Better to see Chapter 11 Occupant protection 


Chapter 12 Airbag benefits, airbag costs of Traffic Safety  (2004)

Chapter 9.  EFFECTIVENESS OF OCCUPANT PROTECTION DEVICES WHEN USED  (From 1991 book Traffic Safety and the Driver)


Words only (no formatting, figures, tables, or photographs) from 1991 book



Paperback copy of complete unchanged book available from , list price $29.95



            Much of the focus in previous chapters has been on factors related to crash involvement.  Interventions to prevent crashes have been referred to as pre-crash phase interventions, or primary safety.  Here we address questions of secondary safety -- occupant protection devices designed to reduce or prevent injury, given that a crash has occurred.  Countermeasures in the crash phase aim at managing the energy dissipated in the crash in such a way as to reduce harm [Haddon 1980].  Occupant protection devices are active or passive.  Active devices can provide protection only if the user performs some specific act, usually each trip, such as fastening a safety belt or wearing a helmet.  Passive devices, such as airbags or passive safety belts, offer protection without the user having to do anything, or in some cases, even without the user being aware of their existence.

            Effectiveness of a device is defined as the fractional, or percent, reduction in the occurrence of some specified level of injury (such as fatality) if a population of occupants changes from all not using the device to all using the device, without any other factors changing.  Equivalently, effectiveness is the percent reduction in risk an average occupant obtains when changing from non-use to use of the device without otherwise changing behavior.  Three distinct effectiveness measures must be considered:

 1.                    Severity-specific effectiveness, defined as the percent reduction in injuries in crashes of a specific severity, or within a narrow range of severities.

 2.                    When-used effectiveness, defined as the percent reduction in injuries that occurs when the device is used.

 3.                    Field effectiveness, defined as the percent reduction in injuries taking into account the use rate for the device.

            The severity-specific effectiveness depends only on the engineering of the device and on the biomechanical properties of the human body.  The when-used effectiveness depends on the types of crashes that occur in actual traffic.  These two effectiveness measures apply to active and passive devices.  Field effectiveness is identical to when-used effectiveness only when the device is always used.  When the device is not used at all times, field effectiveness is less than when-used effectiveness.  This is the case for manual safety belts, but also may apply to nominally passive devices which users defeat by disconnecting, modifying, or improperly using, or by not replacing deployed airbags.  In the present chapter we discuss only the first two of these measures of effectiveness -- that is, the effectiveness of occupant protection devices when used.  Chapter 10 addresses use rates.


Injury biomechanics and occupant restraints


            In order to understand better how restraints protect occupants, we overview the basics of injury biomechanics [Viano 1988; Viano et al. 1989] and vehicle dynamics in crashes [Mackay 1987; Grime 1987].  Vehicle dynamics deals with the mechanical forces and subsequent crushing of vehicles in crashes.  Injury biomechanics deals with the relationship between physical forces and injury.  In the 1950's, Colonel John Stapp, MD, of the US Air Force, provided pioneering information on the forces the body could withstand by subjecting himself to deceleration forces in sled tests.  Since then, biomechanical knowledge has been expanded by, for example, university studies in which human
cadavers have been subjected to the more severe forces that occur in crashes.  From such information, instrumented anthropomorphic dummies have been designed to simulate human responses to crash forces; these are used in vehicle and restraint system evaluation and certification.

            Impact between a human body and a physical object can cause compression, stretching, and other deformation of tissues beyond recoverable limits.  In non-biological systems this would be called mechanical failure; in the human system it is called trauma.  Trauma can be either penetrating or blunt.  Penetrating trauma is most typically caused by high-speed projectiles, such as bullets, or by sharp objects moving at low speeds, such as knives or daggers.  Although penetrating trauma can occur in traffic crashes, blunt trauma is by far the more common cause of injury.  Blunt trauma occurs when the human body strikes, or is struck by, a blunt object such as an instrument panel, a windshield, or the roadway.

            Let us consider what happens to an unrestrained front-seat occupant in a car crashing head-on at a speed of 30 mph into an immovable barrier.  This crash is equivalent to two identical cars crashing head-on into each other at 30 mph, as is apparent when one considers that if the barrier were an unbreakable mirror, an observer could not distinguish between the two situations (except for the steering wheel being on the passenger side); this is further equivalent to one car travelling at 60 mph crashing into an identical stationery car.  On impacting the barrier, the car's speed will change rapidly from 30 mph to near zero (in fact a negative value of about 10% to 20% of initial speed as it bounces back from the barrier).  The change in speed, perhaps about 35 mph, is an important measure of crash severity, and is often referred to in the literature as "delta-v"; a 60 mph car crashing into a stationery car would have the same delta-v.

            The speed of the occupant's body is not materially affected just after the instant of impact, and in accord with Newton's first law of motion, will continue to move at about 30 mph until acted upon by a force.  Factors such as frictional forces between seat and occupant, and arm and leg muscles, will reduce the occupant's speed by small amounts.  No major reduction will occur until some part of the occupant's body strikes some part of the vehicle, perhaps a knee impact which might reduce the speed by about a third.  The major portion of the speed reduction occurs when the occupant's head or chest strikes the steering wheel or instrument panel, typically about 120 ms (a little more than a tenth of a second) after car/barrier contact, by which time the occupant compartment is nearly stationery.  To keep the situation in perspective, the occupant approaches the vehicle structure at a speed similar to that attained by an object dropped from a four storey building.  Any notion that the occupant can avoid harm by muscular effort alone is as unrealistic as the hope of avoiding harm in a fall from a four story building by correctly bracing one's fall.

            The impact between the occupant and the vehicle interior has been referred to as the "second collision", to distinguish it from the first collision, the one in which the car hit the barrier.  There is an analogous "third collision" when soft tissue impacts decelerating skeletal structure inside the body [Viano and Lau 1990].

            If a car were perfectly rigid, the occupant compartment would change from a forward speed of 30 mph to a negative speed almost instantly.  In fact it takes about 150 ms because of the time that it takes for the front structure to crush; the greater the time the vehicle takes to crush, the greater is the potential for occupant protection.  An occupants restrained in place relative to the occupant compartment would then likewise take a similar time to stop; spreading the change in speed over as long a time as possible minimizes the
maximum decelerations, which, from Newton's second law, likewise minimize the injury-producing forces.

            Occupant restraints spread the occupant's change in speed over longer  times, and thereby reduce maximum forces.  In the example above, if the occupant had been wearing a lap/shoulder belt, this would have applied forces keeping the body more fixed to the occupant compartment.  The belt helps the occupant "ride-down" the crash, and impact with the steering wheel or instrument panel is less likely and less severe.  Safety belts also prevent occupants ejecting from vehicles during crashes.  An ejected occupant might travel outside the vehicle at close to the vehicle's pre-crash speed, and would continue at that speed until stopped by striking something. 

            Airbags are restraint systems consisting of a bag in front of the occupant which inflates rapidly when a frontal crash occurs with severity exceeding some pre-set limit [Passell 1987; Maugh 1986], typically a delta-v in the range 10 to 20 km/h.  Instead of striking the steering column or instrument panel, the occupant rides down the crash in contact with the airbag, which additionally spreads the impact forces over a larger area. A lap/shoulder- belted occupant can receive additional protection from an airbag because it may reduce loading forces on the belt, and further reduce the probability of injury from impact with the steering wheel or instrument panel.  The airbag is a supplemental system -- it is designed to be used in conjunction with a lap/shoulder belt.

            There are many field studies illustrating the various ways in which restraint systems have prevented or reduced injury in real world crashes [Huelke and Sherman 1987], and on specific questions such as improper use [States et al. 1987]; here we confine attention to the narrower question of the overall effectiveness of the devices in preventing injury.


Conceptual considerations common to all occupant devices


            A number of general concepts apply for all devices aimed at preventing or reducing occupant injury [Evans 1987a; Horsch 1987; Malliaris, Hitchcock, and Hansen 1985; Mertz and Marquardt 1985].  In formally examining these concepts, Evans [1987a] assumes probabilities of injury as a function of crash severity for occupants using and not using protection devices.  For expository convenience the formalism is summarized below in terms of driver fatalities, although the results are, in general, equally applicable to all occupants and levels of injury.  Safety of drivers using and not using some occupant protection device is compared; the terms "protected" and "unprotected", and "users" and "non-users" denote use, or non use, of the occupant protection device.

            At a given crash severity, the probability of a driver fatality is defined as the number of drivers killed in crashes of that severity divided by the number of crashes at that severity.  Because of the variation in survivability, and other variability, whether the driver is fatally injured or not is stochastic in nature.  Crash severity, s, is considered to be a univariate variable with the property that as s increases, the probability that the driver is killed also tends to increase.  For example, s could be equal to delta v.  We need not specify s in detail beyond requiring that as it increases, then so, in general, does the probability of driver fatality.

            The leftmost curve in Fig. 9-1 shows how the probability of death to an unprotected driver might depend on crash severity.  There is some threshold crash severity value, say s1, at or below which no drivers are killed; the region s £ s1 contains all minor crashes.  As s increases beyond s1, the probability that the driver is killed increases until at some value, say s2, it reaches unity.  That is, no unprotected drivers can survive crashes in the
range s ³ s2.  The other curve, for the protected driver, assumes that the occupant protection device reduces fatality risk at all severities.  When severity just exceeds s1, the unprotected driver's probability of being killed just exceeds zero.  If the protection device has any efficacy at this severity, then its use must reduce this probability to zero.  As s increases it reaches a threshold value, s3, at which even the protected driver's fatality risk is no longer zero.  When s = s2, the unprotected driver's risk of fatality is 100%.  If the protection device has any efficacy at this severity, then the probability that the protected driver is killed must be less than 100%.  However, as severity increases, it reaches a new threshold value, s4, at which even the protected driver has zero chance of survival.


Fig. 9-1 about here


            Dividing the lower curve in Fig. 9-1 by the upper enables us to define a severity-specific effectiveness, e, which gives the percent by which the device reduces driver fatality risk in crashes of severity s, as follows:


                      Probability that protected driver is killed

e(s)  =  100 [ 1  -  ───────────────────────────────────────────── ] . Eqn 9-1

                     Probability that unprotected driver is killed


            When s £ s1, e is undefined (Fig. 9-2).  In this severity range, the probability is zero that the unprotected driver will be killed, so the protection device has no opportunity to reduce this probability further.  It is important to stress that the device may prevent (or reduce) non-fatal injuries in this crash severity range, but if the fatality risk is already zero, the device cannot have any efficacy in preventing fatalities.  When s just exceeds s1, the protected driver's fatality risk is zero while the unprotected driver's fatality risk slightly exceeds zero, so that e must be
100%.  When severity reaches s3, fatality risk for the protected driver becomes greater than zero, so that e becomes less than 100%.  When s > s2, all unprotected drivers are killed, so e is identical to the probability that the protected driver will survive.  When s ³ s4, all protected users are killed, so e = 0.  The functional form of Fig. 9-2, derived from the most basic formal considerations, is supported by empirical data [Campbell 1987a; 1987b].


Fig. 9-2 about here


            The severity-specific effectiveness relationship does not enable us to estimate the effectiveness of the device in preventing fatalities in actual use.  If all crashes had s £ s3, then the overall effectiveness would be 100%.  On the other hand, if all crashes had s ³ s4, the identical device would have 0% overall effectiveness.

            In most of what follows we focus not on the severity-specific effectiveness, but on the when-used effectiveness, E, which is essentially the severity-specific effectiveness weighted by the number of crashes that occur at each severity level.  More simply and directly, it is defined as the reduction in the number of occurrences of some level of injury to a population of occupants if all change from being non-users to users, assuming that all other factors remaining the same.  Formally, for equal numbers of non-user and user occupants,


            No. of non-user injuries  -  No. of user injuries

 E  =  100  ─────────────────────────────────────────────────  .       Eqn 9-2

                     Number of non-user injuries


The when-used effectiveness of any occupant protection device depends on two factors:

1.                     The specific dependence of effectiveness on severity (Fig. 9-2) in crashes, which flows from the engineering of the device and its relation to human biomechanics.

2.                     The actual distribution of crashes by severity that occurs in real traffic.

            It is the severity-specific dependence at some chosen level of severity that tends to be investigated in laboratory evaluations of occupant protection devices.  As crash tests are difficult and expensive, the chosen level tends to be at severities for which the device was primarily designed; tests are less likely to be conducted at substantially higher or lower severities.  Considerations such as these may have contributed to a history of disappointing field results relative to expectations based largely on laboratory tests, because in actual use there are likely to be many crashes at such extreme levels of severity that there is little opportunity for mitigation of injuries.  In addition, a surprisingly large number of fatal crashes are of a bizarre nature not readily encompassed in any laboratory testing program (foreign objects entering the passenger compartment, cars being dragged for long distances along a railroad tracks, etc.).  Based on examining 101 fatally-injured front seat occupants, Huelke et al. [1979] estimated that approximately 50% of the crash fatalities were unpreventable by available occupant restraints.

            When a crash is of such extreme severity that death cannot be prevented, then the force reductions produced by occupant protection devices provide no benefits.  In this regard, fatality is a unique level of injury, because for all other levels of injury, reduction in the forces leads to reductions in injuries.  This general consideration suggests that the effectiveness of an occupant protection device is likely to be higher for injuries at severity levels less than fatality, though specific factors might lead to an opposite
result.  When effectiveness estimates are available only for fatalities, it is worth keeping in mind that effectiveness for lower levels of injury is more likely to be higher than lower.

            Effectiveness, as defined in Eqn 9-2, implies only a reduction in injuries at some specific level -- it does not imply the elimination of injury.  If a device is 40% effective at eliminating fatalities, this means that for every 100 non-users killed, 60 would still have been killed, but 40 not killed, had they been using the device.  However, the 40 not killed would still likely sustain injuries, many of high severity. Thus, by reducing fatalities, the device would actually generate increases in the numbers of injuries at lower severity levels.  These increases are expected to be small compared to the reductions generated by the device for injuries at these lower injury levels, assuming effectiveness at such levels.  This is because the number of injuries increases steeply with declining injury severity, in keeping with many more crashes of minor than major severity [Ricci 1980].


Difficulties in determining effectiveness from data


            It might seem straightforward to estimate the effectiveness of an occupant protection device by simply comparing the percent of fatally injured drivers police reports indicate were wearing the device to the percent of all drivers wearing the device, as determined in independent observations.  Although such calculations appear in the literature, the results they provide (in some cases effectiveness estimates as high as 90%) are grossly in error.  The problems are many.  There is copious evidence that belt wearers are more careful drivers than non-wearers [Ashton, Mackay, and Camm 1983; Deutsch, Sameth, and Akinyemi 1980; Evans, Wasielewski, and von Buseck 1982; Evans and Wasielewski 1983; Evans 1987b; Hunter et al. 1988; O'Neill et al. 1985].  When belted
drivers have crashes, they are of lower severity than crashes of unbelted drivers [Campbell 1987a; 1987b; Ricci 1980].  Thus the simple calculation incorrectly attributes to the safety belt injury reductions that are in fact caused by belted drivers being in fewer, and less severe, crashes.  In an insightful paper, O'Day and Flora [1982] address the overall question of interactions between safety belt use and other factors relevant in estimates of effectiveness, and conclude that the general answer to the question, "What other things are different between persons who wear restraints and those who don't is `Nearly everything'."

            Most of the biases resulting from differences between belted and unbelted occupants can be circumvented by applying the double pair comparison method described in Chapter 2 to FARS data.  Because subject and control occupant are involved in the same crash, the differences in involvement rates and severities between belted and unbelted occupants are no longer of central importance [Evans 1986a].  Even with the double pair comparison method, a number of problems remain -- problems common to all attempts to estimate belt effectiveness using field data.

            One of the most difficult problems, and one which has become more difficult in recent years, is reliability of reported belt use.  There is little reason to expect any bias in restraint use reporting for occupants fatally injured at the scene of the crash, especially as physical evidence is often readily visible.  However, for surviving occupants, especially those who manage to get out of the vehicle, obvious physical evidence may be lacking.  The coded restraint use is based largely on the occupants' own responses to police officers' questions, and may be subject to bias.  There is clear evidence in FARS of large biases since the advent of mandatory safety belt wearing laws; indeed, to admit non-wearing is to confess commission of an offense to a police officer.  Police officers may, understandably,
uncritically accept the occupant's claim of belt wearing in order to avoid the distraction of having to write a safety-belt violation when there is more vital work to be done at the crash scene.  Because of these biases, the estimates of driver and right-front passenger effectiveness [Evans 1986b; 1990a; Evans and Frick 1986] are all based on FARS data only up to 1983 to avoid any influences from mandatory laws, the first of which in the US was implemented in New York State in 1984.  When later data are included [Partyka 1989], higher effectiveness estimates are calculated.

            Another general problem in inferring effectiveness of occupant protection devices is that the precise device used is often unknown.  Knowing that an occupant was "belted" does not indicate whether a lap belt, shoulder belt, or lap and shoulder belt system was used.  As each system has different effectiveness, it is crucial to know which one was used.  This problem is particularly acute for the case of young children where child seats present additional opportunities for miscoding.  Although there is strong evidence that child seats substantially reduce fatality risk [Partyka 1984; Kahane 1986], quantitative estimates in form parallel to those presented below for other devices are unavailable.  In what follows all effectiveness estimates apply exclusively to adult occupants, defined as age 16 years or older.




            The FARS variable indicating which specific occupant protection device was used is often incorrectly coded, especially in some states and early FARS years; unmistakable evidence of such miscoding is provided by indications of use of restraint systems not available in the specified vehicles.  Such coding uncertainties make it impossible to determine the effectiveness of specific restraint systems for outboard-front occupants (drivers and right-front
passengers) using data for cars manufactured before 1974.  Many of these cars had separate shoulder and lap belts in the outboard-front seating positions.  Some users fastened one, or the other, or both; ideally one would want to know effectiveness for all combinations.  However, as it is not possible to know which was used, the data on pre-1974 model-year cars cannot be used to estimate effectiveness of specific restraint systems.  All cars of model year 1974 and later were required to be equipped with integrated three-point lap/shoulder belts in outboard-front seating positions.  In determining the effectiveness of lap/shoulder belts, Evans [1986b] assumes that any outboard-front occupant in a car of model year 1974 or later coded as using any restraint system was in fact using the lap/shoulder belt.  Even for model year 1974 or later cars, it is possible that in some cases occupants modified the restraint system, or used it improperly, so that a FARS coding of, say, lap only, could have been correct.  There are probably few errors from assuming that coding the use of any belt system implies use of a lap/shoulder belt; the assumption does require discarding all data for pre 1974 model-year cars, which does substantially reduce sample sizes, especially since FARS data for calender years 1984 and later are also discarded.


Overall effectiveness in reducing fatalities


            The effectiveness of lap/shoulder belts in preventing fatalities to outboard-front occupants is determined by Evans [1986b] by applying the double pair comparison to occupants in cars of model year 1974 or later coded in FARS data 1975-1983.  The study used 711 belted driver and 716 belted right-front passenger fatalities, together with over 30 000 fatally injured unbelted occupants.  Following the procedures described in Chapter 2, cars containing, say, belted drivers as subject occupants and unbelted right-front passengers
as control occupants were extracted from the FARS data, and the ratio of belted drivers killed to unbelted passengers killed computed.  From a second set of crashes, the ratio of unbelted drivers to unbelted right-front passengers is computed.  From the ratio, R, of these two ratios, the effectiveness of the lap/shoulder belt is computed as


         E  =  100 (1 - R) .           Eqn 9-3


The subject and control data are disaggregated into three age categories, and occupants in all car seats (front and rear, and in center seats) are used as control occupants.  In using the method to estimate restraint effectiveness it is crucial that the control occupant be disaggregated by restraint use.  If this were not done, then the control occupant accompanying a restrained subject occupant would be more likely to survive a crash than a control occupant accompanying an unrestrained subject occupant, in violation of the assumptions of the method, because restraint use by one occupant in a vehicle is highly correlated with use by other occupants.

            The combination of control occupants used led to 46 estimates of E.  Computing weighted averages provides the following estimates of fatality- reducing effectiveness:-


            (42.1 + 3.8)% for drivers

            (39.2 + 4.3)% for right-front passengers


The composite estimate for outboard-front occupants weighted in accord with occupancy rates is (41.4 + 3.8)%, where the error is assumed identical to the smaller of the above errors (if driver and right-front passenger estimates are derived from independent, rather than the same, data the composite error would
be calculated as 2.8%).  Although the estimates are obtained using data containing no cars of model year later than 1984, we assume that they apply to all three-point integrated lap/shoulder belt systems, including passive ones introduced later; these passive systems automatically move belts into place around the occupant when, for example, the door is closed.

            The above estimates are lower than those obtained by most other methods.  Arguably, the best study using field data other than FARS is that of Campbell [1987a], who examined the probability that a crash at a given level of severity proved fatal to belted and unbelted drivers using police reported data from North Carolina.  Severity was measured using the Traffic Accident Data (TAD) scale [National Safety Council 1984] in which the police officer located the crash on a severity scale defined by a series of photographs of crashes of increasing severity.  By controlling for severity, and examining fatalities per crash, this study eliminates two of the sources of bias -- the higher crash rate and the higher crash severities of unbelted compared to belted drivers.  The results showed belt effectiveness not defined for the lowest of 10 severity levels used, and then varying from a nominal indication of 100% at low levels of severity, and declining systematically to 36% at the highest of the 10 severity levels (Fig. 9-2).  The weighted average, which corresponds to when-used effectiveness, is 65%.  This value is larger than Campbell's estimate of 52% for serious injuries, whereas the discussion above suggests effectiveness in fatality preventing is likely less than effectiveness for lower severity injuries.  Effectiveness for right-front passengers is estimated at 54% for fatalities and 44% for serious injuries.  The lack of agreement between the Campbell [1987a] and double pair comparison estimates suggests the operation of large biasing effects.


Possible biases in estimates


            There are a number of possible sources of bias in double pair comparison estimates of belt effectiveness.  It is assumed that estimates derived for drivers accompanied by right-front passengers apply also to unaccompanied drivers, who in fact constitute the majority of driver fatalities.  There are two mechanisms by which belt effectiveness might be different for accompanied drivers than for lone drivers.  First, in a right-side impact an unaccompanied unbelted driver may strike the right interior of the vehicle; the presence of a right-front passenger will cushion the severity of this impact.  The presence of the passenger reduces the risk to the unbelted driver, but does not affect the risk to the belted driver.  Hence, the effectiveness of the belt, which is calculated relative to the unbelted risk, necessarily declines, and the double pair comparison method will therefore underestimate belt effectiveness for unaccompanied drivers.  The magnitude of this effect is estimated to bias the overall effectiveness estimate downwards by less that 1% [Evans 1988a]; various other biasing mechanisms relating to missing data and occupants being struck by other occupants generate even smaller effects.

            The second real difference between effectiveness for accompanied and unaccompanied drivers is that each of these groups of drivers has a different distribution of crashes by direction of impact; for example, the unaccompanied driver who crashes has greater probability of a frontal crash, but lower probability of a rollover crash, than does the accompanied driver.  A calculation using estimates of belt effectiveness by direction of impact (discussed later) and the distribution of crashes by impact direction for accompanied and unaccompanied drivers indicates no difference in overall belt effectiveness (the nominal indication is to bias the estimate upwards by 0.1%).

            The largest potential for bias in the double pair comparison is from possible miscoding of belt use.  Surviving drivers may tend to indicate that a belt was used when it was not.  All belt effectiveness estimates based on field data are vulnerable to biases from this source, which will tend to bias estimates upwards.  However, for two reasons the effects are likely to be less for estimates derived using the double pair comparison method than those derived using other methods.

            First, the double pair comparison method uses only fatal crash data, whereas examining the fraction of drivers killed in crashes of a given severity requires use mainly of data in which no fatality occurred.  It seems plausible that the less serious the crash, the greater is the likelihood that belt non-use might be recorded as use.

            Second, the double pair comparison effectiveness estimates depend mainly on individual crashes in which one occupant is coded as belted, while another occupant in the same crash is coded as unbelted.  The increased potential of serious criminal proceedings (for example, for negligent homicide) against a surviving driver could provide increased motivation to avoid perjury by falsely claiming use, especially as non-wearing was not an offense in the 1975-1983 period in which the data were collected.  The more objectively determined belt use of fatally injured occupants, together with the extreme seriousness of the situation, might discourage surviving occupants from compounding their problems by falsely indicating belt use.  The presence of additional occupants is itself an additional incentive to provide truthful information.

            Data in the study of Campbell [1987a] provide remarkably clear evidence supporting the above comments.  These data provide the fraction of drivers and passengers coded as using a restraint for crashes categorized into 10 levels of severity.  For drivers and passengers the indicated restraint use shows a
strictly monotonic decrease for every increase in severity level; for drivers the percent indicated as belted declines from 11.1% at the lowest severity level to 6.5% at the highest; for right-front passengers the corresponding values are 8.2% and 3.7%.  There is no basis to expect such large declines with increasing severity to be real.  A much more plausible explanation is that the influence of miscoding non-users as users becomes greater as severity level decreases; in other studies free of crash-related tension, drivers responding to telephone surveys and questionnaires consistently self-report substantially higher use rates than are observed [Streff and Wagenaar 1989].  Even at the second most severe of the 10 severity levels, more than 97% of the crash-involved drivers were not fatally injured; given the mix of injuries involved in these crashes it is not improbable that many drivers were outside the vehicle before police arrived; the data for this case [Campbell 1987a] show that if 4% of surviving unbelted drivers had been incorrectly coded as belted, the effectiveness estimate for this level would be 39.7% rather than the 70.8% reported.

            The systematically higher use rates for drivers compared to passengers in Campbell [1987a] supports that accompanied occupants are more likely to provide correct indications of belt use.  There are no observational or fatality data to support large differences. The large difference between the effectiveness estimates for drivers and passengers in Campbell [1987a] (65% - 54% = 11%) seems more likely a reflection of larger biases for drivers (mainly unaccompanied) than for passengers (always accompanied) than a real difference in effectiveness.  The corresponding difference for the double pair comparison estimates is 42.1% - 39.2% = 2.9%.

            Other possible biases can arise from comparing the outcomes from different crashes.  Even after controlling for crash severity, there may be many other crucial differences between belted and unbelted occupants.  For example, those
with high levels of alcohol are less likely to be wearers, but are also more likely to die in a crash of the same severity (Chapter 7), an effect which would bias estimates upwards.  This is unlikely to influence double pair comparison estimates as both occupants are likely to have similar alcohol consumption.

            Although double pair comparison estimates are not free from potential biases, they appear to be far more free from them than estimates based on any other method.  Although a few effects could bias such estimates downwards by small amounts, by far the main concern is of miscoded belt use, which would systematically bias estimates upwards.  In what follows we focus mainly on double pair comparison estimates of effectiveness in reducing fatalities.


Fatality-reducing mechanisms


            Safety belts protect vehicle occupants in two ways; they prevent ejection, and they reduce the frequency and severity of occupant contact with the vehicle's interior.  The when-used effectiveness, E, can be written as the sum of two components,


            E  =  F  +  I ,    Eqn 9-4


where F is the percent reduction in fatalities to an unbelted population of occupants if ejection were eliminated, assuming that those prevented from ejecting would acquire the same fatality risk as those not ejected in similar crashes, and I represents the percent reductions in fatalities from preventing the occupant from impacting the interior structure of the vehicle and reducing the severity of such impact.  The equation assumes that safety belts eliminate ejection, a more than adequately correct assumption for present purposes, even
though some belted occupants may be ejected.  For example, a study [Green et al. 1987] of 919 crashes in the UK finds 2 cases of completely ejected belted occupants among a predominantly belted population; the authors conclude that belt use reduces the rate of ejection by a factor of 39.  FARS data show that only 0.2% of fatally injured ejected occupants are coded as using any type of restraint.

            The fraction of fatalities which would be eliminated if ejection were prevented is estimated by Evans and Frick [1989] by applying the double pair comparison method to 1975-1986 FARS data to estimate the ratio of the risk of death if ejected compared to the risk of death if not ejected.  The ratio is investigated for various occupants as a function of many variables.  For drivers, the overall value is that the risk of death if ejected is 3.82 times the risk of death in the same crash if not ejected.  The FARS data show that 25.27% of unbelted drivers who were killed were ejected.  If these drivers had not been ejected, then F =  (1 - 1/3.82) x 25.27% = 18.7% of all fatally injured drivers would not have been killed.  Substituting this value into Eqn 9-4 gives that the interior impact reduction component of lap/shoulder effectiveness is 23.4%.  These values and their associated errors, together with the corresponding information for right-front passengers, are presented in Table 9-1.  Almost half of the effectiveness of the lap/shoulder belt in preventing fatalities comes from preventing ejection.


Table 9-1 about here



Effectiveness by direction of impact


            Table 9-2 shows belt effectiveness by direction of impact derived from 1975-1983 FARS data, and the contribution to that effectiveness of eliminating ejection derived from 1975-1986 FARS data [Evans 1990a].  The effectiveness values show the same pattern as, but are lower than, those derived by Partyka [1988] using 1982-1985 FARS data, as is expected based on the earlier discussion of increasing miscoding biases in later FARS data.  Lap/shoulder belts reduce fatalities for all principal impact points, much of this effectiveness being due to preventing ejection.  Even for rear impacts, lap/shoulder belts substantially reduce fatalities to drivers and right-front passengers; although the effectiveness estimate has large uncertainty, the estimate that over 20% of fatalities from rear impact are prevented by ejection elimination is more precise.  For far-side impacts (right for driver, left for right-front passenger), elimination ejection prevents 17% of driver fatalities and 16% of right-front passenger fatalities; for near-side impacts the corresponding reductions, 6% and 8%, are less, in part because the fatality risk inside the vehicle is so much greater in a near-side impact [Evans and Frick 1988a].


Table 9-2 about here


            Lap/shoulder belts are (77 + 6)% effective in preventing driver fatalities in "non-collisions".  When the 63% ejection prevention component is subtracted from this effectiveness, a value of I = (14 + 6)% remains; for the right-front passenger the corresponding value is I = (8 + 8)%.  Non-collisions normally imply rollover not initiated by striking a clearly identifiable object, such as a tree or another vehicle.

            From 1978 onwards vehicles in FARS are coded according to whether the first event was rollover, whether rollover was an event subsequent to striking some other vehicle or object, or whether no rollover was involved.  The results in Table 9-3 use 1978-1983 FARS data for the estimates of lap/shoulder safety belt effectiveness, E, and 1978-1986 for the estimate of the ejection prevention component, F.  Note the high effectiveness of safety belts when rollover is the first event (82% for the driver and 77% for the right-front passenger), and that 64% is due to preventing ejection.


Table 9-3 about here



Influence of various other factors


            The upper data in Fig. 9-3 show lap/shoulder belt effectiveness and the lower data the fraction of traffic fatalities preventable by eliminating ejection as a function of driver age.  The fit to the safety belt data is a line parallel to the fit to the ejection data.  This figure, and the discussion in Evans [1989a], support that front-seat lap/shoulder belt effectiveness is higher at younger adult ages.


Fig. 9-3 about here


            Driver lap/shoulder belt effectiveness versus car mass (Fig. 9-4) shows no large systematic trends, in keeping with the result derived in Chapter 4 (Fig. 4-7) using the pedestrian exposure approach.  A higher value at very small mass is supported by Partyka [1989], who also finds higher effectiveness for cars under 880 kg.  It is possible that the higher effectiveness for younger
drivers (Fig. 9-2) elevates effectiveness for the smallest cars.  There are insufficient data to be justify any strong conclusions.


Fig. 9-4 about here


            None of the values of effectiveness versus car model year (Fig 9-5) indicates any departure from the overall driver effectiveness of 42%.  If effectiveness for cars of fixed model year is examined using more recent FARS data than the 1975-1983 data used to produce Fig. 9-4, larger effectiveness estimates are obtained.  This provides clear evidence of the increasing miscoding of belt use discussed earlier, because there is no other plausible reason why belt effectiveness of, say, a 1980 model-year cars should be measured as being higher in 1987 than in 1981.


Fig. 9-5 about here


            In examining the influence of various other factors, Evans and Frick [1986] find effectiveness higher in single- than in two-vehicle crashes; this reflects the larger role of rollover crashes in single-car crashes.  Effectiveness is higher for 2-door than 4-door cars, probably reflecting that the drivers of the 2-door cars were younger and that the 2-door cars were involved in more rollover crashes.  FARS contains a variable indicating degree of deformation.  The nominal indication is of a declining effectiveness as deformation changed from minor, to moderate, to severe, in keeping with Fig. 9-2.  However, as nearly all the data are in the severe category, the uncertainty in the effectiveness estimates at the two lowest severities is large.  Effectiveness is not systematically affected by posted speed limit
(generally known), travel speed (generally not known), available lighting, season of year, type of roadway or urban compared to rural.




            Estimates of restraint system effectiveness for rear seats are more uncertain than those for front seats because two effects combine to reduce greatly the quantity of data.  First, occupancy rates, and therefore fatalities, are considerably lower in rear than in the front seats (Fig. 3-4).  Second, restraint system wearing rates were even lower in rear than in front seats.  Because data are so few, 1975-1985 FARS data are used in the study by Evans [1988b].  Rear seat occupants did not tend to be covered by mandatory wearing laws, and biasing effects are less important in the context of the much lower precision necessitated by small sample sizes.  It is assumed that any occupant coded as using any rear restraint is using a lap belt only.  For the cars on the road in 1975-1985 this assumption will introduce very little error.  Cars of all model years are included in the analysis.

            The overall effectiveness estimates found by Evans [1988b] together with the fraction of fatalities preventable by eliminating ejection from Evans and Frick [1989] are shown in Table 9-4, which may be compared to the corresponding information for lap/shoulder belts in front seats in Table 9-1.  Combining the estimates for right and left occupants gives a composite estimate of (18 + 9)% for effectiveness of lap belts in preventing fatalities to outboard-rear occupants.  This is in close agreement with the value of (17 + 8)% derived by Kahane [1987] using similar methods and data.


Table 9-4 about here


            For lap-only belts in rear seats, the impact severity reducing component, I, is (3 + 10)% for the left-rear passenger and (-1 + 9)% for the right-rear passenger, for a weighted average of (1 + 9)%.  The 18% effectiveness thus appears to flow mainly from preventing ejection, though the high level of uncertainty precludes any more definitive conclusion.  Because of the paucity of data, it is not possible to examine lap-belt effectiveness as a function of the factors examined for lap/shoulder belts.  However, the discussion in Evans [1988b] indicates that effectiveness in frontal crashes is likely lower than overall effectiveness, again suggesting that the main fatality reduction mechanism of the lap belt in the rear seat may be ejection prevention.

            The composite estimate, (18 + 9)%, refers only to fatality reductions to wearers of the belt; it does not include any possible benefits that belt use by rear-seat occupants might generate for front-seat occupants.  Park [1987] estimates that the presence of an unbelted rear-seat occupant increases the fatality risk to front-seat occupants by (4 + 2)%, presumably because of the increased loading force that the unbelted rear occupant imposes on the front occupant.




            Although the shoulder belt alone has never been offered as an occupant protection device, it is a separate component of some passive motorized restraint systems.  Because some occupants will use only this component of the system, it is desirable to estimate its effectiveness, which is also of interest in augmenting overall understanding of occupant protection device effectiveness.  Motorized two-point safety belts automatically move a shoulder belt into place around the occupant when, for example, the door is closed and the ignition circuit is switched on.  Occupants are advised to fasten the
manual lap belt.  When this is fastened, the restraint system consists of a shoulder belt and a lap belt, and is consequently expected to provide occupant protection similar to that provided by integrated three-point belts in the form of manual lap/shoulder belts or automatic lap/shoulder belts.

            Using the assumption that three-point lap/shoulder belts essentially eliminate ejection, Evans [1990b] uses published data [Esterlitz 1987] that 10 out of 56 fatally-injured occupants were ejected from one specific model car equipped with motorized belts to estimate the difference in effectiveness between the two-point motorized system and the lap/shoulder belt system.  Although the number of data are small, they focus specifically on the difference in effectiveness between the two-point motorized system and the three-point system, thus allowing reasonably precise estimates to be made. From this difference, effectiveness of the two-point motorized belt system, in conjunction with whatever lap-belt use occurred in traffic, is inferred.  From independently observed lap-belt use, an estimate is obtained of the effectiveness of the shoulder belt alone in preventing fatalities to outboard-front occupants.

            It is found that the two-point motorized system, in conjunction with the lap-belt use that occurred in traffic, is (9 + 3)% less effective in the field than the when-used effectiveness of three-point belts.  As the three-point system when-used effectiveness for outboard-front occupants is 41%, this gives a field effectiveness of (32 + 4)%, with the larger error than for the difference calculation flowing from a more direct dependence on the effectiveness of the three-point system than is so for the difference.  Based on observational use rates for lap belts, and for the motorized portion (some users defeated the system)  [Williams et al. 1989], the effectiveness of the shoulder belt only is estimated as (29 + 7)%.




            As is the case for shoulder belts, indirect means must be used to estimate airbag effectiveness in preventing fatalities because it will be many years before there are sufficient field data to do it using the methods applied to lap/shoulder belts and lap belts.  All manufacturers advise that a lap/shoulder belt should always be worn in conjunction with an air bag.  The estimation of effectiveness described here [Evans 1990a] is for the airbag alone, without the use of belts, and depends on the following three assumptions:

1.                     Airbags deploy only in frontal, or near frontal, crashes.

2.                     Airbags provide the same interior impact reduction effectiveness as lap/shoulder belts.

3.                     Airbags do not influence ejection risk, whereas lap/shoulder belts eliminate ejection risk.

These assumptions will be discussed further after the effectiveness estimate is described.

            The first assumption is that airbags work only in frontal, or near frontal, crashes, which we define as those with principal impact points at 10, 11, 12, 1 or 2 o'clock (Fig. 9-6).  The preponderance of left over right side deaths follows because impacts on the left, being closer to the driver, are more likely to produce driver fatalities [Evans and Frick 1988a]; the pattern for right-front passengers is approximately the mirror image of Fig. 9-6.


Fig. 9-6 about here


            Fig. 9-7 shows driver lap/shoulder belt effectiveness, and the portion of this that is due to eliminating ejection, based on the data in Table 9-2. 
While the lap/shoulder belt reduces fatality risk for crashes for all impact directions, and is particularly effective for non-collisions (essentially rollovers), the airbag is designed to deploy only in frontal, or near frontal, crashes.  Because of the second and third assumptions, its effectiveness in such crashes is given by the unshaded portions in Fig. 9-7 (that is, I in Table 9-2).


Fig. 9-7 about here


            In order to estimate airbag effectiveness for drivers we require the fraction of crashes that are frontal, or near frontal, and safety belt effectiveness in reducing interior impact, which is assumed to be the same for the airbag, by the same impact directions (Table 9-5).  From these values, effectiveness is calculated as 0.415 X 34 + 0.175 X 24 = 18.3, with a standard error of 4.2; that is, airbags are (18 + 4)% effective in reducing driver fatalities.  Similarly, Evans [1990a] estimates the effectiveness for right-front passengers as (13 + 4)%.  By weighting these values by 0.75 and 0.25 to reflect the approximate proportions of driver and right-front passenger fatalities (Fig. 3-4), we obtain an estimate of airbag effectiveness in reducing outboard-front fatalities of (17 + 4)%.


Table 9-5 about here


            Automobile manufacturers advise that the airbag is a supplemental restraint, and a lap/shoulder belt should always be worn.  Nevertheless, it is still possible that some lap/belt wearers might stop wearing them when they obtain airbags.  Drivers so doing exchange a 42% effectiveness for a 18%
effectiveness, and thereby increase their fatality risk by (1 - 0.18)/(1 - 0.42) - 1 = 41%; the corresponding calculation for right-front passengers estimates that switching from lap/shoulder belt use to airbag only protection increases fatality risk by 43%.

            Each of the three assumptions on which the airbag effectiveness calculations are based have the potential to introduce bias, as discussed in Evans [1990a].  The assumption that the device deploys only in frontal crashes is the least uncertain, although the definition used in the calculation is probably more inclusive than the common design goal of plus or minus 30 degrees from straight ahead, thus biasing the estimate upwards.  The assumption that safety belts eliminate ejection is more than adequately correct for the estimate.  The assumption that airbags do not materially affect ejection risk is based on the absence of any clear mechanism of ejection prevention.  Intuition can conjure up mechanisms by which an airbag could either hinder or facilitate ejection.  Eliminating all ejection in frontal crashes reduces driver fatalities by 9% (Table 9-2), so that if, for example, airbags prevented 10% of such ejections, this would increase the overall effectiveness by 0.41 X 0.1 X 0.9 = 0.4%.  Ejection plays a larger role in near frontals, but as most such ejections are through side glass, it seems unlikely that airbags could make major contributions to ejection prevention. 

            The assumption that airbags provide protection against impact with the vehicle interior equal to that provided by the lap/shoulder belt is made in the absence of firmer quantification, although this assumption seems more likely to bias the estimate downwards than upwards.  However, there are general considerations [Evans 1987a; Horsch 1987] why it is unlikely that any device can have very high effectiveness in impact protection over a wide range of crash severities.  If one made the substantially different assumption that
airbags provide 50% more impact protection than lap/shoulder belts (so that frontal effectiveness would become 34% X 1.5 = 51% for the driver -- a value higher than this seems highly implausible) then the overall effectiveness values would similarly increase by 50%, to 27% for drivers and 19% for right-front passengers. 

            It is possible that some occupants, especially older ones, could be fatally injured at crash severities below the threshold at which airbags deploy, typically designed to occur at a perpendicular barrier crash equivalent of about 12 mph  [Passell 1987; Maugh 1986].  There will be many just-below-threshold-severity crashes because the number of crashes at a given severity increases steeply with declining severity.  On the other hand, deployment can cause or increase injury, especially to out of position occupants [Passell 1987].  Assuming no additional fatalities from either of these effects biases effectiveness estimates upwards.

            There has been only one estimate of effectiveness based on field data, by Pursel et al. [1978], who find an effectiveness in preventing severe injuries (AIS³3) of 9%.  This is based on comparing injuries sustained by 180 occupants in a fleet of airbag-equipped cars introduced in the early 1970s with those sustained in matched crashes of non-equipped cars.  Design approaches to increasing effectiveness beyond this value are discussed by Mertz [1988]. 

            The standard errors in the (18 + 4)% and (13 + 4)% effectiveness estimates for driver and right-front passenger arise only from the errors in the quantities from which they are calculated.  Violations of the assumptions constitute additional sources of error.  As it is not possible to quantify these, one has only judgment to rely on.  I do not consider that the assumptions, collectively, generate any obvious systematic bias in the estimates, nor that the collective effect is to increase the stated errors substantially beyond those quoted.

            The deployment of the airbag only in essentially frontal crashes raises the possibility that it could be more effective for some occupants and situations if these were characterized by a greater tendency towards frontal crashes.  The analysis of Evans [1989a] shows that although the distribution of impact directions does depend on driver age and alcohol use, the fraction that are frontal is relatively unaffected, suggesting in turn that airbag effectiveness is relatively unaffected by driver age or alcohol consumption.  Effectiveness is found to be higher in two-car crashes than in single-car crashes (21% compared to 16%).




            Helmet effectiveness in preventing fatalities to motorcycle drivers and passengers is determined by Evans and Frick [1988b] who applied the double pair comparison method to FARS data for 1975-1986.  Motorcycles with a driver and a passenger, at least one of whom was killed, were used.  In order to reduce as much as possible potentially confounding effects due to the dependence of survivability on sex and age, the analysis is confined to male drivers (there were insufficient female driver data), and to cases in which the driver and passenger age do not differ by more than three years.  Motorcycle helmet effectiveness estimates are found to be relatively unaffected by performing the analyses in a number of ways different from that indicated above.  It is found that helmets are (28 + 8)% effective in preventing fatalities to motorcycle riders (the error is one standard error), the effectiveness being similar for male and female passengers, and similar for drivers and passengers.  By applying essentially the same method to 1982-1987 FARS data, Wilson [1989] obtains a near identical effectiveness estimate of 29%.




            For lap/shoulder belts in front seat and lap-only belts in rear seats, the nominal indications are for higher effectiveness for the left occupant than the right occupant.  Although the differences fall short of statistical significance, they are likely real and flow from the unsymmetrical nature of traffic.  The data in Evans and Frick [1988a] suggest 38% more impacts of high severity from the right than from the left, so that right occupants receive more near-side impacts, for which belt effectiveness is low.  For the airbag, the effectiveness estimate is also larger for the left occupant, though in this case the difference is mainly due to cars with right-front passengers being less likely to be involved in frontal crashes.

            Consistent though the nominal differences are between effectiveness for left and right occupants, they are small.  It is accordingly convenient to focus on the combined estimates for outboard occupants, especially as in some cases only such combined estimates are available.  The previously determined effectiveness estimates for all the devices are summarized in Table 9-6; the derivation for the estimate for airbag plus lap/shoulder belt will be explained in the next section.


Table 9-6 about here


            If one thinks of the front-seat shoulder belt as having a 29% effective­ness, and the lap belt as having an 18% effectiveness, then, if they operated independently, their combined effectiveness would be 1 - (1 - 0.18)(1 - 0.29) = 42%, a value remarkably close to the observed 41% effectiveness of the lap/shoulder belt.  This indicates that the shoulder belt may be mainly
preventing impact with the interior of the vehicle and the lap belt mainly preventing ejection.  Indeed, the component of lap/shoulder belt effectiveness found associated with interior impact reduction is 23% [Evans 1990a]; if the same value applied for the shoulder belt, this would imply that the remaining 6% arose from preventing ejection, a value about one third of that for the lap/shoulder belt.

            When the analysis of Evans [1986b] is applied to cars of all model years (not just 1974 or later), outboard-front effectiveness of (33.6 + 3.6)% is computed.  This lower value is compatible with a mix of lap/shoulder, lap only, and shoulder only belts, as might have been used when pre-1974 model-year cars are included.


Relation to prior estimates


            Because of the large biases previously discussed, nearly all prior estimates are much larger than those shown in Table 9-6.  We therefore confine this comparison to three studies which appear to be relatively free from large biases, discussing first the extensive Final Regulatory Impact Analysis performed by the National Highway Traffic Safety Administration [1984] in connection with Federal Motor Vehicle Standard 208.  Estimates of effectiveness, shown under the NHTSA heading in Table 9-7, were derived by synthesizing the results of many analyses, mainly based on NHTSA's national traffic crash files.  This detailed and thorough study used the best methods and data then available.


Table 9-7 about here


            Taking the mid-points of the ranges of the NHTSA values indicates an effectiveness of 50% for lap/shoulder belt plus airbag, which is five percentage points higher than their 45% value for lap/shoulder belts alone.  Another way to express these values is that a lap/shoulder belted occupant obtains a 9.1% reduction in fatality risk by the addition of an airbag (calculated as 1 - 0.50/0.55).  The value of 46% in Table 9-6 was obtained by assuming that the same five percentage point difference could be applied to the lap/shoulder belt effectiveness estimated using the double pair comparison method; if we are considering drivers only, rather than outboard occupants, the 5% would be added to a 42% effectiveness for drivers [Evans 1986b] to give a 47% effectiveness for lap/shoulder belt plus airbag for drivers.  The effect of adding the airbag now reduces fatality risk to the lap/shoulder belted outboard occupant by 8.5% (calculated as 1 - 0.54/0.59 = 0.0847). (The corresponding calculation, 1 - 0.53/0.58, for the driver gives 8.6%).  This lower value (8.5% rather than 9.1%) is appropriate in view of the general pattern of the values in Table 9-6 being lower than the NHTSA estimates.  The equation given in Evans [1989b] computes that a 8.5% fatality reduction reduction can be achieved by choosing a vehicle with mass larger by 80 kg (170 pounds).  The 17% fatality reduction the airbag provides the unbelted occupant can likewise be obtained by choosing a car with a mass larger by 160 kg (360 pounds).  A car with increased mass provides increased protection to all occupants, whereas each airbag protects only one occupant.

            If the-injury reducing mechanisms of airbag and lap/shoulder belt were independent, the combined effectiveness would 51% (calculated as 1 - (1 -0.17)(1 - 0.41).  On the other hand, if the airbag did only what the lap/shoulder belt did, it would not add to the lap/shoulder belt effectiveness, so the net effectiveness would remain at 41%.  The average of these two is 46%, the same as the estimated effectiveness.

            The studies of Wilson and Savage [1973] and Huelke et al. [1979] are free from all the types of biases present in all the previously-discussed studies; this is not to say that they are free from all possible biases, but only that their biases have different sources than those in all the other studies.  In the Wilson and Savage [1973] study a panel of four expert engineers examined in detail a sample of fatal crashes in which 706 occupants were killed; 74 of these were using some type of restraint system.  Using crash reports, medical and/or autopsy reports, photographs and other such information, the panel discussed the injury mechanisms for each fatally injured occupant, and arrived at a judgement about whether different restraint systems would have prevented the fatality.  Their effectiveness estimate for lap belts is an average for all seating positions, so that it refers mainly, but not exclusively, to front seating positions.  The nature of the method made it difficult to identify cases in which the use of a restraint would have increased, rather than decreased, injuries.  So in this sense, there is a mechanism generating a systematic upwards bias.  The 706 fatalities are a sample of convenience, so that they may not have been representative of fatal crashes in general.  The estimate for airbags and lap belts are remarkably similar to those in Table 9-6, whereas the estimate for lap/shoulder belt effectiveness is lower.

            The study by Huelke et al. [1979] is similar in approach.  It is based on analysis of 101 front-seat occupants fatally injured in 80 crashes in Washtenaw County, Michigan; four occupants were wearing belts.  The potential of different restraint systems to prevent the fatality is estimated by three of the authors of the study.  The smaller sample and narrower geographic location of crashes probably renders the results less typical of effectiveness over all crash configurations.  The lower value for lap belt, and higher value for airbag effectiveness, probably indicates fewer than average numbers of
crashes involving ejection.  The results of Evans and Frick [1989] indicate that preventing ejection alone reduces fatality risk by 18%.




            When a vehicle crashes, it undergoes a rapid change in speed.  Occupants continue to move at the vehicle's prior speed until stopped, either by impact with objects external to the vehicle if ejected, by striking the interior of the vehicle, or by being restrained in some other way.  A number of occupant protection devices have been developed which reduce the severity of impact with the vehicle's interior (for example, lap/shoulder belts and airbags) or reduce the risk of ejection (for example, lap belts).  Three distinct measures of effectiveness must be defined to address the effectiveness of such devices; 1) Severity-specific effectiveness, defined as the percent reduction in injuries in crashes of a specific severity, or within a narrow range of severities; 2) When-used effectiveness, defined as the percent reduction in injuries that occurs when the device is used; and 3) Field effectiveness, defined as the percent reduction in injuries taking into account the use rate for the device.

            It is no simple matter to estimate when-used effectiveness of safety belts from field crash data because wearers and non wearers differ in so many ways.  In particular, drivers who wear have fewer, and less severe, crashes.  Non wearers who survive are more likely than those who die to be incorrectly coded in data files as wearers, especially since the advent of mandatory wearing laws.  All these effects can bias estimates of effectiveness upwards, leading to many unrealistically high published values.  The double pair comparison method, which is less influenced by these biases, was applied to FARS data to estimate the fatality-reducing effectiveness of various devices, giving
(41 + 4)% for lap/shoulder belts in front seats and (18 + 9)% for lap belts in rear seats.  Nearly all of the lap belt effectiveness, and almost half of the lap/shoulder belt effectiveness, is due to preventing ejection. 

            By assuming that airbags do not influence ejection risk, and that they provide interior impact reduction effectiveness equal to that of lap/shoulder belts, effectiveness is estimated at 17%; a different, and somewhat extreme, assumption that the airbag has an interior impact-reducing effectiveness one and a half times that of the lap/shoulder belt gives an airbag effectiveness estimate of 25%.  Lap/shoulder belt plus airbag effectiveness is estimated to be (46 + 4)%.  All these results are for fatalities only, and cannot be extrapolated to lower levels of injury although there are general reasons to expect effectiveness to be higher at lower severity levels.  The occupant protection device expected to provide the highest level of protection is a lap/shoulder belt supplemented by an airbag.




Ashton, S.J.; Mackay, G.M.; Camm, S.  Seat belt use in Britain under voluntary and mandatory conditions. American Association for Automotive Medicine, 27th Annual Proceedings, San Antonio, TX, p. 65-75; 1983.

Campbell, B.J.  Safety belt injury reduction related to crash severity and front seated position. Journal of Trauma 27:733-739; 1987a.

Campbell, B.J.  The effectiveness of rear-seat lap-belts in crash injury reduction.  SAE paper 870480. Warrendale, PA: Society of Automotive Engineers; 1987b. (Also included in Restraint technologies -- rear seat occupant protection. SAE special publication SP-691, p. 9-18; 1987b.

Deutsch, D.; Sameth, S.; Akinyemi, J.  Seat belt usage and risk-taking behavior at two major traffic intersections. American Association for Automotive Medicine, 27th Annual Proceedings, p. 415-421; 1980.

Evans, L.  Double pair comparison -- a new method to determine how occupant characteristics affect fatality risk in traffic crashes. Accident Analysis and Prevention 18:217-227; 1986a.

Evans, L.  The effectiveness of safety belts in preventing fatalities. Accident Analysis and Prevention 18:229-241; 1986b.

Evans, L.  Occupant protection device effectiveness -- some conceptual considerations. Journal of Safety Research 18:137-144; 1987a.

Evans, L.  Belted and unbelted driver accident involvement rates compared. Journal of Safety Research 18:57-64; 1987b.

Evans, L.  Examination of some possible biases in double pair comparison estimates of safety belt effectiveness. Accident Analysis and Prevention 20:215-218; 1988a.

Evans, L.  Rear seat restraint system effectiveness in preventing fatalities. Accident Analysis and Prevention 20:129-136; 1988b. (Also see Evans, L. Rear compared to front seat restraint system effectiveness in preventing fatalities. SAE paper 870485; also included in Restraint technologies -- rear seat occupant protection. SAE special publication SP-691, p. 39-43; 1987).

Evans, L.  Airbag effectiveness in preventing fatalities predicted according to type of crash, driver age, and blood alcohol concentration. Association for the Advancement of Automotive Medicine, 33rd Annual Proceedings, Baltimore, MD, 307-322; 2-4 October 1989a.

Evans, L.  Passive compared to active approaches to reducing occupant fatalities. Paper No. ESV 89-5B-0-005, presented to the Twelfth International Technical Conference on Experimental Safety Vehicles, Gothenburg, Sweden; 29 May -1 June 1989b. To be published in Proceedings of the meeting.

Evans, L.  Restraint effectiveness, occupant ejection from cars and fatality reductions. Accident Analysis and Prevention 22:167-175; 1990a.

Evans, L.  Motorized two-point safety belt effectiveness in preventing fatalities. For presentation to the 34th Annual Meeting of the Association for the Advancement of Automotive Medicine, Scotsdale, AZ; 1-3 October 1990b, and for publication in proceedings of the meeting.

Evans, L.; Frick, M.C.  Safety belt effectiveness in preventing driver fatalities versus a number of vehicular, accident, roadway and environmental factors. Journal of Safety Research 17:143-154; 1986.

Evans, L.; Frick, M.C.  Seating position in cars and fatality risk. American Journal of Public Health 78:1456-1458; 1988a.

Evans, L.; Frick, M.C.  Helmet effectiveness in preventing motorcycle driver and passenger fatalities. Accident Analysis and Prevention 20:447-458; 1988b.

Evans, L.; Frick, M.C.  Potential fatality reductions through eliminating occupant ejection from cars. Accident Analysis and Prevention 21:169-182; 1989.

Evans, L.; Wasielewski, P.  Risky driving related to driver and vehicle characteristics. Accident Analysis and Prevention 15:121-136; 1983.

Evans, L.; Wasielewski, P.; von Buseck, C.R.  Compulsory seat belt usage and driver risk-taking behavior. Human Factors 24:4l-48; 1982.

Esterlitz, J.R.  A comparison of rates of fatal ejection from manual and automatic belt cars. Arlington, VA: Insurance Institute for Highway Safety; June 1987.

Green, P.D.; Robertson, N.K.B.; Bradford, M.A.; Bodiwala, G.G.  Car occupant ejection in 919 sampled accidents in the U.K. -- 1983-86. SAE paper 870323. Warrendale, PA: Society of Automotive Engineers; 1987. (Also included in Restraint technologies: front seat occupant protection. SAE special publication SP-690, p. 91-104; 1987).

Grime, G.  Handbook of road safety research. London, UK: Butterworth; 1987.

Haddon, W., Jr.  Advances in the epidemiology of injuries as a basis for public policy. Public Health Reports 95:411-421; 1980.

Horsch, J.D.  Evaluation of occupant protection from responses measured in laboratory tests. SAE paper 870222. Warrendale, PA: Society of Automotive Engineers; 1987. (Also included in Restraint technologies -- front seat occupant protection SP-690, p. 13-31; 1987).

Huelke, D.F.; Sherman, H.W.  Seat belt effectiveness: case examples from real-world crash investigations. Journal of Trauma 27:750-753; 1987.

Huelke, D.F.; Sherman, H.W.; Murphy, M.J.; Kaplan, R.J.; Flora, J.D.  Effectiveness of current and future restraint systems in fatal and serious injury automobile crashes. SAE paper 790323. Warrendale, PA: Society of Automotive Engineers; 1979.

Hunter, W.H.; Stewart, R.J.; Stutts, J.C.; Rodgman, E.A.  Overrepresentation of non-belt users in traffic crashes. Association for the Advancement of Automotive Medicine, 32nd Annual Proceedings, Seattle, WA, p. 237-256; 12-14 September 1988.

Kahane, C. J.  An evaluation of child passenger safety: the effectiveness and benefits of safety seats. Washington, DC: National Highway Traffic Safety Administration, report DOT HS-806 890; February 1986.

Kahane, C.J.  Fatality and injury reducing effectiveness of lap belts for back seat occupants. SAE paper 870486; Warrendale, PA: Society of Automotive Engineers; 1987. (Also included in Restraint technologies: rear seat occupant protection. SAE special publication SP-691, p. 45-63; 1987).

Mackay, G.M.  Kinematics of vehicle crashes. Advances in Trauma 2:21-42; 1987.

Malliaris, A.C.; Hitchcock, R.; Hansen, M.  Harm causation and ranking in car crashes. SAE paper 850090. Warrendale, PA: Society of Automotive Engineers; 1985.

Maugh, R.E.  Supplemental driver airbag system -- Ford Motor Company Tempo and Topaz vehicles. Proceedings of the Tenth International Technical Conference on Experimental Safety Vehicles, National Highway Traffic Safety Administration, DOT HS 806 916, p. 59-63; February 1986.

Mertz H.J.  Restraint performance of the 1973-76 GM air cushion restraint system. SAE paper 880400. Warrendale, PA: Society of Automotive Engineers; 1988.

Mertz, H.J.; Marquardt, J.F.  Small car air cushion performance considerations. SAE paper 851199. Warrendale, PA: Society of Automotive Engineers; 1985.

National Highway Traffic Safety Administration.  Final regulatory impact analysis, Amendment of FMVSS 208, passenger car front seat occupant protection. Washington, DC; 11 July 1984.

National Safety Council.  Vehicle damage scale for traffic accident investigators. 3rd edition. Chicago, IL; 1984.

O'Day, J.; Flora, J.  Alternative measures of restraint system effectiveness: interaction with crash severity factors. SAE paper 820798. Warrendale, PA: Society of Automotive Engineers; 1982.

O'Neill, B.; Lund, A.K.; Zador, P.; Ashton, S.  Mandatory belt use and driver risk taking: an empirical evaluation of the risk-compensation hypothesis. In: Evans, L.; Schwing, R.C., editors. Human behavior and traffic safety. New York, NY: Plenum Press, p. 93-107; 1985.

Park, S.  The influence of rear-seat occupants on front-seat occupant fatalities: the unbelted case. General Motors Research Laboratories, Research publication GMR-5664; 8 January 1987.

Partyka, S.C.  Restraint use and fatality risk for infants and toddlers. National Highway Traffic Safety Administration. Washington, DC; May 1984.

Partyka, S.C.  Belt effectiveness in pickup trucks and passenger cars by crash direction and accident year. Washington, DC: National Highway Traffic Safety Administration; 13 May 1988.

Partyka, S.C.  Belt effectiveness in passenger cars by weight class  In: Papers on car size -- safety and trends. Washington, DC: National Highway Traffic Safety Administration, report DOT HS 807 444, p. 1 - 35; June 1989.

Passell, P.  What's holding back air bags? In: Viano, D.C., editor. Passenger car inflatable restraint systems: a compendium of published safety research. Warrendale, PA: Society of Automotive Engineers, publication PT-31, p. 3-7; 1987.

Pursel H.D.; Bryant R.W.; Scheel J.W.; Yanik A.J.  Matched case methodology for measuring restraint effectiveness. SAE paper 780415. Warrendale, PA: Society of Automotive Engineers; 1978.

Ricci, L.L., editor.  NCSS Statistics: passenger cars. Report UM-HSRI-80-36, Highway Safety Research Institute, University of Michigan. Ann Arbor, MI; June 1980.

States, J.D.; Huelke, D.F.; Dance, M.; Green, R.N.  Fatal injuries caused by underarm use of shoulder belts. Journal of Trauma 27:740-745; 1987.

Streff, F.M.; Wagenaar, A.C.  Are there really shortcuts? Estimating seat belt use with self-report measures. Accident Analysis and Prevention 21 509-516; 1989.

Viano, D.C.  Causes and control of automotive trauma. Bulletin of the New York Academy of Medicine 64:376-421; 1988.

Viano, D.C.; Lau, I.V.  Biomechanics of impact injury. International trends in thoracic surgery -- surgical management of chest injuries, Chapter 2, Volume 7; 1990 (in press).

Viano, D.C.; King, A.I.; Melvin, J.W.; Weber, K.  Injury biomechanics research: an essential element in the prevention of trauma. Journal of Biomechanics 22:403-417; 1989.

Williams, A.F.; Wells, J.K.; Lund, A.K.; Teed, N.  Observed use of automatic seat belts in 1987 cars. Accident Analysis and Prevention  21:427-433;  1989.

Wilson, D.C.  The effectiveness of motorcycle helmets in preventing fatalities. Washington, DC: National Highway Traffic Safety Administration, report DOT HS-807 416; March 1989.

Wilson, R.A.; Savage, C.M.  Restraint system effectiveness -- a study of fatal accidents. Proceedings of Automotive Safety Engineering Seminar, sponsored by Automotive Safety Engineering, Environmental Activities Staff, General Motors Corporation; 20-21 June 1973

Table 9-1.        Fatality reductions from lap/shoulder belt use and from eliminating ejection for outboard-front occupants. From Evans [1990a].




                    Fatality    Fatality reduction, %   






                    source*                  passenger 





                         E     42.1 + 3.8  39.2 + 4.3 




                         F     18.7 + 0.5  16.9 + 0.6 




                    I = E - F 23.4 + 3.8  22.3 + 4.3 





*                  E is lap/shoulder safety belt when-used effectiveness

                   F is fatality reduction from preventing ejection

                   I is fatality reduction from reducing impacts with vehicle interior

Table 9-2.        Comparison of lap/shoulder belt when-used effectiveness, E, with fatality reductions from ejection elimination, F, according to principal impact point. Plus or minus one standard error is indicated under each estimate. From Evans [1990a].




           Principal                                 Right-front   

           impact                     Driver         passenger    

           points  Description  ├────────┬────────┼────────┬────────┤

                                    E, %    F, %    E, %    F, % 


              12       Front         43       9      39       8  

                                    + 8     + 1     + 9     + 1  


             1,2    Front right     41      21      30      14  

                                  + 18     + 1   + 20     + 1  


               3       Right         39      17      27       6  

                                  + 15     + 1   + 19     + 1  


           4,5,6,7,8   Rear          49      22      45      21  

                                  + 14     + 1   + 20     + 2  


               9       Left          27       8      19      16  

                                  + 17     + 1   + 20     + 1   


             10,11  Front left      38      12      23      16  

                                  + 15     + 1   + 20     + 1  


              13        Top          59      41      46      41  

                                  + 10     + 1   + 15     + 1  


               0    Non-collision    77      63      69      61  

                                    + 6     + 1     + 8     + 1  



            All principal impact      42      19      39      17  

              points combined        + 4     + 1     + 4     + 1  


           └────────────────────────┴────────┴────────┴────────┴────────┘                  └────────────────────────┴────────┴────────┴────────┴────────┘

Table 9-3         Results according to rollover status.  From Evans [1990a].





                Rollover    Fatality  Fatality redns, %  



                status      reducing            Right-  


                       a               Driver   front   

                (Distn)     source                       





                                 E       82 + 5    77 + 7 

              Rollover is                                



              first event       F       64 + 1    64 + 1 




                            I = E - F   18 + 5    13 + 7 






                                 E       55 + 10   57 + 11

              Rollover is                                



              subsequent        F       42 + 1    43 + 1 



              event (16.4%)                              

                            I = E - F   13 + 10   14 + 11






                                 E       69 + 6    67 + 6 




              rollovers         F       50 + 1    50 + 1 




                            I = E - F   19 + 6    17 + 6 






                                 E       31 + 8    23 + 9 




              rollover          F        7 + 1     6 + 1 




                            I = E - F   24 + 8    17 + 9 






a Distribution of driver fatalities by rollover status based on same data used to determine F; distribution for right-front passengers is similar.


b Calculated from combined raw data for first event and subsequent event cases

Table 9-4.        Fatality reductions from lap belt use and from eliminating ejection for outboard-rear passengers. From Evans [1990a].





                     Fatality    Fatality reduction, %  




                                   Left        Right    


                     source*     passenger  passenger 





                          E     19.4 + 10.0 17.3 + 8.7




                          F     16.1 + 0.8  17.7 + 0.7




                      I = E - F   3.3 + 10.0 -0.6 + 8.7






*                  E is safety lap belt when-used effectiveness

                   F is fatality reduction from preventing ejection

                   I is fatality reduction from reducing impacts with vehicle interior

Table 9-5.        Summary of driver results for frontal and near frontal crashes used to infer airbag effectiveness.





                                   Fatality reductions, % 

     Principal impact  Distri-                          


       (clock points)   bution                         

                                    E        F       I  





                        41.5%   43 + 8   9 + 1 34 + 8

        (12 o'clock)                                    




        Near frontal                                    

                        17.5%   39 + 11 15 + 1 24 + 11

     (10+11+1+2 o'clock)                                




Table 9-6.        The when-used effectiveness of various devices in preventing fatalities to outboard occupants.  Except for the lap belt, which is for rear seats, all other values are for drivers and right-front passengers.





                                      Effectiveness in  


                                    preventing outboard


                                    occupant fatalities







                Airbag plus                             

                                           46 + 4       

                lap/shoulder belt                      




                Lap/shoulder belt         41 + 4       





                Shoulder belt             29 + 8       




                Lap belt                               

                                           18 + 9       

                (rear seat)                            




                Airbag only               17 + 4       




Table 9-7.        Comparison of estimates of fatality reducing effectiveness for outboard occupants presented in Table 9-6 with those in three prior studies.  The lap belt effectiveness in Table 9-6 is for rear seats only, whereas the other lap belt estimates are for mainly front seats.  All other estimates are for front seat occupants, mainly drivers.




                      Fatality reducing effectiveness estimate (percent)  




                      Value in                Wilson and   Huelke et  

  device                         NHTSA [1984]                          

                      Table 9-6               Savage [1973] al. [1979] 





Airbag plus                                                           

                       46 + 4       45 - 55         -           -      

lap/shoulder belt                                                     





Lap/shoulder belt     41 + 4       40 - 50        31           32     





Airbag plus                                                           

                         -          40 - 50        29           34     

lap belt                                                              





Shoulder belt         29 + 8          -           -            28     





Lap belt              18 + 9       30 - 40        17           13     





Airbag only           17 + 4       20 - 40        18           25