Accident Analysis and Prevention 35 (2003) 361–367

The paradox of driving speed: two adverse

effects on highway accident rate
David Navon∗
Department of Psychology, University of Haifa, Haifa 31905, Israel
Received 17 July 2001; received in revised form 15 November 2001; accepted 26 November 2001

Abstract
Whereas speeding is known to be a substantial risk factor in driving, there is no unequivocal evidence that accident rate on limited-access
motor highways is considerably affected by average speed or by speed limits meant to regulate it. It is suggested here that the seeming
puzzle actually may have a straightforward explanation: accident–prone interactions (APIs) between cars occur when they pass each
other—mostly moving in the same directions or in opposite ones. Such interactions are shown here to happen more frequently, the lower
average speed is. To the extent that high speed limits contribute to increase in average speed, they serve to reduce the number of such
interactions, thereby to moderate at least part of the negative effect of speed on the driver’s ability to avoid an impending accident.
© 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Speed limit; Driving speed; Average speed; Relative speed; Speed variance; Speeding; Highway accidents; Crash rate; Accident rate

1. Introduction Likewise, the effect of speed limit regulations meant to

decrease average speed is far from being clear. To the ex-
Taking measures to reduce average driving speed in motor tent that speed limit determines actual driving speed, it is
highways, like enacting low speed limits for them, is often expected to affect accident rate. However, in a study that
justified by the slogan “speed kills”. compared speeding behavior and accident rates under dif-
This slogan is founded on a sound empirical basis. It is ferent speed limits (Parker, 1997), it was found that the
obvious that an increase in speed greatly increases percep- rate of accidents was not consistently related with speed
tion distance (viz. the distance traveled by the car from the limit. Drawing a conclusion from that study should be sus-
moment a stimulus is visible to the moment it is perceived pended until further corroborated, and its data might possi-
by the driver), decision distance, braking distance and the bly be due to some sort of artifact (Shinar, 1999). On the
physical impact of an accident on the body. It is equally clear other hand, Parker’s results seem compatible with the fact
that excessive speed may compromise the control of the car that accident rates on motor highways that have no speed
by the driver and it is well documented that excessive speed limit (as the case is in many German autobahns) are surpris-
considerably increases crash involvement rate, let alone rate ingly low. That might be attributed to the heavy traffic vol-
of injuries, per a given mileage of travel within any given ume on those highways which acts to reduce average speed
road-type (e.g. Cirillo, 1968; Fildes et al., 1991; Munden, which in turn diminishes accident rate (e.g. Elsner et al.,
1967; Solomon, 1964; Treat et al., 1977; Viano and Ridella, 2000). However, the indirect effect of traffic volume on ac-
1996). cident rate might be mediated in other ways, e.g. via driver’s
On the other hand, the effect of average speed of traffic vigilance.
is far less clear. Some extensive analyses of accident data In any event, the results cited here do not appear quite
yield no effect for average speed (e.g. Cirillo, 1968; Fildes commensurate with the expectation that average driving
et al., 1991; Munden, 1967; Solomon, 1964; TRB, 1984; speed must be a major determinant of accidents. One is led
Treat et al., 1977; Viano and Ridella, 1996). Some others to wonder whether there is no inherent difference between
point to both positive and negative effects, depending on the aggregate effect of speed on the one hand and the effect
accident-type (Shinar, 1999). of excessive speeding within a given speed limit regime
on the other hand. Indeed, there is some indirect causal
∗ Tel.: +972-4-8240921; fax: +972-4-8240966. mechanism that moderates the harm potential of average
E-mail address: dnavon@[Link] (D. Navon). speed.

0001-4575/02/$ – see front matter © 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 0 1 - 4 5 7 5 ( 0 2 ) 0 0 0 1 1 - 8
362 D. Navon / Accident Analysis and Prevention 35 (2003) 361–367

1.1. Accident–prone interactions where E(D|A) is the damage that an accident is expected to
cause if it occurs.
Sometimes the same factor has opposite effects on dif- Hence, the number of accidents within a period of M time
ferent links in a causal chain. For example, it might be that units depends on the number of APIs (i.e. M × p(API)) as
some factor increases susceptibility to an illness yet reduces much as it depends on how accident–prone they actually are
the prevalence of its agents. That is probably the case with (i.e. p(A|API)), and the potential damage of those accidents
driving speed. also depends heavily on the number of APIs.
Most fatal or severe accidents in limited-access motor Ironically, there is a reason to suspect that severe speed
highways occur after two or more cars have entered a state limits act to considerably increase the number of APIs.
of high proximity. Such a close encounter need not be the APIs occur mainly in four types of situations: (a) when
primary cause, but it definitely serves as a necessary con- cars moving on intersecting roads come at about the same
dition for some types of accident and as a facilitative agent time to an intersection, (b) when cars moving in opposite
for others. Let such an encounter be called accident–prone directions on the same undivided road pass each other, (c)
interaction (API). when cars moving in the same direction on different lanes
Given that an API arises, speeding tends to increase both of the same road momentarily drive next to each other, and
the probability that the API in turn develops into an actual (d) when a car coming from the rear of another one switches
accident and the damage that the latter is expected to cause lanes next to the other one (either by switching to an adjacent
when it occurs. However, since the very existence of an API lane to overtake it, or by returning to the original lane).
is a necessary condition for the occurrence of an accident of
the sort in question, the likelihood of such an accident per Case (a). The frequency of type-a APIs is unaffected by
unit time, p(A), is a product of the likelihood of the occur- driving speed, by the same token that the amount of rain
rence of an API per unit time, p(API), and the conditional drops falling on a person walking in the rain by and large
probability that the API in turn develops into an actual ac- does not depend on her pace of walking. However, that case
cident, p(A|API), namely is mostly irrelevant for limited-access highway driving.
p(A) = p(API) × p(A|API) (1) Case (b). The frequency of type-b APIs is greatly reduced
by driving speed. The shorter is total time on the road for
Similarly, the equation for the expected damage across such
any given car, the smaller the number of its encounters with
accidents, E(D), isvspace*1pt
cars going in the opposite direction. Imagine, for example,
E(D)=p(A)×E(D|A)=p(API)×p(A|API)×E(D|A) (2) a 60 km countryside road in which the mean number of cars

Fig. 1. An illustration for the effect of speed on the frequency of type-b APIs. The path of a car is represented by an arrow from A to B or vice versa.
Each crossing of two arrows is an API.
 D. Navon / Accident Analysis and Prevention 35 (2003) 361–367 363

Fig. 2. An illustration for the effect of speed on the frequency of type-c APIs. The path of a car is represented by an arrow from A to B. The arrow
corresponding to the early car is solid; the arrow corresponding to later cars is dashed. Each crossing of two arrows is an API.

going in each direction in an hour is 1. If two cars setting time lag that still enables the late car to catch up with the
out from both ends move at a speed of 60 km/h, they are early one (see Fig. 2).
bound to pass each other provided that their starting times Putting it more intuitively, for any given time lag, the
are not more than 1 h apart. If, however, they both move at lower the speeds, the smaller the initial inter-car distance,
a speed of 120 km/h, they are bound to pass each other only and the longer the time-yet-to-stay on the road for the early
provided that their departure times are not more than 1/2 h car at the start of the late one, which gives more time for
apart. diminishing the initial inter-car distance to 0.

Putting it differently, a car departing from A on t1 and Case (d). Like type-c APIs, type-d ones involve the ap-
arriving at B on t1 + x encounters any same-speed car that proaching of cars moving in the same direction with differ-
departs from B between t1 − x and t1 + x. Since x, namely ent velocity. Hence, their frequency is similarly reduced by
duration of stay on the road, equals distance/speed, increas- average driving speed.
ing the speed by a factor of g will reduce the number of
encounters by 1/g (see Fig. 1).

Case (c). The frequency of type-c APIs is also greatly 2. Simulation results
reduced by driving speed.
To demonstrate the effects of average driving speed on the
Suppose two cars set out from A to B with some time lag, frequency of APIs, as well as provide some idea about their
∆, where the early one moves at a constant velocity, v, and magnitude, I present here frequencies of simulated APIs for
the late one moves p times faster, namely with velocity pv. four simplistic speed distributions differing in average, vari-
The early car departs from A on t1 and arrives at B on t1 +x, ability or both. This was done by simulating traffic mov-
and the late one departs from A on t1 + ∆ and arrives at B ing on a 60 km highway connecting A and B. Car departure
after y min, namely on t1 + ∆ + y. The late car will catch up was governed by a Poisson process, with a probability of
with the early one whenever x ≥ ∆ + y, namely whenever 0.05 s−1 . The simulated traffic within a restricted sector of
∆ ≤ x − y. Since x and y are quotients of the same distance, time–space is illustrated in Fig. 3.
d, and the respective speeds, v and pv, the left wing also The actual speed of any departing car could assume any
equals d(1 − 1/p)/v. Hence, increasing the speeds of both of five values with equal probability. The speed of cars was
cars by a factor of g will curtail by a factor of g the maximal set to be constant throughout travel (assuming for simplicity
364 D. Navon / Accident Analysis and Prevention 35 (2003) 361–367

Table 1
Frequencies of APIs of various categories as a function of average speed (in km/h) and speed variability (making up four speed sets), for cars departing
over a 1 h period with embedding traffic
Average Variability

Low (range/mean = 1/7) High (range/mean = 4/7)

Low speed (mean = 70)

Speed set (65, 67.5, 70, 72.5, 75) (50, 60, 70, 80, 90)
Frequency of b-type 73534 77938
Frequency of simple c(d)-type 1430 6858
Frequency of co-occurrence of b- and simple c(d)-type 38 204
Frequency of risky c(d)-type 43 195
Frequency of co-occurrence of b- and risky c(d)-type 32 38
Total 75077 85233
High speed (mean = 112)
Speed set (104, 108, 112, 116, 120) (80, 91, 112, 128, 144)
Frequency of b-type 42087 45202
Frequency of simple c(d)-type 924 4461
Frequency of co-occurrence of b- and simple c(d)-type 15 81
Frequency of risky c(d)-type 18 81
Frequency of co-occurrence of b- and risky c(d)-type 7 9
Total 43051 49834
The five API categories listed are mutually exclusive. The term “risky c(d)-type” denotes the coincidence of at least three cars moving in the same direction.

separately. In addition, c(d)-type APIs were broken down to

coincidences of two cars going in the same direction (termed
simple c(d)-type) and coincidences of more than two cars
going in the same direction (termed risky c(d)-type). The
latter type of coincidence must be particularly risky, when
there are no more than two lanes for the same-direction traf-
fic. No simulation of lane choice was attempted, but any
coincidence of at least three cars going in the same direc-
tion on the same road segment was counted as a c(d)-type
API of the risky version. In addition, the frequency of the
extremely dangerous co-occurrences of that risky version of
a c(d)-type API with an API of type-b was also counted.
APIs were counted for all cars departing over a 1 h period.
To simulate embedding traffic, car departure governed by
Fig. 3. An illustration for the simulated traffic for a speed set of (80, the same Poisson process was allowed both up to 1 h before
91, 112, 128, 144). The diagram presents traffic within a 5 km road and up to 1 h after the observed period.
segment taken from the middle of the road (the ordinate), and within a The results of those calculations (averaged across 50 runs)
time window of 320 s taken from the middle of the measured period (the
are presented in Table 1.
abscissa). The path of a car is represented by a single slanted line: solid
line for cars going from A to B; and dashed line for cars going from B As can be seen in Table 1, the number of APIs is in-
to A. Each crossing of two or more lines is an API. versely related with average speed. This is especially pro-
nounced with respect to the most risky APIs resulting from
the co-occurrence of at least three cars on the same road seg-
no mutual impeding, let alone jams1 ), but any coincidence ment. A 60% increase in mean speed was found to decrease
of cars on the same road segment was counted as an API, the total number of the latter by a factor of 2.6 or 2.8 (for
of types b, c (or2 d) or of their co-occurrence as the case high and low variability speed distributions, respectively).
may be. A co-occurrence of APIs was considered to have Thus, unlike speed variability, high average speed does not
occurred when its constituent APIs coincided within a dis- seem to cause accident–prone situations. On the contrary, it
tance of 10 m. Its constituent APIs were not counted then acts to considerably reduce their number.
Table 2 presents the frequencies of APIs of the various
1 Which requires that the conditions on the road, including times when
types as a function of speed within each of the speed distri-
APIs occur, allow unobstructed travel. That premise is, of course, unre-
alistic, but it simplifies analysis.
butions. The frequency of APIs for any given speed includes
2 The difference between types c and d depends on pre-coincidence all encounters in which at least one car has that speed. APIs
lane choice. That was not simulated here. with cars having different speeds are counted once for each
 D. Navon / Accident Analysis and Prevention 35 (2003) 361–367 365

Table 2
Frequencies of APIs of various categories as a function of speed (in km/h), average speed and speed variability, for cars departing over a 1 h period with
embedding traffic
Average Variability

Low (range/mean = 1/7) High (range/mean = 4/7)

Low speed (mean = 70)

Speed 65 67.5 70 72.5 75 50 60 70 80 90
Frequency of b-type 27720 26942 26497 26319 24902 33801 29854 27315 25802 23518
Frequency of simple c(d)-type 748 500 426 507 679 4132 2309 1954 2354 2968
Frequency of co-occurrence of 25 18 18 19 22 150 101 89 95 110
b- and simple c(d)-type
Frequency of risky c(d)-type 29 26 24 26 27 148 116 104 107 114
Frequency of co-occurrence of 23 22 22 21 20 29 27 24 25 22
b- and risky c(d)-type
Total 28545 27508 26987 26892 25650 38260 32407 29486 28383 26732
High speed (mean = 112)
Speed 104 108 112 116 120 80 91 112 128 144
Frequency of b-type 15844 15357 15165 15101 14305 19339 17596 15760 14982 13691
Frequency of simple c(d)-type 483 323 278 328 437 2456 1644 1323 1559 1941
Frequency of co-occurrence of 10 8 7 8 9 56 42 36 38 44
b- and c(d)-type
Frequency of risky c(d)-type 12 11 10 11 12 59 51 43 46 47
Frequency of co-occurrence of 5 5 5 5 5 7 6 6 5 5
b- and risky c(d)-type
Total 16354 15704 15465 15453 14768 21917 19339 17167 16631 15728
The five API categories listed below are mutually exclusive. The term “risky c(d)-type” denotes the coincidence of at least three cars moving in the
same direction. Frequencies of APIs for any given speed include all encounters in which at least one car has that speed.

speed. Hence, the frequencies of APIs for different speeds 1.1), these results demonstrate an unmistakable effect. How
do not sum to the overall number of APIs presented in does it square with the finding that excessive speeding is
Table 1. associated with a rise in the incidence of severe accidents?
As can be seen in the table, within each speed distribution The answer calls for noting two important distinctions.
the total number of APIs across all categories monotonically
decreases with speed. That is also true of the number of APIs 3.1. Relative versus average speed
of type-b or, by and large, of the number of co-occurrences
of an API of that type with a risky c(d)-type API. The num- The effect of speeding relative to other cars in the traf-
ber of c(d)-type APIs, on the other hand, is minimal at some fic flow, holding average speed constant, should be distin-
medium speed (cf. a similar result for two cars c(d)-type guished from the effect of any factor, like speed limit, that
APIs in Hauer, 1971), as is the co-occurrence of a simple affects average speed of the entire traffic flow.
c(d)-type API with a b-type API. The latter curvilinear rela- Within every speed limit regime, excessive speeders are
tionship is probably due to the simplifying assumption that more liable to be involved in divided-highway accidents than
overtakes are completed instantaneously. In the real world, drivers who stay close to median speed (see, e.g. Shinar,
however, c(d)-type encounters may take quite a long time, 1999). As can be seen in Table 2, although the number of
especially when the conditions of the road do not allow type-b APIs decreases with speed, type-c APIs behave dif-
unobstructed travel, thus, allowing more co-occurrences ferently. Driving at median speed minimizes the sum of over-
with b-type encounters. In that case, the number of such taking interactions—number of passing cars plus number of
co-occurrences was found to monotonically decrease with cars being passed (Hauer, 1971). Consider now type-d APIs.
speed. A driver who intends to drive at high speed is likely to be
In any event, for every category of encounters, the slowest more frustrated by having to drive behind a slow car than
speed is associated with the highest number of APIs. a driver who cares less for speed, hence, the former would
be more liable to overtake, sometimes even when the traffic
coming from the rear on the overtaking lane is too close for
3. Discussion comfort.
Note, however, that this scenario describes the effect of
The results of the simulation are just illustrative, so order speeding within a given local environment in which other
of magnitudes should be taken with a grain of salt. However, cars drive more slowly. The case is different with the ef-
coupled with the mathematical reasoning above (see Section fect of any factor that modulates the average of the entire
366 D. Navon / Accident Analysis and Prevention 35 (2003) 361–367

distribution of speed. Whereas relative speeding is clearly find reasonable. That proportion is probably negatively re-
risky for any given environment, it is less clear that there is lated to speed limit. Hence, a reasonably high speed limit
much harm to motorists in affecting the environment itself by could serve to diminish the motivation to attempt dangerous
any measure that increases average speed. Table 1 clearly il- attempts of lane switching.
lustrates that the number of APIs considerably declines with
average speed. Table 2 further illustrates that that advan- 3.3. Adverse effects of speed
tage is not restricted to medium-speed drivers. High-speed
drivers benefit from it as well. For example, the number of Even when the total risk across all cars within an entire
APIs expected for a driver who drives at 90 km/h when the traffic flow is considered, speed acts in two adverse ways:
speed set ranges from 50 to 90 km/h would be considerably on the one hand, it compromises drivers’ ability to avoid
smaller if she drove at the same speed when the speed set impending accidents, thereby increasing p(A|API), and it
ranged from 80 to 144 km/h, and even smaller than that if she as well increases the expected damage per such accident,
elected to use the change in speed set to drive at 144 km/h. E(D|A). On the other hand, average speed considerably
Furthermore, the number of APIs of type-b does mono- reduces p(API).
tonically decrease with speed. Accordingly, speeding within Hence, to get an idea of the net effect of average speed on
a given local environment reduces at least the likelihood of p(A) and E(D), it remains to compare its negative effect on
being involved in an accident–prone interaction caused by p(API) with its positive effects on p(A|API) and E(D|A). As
traffic coming from the opposite direction. That sort of in- can be seen in Eqs. (1) and (2), the former must counteract
teraction is likely mostly in undivided highways, but not to- the latter two, yet it is not clear whether that compensation
tally impossible in divided highways (e.g. due to dazzling). is small or considerable. In any event, the incompatibility
between the large predicted effect of speeding and the com-
3.2. The effect of speed limit on number of APIs paratively mild gross empirical effect of average speed may
not be as disturbing as it intuitively appears to be.
Raising legal speed limit clearly acts to reduce the number As for the net effect of speed limit, it should be noted
of APIs by increasing average speed. Could that effect be that the reduction in p(API) with higher speed limit affects
partly diluted by an increase in speed variability? Note that any individual driver, whereas the effect of speed limit on
the frequency of overtakes is directly related to the variabil- p(A|API) depends mainly on the extent that the raise in speed
ity of intended speed (Lave, 1985; Rodriguez, 1990; Garber limit actually have induced the particular drivers involved
and Gadiraju, 1988; see Shinar, 1999 for a review): If all in that API to increase driving speed. In other words, when
drivers drove at the same speed, no car would catch up with the speed limit is raised, a driver whose actual driving speed
any one that moves ahead of it. The effect of speed variabil- remains the same can benefit from lower expected number
ity is also illustrated in Table 1. However, speed variability of APIs (see Table 2) yet not suffer much increase in the
was not found to be consistently related across studies to probability that an API she happens to be involved in results
average speed (Brown et al., 1990; Freedman and Williams, in an accident.
1992; Garber and Gadiraju, 1988; Levy and Asch, 1989; Note, however, that the reduction in traffic volume due to
McCarthy, 1991). That might appear strange, considering raising speed limit might lead to a decline in the vigilance
the common correlation between means and standard devi- of at least some drivers in sparse-traffic roads. That might
ations. The explanation could be that low speed limits in- cause some rise in p(A|API) that would somewhat counteract
crease the difference between intended speed and prescribed the effect of the reduction in p(API) on p(A).
speed, which must be a potent source of variability in ac-
tual speed, since while some drivers comply with the limit, 3.4. A brief summary
some others violate it or adjust their momentary speed to
the momentary likelihood of an impending meeting with an In view of these adverse effects of speed, individual choice
enforcer of the law. of driving speed constitutes an intriguing social dilemma: if
Consider now another effect of speed limit on the fre- any driver could be certain that all other ones chose the same
quency of dangerous lane switching. The likelihood that a speed as she did, there probably would be no much harm in
type-d API turns into lane switching presumably correlates choosing high speeds. However, since practically she cannot
highly with the extent of the discrepancy between the in- presume that, she would do well to drive at medium speed, at
tended speed of the driver of the car coming from the rear least in divided highways, both for minimizing her own risk
and the momentary speed of the car moving in front of and for refraining from risking others. Whereas this mes-
it. That discrepancy must have to exceed some frustration sage is worth being publicly delivered to drivers, it is more
threshold before it turns into an action of lane switching. doubtful that there is much utility in resorting to enacting
The frequency of cases in which drivers are frustrated low speed limits for limited-access motor highways. That
enough by the low speed of cars driving in front of them must increase, to the extent that it affects driving speed, the
to attempt lane switching must increase with the proportion number of accident–prone interactions, which might com-
of cars moving more slowly than what most other drivers pensate for the increase in the expected loss from any such
 D. Navon / Accident Analysis and Prevention 35 (2003) 361–367 367

interaction. The actual extent of this moderating effect is a Hauer, E., 1971. Accidents, overtaking, and speed control. Acc. Anal.
matter for further, extensive studies. Prev. 3, 1–13.
Lave, C., 1985. Speeding, coordination, and the 55-mph limit. Am. Econ.
Rev. 75, 1159–1164.
Levy, D.T., Asch, P., 1989. Speeding, coordination, and the 55-mph speed
Acknowledgements
limit: comment. Am. Econ. Rev. 79, 913–915.
McCarthy, P.S., 1991. Highway safety and the 65-mph speed limit.
I am indebted to Ziziana Lazar for programming the sim- Contemp. Policy Issues 9, 82–92.
ulations, to Ronen Kasten for his assistance in analyzing the Munden, J.W., 1967. The relation between a driver’s speed and his
results, to Jeff Miller and David Shinar for helpful comments accident rate (Report LR 88). Road Research Laboratory, Crowthorne,
on an early draft of the concept, and to the editor and two Berkshire, England.
Parker, M.R., Jr., 1997. Effects of lowering and raising speed limits on
anonymous reviewers for comments on a later draft. Corre- selected roadway sections (Report FHWA-RD-92-084). US Department
spondence should be addressed to dnavon@[Link]. of Transportation, Federal Highway Administration.
Rodriguez, R.J., 1990. Speed, speed dispersion, and the highway fatality
rate. Southern Econ. J. 10, 349–356.
References Shinar, D., 1999. Speed and crashes: a controversial topic and an elusive
relationship. Traffic Eng. 41, 52–55.
Brown, D.B., Maghsoodloo, S., McArdle, M.E., 1990. The safety impact Solomon, D., 1964. Accidents on main rural highways related to speed,
of the 65-mph speed limit: a case study using Alabama accident driver, and vehicle. US Department of Commerce, Bureau of Public
records. J. Safety Res. 21, 125–139. Roads, USA, July.
Cirillo, J.A., 1968. Interstate system accident research: study II, interim TRB., 1984. A Decade of Experience (Special Report 204:55). National
report II. Public Roads 35, 71–75. Research Council, Washington, DC.
Elsner, A., Reichwein, S., Schepers, A., 2000. Verkehrssicherheit auf Treat, J.R., Tumbas, N.S., McDonald, S.T., Shinar, D., Hume, R.D.,
europrischen Autobahnen. Strasse und Autobahn 51, 65–72. Mayer, R.E., Stansifer, R.L., Castellan, N.J., 1977. Tri-level study of
Fildes, B.N., Rumbold, G., Leening, A., 1991. Speed behavior and drivers’ the causes of traffic accidents, Vol. I. Causal factor tabulations and
attitude to speeding (Report). Monash University Accident Research assessment (DOT-HS-805-085. NHTSA). US Department of Transpor-
Center, Monash, Vic., Australia, 16 June. tation.
Freedman, M., Williams, A.F., 1992. Speeds associated with 55-mph and Viano, D.C., Ridella, S., 1996. Crash causation: a case study of fatal
speed limits in northeastern states. ITE J. 62 (2), 17–21. accident circumstances and configurations (no. 960458). Society of
Garber, N.J., Gadiraju, R., 1988. Speed variance and its influence on Automotive Engineers, International Congress and Exposition, Detroit,
accidents. University of Virginia, Charlottesville. MI, 26–29 February.