A field experiment to test the effects of automated feedback and monetary incentive on speeding behavior
vi TABLE OF CONTENTS Page LIST OF TABLES viii LIST OF FIGURES ix INTRODUCTION 1 SPEEDING 3 DOES SPEEDING AFFECT TRAFFIC SAFETY 13 WHAT COUNTERMEASURES EXIST FOR ADDRESSING THE SPEEDING PROBLEM 18 DATA RECORDING, DRIVER MONITORING AND ISA 20 FOLLOW-UP STUDIES TO SWEDEN'S LARGE SCALE FIELD TEST 36 SUMMARY OF EXISTING ISA RESEARCH 44 FEATURES OF THE TESTED ISA SYSTEM: AUTOMATED FEEDBACK AND MONETARY INCENTIVE 48 HYPOTHESES 61 HYPOTHESES ASSOCIATED WITH SPEED 61 HYPOTHESES ASSOCIATED WITH SUBJECTIVE MENTAL WORKLOAD 62 HYPOTHESES ASSOCIATED WITH ISA ACCEPTANCE 63 METHOD 65 EXPERIMENTAL DESIGN 65 DEPENDENT VARIABLES 67 PARTICIPANTS 68 MATERIALS 70 PROCEDURE 79 RESULTS 86 DATA INSPECTION AND CLEANING 86 DESCRIPTIVE ANALYSES 87 PERCENTAGE OF TIME SPEEDING BY AND ACROSS SPEED LIMIT ZONES 90 ANALYSIS OF VARIANCE FOR MEAN SPEEDS BY SPEED LIMIT ZONE 119 MILES DRIVEN PER WEEK 124 PERCEIVED MENTAL WORKLOAD 125 TRUST AND ACCEPTANCE OF AUTOMATED FEEDBACK 132 TRUST AND ACCEPTANCE OF MONETARY INCENTIVE 134
vi i DEBRIEFING SURVEY RESULTS 136 TESTS OF CORRELATIONS 140 SUMMARY OF RESULTS 144 DISCUSSION 148 EFFECTS OF MONETARY INCENTIVE AND AUTOMATED FEEDBACK 148 PERCEIVED MENTAL WORKLOAD 157 TRUST AND ACCEPTANCE 160 PERCEIVED USEFULNESS AND WILLINGNESS TO KEEP 163 IMPLICATIONS FROM CORRELATIONAL ANALYSES 165 FUTURE RESEARCH 166 CONCLUSION 172 REFERENCES 173 APPENDICES 186 VITA 222
Vl l l LIST OF TABLES Table Page 1. Literature Review Summary Table 60 2. Summary Table for Experimental Hypotheses 64 3 Experimental Design 66 4. Number of males and females and average age by experimental group 69 5. Descriptive statistics for demographic variables by experimental group 87 6. Self-report items about sensation seeking, automation and driving behaviors 88 7. Mean percentage of time driving within speed ranges by Week and Ml group 116 8. Analysis of Variance for the Six Dimensions of the NASA-TLX 131 9. Mean trust ratings and t-coefficients for trust and acceptance of the AF system 134 10. Mean ratings and t-coefficients for trust and acceptance of the Ml system 135 11. Descriptive analysis of responses to debriefing survey, no-MI participants 137 12. Descriptive analysis of responses to debriefing survey, Ml participants 137 13. Sample response to open-ended debriefing questions 139
IX LIST OF FIGURES Figure Page 1. GPS system used for project (dollar bill included for size 73 2. Potential speed scenarios 76 3. AF and Ml display showing a bonus amount ($23.35) 77 4. Data logger and cellular modem 79 5. Percentage of time driving at or below the speed limit as a function of monetary incentive and advisory feedback 92 6. Percentage of time driving 5 to 8 mph over all speed limits as a function of monetary incentive and advisory feedback 93 7. Percentage of time driving 9 or more mph over all speed limits as a function of monetary incentive and advisory feedback 95 8. Percentage of time driving at or below 25 mph speed limits as a function of monetary incentive and advisory feedback 96 9. The effect of monetary incentive and automated feedback on the percentage of time driving 5 to 8 mph over 25 mph limits 98 10. The effect of monetary incentive and automated feedback on the percentage of time driving at or below a 30 mph limit 99 11. The effect of monetary incentive and automated feedback on the percentage of time driving 5 to 8 mph over a 30 mph limit 101 12. The effect of monetary incentive and automated feedback on the percentage of time driving 9 or more over a 30 mph limit 102 13. The effect of monetary incentive and automated feedback on the percentage of time driving at or below a 35 mph limit 104
X Figure Page 14. The effect of monetary incentive and automated feedback on the percentage of time driving 1 to 4 mph over 35 mph roads 105 15. The effect of monetary incentive and automated feedback on the percentage of time driving 5 to 8 mph over 35 mph roads 107 16. The effect of monetary incentive and automated feedback on the percentage of time driving 9 or more over 35 mph roads 108 17. The effect of monetary incentive and automated feedback on the percentage of time driving at or below 40 mph roads 109 18. The effect of monetary incentive and automated feedback on the percentage of time driving 5 to 8 over 40 mph roads 110 19. The effect of monetary incentive and automated feedback on the percentage of time driving at or below the 55 mph speed limit 113 20. Average speed on 25 mph roads as a function of monetary incentive and advisory feedback 120 21. Average speed on 30 mph roads as a function of monetary incentive and advisory feedback 122 22. Average speed in 35 mph zones as a function of monetary incentive and automated feedback 123 23. Average total miles driven per week by incentive group 125 24. Perceived mental demand as a function of monetary incentive and automated feedback 127 25. Perceived temporal demand as a function of Monetary Incentive and Automated Feedback 128 26. Perceived effort as a function of monetary incentive and automated feedback 129 27. Perceived frustration as a function of Ml and AF 130
1 INTRODUCTION Each year, traffic crashes claim 1.2 million lives across the world (World Health Organization, 2004) and more than 40,000 lives in the United States (National Highway Traffic Safety Administration (NHTSA), 2007). These events are the number one cause of death for Americans under age 35 (Centers for Disease Control, 2006). Crashes occur for a variety of reasons, including distraction, speed, aggressive driving, impaired driving, perceptual errors, and fatigue. Many emerging in-vehicle technologies are designed to improve safety by preventing contributing factors from occurring or minimizing the adverse effects of their occurrence. For example, adaptive cruise control (ACC) reduces tailgating (Rudin-Brown & Parker, 2004). With the system engaged, drivers maintain following distances based on distances to lead vehicles. Although the safety advantages of such automation are clear, systems require thorough testing to validate benefits and assess unintended negative consequences such as over-reliance on the system (Lee & See, 2004; Parasuraman & Riley, 1997). Rudin-Brown and Parker reported that the tested adaptive system reliably increased following distance, yet drivers were more likely to engage in a distracting secondary task when driving with ACC than driving without it. The focus of the current project, Intelligent Speed Adaptation (ISA), is similar to ACC because it reacts to the environment in a dynamic manner to reduce an unsafe driving behavior. However, the basic notion of ISA is dynamic reaction to changes in speed limits, whereas ACC adapts to the speed of a lead vehicle. ISA uses Global Positioning Systems to compare the speed of the vehicle to the speed limit of the roads on which the driver is travelling. In areas where databases of speed limits exist, ISA
2 systems are feasible, and the systems provide an opportunity to reduce speeding by informing the driver about speed, introducing mechanical resistance to make speeding quite effortful, or preventing the driver from speeding (Carsten & Tate, 2005). There is empirical support for the effectiveness of ISA (Regan, et al., 2006; Varhelyi, Hjalmdahl, Hyden, Draskoczy, 2004), but some researchers have noted the potential for driver habituation (Dingus, Klauer, Neal, Petersen, & Lee et al. 2006). A possible factor related to this habituation is an inadequate feedback structure (Toledo & Lotan, 2006). For example, feedback provided by the system only when the driver chooses to access it may receive initial attention by the driver and then a gradual decrease in attention as time progresses. The current project evaluated a strategy for decreasing speeding behavior by manipulating auditory and visual feedback and behavior-based incentives. The feedback and incentive schedules were structured so that drivers experienced real-time warnings, received an economic incentive, or received warnings and the incentive for keeping within preset speeding parameters. The speeding of these drivers was compared to a control group that experienced neither warnings nor economic incentives. In addition to speeding behavior, driver workload and acceptance of the system was analyzed. Acceptance and workload are important: if drivers dislike the system they will be unlikely to use it voluntarily. Alternatively, if some aspect of workload, such as temporal demand, increases when driving with ISA, then a reduction in speeding may be offset by a negative behavioral adaptation, such as tailgating. Theoretically, there is concern among human factors researchers that
3 exposure to automation may lead to unintended behavior change (Comte, 2000; Parasuraman & Riley, 1997). ISA systems are a relatively new technology. They have been field tested in Europe and Australia, but not in the United States. Evaluations indicate that the systems are promising, but the benefit may not generalize to the United States driving population. In addition to the applied nature of the study, theoretical perspectives helped to guide the proposed feedback and incentive structures, to make predictions about potential negative consequences such as increased mental workload, and to explain why drivers speed. Speeding Speeding appears to be a universally accepted behavior in several Westernized countries. Shinar, Schechtman, and Compton (1999) analyzed self-report data that were collected annually over 11 years. The sample size of adults from the United States was relatively large, 1,250 respondents per year, and the questions asked related to general health behaviors such as diet and smoking and to traffic safety behaviors such as wearing seat belts, drinking and driving, and speeding. In general, the authors reported that the respondents in the later years of the study period placed more importance on health habits than respondents in the first years of the study period. Similarly, the respondents indicated the importance of buckling seatbelts and driving sober increased over time. However, the respondents' attitude toward speeding did not follow this pattern. At the beginning of the measurement period, respondents indicated that speeding was less of a threat to safety than driving while impaired or unbuckled.
4 Moreover, Shinar et al. (1999) report that the perceived relevance of speeding to safe driving continued to decrease as the 11 year period progressed. In a separate analysis of self report data Blincoe, Jones, Sauerzapf, & Haynes (2006) collected over 500 surveys of English drivers who were ticketed for speeding through an area with an automated speed camera. One line of questioning concerned reasons for speeding. Frequent answers were that speeding was not dangerous and that "everybody else speeds." Similarly, self report data collected by Fleiter and Watson (2006) indicate that Australian drivers view speeding as acceptable. They asked drivers to indicate the extent to which they speed on urban and "open" roads with respective speed limits of 60 and 100 kilometers per hour. A third of the sample indicated that they speed on the slower road, whereas more than half indicated that they choose to speed on the faster road. Fleiter and Watson (2006) also asked drivers to indicate what speed should be permissible at both limits. The drivers overwhelmingly selected speeds that were greater than the posted limit. Similar findings are reported by Shinar (2001), who asked drivers to indicate what speed they typically drove on roads with different speed limits. Based on the hypothesis that different conditions affect speed choice, Shinar asked drivers to indicate the speed they would choose to drive when they were driving in situations such as driving alone, driving with family, or driving for fun. Drivers consistently indicated that the speed they drive when alone was higher than posted limits, and the "fun" speed was generally greater than the typical speed. These findings were based primarily on self-report data, and the validity of subjective results is always a concern. However, the work of Haglund and Aberg (2000)
5 indicates that there is value to self report data. These researchers tested the relationship between observed speed and drivers' self reports by using a hidden speed camera and an interview that occurred down road from the camera. Researchers asked the drivers, who were unaware that speed was recorded, to indicate how fast they were driving before they were stopped and how fast they would normally drive on the road. The correlations between observed and reported speed and observed and normal speed were .58 and .5, respectively. These data indicate that drivers' estimates were in general agreement with observed speeds. However, other researchers report more moderate relationships between observed and reported speed (see Corbett, 2001). In contrast to debate about the relationship between observed and reported speed, there is little doubt that drivers choose to speed in free flow traffic, particularly where the roads are in good repair and enforcement is absent. For example, Freedman, De Leonardis, Poison, Levi, and Burkhardt (2007) completed an evaluation that measured the effect of rational speed limits. The notion behind rational speed limits is that the overwhelming majority of drivers rationally select safe speeds based in part on road design, and therefore, speed limits should correspond to these selected speeds. Typically, the rational speed is defined as the speed at which 85% of the traffic selects during free flow conditions. Freedman et al. divided several roadways into 7 mile sections that contained multiple speed limit changes, and then measured the speeds that drivers selected in free flow traffic, defined as flow conditions in which five seconds separated the own vehicle from a lead vehicle. When driving in free flow, the authors found that 50 to 90% of drivers exceeded the speed limits before speed limits were
6 changed to the rational speed. After implementing the rational speed limit program, as many as 50% of drivers still violated the speed limit. In addition, the average speed did not change — drivers were traveling as fast after the limits increased as they were before they were changed. They simply were not violating the speed limit as frequently because the limit increased. In sum, an abundance of self report data indicates that speeding is commonplace, and high correlations between observed and reported speed suggest that self report data are indicative of drivers' true speed. Finally, the Freedman et al. study indicates that exceeding the speed limit may be the norm in certain locations. Several questions follow from these findings about the prevalence of speeding: 1) Who is speeding? 2) Why are they speeding? 3) Is speeding truly dangerous? and 3) If it is dangerous, why? Who is speeding? From a demographic perspective, there are several driver characteristics that covary with speeding. Many authors report that younger drivers are more likely to speed than older drivers (Quimby, Maycock, Palmer, & Buttress, 1999; Wasielewski, 1984), as suggested by fatality statistics (NHTSA, 2007.) Other individual difference variables include income level, with wealthier drivers speeding more frequently than their less wealthy counterparts; sex, with males driving faster than females; and vehicle size, with drivers of large vehicles driving faster than those with smaller vehicles (Shinar, 2007.) Why are drivers speeding? Explaining why individuals speed is vexing because drivers speed for different reasons, and there is considerable within group variance. An individual may speed in a certain condition on one day (being late for work) but choose
7 not to speed the next time the condition is present. Drivers may be rushing to work, speeding because they are within a group of drivers who are exceeding the limit, or speeding for emotional reasons (thrill or anger) (McKenna, 2005). McKenna describes these willful acts of speed as "rebellious" or "pragmatic" speeding. In addition to willful speeding, drivers may not be aware that they are speeding. That is, they may have a lapse of attention during which their speed may increase without their knowledge. A second cause of unintended speeding may be perceptual speed adaptation. Unintended speeding may also result from a failure to see a speed limit sign. When trying to explain these reasons more completely, researchers have applied theories from social psychology, sensation and perception, and information processing. From an information processing approach, speeding increases risk because of time pressure. Researchers quantified the time needed to sense, perceive, decide, and act to various stimuli using methods such as the additive method (Sternberg, 1969). In fact, researchers succeeded in parsing out the processing time associated with certain stages and factors that affect processing in laboratory settings. When movement through space becomes a factor, the equation Distance = Rate * Time dictates that an individual driving at a faster speed has less time to react to a given stimulus than an individual driving at a slower speed who encounters the same stimulus. In addition to the distance equation, stopping distance increases exponentially with speed. Finally, environmental factors such as darkness, weather, and roadway characteristics can add further constraints on the amount of time that drivers have to process and react. However, humans do not make decisions based on the computation of complex
8 formulas, and it is reasonable to assume part of the speeding problem is a result of naivety about the true risks. This situation may be further compounded by perceptual speed adaptation. Perceptual Effects on Speed - Speed Adaptation. This theoretical perspective is helpful for understanding why some violations are committed unknowingly. The vestibular system is attuned to sudden changes in movement; however, humans quickly adapt to steady movement. When angular or rotational acceleration occurs, fluid shifts in the vestibular canals, and the moving individual experiences the sensation of acceleration. When movement through space is at a constant velocity, the vestibular fluid will not shift, and individuals adapt to the speed at which they are moving. When individuals adapt to a set speed, e.g., 45 mph, and suddenly change to a new constant velocity, e.g., 25 or 65 mph, what they experience differs from what they would have perceived had they not adapted to 45 mph velocity. However, individuals experience significantly different perceptions depending on whether the new constant velocity is greater or less than the adapted speed. When individuals experience an increase in true velocity, the perceived change in velocity is greater than the true change, i.e., they feel that they are going faster. In contrast, and of particular interest to the current study, when there is a reduction from a high speed to a lower speed, individuals perceive that the reduction in speed is significantly slower than the true reduction. Psychophysical studies completed several decades ago support this perceptual phenomenon (Denton, 1966; Matthews, 1978; Schmidt & Tiffin, 1969). For example, Schmidt and Tiffin had participants make several estimates when a vehicle
9 they were driving reached 40mph. When participants accelerated from 0 to 40, the average speed that they estimated was 41mph. However, when the individuals maintained a speed of 70mph for a period of 20 minutes and then slowed to what they perceived to be 40mph, the perceived 40mph was 50mph. Schmidt and Tiffin (1969) showed that this underestimation of perceived speed increased as a linear function of time spent at the higher constant velocity. Denton (1966) completed a similar investigation. In addition to showing that drivers underestimate reductions from a higher to a lower speed, he also showed that participants overestimated increases from lower to higher speeds. Finally, Matthews (1978) recorded radar readings of vehicles traveling on a roadway that had the same speed limit in both directions. However, in one direction vehicles traveled on a road that was 15mph higher than the test road, whereas vehicles traveling in the opposite direction had previously driven on a lower speed road. Matthews (1978) reported that drivers who traveled on the higher speed road had significantly greater travel speeds than those traveling in the opposite direction. These studies provide empirical support to the frequent anecdotal reports of perceptual speed adaptation. Thus, the speed adaptation phenomenon may contribute to the prevalence of and dangers associated with speeding. An individual who adapts to a high speed of 70 mph and does not slow sufficiently due to faulty perception may underestimate the time available to safely respond to a hazardous event. Unfortunately, required information processing time does not change because drivers choose a higher speed. Further, individuals generally make decisions based on heuristics and past experience
10 rather than the computations needed to determine decision time relative to stopping distance; therefore, it seems likely that drivers may underestimate the increased risks of speeding. Decision Making: Heuristics and Prospect Theory. The classical approach to decision making would suggest that when choosing to speed or comply with the limit, drivers weigh the probabilities of costs and benefits. These costs might include lost time, lost money associated with a speeding citation, and losses associated with a crash. Drivers may use confirmation bias and the availability heuristic when determining the probabilities associated with the possible outcomes. For example, drivers may discount speed related crashes and recall only those related to impaired driving. Drivers who see a driver adhering to the speed limit may look to confirm that the driver is "old" and dismiss information that disconfirms an assumption that only certain populations drive the speed limit. Speeders may recall that they always drive over the limit and never get stopped. Drivers may further downplay the threat of sanction if a judge dismissed or reduced a ticket when they went to court. Given the low perceived probabilities of costs associated with sanctions and crashes, classical decision making theory would predict that drivers weigh the potential losses accordingly when they select speed. The one cost that drivers might perceive as probable if they comply with the limit is the loss of time. Prospect theory explains an interesting phenomenon associated with the differences between subjective ratings of losses and gains (Kahneman, Slovik, & Tversky, 1982.) Kahneman et al. report a body of work that indicates individuals perceive a loss of a certain amount as more important than a gain of the same amount.
11 Prospect theory would predict that individuals who speed because they are rushing do so because they feel they are losing time rather gaining time. Prospect theory would provide an explanation for why drivers who speed for pragmatic reasons account for a significant portion of speeders, particularly given findings that indicate faster speeds do not ensure faster commutes (Regan et al., 2006). Theory of Planned Behavior. In contrast to information processing and perceptual perspectives associated with speed, the Theory of Planned Behavior (TPB) offers an explanation as to why individuals purposefully speed. TPB is rooted in social psychology and is a frequently cited theory used to explain various behavioral phenomena. This theory was proposed by Ajzen and reviewed by De Pelsmacker and Janssens (2007). The central features of the theory include "norms", "attitudes", and "intentions." Norms refer to the rules and beliefs that an individual has about a certain behavior. These norms can be personal, normative, descriptive, or subjective. There is a "moral" component to personal norms; this component captures the extent to which the individual thinks the behavior is "right" or "wrong." A second component of personal norms described by De Pelsmacker and Janssens (2007) refers to the regret an individual would experience if the behavior were or were not manifested. Norms may also be relative or normative. Normative norms are those that describe what the individual thinks others believe about the behavior. These norms include "everyone else is doing it" reasoning for engaging in a behavior. Subjective norms refer to the extent to which the individual feels pressure from valued individuals, e.g., friends, family, peer groups, to engage in the behavior.
12 According to TPB, these four norm types directly affect attitudes about the behavior. Attitudes are comprised of affective and cognitive components. The affective aspect of attitudes capture the emotion associated with the behavior: "I am anxious to get home and see my family." The cognitive component describes the logic associated with the attitude: "I know that speeding is technically illegal but the police allow 10 mph over the limit." Researchers who refer to TPB suggest that attitudes directly affect intentions to engage in the behavior. Two additional important factors in this theory are "perceived behavioral control (PBC)" and "habits." PBC and habits moderate intention such that in situations when individuals believe they have behavioral control, they are more likely to realize their intention. Similarly, Azjen suggests that the intention to engage in a behavior is stronger when the behavior is relatively more habitual. TPB may explain a variety of behaviors. With regard to speeding, De Pelsmacker and Janssens (2007) completed a study that used structural equation modeling to test a model based on the factors contained in the theory. The authors devised Likert-type survey items that were written to reflect the latent factors described above, i.e., norms, attitudes, intentions, habits, and perceived behavioral control. In addition, they asked participants to estimate how often they speed. The results of the factor analysis indicated that there were significant loadings for items associated with each latent construct. The constructs with the strongest effect size were habits, intentions, and personal norms. TPB is reviewed here because it has received significant attention from traffic safety researchers. However, hypotheses based on the theory will not be included in the
13 current field experiment because there seems to be little explanatory power beyond what could be gained by asking individuals about their intentions and habits. Intuitively, to the extent individuals have opportunities to manifest a behavior and intend to do so, the behavior will be realized. This is particularly so if the behavior is habitual. The results of the factor analytic study reported by De Pelsmacker and Janssens (2007) suggest that asking drivers about their intentions and habits alone would yield nearly the same power for predicting self reported speeding behavior as using the full theoretical model. A final problem with the model used by De Pelsmacker and Janssens is the failure to consider the possibility that habits influence attitudes. Clearly, attitudes toward a behavior become more favorable as the behavior becomes more habitual. In sum, TPB describes some components involved in purposeful speeding but does not appear to sufficiently explain how or why norms are formed. The concept of habit, admittedly a strong component in an individual's decision to engage in a behavior, appears to be introduced as an afterthought. Further work may improve the explanatory power of TPB, and researchers would welcome a theory that facilitates the understanding of why drivers speed. Although some individuals indicate that speeding is not dangerous, evidence suggests speed increases the frequency and severity of vehicle crashes. Does Speeding Affect Traffic Safety? NHTSA publishes an annual report entitled Traffic Safety Facts. The document provides a wealth of data about crash related variables. The most recently published version of the document (2007) ranks a set of 16 variables by relative frequency of
14 "Related Factors for Drivers and Motorcycle Operators Involved in Fatal Crashes." The most frequent factors in the list of variables for fatalities occurring in 2005 were driving too fast for conditions, inattentiveness, driving recklessly, and failing to keep in the proper lane (NHTSA, 2007). "Driving too fast for conditions" is listed as a factor for 21% of traffic fatalities for 2005. This translates to nearly 12,000 deaths in one year. Further, 36% of the total fatalities in 2005 were associated with unknown or unlisted factors. Some portion of this 36% was likely due to speed, which would increase the overall number of speed related fatalities. These recent data are supported by Treat, Tumbas, McDonald, Shinar, Hume, et al., (1977) who identified speed as a serious safety matter. Treat et al. completed in-depth analyses of approximately 2,000 fatal crashes. The authors used a strict definition of "causal:" a crash would not have occurred if the factor were absent. They determined speed was the causal factor in 8% of crashes and the probable cause of an additional 15% of the crashes. The Treat et al. data are dated, but, their conclusions support the fatality statistics reported by NHTSA (2006a) and recent work that captured driver behavior in naturalistic settings. The 100-car naturalistic driving study provides further underscores the need to establish effective speed countermeasures (Dingus et al., 2006; Klauer, Dingus, Neale, Sudweeks, & Ramsey, 2006; Klauer, Sudweeks, Hickman, & Neale, 2006b.) The naturalistic study was a significant endeavor. One hundred vehicles were instrumented with several video cameras and data loggers to capture images of the drivers, their vehicles, their environments, and several other parameters including acceleration, vehicle speed, yaw rate, and forward time to collision. The vehicles were tracked for one