluated all 79 cases submitted, and identified seven new possible air bag deployment related crashes. Two of these seven are from 1970's vintage GM air bags. Additional information is being gathered to determine whether these six cases should be added to the SCI count. Forty-two of the remaining 72 cases are already included in the SCI count, 13 are high speed crashes and 18 do not have sufficient information to adequately evaluated the crash circumstances.

  12. On January 8, 1997, AAMA supplied the agency with estimates of the benefits and disbenefits of depowering. On January 29, 1997, AAMA supplied new estimates of the benefits and disbenefits of depowering. AAMA estimated the disbenefits of depowering for adults to be 0 to 55 fatalities; benefits included 39-83 children (using NHTSA's estimate), 215 to 220 unbelted adult out-of-position fatalities, and 643 belted fatalities (potential benefit from the NHTSA discussion of the Holden analysis). On page 6 of the January 30, 1997 submission, AAMA added some different estimates. AAMA estimated the disbenefits of depowering for adults to be 7 to 47 fatalities; benefits included 39-83 children (using NHTSA's estimate), 215 to 330 unbelted adult out-of-position driver fatalities, and 80 passenger side adults. AAMA stated "This results in a net safety benefit to all occupants of 327 to 446 lives per year by depowering." The agency has examined these estimates and has the following comments:

    1. AAMA estimated that depowering could save 215 to 330 small out-of-position, unbelted adult drivers. This estimate was based on estimates of the number of small drivers that would be unbelted, estimates of the percent of crashes in which braking or other factors would cause those unbelted drivers to get close to the air bag and test data by Transport Canada on 5th percentile female dummies showing a high percentage of potentially fatal neck injuries to dummies close to the air bag.

      Response: NHTSA has examined as many low speed air bag fatality cases as it can find. Based on the cases found, NHTSA is not aware of hundreds of fatalities from air bags in low speed crashes. Since there are many more low speed crashes than high speed crashes, and since the air bag deploys at the same speed whether it be a 15 or 30 mph impact, one would expect to find a much higher number of air bag induced fatalities in low speed crashes than in high speed crashes. Thus, how could we claim that depowering would save hundreds of drivers?

    2. The agency does not agree with several of AAMA's analyses and assumptions. Some of the assumptions have been discussed previously in this response to comments section. The following table shows an analysis of AAMA's numbers provided on page 5 of the January 29, 1997 submission. Without regard as to whether we agree with the assumptions or not, the agency finds significantly more crashes in the target population than estimated by AAMA. Based on this comparison, the agency would estimate about 4.5 times (3,748/838) as many fatalities as AAMA would predict, using AAMA's methodology, as the disbenefits of depowering.

Table I-6
NHTSA Analysis of AAMA Estimates

NHTSA % NHTSA Estimates AAMA % AAMA Estimates

% nonrollover 93.9% 2,065,931

% GAD = F 58.1% 1,200,306

% DOF = 12 63.3% 759,793
% delta V = 30 mph + 2.6% 19,755 1.9% 8,934
% SHL = D, Extent 3-4 36.7% 7,250 21.1% 1,885
% unbelted 51.7% 3,748 44.5% 838
GAD = F = general area of damage front

DOF = 12 = direction of force = 12 o'clock

SHL=D and EXTENT 3-4 = specific horizontal location is distributed damage, extent zone is how far back the hood crushed; extent zone 3-4 extends from 40 to 80 percent maximum crush of the hood.

Notes: NHTSA estimates were generated by a SAS job run January 16, 1997.

AAMA estimates are those submitted with their memo of January 29, 1997.

Response to IIHS's docket comments, February 5, 1997, (74-14-N108-135) about the PRE.

  1. IIHS argues that the real-world crash event is often much more complex than is accounted for in NHTSA crash tests and simulations. In crashes with a delta V > 10 mph, IIHS found that 35 percent had braking prior to impact and 12 percent had some other maneuver that might have positioned an unbelted occupant close to the air bag at the time of deployment. Thus, IIHS argues that the PRE ignores many unbelted occupants who would benefit from, rather than be harmed by, depowered air bags.

    Response: The agency agrees that there will be some occupants out-of-position in high speed crashes that would benefit from a depowered air bag. However, that number is impossible to determine from crash data and the agency believes it would be very small. It is very difficult to determine whether someone died from the air bag or would have died without the air bag. Cases can be found where people have died in similar situations with and without an air bag deploying. The reasons are not documented or obvious. In high speed crashes, the air bag senses the crash quickly and the occupant usually doesn't have time to get close to the air bag before deployment. In low speed crashes and in some pole impacts, the sensors can take more time to determine whether deployment is necessary and this late deployment sometimes allows occupants to get too close to the air bag before deployment. Since there are many more low speed crashes than high speed crashes, the agency believes the number of high speed crashes with occupants too close to the air bag will not be larger than the number of fatalities found in low speed crashes and will not significantly change the results of the analysis.

  2. IIHS argues that Method 1 in the PRE does not recognize the difference between the belt and air bag injury curves. IIHS also argues that Method 2 is very complicated and that mistakes can be magnified many times in the analysis. IIHS argues that the probit model poorly approximates observed probabilities of injury, especially at higher crash severities.

    Response: As discussed above in response to the AAMA comments, the agency is aware that Method 1 does not recognize the difference in the belt versus air bag curve. However, there are increasing injuries with increasing chest g's in both curves. Method 2 adjusts for these differences.

    The IIHS characterization of Method 2 as a 'multistep procedure' with 'possibility of errors' which 'multiply with each step' to render the final result uncertain (p.6, last paragraph) is not accurate.

    As to the estimation of the change in injury numbers due to depowering, the procedure does not involve more than relating two injury curves: one based on cadaver tests, the other based on crash data. There are clearly statistical errors in estimating either of these curves, but no further error is involved, if we accept the assumptions of the relationship between probability of chest injury, chest acceleration, and crash severity, and the assumptions about the change in chest acceleration due to depowering of the air bag. While the assumptions can be subject to dispute, they do not contribute to the statistical error involved in Method 2.

    The curves are estimated using standard statistical procedures. The only other statistical error involved in estimating the change in injury numbers is in the estimation of the distribution of individuals involved in crashes by crash severity (the exposure population). This is estimated directly from the data base and is not likely to be substantially biased, given the quality of the NASS-CDS data system.

    The method used to predict the change in the number of fatalities involves two additional sources of statistical error: prediction of the number of fatalities from the two highest AIS scores and the use of imputed chest injuries in place of the actual chest injuries. Both of these procedures are readily validated by comparing the estimators with the actual fatality numbers. This is done in Table IV-10 and Table IV-12 of the PRE. It is apparent from these tables that no major error is introduced through these procedures. (When examining Table IV-10, bear in mind that the procedure generating the predicted number of fatalities was designed to give a good estimate of the totals only.)

    In summary, the above analysis shows that the main source of statistical error in the procedure comprising Method 2 is the estimation of the injury curves at the initial stage. There is no 'multiplication of errors' on various steps of the computation, although some steps depend on uncertain assumptions. However, every statistical analysis is based on some model assumptions. The IIHS critique does not show that the assumptions underlying Method 2 are scientifically unsound.

    The remaining IIHS criticism of Method 2 is that it relies on the estimated relationship between delta-v and the probability of chest injury, which is claimed to be inadequate because of the use of the probit model.

    The probit model is a standard approach to modeling probabilities as a function of covariates. It uses the cumulative normal distribution function which allows flexibility to model both the location of the curve and its slope, thus providing a wide range of possible shapes of the curves which approximate the observed probabilities. Alternative standard models, such as the logistic or log-log curves lead to essentially the same results with the type of data in this problem.

    It is quite clear that some smoothing of the raw data is required since the raw estimated probabilities are quite irregular, particularly in higher delta-v ranges. The problem is that in these ranges there are relatively few observations. These curves are the best that can be done with the data available. The shape of the curves is to a large degree determined by the observations in the lower delta-v ranges, where the number of observations is greater. An attempt to approximate the observed probabilities more precisely at higher crash severities would lead to non-standard (perhaps bimodal) shapes of the approximating curves, which does not seem justified.

    Contrary to what IIHS suggests, this lack of fit at the highest delta-v ranges for AIS 3 (and perhaps AIS 4) injury curves does not result in an upward bias in the estimation of injury and fatality change due to depowering. This is clear from Table IV-11 of the PRE, which shows that the modified probability of AIS 3 injury decreases for both '31-40' and '41 and higher' delta-v ranges (and for AIS 4 injuries there is only a very small increase in the '41 and higher' delta-v range). Actually, the IIHS statement that 'much of the imputation takes place' at higher crash severities, is not correct, since the increase in fatalities in the example presented in the PRE comes more from the increase in AIS 5 injuries (which are very well approximated by the probit model) than from the change in AIS 3 (or AIS 4) injuries.

1. Defined for fatalities from FARS as principal or initial impact 10 o'clock to 2 o'clock impacts; for injuries from NASS as General Area of Damage as Frontal; or General Area of Damage as Right or Left and specific lateral damage location - most severe impact as frontal and direction of force as 10 to 2 o'clock; or General Area of Damage as Right or Left and specific lateral damage location - most severe impact not frontal and direction of force as 11 to 1 o'clock.