The effectiveness estimates from the two sets of data are relatively close, with the exception being lap belt fatalities, which is probably the result of small sample size in RSEP. The results of the raw data in Table IV-2 are that lap/shoulder belts are more effective than lap belts in reducing moderate to fatal injuries; again, the exception is RSEP fatalities. These data are considered "raw" data because they have not been "controlled" for various factors. For example, an examination of the data shows that occupants wearing lap or lap/shoulder belts were generally involved in less severe accidents, in terms of damage extent zones and Delta V, then unrestrained occupants.

Delta V is the instantaneous velocity change during the impact. Delta V data are shown in Table IV-3.

Another way to examine Delta V by restraint usage is shown below using NCSS data.⁴

Since the effectiveness of belts is believed to be higher in the lower severity crashes, the effectiveness estimates from the raw data would be overestimated. One theory is that present belt wearers are a special set of drivers who are more cautious and less prone to severe accidents. These factors must be controlled for, since a mandatory seat belt use law or an automatic belt requirement would result in a new set of belt wearers with driving characteristics more like the average driver.

In the Restraint System Evaluation Project considerable statistical analyses were performed by the contractor to control for four factors which could bias the effectiveness estimates taken from the raw data. These four factors were: age of occupant, accident severity, impact mode (front, side, etc.), and size of car. The results were as follows for AIS 2 or greater injuries, including fatalities.
 

While the agency did not control for all four of the variables in its in-house-analysis of NCSS/NASS as was done by the contractor with the RSEP data, a substantial effort went into assessing the impact of damage type and accident severity on overall effectiveness. Accident severity by impact mode was found to be significantly different between restrained and unrestrained occupants.

Initially, the agency examined the impact that Delta V has on effectiveness. Using the NCSS file, it was found that restrained occupants were involved in less severe accidents to such an extent that the severity of the accident by itself could explain most of the apparent fatality effectiveness of restraints and nearly half of the apparent injury festiveness (see footnote 4 on page IV-6). Due to the large number of cases of unknown Delta V in the file (55 percent of the cases have unknown Delta V), it was realized that Delta V, by itself, was not a good control factor. This is especially true since Delta V is unknown in most rollover accidents, where seat belts are particularly effective.

The agency then examined two other methodologies to control for accident severity. The first methodology examined Delta V, when known, and the collision deformation classification (CDC) or damage extent zone, when Delta V was unknown, by crash mode using the NCSS data. The results of this analysis are adjustment factors for lap/shoulder belts of 28.4 percent for fatalities and 18.0 percent for AIS 2-5 Injuries. By breaking up the accidents into 21 categories, this methodology has a problem with sample size in the severe accident groups (Delta V > 30 mph and CDC > 5); thus the results are somewhat tenuous. Using the formula: (1-Real Effectiveness) = (1-Observed Effectiveness) / (1-The Adjustment Factor), and applying this formula to the NCSS/NASS data in Table IV-2, results in the following controlled effectiveness estimates:
 

The second methodology examines only the collision deformation classification by crash mode and damage extent.⁵ Three separate analyses were performed using this methodology on the combined NCSS, NCSS NASS, and NASS files. First, unrestrained occupants were adjusted to match the frequency of damage area and extent that were observed for restrained occupants. This is the same methodology used in the two previous analyses discussed and probably best represents the effectiveness for current belt users. Second the restrained occupants were adjusted to match the unrestrained occupants. Third, the restrained and unrestrained occupants were each adjusted to match the entire population of occupants, restrained and unrestrained.

These last two analyses were performed to see how effectiveness might change if a mandatory belt use law turned a large proportion of current nonusers into belt users. The results are shown in Table IV-4. This third analysis (restrained adjusted to all occupants) may best represent the seat belt effectiveness for a group of current non-users who would accept wearing belts.

Comparing the two right columns shows very little difference between these two analyses. However, comparing the left column to the two right columns indicates that restraints are more effective for current users than they would be for current non-users since the current non-users tend to be in more severe accidents, where belts are not as effective.

Finally, the third analysis - where restrained and unrestrained occupant counts are both adjusted to reflect the damage distribution of the entire population - was performed on RSEP alone, RSEP adjusted to NCSS/NASS, and a combination of all of the previous files: RSEP, NCSS, NCSS/NASS and NASS.⁶ Moreover, 90 percent confidence bounds were calculated for the effectiveness estimates (by a technique that generates asymmetric bounds).

The results, which are shown in Table IV-4a, employ the largest available probability sample of accident data collected by the agency. Moreover, the adjustment procedure, as explained in the report, makes RSEP data comparable with the other files.

Having examined all of the results of the above analyses, the agency believes that it is appropriate to rely on the controlled data more heavily than the raw data in deriving an effectiveness range. The controlled data give an indication of the direction and possible magnitude of the adjustment, but the agency does not believe that the controlled data can be used to pinpoint an exact effectiveness estimate. Instead, an effectiveness range is seen as the best approach to estimating uncertain variables. The final estimates of the agency are as follows: lap and shoulder belts are estimated to reduce fatalities by 40-50 percent and AIS 2-5 injuries by 45-55 percent, with fairly narrow confidence bounds. Lap belts are estimated to reduce fatalities by 30-40 percent and AIS 2-5 injuries by 25-35 percent, with substantially greater statistical uncertainty. (See the 90% confidence bounds in Table IV-4a; lap belts have a wider confidence bound than lap/shoulder belts mainly due to a smaller sample size, see Table IV-2.)

Several SNPRM commenters, notably Ford (74-14-N35-065), Chrysler (74-14-N35-036), Renault (74-14-N35-050), the American Seat Belt Council (74-14-N35-044), and Professor Nordhaus (74-14-N35-079), argued that the Department's effectiveness estimates for manual belts were too low. Chrysler stated that the correct range should be 50-70 percent. Renault stated that according to an analysis of accidents in France, effectiveness is around 60 percent. Renault supplied a graph showing that Delta V for unbelted occupants was only slightly higher than Delta V for belted occupants and stated that belted drivers may feel better protected and therefore drive faster. However, France has a much higher belt usage rate than the U.S. (see Table IV-4b; data used in the Delta V graph implied 39 percent belt usage). Professor Nordhaus stated that the Department adjusted the effectiveness estimates too low. He apparently believes the Department determined the level of adjustment by assuming that all of the more severe accidents will involve restrained occupants, when no analysis in the record predicts 100 percent usage. The Department considered this very point raised by Professor Nordhaus and for that reason Table IV-4 includes the third column -- restrained and unrestrained adjusted to all occupants. This is one of the reasons the Department chose a range of values for effectiveness.

Ford believes that the Department should rely on the combined NCSS/NASS/RSEP data that indicate a confidence interval of 37-68 percent for fatality effectiveness of manual lap/shoulder belts. Ford believes this indicates a range of 50-60 percent effectiveness rather than 40-50 percent. The Department based its estimates on several analyses, rather than just the one combined analysis cited by Ford, and took into account the best estimates and confidence bounds derived from these analyses.

Ford further justified a 50-60 percent range by quoting B. J. Campbell's analysis of North Carolina State accident data (Safety Belt Reduction Related to Crash Severity in Front Seated Position, HSRC-PRI29, March 1984, Docket No. 74-14-N35-065) which found a 62 percent fatality reduction for belts even after the data had been controlled for differences in TAD Crash Severity and other factors. Based on NHTSA's extensive experience in statistically analyzing State data, as for example in the evaluation of several existing standards, the control variables available in state data are inadequate for adjusting the populations for differences in crash severity. In other words, analyses of State data using control variables yield exaggerated effectiveness estimates. In particular, the 62 percent estimate in Campbell's study appears to be overstated. Another reason for manual lap/shoulder belt selecting the 40-50 percent range for manual effectiveness was because C. J. Kahane's study of the potential effectiveness of air begs and seat belts (Estimates of Fatality Reduction of for Air Begs and Lap/Shoulder Belts, February 1984), examined the unrestrained front-seat occupant fatalities in NCSS and concluded that 51 percent of those fatalities were likely to have been prevented by belts. Many of the other fatalities involved circumstances that would have rendered any restraint system of little value.

Ford further stated that the actual data presented, historical literature and the Campbell study, indicate lap/shoulder belts are more effective in preventing fatalities than injuries, not the other way around as estimated by the Department. The Department's conclusion that injury effectiveness is 5 percentage points higher than fatality effectiveness for belts, was largely based on the NCSS/NASS adjusted data - the latest data source - which show that AIS 2-5 injury effectiveness is 48 percent while fatality effectiveness is 39 percent. The RSEP adjusted data indicate the effectiveness is about the same. While it is true that many historical estimates and the Campbell study indicate higher fatality effectiveness than injury effectiveness, these studies are not typically comparable with this analysis because of the AIS 2-5 injury criteria used here. If AIS 1 inju'ries were included, fatality effectiveness for lap/shoulder belts would be much higher than injury effectiveness.

The agency also examined the effectiveness of belts as derived from a review of the experiences in a number of countries after implementation of mandatory usage.laws. Sufficient data are available to compute effectiveness in 11 locations. The fatality effectiveness of belts ranged from a low of 20 percent in Quebec, Canada, to a high of 77 percent in Sweden. The 11 location average effectiveness was 47.1 percent. This includes some unknown combination of lap belts and lap/shoulder belts, although most of these countries required lap/shoulder belts as of the early 1970's. While this appears to confirm the results of the NCSS, NASS and RSEP studies, the agency did not consider these results in its final determination of belt effectiveness. The agency has no way of verifying the validity of the data or the statistical techniques employed in the various locations. The details of the effectiveness computations for each location are contained in Table TV-4b.
 

In general, the agency has less confidence in effectiveness estimates for AIS 1 injuries than for more severe injuries due to reporting problems. Many people don't report minor injuries or don't know they are injured until the next day. While these reporting problems should not impact the relative effectiveness of lap and lap/shoulder belts, there is some doubt about whether the overall level of effectiveness is accurate.

Based on the data presented above, the agency estimates the effectiveness of lap and lap/shoulder belts to be 10 percent in reducing AIS 1 injuries.

1.RSEP data in Table IV-2 include 783 more cars than were available when the following reports were completed, and when the controlled estimates which appear on page IV-7 were made, thus, the effectiveness estimates for the raw data are slightly different between the two tables for lap/shoulder belts (59% vs. 61%). "Fact Book: A Summary. of Information About Towaway Accidents Involving 1973-1975 Model Cars," Robert G. Hall, Highway Safety Research Center, University of North Carolina, May 1976. "A Statistical Analysis of Seat Belt Effectiveness in 1973-75 Model Cars Involved in Towaway Crashes" Highway Safety Research Center, University of North Carolina, May 1976.

2.Data from the Fatal Accident Reporting System (FARS) are not utilized here for two reasons: 1) FARS only includes fatal accidents, thus the number of accidents which did not result in a fatality due to seat belt usage would have to be estimated. 2) Restraint system usage in FARS is not considered as reliable as in NCSS or NASS. In comments to the SNPRM, Volkswagen (74-14-N35-046) disagreed with 1) above and provided a formula to calculate effectiveness from FARS. However, the formula is sensitive to belt usage in a potentially fatal accident. Given the Department's findings that belted occupants are included in less serious accidents, one can not use a general indication of belt usage (e.g. observed usage) as a proxy measure for belt usage in potentially fatal accidents.

3."Restraint Use and Effectiveness as Estimated From U.S. Accident Files and Observational Survey" Van Dyke and Springer, NHTSA, November, 1982.

4.Effects of Different Crash Severities for Restrained vs. Unrestrained Occupants," Conrad Cooke, Engineering Systems Staff, NHTSA, 12/l/83.

5.Seat Belt Effectiveness Estimates Using Data Adjusted for Damage Type," Susan C. Partyka, Mathematical Analysis Division, NHTSA, January 1984.

6."Addendum to Seat Belt Effectiveness Estimates Using Data Adjusted for Damage Type, Charles J. Kahane, Office of Program Evaluation, NHTSA, February 1984.

7."Effectiveness of Safety Belt Usage Laws," Dr. Franklin B. Fisher, May 1980 (Peat, Marwick, Mitchell & Co.).

8."Seat Belts: Effectiveness of Mandatory Use Requirements," Roger L. McCarthy, et al., SAE 840329, Failure Analysis Associates.

9.The Effectiveness of the Canadian Mandatory Seat Belt Use Laws," Brian A. Jonah, Transport Canada.

10.Task Force Report on Safety Belt Usage Laws," Livingston, et al, NHTSA, June 1978.

11."Patterns of Safety Belt Usage Following Introduction of Safety Belt Wearing Law", Hakkart, A., Ziedel, D., Technion, Israeli Inst. Tech, June 1983.

12."Legislation for Seat Belt Use in Britain," Murray Mackay, University of Birmingham, SAE 840328.

13.Seat Belt Use in Sweden and Its Injury Reducing Effect," Hans Norin et al, SAE 840194. Data on Volvo cars alone indicates a belt effectiveness of 74.5%'.

14.Calculated as follows: E = F_R / [U_a - (1-F_R) (U_B)]
where
 E  = Effectiveness
F_R = Fatality Reduction
U_a = Usage After the Law
U_B = Usage Before the Law

15."Usage and Effectiveness of Seat and Shoulder Belts in Rural Pennsylvania Accidents," C. J. Kahane, NHTSA, 1974, DOT-HS-801 398.