Accid. Anal. and Prev. Vol. 25, No. 5, pp. 555-559, 1993 Printed in the U.S.A.
OOOl-4575/93 $6.00 + .oO 0 1993 Pergamon Press Ltd.
MEASURING CAR DRIVERS’ SKILLS-AN ECONOMIST’S VIEW FINN JORGENSEN Department
of Economics,
Nordland
(Received
College, Bode, Norway
12 May 1992)
Abstract-This paper models driver behaviour in a way that allows us to define the driver’s overall driving skills in a fruitful manner. Thus, the model enables us to capture the main influence of training courses on driving. The model’s results are seen in the light of a Norwegian investigation that concluded that slippery surface driving courses have increased traffic accident rates significantly. Norwegian car drivers seem to overestimate the driving risk before they pass these courses, which reduce drivers’ risk perception relatively more than they reduce the objective risk of skidding.
1. INTRODUCTION In 1979, the Norwegian authorities launched a new plan for the compulsory driving training course. One significant change from the old programme was that aspiring car drivers have to pass a slippery surface driving course (SSDC) in order to obtain a permanent driving licence. Many reckoned this course to be of great importance with regard to reducing traffic accidents. The effect on accident rate of the SSDC has been analyzed by Glad (1988). His results were quite astonishing to many people and received considerable publicity in the media. He concluded that the SSDC had significantly increased the accident rate per distance driven among male car drivers-especially in the first couple of years following the course. For female drivers the SSDC had no significant influence on their accident rate. All types of accidents were reported in Glad’s investigation. Hence, the majority were damage only. For a detailed description of Glad’s methodological approach and results, refer to Glad (1988). Glad’s findings led people to believe that the SSDC certainly did not improve car drivers’ skills and that consequently it should be abandoned or its content changed substantially. The aim of this paper is to show that the above conclusion may well be wrong. This is done by modelling driver traffic safety behaviour in a way that I. allows us to define a driver’s overall driving skills in a fruitful manner and 2. that captures the main impact of driving training courses.
2. A MODEL
OF DRIVER
BEHAVIOUR
The model of driver’s behaviour presented in this paper is similar to those presented in Peltzman (1973, O’Neill(1977), Blomquist (1986) and Janssen and Tenkink (1988) in the sense that the car driver is assumed to be a risk neutral utility maximizer. To model properly the influence of the SSDC an important extension of these models is, however, made: among traffic psychologists, a car driver’s perceptual skills are regarded as important determinants when describing his overall driving skills (Wilde 1982). Therefore, I will make a distinction between drivers’ perceived level of risk and accident loss on one hand and their respective objective values on the other hand. By specifying the driver’s behaviour with his selected speed (x), more restrictions are placed on the actual functions than in previous works. This enables me to come up with clearer conclusions as far as driving skills are concerned. Perceived und objective probability of an uccident In the following section I suppose that a driver will experience two states of the world per unit of distance driven; either an accident does occur with probability p” or it does not occur with probability (1 - p”). The probability that the driver will be involved in an accident depends on his own selected speed, such that the derivatives py, p,:., > 0* hold. The p”(x) relationship is influenced by the driving characteristics of the car, the quality of the roads, *These assumptions are in line with ordinary assumptions in moral-hazard models: an increase in the safety effort of an agent (reduction in x) reduces the probability of an accident at a diminishing rate (see for instance Viscusi [1984]).
Measuring
Using eqns l-6, eqn 7 may be written Min,TC” = tlx + yhx”
y = c$, b = sob,, c = a, + 6,.
(8)
The optimal value of .Xis: x* = [tiybc]”
where d = l/(1 + c).
(9)
From earlier assumptions it follows that c > 2. Thus, it is straightforward to show from eqn 8 that TC:, > 0, such that x8 is a global optimal value. The objective expected loss per unit of distance (Q) is: QO(x) = p”(x)L”(.u) = hx’
(10)
Assuming that technical improvement of roads and cars will reduce the product h = a, . b, onlyS, the impact of such improvements may be analyzed by using eqns 9 and 10. From eqn 9 we can deduce the following elasticity: EL&+
= -d
Equation 9 in combination
=c0.
(11)
with 10 gives:
Q: = blt/bcy]‘d.
(12)
From eqn 12 follows: EL,&:=
1 - cd= d>O.
(13)
Thus, according to the model specification presented here, better driving conditions will increase speed but reduce objective expected accident loss. It is worth noting that EL/,X* and EL,,@ are independent of a and p. This means that relative changes in speed and expected loss from changing driving conditions are not influenced by the extent to which adriver underestimates or overestimates driving risk and accident loss. 3. A CAR DRIVER’S
SKILLS
A common view among traffic psychologists is that a driver’s overall driving ability comprises three types of skill: (Wilde 1982) first, perceptual skills, which determine the extent to which subjective estimates of the probability of an accident and accident
$A thorough (1991).
discussion
of this matter
is given in J@rgensen
car drivers’
557
skills
loss agree with their respective objective values; second, decisional skills, which refer to the driver’s ability to judge what should be done in different traffic situations; third, vehicle handling skills, which determine whether the driver effectively can implement his desired actions. Let us discuss in more detail how these types of driving skills can be interpreted in the proposed model. By using eqns 2 and 4 I assume that a driver’s perceptual skills are defined by Y = a/3
y > 0.
(14)
If y 5 1, the driver underestimates, correctly estimates, or overestimates, the real expected loss per unit of distance. A driver’s perceptual skills increase when y -+ 1. Then the difference between perceived and objective accident loss is reduced at any speed. Both the decisional and vehicle handling skills of a driver influence primarily the relationship between objective probability of an accident and speed; when such qualities of driving skills improve, the p”(x) relationship shifts downwards. In the following 1 assume that these types of skills influence a, and not a, in eqn 1. This is a sensible assumption (Jorgensen 1991). Consequently, a change in the driver’s skills will alter his objective probability of an accident with the same percentage at any speed. Similar assumptions are made concerning the L”(h) relationship; an improvement of the driver’s decisional and vehicle handling skills will reduce 6, in eqn 2 only. However, the influence on b, of changing driving skills is likely to be small (Jorgensen 1991). In summary, a reasonable indicator of a driver’s decisional and vehicle handling skills is: b = a,b, .
(15)
When these kinds of driving skills improve such that b decreases by y percent, the expected accident loss per unit of distance will decrease by y percent at any speed. The driver’s time cost per unit of time (t) is not influenced by his driving skills. From eqn 6 it thus follows that the T(x) function is invariant of the driver’s skills. Hence, in the model, a driver’s overall driving skills are determined only by the values of y and b. One index of overall
driving
skills
To discuss in more detail how perceptual skills on the one hand and decisional and vehicle handling skills on the other influence a car driver’s overall driving skills, let us take the sum of his time costs
558
F. J~RCENSEN
and real expected point. Using eqns Tcyx)
=
accident loss (K’“) as a starting I, 3, and 6, TC’” may be written:
f/x + bx’
b) =
t(tlybc.) -” + b(tlybc)‘-”
(16)
(17)
TC,” denotes the sum of the driver’s time costs and real expected accident loss at his chosen speed. It obviously says something about his driving skills; at a fixed value oft, the lower the value of Tc’:, the better the driver. According to previous discussions. t and c are assumed not to be influenced by the car driver’s skills. Thus a driver’s skills will influence TC,O only through their influence on -y and b. After some transformation following elasticities: EL,TC:‘=
SLIPPERY SURFACE DRIVING COURSE
The influence b = aoh<,, L’ = uI + b,
Note that TCV) # TC‘(x) when y # 1, that is when the driver has incorrectly estimated expected loss per unit of distance. As the driver chooses a speed that minimizes TC’, it follows when using eqn 9 in combination with eqn 16 that: TP(y,
4. THE
to eqn 17 we get the
on overall
driving
skills
What can we now say about the impact of the SSDC? The SSDC includes theoretical as well as practical training. The main objective of the theoretical part is to inform drivers about tyre adhesion to the road under different types of winter road conditions. It is reasonable to believe that this learning will improve drivers’ perceptual skills in identifying slippery surfaces; that is y approaches to I. Thus, y will increase if the driver underestimates winter driving risk before the learning process starts (y < I), whereas y will decline if he overestimates these risks prior to learning (y > I). During the practical part of the SSDC the drivers practise various emergency manoeuvres under winter driving conditions. These exercises take place on special tracks specifically designed for driving training purposes. Thus, the practical part of the SSDC will improve er’s decisional skills and vehicle handling plying a decrease of the value of b in the The above conclusion, combined with 18 and 19, suggests that SSDC has led to a in TC:’ indicating that the SSDC really has
the drivskills, immodel. formulae reduction
improved
the driver’s overall driving skills.
(I - ci)y’ c-y + 1
Glad’s (1988) results indicate that objective expected accident loss per unit of distance (Q? is greater after rather than prior to the SSDC for male
and EL,,TC” = d. l
(19)
From eqn 18 it follows that EL, TCP is negative when y < I and positive when y > I. Hence the TC,O(y) relationship is U-shaped and reaches a minimum when y = 1. This is sensible; when y < 1 and is increasing, the driver’s perceptual skills are improving and he becomes a better driver. Similarly, for y > 1 the driver’s perceptual skills worsen and his driving skills become poorer as y increases. Equation 19 reveals an increasing concave relationship between TC’Z and b (0 -=Id < I). This result, too, is in line with earlier statements; an increase in b means that the driver’s decisional skills and vehicle handling skills worsen. Thus, his overall driving skills worsen. It is worth noting that both EL,TC,O and EL,TCz are independent of t in the model. As the value of c does not differ greatly between car drivers (J@-gensen 1991), it follows that EL,TC: has nearly the same value for all drivers. The value of EL,,TC,“, however, varies between car drivers due to different values of y.
drivers, whilst the course has no significant influence on p for female drivers. From eqn I2 it follows that: EL,Q’:=
pcd=d
-
1
(20)
to eqn I3 EL,,Q: = d > 0. As the SSDC reduces the value of b. this points in the direction
According
of lower Qz for drivers who have passed the SSDC. Hence from eqn 20 it may be concluded that y must be lower uftfipr rather than prior to the course, which leads to an increase in QJ. The latter assertion leads to the interesting conclusion that male as well as female drivers who entered the SSDC overestimated winter driving risks (y > 1). From eqn 12 it follows that: b = b(r) = @‘d(t/c)d-
$/’
e=Q:,
(21)
and EL,b(y)
= L’> 0
(22)
Measuring
The b(y) relationship in eqn 21 denotes combinations of decisional and vehicle handling skills on the one hand (h-value) and perceptual skills on the other hand, which allows Qi, to remain constant. As the c-value is definitely greater than 1, eqn 21 reveals an increasing convex relationship between h and y. A reasonable estimate of c is 4 (Jorgensen 1991). Hence, it follows from eqns 21 and 22 that the SSDC has caused a relative decrease in y which is at least equal to I14 of the decrease in 6. To sum up, one cannot conclude from Glad’s (1988) investigation that the SSDC has worsened car driver’s skills. As y decreases because of the SSDC, the SSDC has, however, reduced driver’s risk perception relatively more than the reduction in objective risk-at least 25% more. Thus, the drivers overestimated objective driving risk before they passed the SSDC.
5.
SUMMARY
The most important message from this analysis is that one cannot judge the quality of a driving training programme by its influence on accident rate alone. One has also to take into account that the course may change drivers’ choice of speed and hence risk acceptance. Both these ideas are captured by looking at the objective total driving costs per unit of distance. This cost index also enables us to
car drivers’
559
skills
discuss the impact of training on different types of driving skills, as defined by traffic psychologists, and on drivers’ overall driving skills. Obviously, this is a fruitful discussion when planning the content and procedure of a driver training programme. Ac,linoM,/rdgemenru-The Lensberg, and Djamel
author thanks PHI Pedersen, Abderrahmane for comments.
Terje
REFERENCES Blomquist, G. A utility maximation model of driver traffic safety behaviour. Accid. Anal. Prev. 18:371-375; 1986. Glad, A. Fase 2 i foreropplaringen. Effekt pa ulykkesrisikoen (Stage 2 in the Norwegian driving training program-its impact on accident rate). In Norwegian. Oslo: Institute of Transport Economics: 1988. Janssen, W. H.; Tenkink, E. Consideration on speed selection and risk homeostasis in driving. Accid. Anal. Prev. 20: 137-142; 1988. Jones-Lee, M. W. The economics of safety and physical risk. Oxford: Basil Blackwell; 1989. Jorgensen, F. Kjorehastighet og trafikksikkerhet (Drivers’ speed selection and traffic safety). In Norwegian. Bode, Norway: Nordland College; 1991. O’Neill. B. A. A decision-theory model of danger compensation. Accid. Anal. Prev. 9:157-165; 1977. Peltzman. S. The effects of automobile safety regulation. Journal of Political Economy. 83:677-725; 1975. Viscusi, W. K. The lulling effect: The impact of childresistant packaging on aspirin and analgesic ingestions. American Economic Review. 74:324-327; 1984. Wilde, G. J. The theory of risk homeostasis. Implications for safety and health. Risk Analysis. 2:209-225: 1982.