This  revision on Oct 30 1999. moved to this location 11 May 2000.

AN EMPIRICAL ESTIMATE OF SEAT BELT
EFFECTIVENESS FROM FARS & NOPUS DATA.

A PRIORI

Here is a thumbnail review of safety belt research and legislation.

A cyclist safety group in Ireland are concerned that belted drivers are more reckless drivers, and are killing cyclists.  They sent a letter to the Minister of State at the Department of the Environment, which I reproduce with their permission.

What follows shows how FARS and NOPUS data can be used to establish that by fastening a seat belt, a driver more than doubles the likelihood of being a fatality.

SYNOPSIS.

The conceptual basis for the argument that follows relies on relating the observed rate of belt wearing by car drivers with the reported rate of belt wearing of car driver fatalities.  If the rate of belt wearing found on driver fatalities was lower than the rate of belt wearing by drivers on the road, then it is obvious that wearing a seat belt is associated with a lower probability of being a driver crash fatality, and vice versa.

The rate of wearing by car drivers was obtained from NOPUS (National Occupant Protection Use Survey) reports commissioned by the US Department of Transport.  The rate of wearing by fatalities was obtained from FARS (The Fatal Accident Reporting System) which is an online database compiled from police reports by the US Department of Transport.

It has only been possible to perform this analysis since the NOPUS reports were published in 1996.  A similar analysis is not possible with Australian data because there is no similar study, the belt wearing rate in Australia is too high and there is too small a population.  However there is no good reason why the US data should not apply to Australia.

What follows is a summary of the more complete documentation on the Data document.

TOTAL POPULATION

As an example consider the data for 1996 which is presented in the Table below.  In 1996 NOPUS observations showed a wearing rate among drivers of 65.1% with 95% confidence that the true value was within 4.2% of that figure.    If seat belts were saving lives, then less than 65.1% of fatalities should have been found wearing seat belts.  In 1996, 59.2% of driver fatalities were found wearing a seat belt.   It appears that wearing a seat belt has significantly (>95% confidence) reduced the likelihood of being a fatality.

ALCOHOL AFFECTED

Within the population of car drivers there is a group of car drivers who are alcohol affected.    Alcohol affected car drivers are known to have the attribute that they wear a seat belt less frequently than non alcohol affected car drivers, and are also known to be high risk.  As a consequence they distort the results previously calculated.  (A variant of Simpson's paradox exists here).  To rectify the distortion they should be placed in a separate stream, and each stream should then be subjected to the analysis carried out in the previous paragraph.

The proportion of drivers who are affected by alcohol is small, estimated at less than 1%.   Discarding small (< 1%) significant groups does not significantly increase the observed ratio of seat belt wearing by people in the main group, or significantly alter the calculated seat belt effectiveness.   If in 1996 alcoholics had been 1% of the main group, the error of discarding the group would have been 2.3% in calculating seat belt Effectiveness.

The proportion of 1996 driver fatalities who were alcohol affected was 24%.  Factoring the alcohol affected driver fatalities out of the main stream reveals that the seat belt wearing ratio of driver fatalities is 71.1%.

Streaming has removed a distortion.  There is 95% confidence that the true proportion of car drivers wearing a seat belt is within 4.2% of 65.1%.   The number of main stream fatalities found wearing a belt was 71.1%.  It is evident that wearing a seat belt is significantly (>95%) associated with the wearer's likelihood of being a fatality.

RURAL AND URBAN STREAMS

Most rural fatalities result from head on collisions or rollovers at speed.  Urban fatalities are more likely to result from side crashes at slower speeds.   Seat belts have been designed for the circumstances of rural crashes.  In 1996 NOPUS observations found the rate of belt wearing among urban car drivers was 67.3% with a 95% confidence value of 3.7%.  When non alcohol affected car driver fatalities are streamed into fatalities on urban roads and fatalities on rural roads, it is found that on urban roads 75.4% of car driver fatalities were wearing a seat belt.   Rural fatalities were found belted 67% of the time.

If seat belts were saving urban car driver lives, then the probability that 75.4% of urban car driver fatalities would be found wearing seat belts when the observed rate of belt wearing by urban car drivers was 67.3% (95% confidence limits is 3.7%) is less that 0.01% (one in ten thousand)

Table 1 Showing the effect of factoring out a numerically small high risk group and of considering only urban roads.

Driver fatality groups 1996 %Drivers Belted
%Fatalities Belted
Effectiveness.
All Drivers
65.1%(4.2%)
59.2%
+22%
Nondrinkers
65.1%(4.2%)
71.1%
-32%
Rural Nondrinkers
64.3%(9%)
67.0%
-12%
Urban Nondrinkers
67.3%(3.7%)
75.4%
-59%

Seat belt “effectiveness” is an estimate of the proportion of lives that would have been saved if occupants who did not wear a seat belt had been wearing a seat belt.

Speeding, like alcohol, is hazardous road behavior.  Like alcohol affected drivers, speeding drivers wear seat belts much less frequently than average.  Extracting the speeders from both groups reveals that belts are even more hazardous than shown.

Follow this link to an empirical study in which an attempt is made to objectively determine the effectiveness of seat belts.

Here is a copy of State v. Hartog, 440 N.W.2d. 852 (1989) the only case I have found on seat belt laws.  If you are aware of other cases I would appreciate notification.

In Hartog the judge found that a state government could make laws that limited the liberty of the individual if in so doing the law improved the safety for other people, and/or reduced the cost to the public at large.

As things stand at the moment:

  1. There is clear and undisputed evidence that increased seat belt wearing is accompanied by increased fatalities among non-occupant collateral people (pedestrians and cyclists).
  2. The evidence that a seat belt improves the safety of the driver and front seat passenger has been refuted.
  3. There is clear unrefuted evidence that by wearing a seat belt a driver increases the risk of being a fatality.
THE CONCLUSION?  Hartog is based on failed evidence. Hartog is about due for an appeal.

BACK TO KILLERBELT INDEX
 

Originally published 28 April 1997.
Most recently upgraded 30 October 1999