Towards a general theory of driver behaviour
Introduction
Reaching a destination is usually the main goal of driving. In the decision-making process to achieve this goal, feedback is usually self-evident as the driver navigates towards and approaches her or his destination. Subsumed under this goal are a variety of secondary goals among which there has been a lasting controversy regarding the role played by risk of collision. In several formulations (e.g., Näätänen and Summala, 1976) this risk has been assumed to be predominantly a zero risk of collision, in others (e.g., Gibson and Crooks, 1938, Wilde, 1982, Adams, 1985) a target level of risk has been proposed. This paper will argue that risk of collision is generally not relevant in the decision-making loop. What is relevant is feedback regarding the difficulty of the driving task.
From the outset, however, it is important to distinguish between three basic uses of the term risk: objective risk, subjective risk estimate and the feeling of risk. In the first usage, objective risk may be defined as the objective probability of being involved in an accident. This is usually determined in a post hoc way from analysis of accident data. This concept of risk has been referred to elsewhere as ‘statistical risk’ (Grayson et al., 2003). Subjective risk estimate refers to the driver's own estimate of the (objective) probability of collision. Such estimates of risk represent the output of a cognitive process, while the feeling of risk represents an emotional response to a threat, a distinction previously clarified, for example, by Haight (1986) and Summala (1986). Under certain conditions, subjective estimate of risk and feelings of risk may be closely associated, such as when a driver has lost control of a vehicle on an icy road and is about to collide with another road user. However, this association may apply only after subjective estimates of risk have exceeded some critical value.
Once a motor vehicle begins to move, collision (or veering off the roadway) is not a matter of some refined estimate of a very low probability: it is inevitable. The probability of crashing is one, unless, of course, the driver more-or-less continuously makes direction and speed adjustments to avoid this otherwise certain outcome. For this reason, an earlier conceptualization of key elements of the driving task focused on avoidance of potential aversive consequences and the conditions for delaying an avoidance response, which had implications for safety (see Fuller, 1984). In that conceptualization, objective risk of collision was assumed to be related to the extent of delay of an avoidance response, once a critical threshold had been passed. An example of a delayed avoidance response might be not slowing down when approaching a turning vehicle, which was expected to be out of the driver's path by the time it was reached. This perspective on driver behaviour was subsequently elaborated into a comprehensive behaviour-analytic model, enabling detailed consideration of the role of antecedent events and consequences in the determination of driver behaviour (see Fuller, 1991a, Fuller, 1991b).
In that model, subjective risk estimates were not a determinant of driver decision making, except in the profound sense of motivating the continuous avoidance of certain catastrophe, and this distinguished the approach in a fundamental way from that of the Risk Homeostasis theory of Wilde, 1994, Wilde, 2001. As is well known, Wilde argued that through weighing up the costs and benefits of alternative actions, drivers arrive at an accepted level of risk which they actively target (target risk), ultimately yielding the road accident toll in the drivers’ jurisdiction over a period of time. Thus, subjective risk estimates and objective risk are coupled in Wilde's theory. But further than this, Wilde also coupled subjective risk estimates and feelings of risk (fear). The experience of fear on the roadway informs estimates of subjective risk and behaviour adjustments are made so as to match these estimates with target risk.
Wilde's coupling of objective risk, subjective risk estimate and feelings of risk is clearly illustrated in his interpretation of a finding reported by Taylor (1964). Taylor found that measures of driver arousal (GSR), associated with particular roadway segments, were correlated with accident probabilities and inversely related to driver speed in those segments. He suggested that drivers were able to maintain GSR levels per unit time approximately constant by adjusting speed over different road segments. GSR rate, he proposed, was the feedback information drivers used to regulate speed. Wilde interpreted this to mean that drivers’ assessments of subjective risk were accurately reflecting objective risk in those segments and were determining their fear response (i.e., GSR) and behavioural adjustment, as represented in heightened arousal and choice of speed. Thus, all three ‘risk’ elements covaried in the theory.
There are a number of problems with this interpretation of Taylor's results, however. The first is to assume that GSR is a measure of fear or of feelings of risk. As mentioned above, and admitted also by Wilde (1994), GSR is also a generalized measure of arousal (specifically as expressed through the sympathetic ANS). Consistent with this is the later finding by Heino et al. (1994) that electrodermal activity was not very specific to changes in perceived level of risk.
Furthermore, GSR reflects both orientation responses and adjustments to temperature fluctuations. Thus, it will covary with attentional demands of a situation as well as motor activity (see, for example, Heino et al., 1994). A related problem has to do with the suggestion that GSR responses provide feedback information since, except in extreme situations, we are typically unaware of the level of activity of our sweat glands. What Taylor showed was that at certain locations historically associated with a higher probability of accident and also associated in his study with observable ‘traffic events’ (by which I presume he means potential conflicts), drivers showed increased electrodermal activity (EDA) and slowed down. By slowing down they spread the EDA over a longer time-base and therefore lowered its level per unit time. Taylor concluded that ‘drivers adopt a level of anxiety that they wish to experience when driving, and then drive so as to maintain it’. Wilde interpreted ‘level of anxiety’ here to mean a fear state coupled to subjective estimates of the probability of collision estimates, which are in turn linked to the objective probability.
An equally plausible explanation of Taylor's observations as that of risk-homeostasis, however, is the proposition that drivers respond to variations in task difficulty rather than feelings of risk and that they respond to these variations both in terms of autonomic arousal and adjustments in speed. EDA then becomes a correlate of task difficulty, an epiphenomenon that may play only an indirect role in mediating driver behaviour. If we replace ‘anxiety’ in Taylor's conclusion with ‘task difficulty’, then we get: ‘drivers adopt a level of task difficulty that they wish to experience when driving, and then drive so as to maintain it’. Taylor indeed found strong evidence in support of this revised conclusion. He showed that the GSR, expressed as a rate per unit time, was negatively correlated with driving experience, providing quite a good fit to a negative exponential function. Taylor tried to argue that over the same route the less experienced drivers must have perceived more risk than the more experienced drivers. But not only is there accumulating evidence to show that inexperienced drivers typically underestimate risk compared with more experienced drivers (e.g., Finn and Bragg, 1986, Delhomme and Meyer, 1998), but surely it is just as likely, if not much more so, that the less experienced drivers would simply have found the task of driving under the same conditions more difficult.
Given that crashing is more-or-less continuously inevitable unless a driver does something about it, it is not surprising that Taylor and subsequently Wilde should have made subjective risk estimate and fear so central to their thesis. However, what I want to propose here is that drivers adjust their speed to deal more easily with some hazard or potential difficulty. Thus, risk estimates linked to risk feelings are not ongoing determinants of driver decision making.
This view is largely concordant with that proposed by Näätänen and Summala (1976), McKenna (1988), and Wagenaar (1992), summarized by Summala (1986), who rejects the concept of risk as a determinant of driver behaviour. Summala argues that in most situations drivers know what they should do or not do to avoid a certain or almost certain accident. Driver behaviour is determined by the maintenance of safety margins, operationalized in his terms as the distance of the driver from a hazard. In a more recent formulation, Summala, 1996, Summala, 1997 describes a ‘lane-tube’, formed by the roadway and lane markings painted on it. If a driver maintains speed and direction, it is the time to crossing the boundaries of the tube (time-to-line-crossing) which provides the control measure for lane-keeping and similarly time-to-collision provides the control measure for headway selection and approach to stationary obstructions. No concern is normally given to risks. As Wagenaar (1992) succinctly states: “… people … run risks, but they do not take them”. What undermines the maintenance of safety margins, however, are motivating conditions which push drivers to higher speeds, an insensitivity to low probability events on the roadway and a growing desensitization to potential threats (because the threats are not realized). Given Summala's position on the determination of driver behaviour, the question then arises as to how drivers determine what is a safe margin in any given driving situation. Summala suggests that estimate of time-to-collision, for example, is a very basic human skill, for which computations can be carried out without cognitive computational processes (by which I presume is meant conscious processing). Safe margins are learned through experience and so most of driving ‘becomes a habitual activity which is based on largely automatized control of safety margins in partial tasks’ (Summala, 1986, p. 10).
Attractive as this model is, being situated firmly in a well-established behavioural paradigm, it is nevertheless vulnerable to the implausible requirement to recognize, and learn how to respond safely to, what is a virtually infinite number of roads and traffic scenarios. A learning model can provide a powerful explanation for which behaviours become established, once emitted. But it is unable to specify with any degree of precision which behaviour will be emitted in the first instance. What is needed is a heuristic, which goes beyond avoidance learning as a means of determining driver decision-making and therefore behaviour. One such heuristic is perceived task difficulty. If we agree that the driver's task is to attain mobility goals while avoiding collision, then most relevant to driver decision-making is the driver's perception of the difficulty of meeting those demands. Given this proposition, the question then arises as to what determines driving task difficulty.
Section snippets
The task–capability interface model
A recent conceptualization of what determines driving task difficulty has been presented in the task–capability interface (TCI) model (see Fuller, 2000 for the initial version of this model and Fuller and Santos, 2002 for a more developed version). In this model, task difficulty arises out of the dynamic interface between the demands of the driving task and the capability of the driver. Where capability exceeds demand, the task is easy; where capability equals demand the driver is operating at
Task difficulty homeostasis
How might the perception of task difficulty determine driver behaviour? The proposition I want to suggest is that at the outset of a journey, and sometimes also during it, a driver will determine a range of task difficulty that she/he is prepared to accept, a kind of target margin or envelope of task difficulty. A key element of this is the upper boundary of difficulty beyond which the driver prefers not to go. That preference may influence in the first place both choice of route and time of
Sensitivity to task difficulty
The concept of task difficulty is not new in the driver behaviour research literature, but it has existed in a different guise, namely that of mental workload. Kahneman (1973) defines mental workload as being a specification of the capacity an operator spends on task performance (see also de Waard and Brookhuis, 1997, de Waard, 2002). As de Waard (2002) states: ‘… in particular the word difficulty reflects mental workload very well’. Brookhuis and de Waard (2001) define mental workload as the
Task difficulty and risk assessment
The evidence reviewed above provides clear support for the notions that drivers are sensitive to task difficulty and attempt to maintain their experienced level of difficulty within a margin of acceptability. But the question remains as to the relationship between driver perceptions of task difficulty and their assessments of statistical risk. Perhaps task difficulty is really only a surrogate for risk assessment and the TDI model is the old wine of RHT relabeled in a new bottle.
In a recent
Risk homeostasis a special case of task difficulty homeostasis
We can tentatively conclude from the above results that Taylor and Wilde were correct in exposing experienced risk (i.e., feelings of risk) as a critical determinant of driver behaviour, but that Wilde was wrong in assuming this was the same as drivers’ estimates of the probability of crashing (or statistical risk) and therefore the fundamental determinant of the accident toll in a jurisdiction. The TDI model argues that experienced risk and subjective estimates of statistical risk will only
Conclusion and some further considerations
Driving task difficulty is inversely related to the difference between driver capability and driving task demand. Drivers appear to be able to make judgements of task difficulty easily and to behave in such a way as to keep the level of task difficulty within target boundaries. The feeling of risk may be an important source of information about task difficulty. However, this risk experience is not the same as the driver's rating of the risk of collision. Thus, although drivers may target a
Acknowledgements
I would like to thank Conor McHugh and Sarah Pender for their data collection, Eddie Bolger for assistance in preparing the digital videos of driving scenarios, and Liisa Hakamies-Blomqvist for valuable comments on an earlier version of this paper.
References (73)
- et al.
The effects of mobile telephoning on driving performance
Accid. Anal. Prev.
(1991) Exposure and experience are a confounded nuisance in research on driver behaviour
Accid. Anal. Prev.
(1982)Hazard and risk perception among young novice drivers
J. Safety Res.
(1999)- et al.
Perception of the risk of an accident by young and older drivers
Accid. Anal. Prev.
(1986) Young drivers’ overestimation of their own skill—an experiment on the relation between training strategy and skill
Accid. Anal. Prev.
(1996)Risk, especially risk of traffic accident
Accid. Anal. Prev.
(1986)Compensation in older drivers as reflected in their fatal accidents
Accid. Anal. Prev.
(1994)Sensation seeking and risky driving: a review and synthesis of the literature
Accid. Anal. Prev.
(1997)- et al.
Accid. Anal. Prev.
(2001) - et al.
Novice drivers’ accidents and violations—a failure on higher or lower hierarchical levels of driving behaviour
Accid. Anal. Prev.
(2001)
Role of primary personality factors in the perception of traffic signs and driver violations and accidents
Accid. Anal. Prev.
Reducing drivers’ mental workload by means of an adaptive man-machine interface
Transportation Res. Part F
Models of driving behavior: a review of their evolution
Accid. Anal. Prev.
Older drivers’ aberrant driving behaviour, impaired activity, and health as reasons for self-imposed driving limitations
Transportation Res. Part F
Accident risk and driver behaviour
Safety Sci.
Are we all less risky and more skilful than our fellow drivers?
Acta Psychol.
Maintaining task set under fatigue: a study of time-on-task effects in simulated driving
Transportation Res. Part F
Cellular phones and traffic accidents: an epidemiological approach
Accid. Anal. Prev.
Risk and Freedom: the Record of Road Safety Regulation
Driver information needs
Highway Res. Rec.
Variable errors set a limit to adaptation
Ergonomics
Assessment of drivers’ workload: performance and subjective and physiological indeces
Drivers’ margins of safety considered as a focus of research on error
Ergonomics
Driver fatigue
Hum. Fact.
Risk perception and decision taking during the transition between novice and experienced driver status
Ergonomics
The Field Guide to Human Error Investigations
Control motivation and young drivers’ decision making
Ergonomics
Can strategic and tactical compensation reduce risk in older drivers?
Age Ageing
Mental workload
On the measurement of driver mental workload
A conceptualization of driver behaviour as threat-avoidance
Ergonomics
The modification of individual road used behaviour
Psychology and the highway engineer
Cited by (657)
Color design method for tunnel wall-type portals considering harmonization with time-varying environments: A study of driving simulation
2024, Tunnelling and Underground Space TechnologyPersonalizing automated driving speed to enhance user experience and performance in intermediate-level automated driving
2024, Accident Analysis and PreventionDevelopment and validation of the Driver Attention Regulation Scale: A measure of the perceived ability to regulation attention on the road
2024, Transportation Research Part F: Traffic Psychology and BehaviourHow drivers perform under different scenarios: Ability-related driving style extraction for large-scale dataset
2024, Accident Analysis and PreventionThe impact of COVID-19 on speed behavior in Wisconsin
2024, Transportation Research Part F: Traffic Psychology and Behaviour