CHAPTER 8: CONCLUSIONS
The analyses reported in this document are the first to evaluate driver inattention immediately prior to a crash and near-crash. These analyses used data collected as part of a large-scale naturalistic driving study. The analytical methods used were applied from epidemiology, empirical research, and qualitative research. The application of these analytical methods demonstrates the power of naturalistic driving data and its importance in relating driving behavior to crash and near-crash involvement.
Driver inattention was operationally defined at the beginning of this report as one of the following:
These four types of inattention, either in isolation or in combination, were used to answer the research questions addressed in this letter report. Some of the important findings addressed as part of these questions are presented below:
Odds ratio calculations, or relative-risk calculations for a crash or near-crash, were conducted in three separate chapters. First, Chapter 2, Objective 1, odds ratios were calculated for three levels of secondary task complexity, two durations of time that eyes were off the forward roadway for driving-related inattention to the forward roadway , two durations of time for non-specific eyeglance away from the forward view , and driver drowsiness (moderate to severe). Odds ratio calculations were calculated in Chapter 3, Objective 2 to determine whether driving while engaging in secondary tasks or drowsy through various types of driving environments produced higher near-crash/crash risks. Finally, odds ratios were also calculated for total length of time eyes were off the forward roadway by increments of 0.5 seconds in Chapter 6, Objective 3 .
Data used to calculate the odds ratios included a subset of the 69 crashes and 761 near-crashes where the driver was at-fault that were collected as part of the 100-Car Study and 20,000 baseline epochs (5,000 baseline epochs for any odds ratios requiring eyeglance data only). Please note that the 20,000 baseline driving epochs were first selected based upon the number of crashes, near-crashes, and incidents that each vehicle (not driver) was involved and then randomly selected across the entire 12 months of data collection. Each baseline epoch was a 6-second segment when the vehicle was in motion. This stratification technique created a case-control data set as those vehicles who were more involved in crashes, near-crashes, and incidents also had more baseline events to compare. Case-control designs are optimal for calculating odds ratios due to the increased power that a case-control data set possesses. Greenberg et al. (2001) argue that using a case-control design allows for an efficient means to study rare events, such as automobile crashes. Thus, the causal relationships that exist for these events can be evaluated by using relatively smaller sample sizes than are used in typical crash database analyses where thousands of crashes may be used.
Table 8.1 presents the odds ratios for the different types of inattention that increase individual near-crash/crash risk. Please note that driving-related inattention to the forward roadway is not in this table as this type of inattention was found to be safer than normal, baseline driving. Tables 8.2 and 8.3 present the odds ratios for the interaction of drowsiness with various environment and road-type conditions and the interaction of complex secondary tasks with environmental conditions, respectively. The odds ratios for the interaction of moderate-secondary-task engagement and environmental variables will not be presented as a majority of these odds ratios were not significantly different from 1.0. Table 8.4 presents the odds ratios for the lengths of total time eyes were off the forward roadway. All tables present only those odds ratios that were greater than 1.0. In all tables, those that were significantly different from 1.0 are in bold font.
The odds ratios presented for the time eyes were off the forward roadway suggests that any time driver’s eyes were off the forward roadway greater than 2 seconds increases near-crash/crash risk by two times (Table 8.4). None of the eyeglances away from the forward roadway that were less than 1.5 seconds were significantly different from 1.0.
A population attributable risk percentage calculation is a measure of the percentage of crashes and near-crashes that could be attributed to the variable being measured. Population attributable risk percentages are useful when interpreting odds ratios, or relative risk calculations for a crash or near-crash. Some odds ratios may have a very high individual risk; however that behavior/situation does not occur frequently in nature and therefore attributes to very few crashes in the population. An example of high odds ratios leading to low population attributable risk percentage includes the secondary tasks of reaching for a moving object, external distraction, reading, applying makeup, and eating. Even though each of these tasks obtained very high individual near-crash/crash risk, these factors did not account for a large percentage of actual crashes and near-crashes as shown by the population attributable risk percentage calculations in Table 8.5. Drowsiness, in contrast, resulted in a high relative near-crash/crash risk value and attributed to between 22 and 24 percent of the crashes and near-crashes in the population. This finding is important since these values are much higher than most crash database research has shown (Campbell, Smith, and Najm, 2003).
Also note that while the odds ratio for t alking/listening to a hand-held device was only slightly above 1.0 and much lower than dialing a hand-held device , the population attributable risk percentage was similar for both actions. This result may be due primarily to the relative frequency of occurrence of both actions. Dialing a hand-held device may be more dangerous but it requires less time whereas talking/listening to a hand-held device occurred frequently and perhaps, for long periods of time. Talking/listening to a hand-held device was the most frequent type of secondary task distraction observed.
An important result from these analyses is that eyeglances greater than 2 seconds contributed to 18 percent of all crashes and near-crashes and eyeglances in general attributed to 18 percent of all crashes and near-crashes that occur in a metropolitan driving environment (Table 8.6). While the purpose or location of eyeglance does matter, the longer the time away from the forward roadway, the more dangerous the activity becomes. It is apparent that many crashes are attributable to long glances away from the forward roadway.
Please note that there are some limitations of the given data set that must be considered when interpreting these results. First, the 100-Car Study was conducted in one geographical area of the country and that location was a metropolitan area; therefore, the odds ratios and the population attributable risk percentages are generalizable to a metropolitan environment and probably less so to the United States driving population at-large.
Further analyses need to be conducted to determine how all of these individual odds ratio and population attributable risk percentage calculations interact with each other. Please note that many of these odds ratios were individually calculated and do not account for any correlations that probably exist between many of these variables, i.e., weather conditions and roadway surface conditions. A logistic regression could be performed to assess the odds ratios and population attributable risk percentages accounting for these naturally occurring correlations. Please note that measures were taken to reduce the amount of correlation by using only those events where one type of inattention was present. For example, the odds ratios that were calculated on drowsiness or one of the levels of secondary task , driving-related inattention , or non-specific eyeglance used only those events that contained a single type of inattention. Therefore, the correlations between these odds ratios are somewhat controlled. The odds ratios that were calculated on each secondary task type (i.e., dialing hand-held device ) are not as controlled and correlations probably do exist among some of these. While this should not detract from the odds ratio calculation itself, these odds ratio calculations and subsequent population attributable risk percentage calculations should not be summed to assess an overall impact of secondary task engagement, for example.
While eyeglance duration was used in two chapters of this report, secondary task duration analysis was not presented. Project resources limited this reduction task primarily because of the difficulties involved in operationally defining task duration. While others have operationally defined secondary task duration (Stutts, et al., 2003), there were many issues in the data collection and reduction procedures that created obstacles for this type of reduction. For example, there were only cameras pointing at the driver which made a length of conversation with passenger difficult to assess. Also no continuous audio channel was present which also hindered a calculation of duration of conversation with passenger, radio usage, and hands-free devices . The use of 90-second segments of crash and near-crash events and 6-second baseline epochs also precluded the determination of length of hand-held device conversations, and sometimes eating, drinking, or more lengthy secondary-task types. While some of these issues could be alleviated with more time (i.e., reducing the entire trip file rather than a 90-second segment), the issues of no audio or view of the passenger seating in the vehicle will be difficult to overcome. Future research may attempt to overcome these issues with either a snapshot of the passenger compartment to determine number of passengers in the vehicle or brief but frequent bursts of an audio channel to help determine conversation length, whether the stereo is in use, etc.
As was repeatedly found throughout these analyses, drivers are inattentive and/or looking away from the forward roadway during a significant portion of the events and baseline epochs. While some of this inattention may be due to systematic scanning of the driving environment or engagement in secondary tasks or drowsiness, any eyeglance away from the forward roadway greater than 2 seconds greatly increases near-crash/crash risk. Developers of collision avoidance warning systems should incorporate these findings into newer generations of warning systems. If the system can incorporate driver eyeglance location prior to a crash, the false alarm rate of these warning systems could be greatly reduced thus increasing their effectiveness.
It is apparent from the results of the analyses in Chapter 3, Objective 2, that there are roadway and traffic environments that are better suited to engage in secondary tasks (Tables 8.3 and 8.5). Generally, it appears that engaging in secondary tasks during more visually cluttered, lower sight-distance, or demanding traffic environments (intersections, entrance/exit ramps, curved roadways), poor weather or roadway conditions (rainy weather, icy or wet road surfaces) are not the optimal locations and/or moments to engage in secondary tasks. This information could be used to better educate young drivers or those drivers who are attending traffic schools about the dangers of distracted driving and how to avoid crashes and near-crashes due to distraction. It was also found that near-crash/crash risk due to drowsiness increased when drivers were on straight/level roadways and less visually demanding environments (i.e., low traffic densities). Drivers should be aware that it may be harder to fight the effects of drowsiness and that near-crash/crash risk does increase despite the less-demanding driving environment.
The strong correlation obtained between involvement in inattention-related crashes and near-crashes and involvement in inattention-related baseline epochs has several implications on driving behavior. First, this strong correlation implies that those drivers who are getting caught, per se, by involvement in inattention-related crashes and near-crashes, are also those who frequently engage in secondary tasks or drive drowsy on a regular basis. This may also indicate that there are not very many drivers who do engage in secondary tasks and/or drive drowsy frequently while driving that are never or rarely involved in inattention-related crashes and near-crashes. This relationship will be further explored in Task 5 of this research contract.