|We appeal to the data of laboratory experiments|
in much the same way that learned men,
centuries ago, appealed to the authority
of Aristotle or of Thomas Aquinas.
There are no conceptual problems in designing a real-life experiment that would test RHT under well-controlled conditions, but there are ethical and practical limitations to the deliberate creation of new conditions in which theoretical expectations can be verified. This is particularly true for the behavioural sciences. For instance, hypotheses about the heritability of individual differences in intelligence or personality traits could be settled compellingly if selective breeding of humans in controlled environments were as practicable and ethically acceptable as are current experiments with animals. Like heritability in humans, risk homeostasis may be difficult to test, but not because of inherent fuzziness in the theory.
An unrestrained experimenter for instance, might develop a simple design as follows. Several different geographical areas are selected and treatments randomly assigned. In Area A, a novel non-motivational intervention is implemented, for instance, in the form of physical changes to the roads. These changes are implemented following mass-media publicity effectively announcing the changes as a major safety benefit. In Area B, the same physical modifications are carried out, but these are announced as a minor safety advantage. Areas C and D receive the same physical treatment of the road conditions, but these are announced as a major and a minor threat to safety, respectively. This design allows the following expectations to be specified: In Area A there will be a greater upward change in accidents than in Area B. Area C will see a greater downward change than Area D. In all areas the accident rates will, after some time, return to previous rates. Public opinion data may be collected to check how long the intended perceptions of the safety or riskiness of the interventions lasted, and roadside observation may be used to monitor behavioural adaptation.
Ironically, we actually do many of these things, for example, mandating seatbelt wearing in Area A and equipping cars with daytime running lights in Area B. We order a change from left-hand to right-hand traffic in Area C and remove subsidies to driver education in high schools in Area D. We do all these things as politicians, administrators, safety advocates or legislators, but not as researchers who would be better equipped to collect evaluation data in a scientifically sound manner. And, of course, researchers might not even obtain ethics approval for some of these interventions because of the hazards involved.
There can be no question that experimentation in the laboratory--that oracle of modernity--offers easier opportunities for including comparison groups, for control over independent variables, and for more precise assessment of the dependent ones. However, laboratory experimentation introduces its own limitations to the unambiguous interpretation of findings. What follows is what a prominent ergonomist had to say on this topic:
Adding further strength to this conclusion is the fact that the experimenter must make use of volunteers to make the proposed experiment pass standards of ethical acceptability. Volunteers enter the experiment with their own motivations, expectations and perceptions. They may respond as they believe the experimenter would like them to, or do the opposite, or be indifferent about their behaviour. People volunteering in experiments have been found to be different from non-volunteers in that they are more often female, younger, higher in intelligence and need for social approval, less conventional, and they have fewer rightist political beliefs.
Add to this, too, that in the laboratory, volunteers are sometimes exposed to conditions that they would not likely expose themselves to of their own accord--especially in real-life situations and in the absence of the protection, or pressure, provided by the experimenter (for example, drinking to a high blood alcohol concentration).
Moreover, the special case of experimentation in the domain of risk taking must be considered. Risk, by definition, cannot be simulated. This is because two main purposes of simulation, and laboratory experimentation in general, are more control afforded to the experimenter and less risk occurring to the participants. Thus, to the extent that some factor under investigation is expected to influence the perception or acceptance of risk as dependent variables, and to the extent that simulation is successful in eliminating risk from the experimental condition, simulation must be judged to be an inappropriate environment for the testing of the effects of that factor upon these dependent variables. In other words, "simulation of risk" is a contradiction on a par with "recreation of an original". Consequently, if risk-taking behaviour is to be studied in the laboratory, risk has to be brought into the laboratory.
Further, it should be noted that RHT is a set of interrelated hypotheses developed to explain the accident rate of large numbers (often millions) of socially-interacting road users over a considerable length of time. Breakdown of control causes some people to have accidents. These accidents subsequently serve as danger signals to others and help the majority to avoid them. In contrast, laboratory experiments usually involve relatively small numbers of participants who participate solo, or an even smaller number of small groups of participants if participants are allowed to interact. The time frame in laboratory experiments is usually limited to a few hours at the most.
In the face of the many potential difficulties, it may seem ridiculous to attempt to create laboratory conditions for the purpose of adding support to RHT. One is reminded of the story about the student who entered the research laboratory one night to find his much-respected professor searching for something in all corners of the lab. When asked what he was looking for, the professor announced that he had lost his gloves in the park. "But, professor," the student asks, "why don't you look for your gloves in the park?" Came the somewhat miffed reply: "You know, young man, in the park it's dark. Here there's light."
Any risk of ridicule would have been zero had my students and I decided not to go ahead with these experiments. But we would also have missed out on the pleasure this experimentation provided to the numerous student-experimenters in undergraduate and graduate education, as well as many of the participating subjects.
Witness the fact that some came back for more, and others spontaneously asked if they could participate. We also think we learned something.
There is no human behaviour that has total certainty of outcome. Any act, because of limitations of skill and other deficiencies in control, may or may not be accomplished in accordance with the intent. And the consequences of the act, even when executed in full accordance with the intent, may be different from what was expected. Because of this dual source of uncertainty, any human act may be labelled as an act of risk taking.
To illustrate with a few examples: A young musician aspires to rendering an impressive performance of Beethoven's piano sonatas. She has set her own performance target in terms of pace, timbre, loudness and melodic expression. She can't be sure if she will perform to her targets. In addition, she can't be certain of the reactions of the audience and the music reviewers. Will they find her rendition too slow or too fast, too harsh or too mellow, too loud or too soft, too romantic or too stark, or just right in each of these dimensions? Similarly, a driver may be uncertain whether he can control his vehicle at a given speed through a particular curve on wet pavement. In addition, if he loses control, he doesn't know how serious the effects will be, whether he might be killed or injured, or walk away without a scratch. To contrast these two sources of risk, the first may be called "uncertainty of performance" and the second "uncertainty of consequence".
Obviously, then, all behaviour is risk-taking behaviour, regardless of whether this is consciously realized by the acting person or not. It is obvious, too, that the challenge of life is not to eliminate risks. "Zero risk" is not a meaningful option, since it can only exist in the absence of behaviour--after death, in other words. Instead, the challenge to the individual is to optimize the level of risk taking in such a way that the overall expected benefits accruing to that person are maximal. This was discussed above, in Chapter 4, and illustrated in Figure 4.2.
The challenge to psychology is not to determine whether a person is a risk-taker or not, because all individuals are risk-takers at all times, but to determine whether a person takes too much risk, too little, or exactly the right amount of risk for the maximal satisfaction of his or her goals. Another challenge to psychology is to provide people with the means to optimize their risk taking.
Consider a driver's behaviour on a highway. Driving faster than the average driver, following more closely, listening to the radio or a passenger, or having one's attention distracted from the driving task may be associated with a greater likelihood and/or severity of an accident. These behaviours may also be associated with increased gasoline consumption and the chances of being charged with a traffic violation. Thus, the sum total of the expected losses would increase. At the same time, however, the driver who engages in these behaviours may expect the benefit of a shorter travelling time and less boredom during the trip. Surely, at zero speed there is zero traffic accident risk. But zero speed also means zero mobility. Therefore, the challenge to any driver who wishes more than zero mobility is to choose an amount of mobility and a manner of driving such that the net benefit of his or her exposure to risk is likely to maximize. That level is at the location of the arrow in Figure 4.2 in Chapter 4.
For the purpose of risk assessment, the interpretation of observational data can be a problem. For instance, a driver who typically drives faster than other drivers is not necessarily at greater-than-average accident risk, if he or she is more skilful in accident avoidance than the average driver. To estimate such a driver's accident likelihood, an independent measure of skill would be needed. It would take further work to establish whether this driver--even if the level or risk taken by her or him were greater than average--took more than optimal accident risk.
To sum up: what we need is a laboratory task that allows separate identification of a person's skill and degree of risk taking. It should be possible to determine whether this person takes too much or too little risk, or just the right amount to maximize net benefit. The task should be difficult enough that the person experiences a considerable degree of performance uncertainty. Some responses should have outcomes that provide for uncertainty of consequence. Some outcomes should be pleasant, so that the person will be motivated to perform the task. Some potential outcomes should be unpleasant, so that the task will be experienced as one of risk taking. Risk taking is defined here as making responses with a given likelihood of unpleasant consequences, this likelihood being determined by the experimenter and communicated to the subject. Finally, unpleasant consequences should be sufficiently unpleasant that the participants will prefer to avoid them, yet innocuous enough to be practical and ethically acceptable.
Below we describe some of the different, but conceptually related, "computer games" that were developed in view of the above considerations. For the sake of convenience, the techniques have been labelled with nicknames like "Brinkmanship" and "Guessmanship". Subjects' participation was always rewarded, either with money earned in return for points scored, or with social recognition in the form of public posting of the names of the players in the order of their performance.
In some cases, people competed for money prizes and only the best three players would win. In others, the participants paid the experimenter a fee for the privilege of playing. In this case, instead of compensating the experimenters for their efforts, the monies were added to the prizes awarded for the best competitive performance and thus ultimately returned to the participants, that is, to some of them. Not surprisingly, it was not uncommon for people to spontaneously request participation in the experiments. Thus, the experimenters had reasons to believe that the subjects were usually keenly interested in their own behaviour during the experiments and tried to perform as well as they could. This may be an important benefit of this manner of subject recruitment.
This is difficult, because you cannot tell exactly when 1500ms have lapsed. A response at 1500ms is rewarded with the maximum number of points. Slower responses earn proportionally fewer points; at 3000ms and beyond, the pay-off is zero. However, responses faster than 1500ms, called "undershoots", may be followed by a penalty that occurs by chance. These penalties occur in a predetermined proportion of the undershoots, for example, in 20%, 50% or 80% of the cases. Before you begin, you have been told what their probability is. The probability changes after separate sets of usually 25 to 100 trials each. Non-penalized responses at 1499ms or faster yield zero points for the trial in question. The computer assures accuracy at 1ms.
Before the experiment begins, you are allowed a number of trial attempts to get used to the situation and to practise the task without pay-off. Each trial is followed by feedback about the trial number, the actual response time on that trial, the average stray from 1500ms on all preceding trials, plus the accumulated numbers of overshoots and undershoots.
During the experimental trials (with pay-off), each separate response is followed by feedback, to which the number of points gained or lost on the trial is added, as well as accumulated net points earned up to that trial. Furthermore, the computer's loudspeaker produces a single beep for undershoots that are not penalized and a double beep for those that are.
This task may be viewed as an experimental analogue of the notion of "brinkmanship", that is, the more you dare, the more you gain--unless you dare too much. You must balance the desire to gain points against the fear of losing points, and try to optimize your mean response time accordingly.
Obviously, the number of game points you earn depends on your response-timing skill, the precision of your "mental clock". Note, however, that responding as fast as you can is not the object of the game. If you do so, you will make risky responses only and collect no points. Obviously, too, the number of points you earn depends on your risk-taking tendency. If you shy away from the "brink" too much, you may never incur a penalty, but your gains will be small. If, on the other hand, you make too many undershoots, you may well collect a great number of points on some trials, but you will also incur many penalties. Response-timing skill can be measured in a variety of ways. One is the average stray around 1500ms during the warm-up trials; another is the average stray around the participant's own mean response time during the trials with pay-off. The lower the degree of stray (also called the average deviation), the better the skill.
Assisted by computer, the mean response time at which the participant's point earnings would be maximized is determined for each participant, while leaving the dispersion (the stray) of the participant's response distribution intact. This is called the participant's optimal mean response time. It is calculated as follows:
First, the computer adds 1ms to the response time in each trial and calculates what the net earnings would have been across all trials. Then it adds 2ms to each actual response time, calculates what the net earnings would have been, then adds 3ms and so on, up to 300ms. It also subtracts values between 1 and 300ms, one at the time, from each actual response time. This is an ideal job for a computer; people would find it much too tedious a task. The mean response time at which the participant would have earned the highest value of potential net point earnings is the optimal mean response time for that participant. By comparing the optimal mean response time of a participant with his or her actual mean response time, we can derive a quantitative measure of risk taking.
The optimal mean response time minus actual mean response time is called "deviation from optimality", or DFO for short. We can use the DFO to identify three types of risk taking. A positive DFO indicates that the participant, on average, responded too fast to maximize net earnings and by how much. This person is called "risk-loving" or "risk-seeking", because he or she sacrifices points by being too daring and erring on the risky side. "Risk-neutrality" or "risk-optimality" is reflected in a DFO = 0 result, since the player does not increase or diminish the amount of risk beyond the amount of risk-taking that is optimal. Finally, "risk-averse" or "risk-avoiding" participants show negative DFOs; they lose points by being too cautious and erring on the safe side of the brink. Note that, in order to maximize net benefit, a person has to take more than zero risk and that the right degree of "brinkmanship" produces the greatest net gain.
How can this task be used to test risk homeostasis theory? The following nine points elaborate on this question:
Questions are chosen such that few people would be expected to know the precise answer, but that many would be able and willing to make a guess. The closer the answer comes to the truth (expressed, for example, as a percentage of the actual figure), the more points participants gain. If, however, the answer given is higher than the correct answer, a die is thrown or a coin is tossed to determine whether a penalty will be applied or not. Once again, the higher your estimate, the more you gain, unless your estimate is too high.
In yet another version, the participant is shown sheets of paper with a vertical line on each. Below the line is a dot and the subject is asked to draw (starting at the dot) a horizontal line of the same length as the vertical. Again, the participant is told, "the closer you come to the truth, the more points you make, but you run the risk of a penalty if your line is longer than the vertical." Other samples of Brinkmanship items are presented in Figure 10.1
The principle of Brinkmanship can be applied to all kinds of skilful performance, on the computer and elsewhere, but unless a computer is used, the calculation of DFO and other variables may be a bit of a problem.
Figure 10.1: Samples of Brinkmanship items. The surface area of the figure on the left equals 100. What is the surface area of the figure on the right?
Points are earned only if the correct answer lies within the uncertainty band--within the safety margins, that is--while the number of points earned on any question is greater to the extent the safety margin is narrower. In one version of this technique, the number of points earned is simply the complement of the safety margin, the sum of the safety margin and the number of points earned always amounting to ten. Thus, if the player chooses a safety margin of "plus or minus three" and the correct answer lies within the uncertainty band, seven points will be awarded. Had the player chosen a safety margin of "plus or minus eight", two points would have been awarded. The largest safety margin the player is allowed is plus or minus nine, the smallest is zero. If the correct answer is outside the uncertainty band, the player does not lose or gain any points whatsoever.
Consequently, participants can reduce their fear of not earning any points by widening the safety margin. On the other hand, in order to satisfy the desire to make a large number of points, the margin must be made narrow--and the correct answer must still lie between the point estimate plus or minus the safety margin. We called this game "Narrow Escape" because the art of playing well is to escape with the truth, but as narrowly as possible.
A computer programme determines from the participant's responses whether, and to what extent, the chosen margins should on average have been wider or narrower for the net number of points to have been maximized. The operation of this programme is analogous to the calculation of the deviation from optimality (DFO) as described above. Risk aversion (underconfidence) is reflected in choosing safety margins that are too wide, while risk seeking (overconfidence) is manifested in safety margins that are too narrow for net points to be maximal. Risk optimization is in evidence if the player's net points earned equal the potential maximum, that is, when the player could not have increased net points earned, either by narrowing or by widening the chosen safety margins by a constant across the questions.
If you wish to explore your own risk-taking tendency or that of your friends or family members, you need an approach that makes it unnecessary to use special computer programmes for the calculation of the deviation from optimality. All you need for the next game is to consult two tables (provided here) and use a bit of simple arithmetic. The game is called "Guessmanship", because the art of playing well is the art of smart guessing in the face of uncertainty. Only yes-or-no answers are allowed. The participant answers a set of, say, 100 questions of the following type:
If the player responds by a correct "no", sure points are earned. If the player responds by an incorrect "no", no points are earned or lost. If the player responds by a correct "yes", sure points are earned. If however, the player responds with an incorrect "yes", a penalty is applied based on chance (that is, in some cases, but not in all). Therefore, a "no" answer is safe; if correct, points are gained, and when wrong, no points are lost. A "yes" answer is risky; if correct, points are earned, but if wrong, points may be lost. A player who always answers "no" never loses any points--but does not gain any on questions where the right answer is "yes". It follows that avoiding "yes" responses altogether is not necessarily in the player's best interest.
Obviously, then, how many points a player will obtain does not only depend on the player's knowledge, but also on strategic skill as to what to do when uncertain: to say "yes" or to say "no". Two people of equal knowledge, therefore, do not necessarily obtain the same game score. It is possible for a more knowledgeable person to get a lower game score than a less knowledgeable competitor, because the latter has a better risk-taking strategy.
To obtain a measure of risk-taking strategy, first make sure that in one-half of all the questions, the correct answer is "yes", and tell the players so, because that simplifies matters. Then assign score points, and tell the players what they are, for instance, as follows:
For the calculation of the player's level of knowledge and risk-taking tendency, please refer to Tables 9.1 and 9.2. These use some statistics jargon, related to the normal "bell-shaped" curve, which does not need any further discussion here. Statistical jargon is strange indeed, as you yourself may have noticed. Statistics has its "normal deviates"; in the rest of the world, deviates are abnormal. Statistically "significant", does not mean important, sizable or meaningful, but merely that the likelihood that a particular finding had occurred by chance was very small, say, less than 1 in 100, or 5 in 100. To the uninitiated, the term "standard deviation", as well as "normal deviate", may sound like an oxymoron (like "thunderous silence", "profound superficiality" or "civil war"), and there is nothing childish in "regression", not even in multiple regression.
But, back to the calculations. The measure of skill or knowledge basically depends on the proportion of questions answered correctly, but it considers the two kinds of questions separately (those where the correct answer is "yes" and those where it is "no"; see Section A in Table 9.1).
Now for the measure of risk-taking tendency. This involves a quantitative comparison between the two types of errors that a person can make: the frequency of saying "no" when the correct answer was "yes", relative to the frequency of saying "yes" when the correct answer was "no" (Steps 6 through 10 in Table 9.1). A cautious person will prefer to give a safe answer when in doubt, and thus the first type of error will be more frequent than the second. That much is obvious. But how safe is safe enough? What is the optimal level of safety? Or, stated differently, what is the optimal level of risk?
In Section 4.1 we argued that the optimal (or target) level of risk depends on the costs and benefits of safe and risky behaviour alternatives. In the Guessmanship game, costs are given in terms of points lost or gained for incorrect "no" and incorrect "yes" answers, while the benefits are given in terms of points gained for correct "no" and correct "yes" answers. Therefore, the optimal level of risk can easily be calculated (as is done in Steps 11 through 13 in Table 9.1) The only thing that remains to be done is to compare the person's risk-taking tendency with the optimal risk-taking tendency (and this is done in Step 14). So, any individual who is seen to take more than optimal risk in the Guessmanship game may be labelled as "risk-seeking". When the amount of risk taken is less than optimal, we may label this as risk avoidance, and whenever the actual amount of risk taken is equal to the optimal amount, we have a case of risk optimization.
|1.||Determine the percentage of times the player responded "yes" to questions where the correct answer is "yes".|
|2.||Look up the "normal deviate" equivalent of this percentage.|
|3.||Determine the percentage of times the player responded "no" to questions where the correct answer is "no".|
|4.||Look up the "normal deviate" equivalent of this percentage.|
|5.||Add up the values obtained in Steps 2 and 4. This measure of knowledge will vary between a minimum of zero and a maximum of 4.6 or so.|
|6.||Determine the percentage of cases in which the player said "no" in response to the questions where the correct answer is "yes".|
|7.||Look up the "ordinate" equivalent of this percentage.|
|8.||Determine the percentage of cases in which the player said "yes" in response to the questions where the correct answer is "no".|
|9.||Look up the "ordinate" equivalent of this percentage.|
|10.||Divide the value obtained in Step 7 by the value obtained in Step 9.|
|11.||Add the points value of a correct "no" to the cost of an incorrect "yes".|
|12.||Add the points value of a correct "yes" to the cost of an incorrect "no".|
|13.||Divide the sum obtained in Step 11 by the sum obtained in Step 12.|
|14.||Subtract the ratio obtained in Step 10 from the ratio obtained in Step 13. Risk-seeking tendency is reflected in positive differences, risk-avoidance in differences less than zero.|
|%||normal deviate||ordinate||%||normal deviate||ordinate||%||normal deviate||ordinate|
The techniques described in this chapter may be useful for the purpose of experiencing risk and realizing that risk should be optimized, not minimized, to obtain maximum benefit. They may also serve empirical testing of specific hypotheses such as the ones that are pertinent to risk homeostasis theory, as well as for the exploration of various other questions. What follows is a list of examples:
Students and I have explored the above questions in the context of tens of student research projects involving some 2500 players/participants. Some of these were students themselves; others were taxi-drivers, prison inmates, shopping mall patrons, motorcycle enthusiasts. We will discuss some of the findings in the next chapter. To our satisfaction, we have found that people were usually quite motivated to apply themselves to the tasks described above, and keenly interested in their performance as well as in their standing comparative to others. We feel, therefore, that we are tapping behaviour of a relatively high level of ego-involvement, behaviour that is of interest to the participants themselves and, therefore, possibly to the study of human behaviour.
Instead of experiencing difficulties in recruiting sufficient numbers of indifferent or reluctant subjects, we have at times been approached by people who spontaneously volunteered their participation. Even when we demanded a few dollars from a person for the privilege of participating in our experiments, we experienced no major difficulty in attracting subjects.
Most of the tasks described above are highly portable. They can be performed in the lab, at home in the family room, in shopping centres, in pubs, on beaches and other locations of the experimenter's imagination.
One major theoretical limitation should be pointed out. With the above methods, the degree of risk optimization by people can only be calculated against a specified criterion (such as monetary gain or social recognition), not against other criteria, for instance, the satisfaction of curiosity, maintaining comfortable psychophysiological arousal, the psychomotor challenge of cancelling the stimulus as close as possible to 1500 ms after onset regardless of points gained or lost, the desire to finish the experiment quickly. It is conceivable, therefore, that a person is actually optimizing the degree of risk taking in his or her behaviour against the composite criterion of all of his or her goals, although the calculated deviation from optimality against the criterion of monetary gain, for instance, is substantially different from zero. We will not know whether the person is actually optimizing risk until we develop techniques that allow us to determine the extent of risk taking in behaviour aimed at the satisfaction of these additional motives and the degree of satisfaction obtained. So, it must be admitted that we cannot even obtain a truly complete measure of a person's risk taking in the laboratory, let alone in real life.
One striking consequence of this is that we cannot be sure whether the amount of risk taking that is reflected in a nation's traffic accident rate is too high, too low or optimal. Are we collectively taking too much risk or too little, and how can we tell? Until we find out, it would seem totally inappropriate to label death on the road as "useless death", as some have done. Death on the road is saddening, shocking, regrettable, but to call it useless is paternalistic, arrogant and insulting to the victims, their survivors and the population at risk.
[TABLE OF CONTENTS] [PREVIOUS CHAPTER] [NEXT CHAPTER]