Background In 1997, NHTSA amended FMVSS 208 to make the unbelted test for air bags less stringent and, in effect, allow "depowered" air bags. Suppliers depowered their air bags by removing some of the propellant. This redesign took place in time for the 1998, or at the latest, the 1999 model year (but some air bags may have had little change from 1997 to 1999). The goal was to offer immediate relief, in new vehicles, from some of the hazards of air bags to out-of-position occupants. NHTSA projected that depowered air bags would benefit out-of-position occupants and reduce drivers' arm injuries, but might conceivably be less effective than pre-1998 air bags for unbelted occupants. Without a statistical analysis of each of these effects, based on actual crashes, it is difficult to assess the net effect of depowering.

ObjectiveVehicles would be grouped according to the characteristics of their air bags. The fatality risk in vehicles with depowered air bags would be compared to the corresponding risk in the same or similar make-models prior to depowering. Effects would be estimated for belted and unbelted occupants; for child passengers, young adults and old adults.

Proposed Approach Statistical analyses of FARS data, similar to some of those in NHTSA's 1996 evaluation of air bags, will be used to compare fatality risk with pre-1998 and depowered air bags. Statistical analyses of NASS data, similar to those in the 2001 Fifth/ Sixth Report to the Congress: Effectiveness of Occupant Protection Systems and their Use, will explore the effect on arm injuries and overall injury risk. The effect of "depowered" bags cannot be accurately studied without knowledge of what bags are depowered and by how much. The manufacturers furnished NHTSA with extensive test data (rise rates, tank pressures) in response to Information Requests. These data, supplemented if necessary by additional test procedures, will be used to classify the air bags. Initial statistical analyses can be completed in 2004, but more time may be needed if (1) additional test data have to be collected or (2) the initial FARS database is not large enough for statistically meaningful results.