The Causes and Consequences of Distraction in Everyday Driving

Abstract

To document drivers’ exposure to potential distractions and the effects of these distractions on driving performance, inconspicuous video camera units were mounted in the vehicles of 70 volunteer subjects. The camera units automatically recorded a closeup view of the driver’s face, a broader view of the interior of the vehicle, and the roadway immediately ahead of the vehicle whenever it was powered on. Three hours of randomly selected data per subject were coded based on a taxonomy of driver distractions (talking on cell phone, eating, tuning radio, etc.), contextual variables (whether vehicle stopped or moving, road type, traffic level, etc.) and observable measures of driver performance (eyes directed inside or outside vehicle, hands on or off steering wheel, and vehicle position in travel lane). Results were analyzed descriptively and using nonparametric bootstrap analysis techniques. The most common distractions in terms of overall event durations were eating and drinking (including preparations to eat or drink), distractions inside the vehicle (reaching or looking for an object, manipulating vehicle controls, etc.), and distractions outside the vehicle (often unidentified). Although many of the distractions were also associated with negative driving performance outcomes, further research is needed to clarify their impact on driving safety.

This paper reports on the second phase of a comprehensive research project examining driver distraction and its contribution to traffic crashes. The initial phase of the project involved analysis of five years of national Crashworthiness Data System (CDS) data to determine the role of driver distraction in U.S. traffic crashes and the specific sources of this distraction (Stutts, Reinfurt, Staplin and Rodgman, 2001). This second phase of the project involved the collection of naturalistic driving data to document drivers’ exposure to specific distracting events and the effects of these events on driving performance.

Driver distraction, and its implicit effects on hazard recognition and vehicle control, has been a prominent topic on highway safety agendas, as well as for the U.S. Congress, state legislatures, the media, and the public at large. Much of this attention stems from the enormous increase in cellular telephone use by drivers and the prospect of similar growth in other in-vehicle technologies such as vehicle navigation systems, wireless Internet capabilities, and wireless messaging.

A growing body of literature addresses the safety implications of cell phones and other wireless in-vehicle technologies. The proceedings of a “Driver Distraction Internet Forum” sponsored by the National Highway Traffic Safety Administration in the summer of 2000 (Llaneras, 2000) provides a good summary of much of this research, as does a recent report by the National Conference of State Legislatures (Sundeen, 2002). Most of these studies have been carried out in controlled settings in laboratories, on test tracks, or using driving simulators. As a group they offer strong evidence that the new technologies can negatively affect some aspects of driving performance.

Although the proliferation of new in-vehicle technologies certainly merits concern, the analysis of national crash data documented in the first phase of this study revealed many things distracting drivers and contributing to crashes. Leading the list were objects or events outside the vehicle, followed by adjusting the radio and other occupants in the vehicle. Manipulating vehicle controls, eating and drinking, cell phone use, and smoking were further down the list (Stutts et al., 2001). Other more recent studies carried out in Pennsylvania (Pennsylvania Joint State Government Commission, 2001), California (California Highway Patrol, 2002), Virginia (Glaze and Ellis, 2003), and in Great Britain (Stevens and Minton, 2002) show similar hierarchies of distracting events (although cell phones were more prominent in the California study). Drawing from this literature, the focus of the current paper is on the full range of events and activities that can draw a driver’s attention away from the task at hand, delaying recognition of safety threats and impairing effective control of the vehicle.

The above studies all involved analyses of crash data. What remains missing from the literature is data on drivers’ exposure to various potentially distracting events while engaged in everyday driving. Without information on the frequencies with which drivers engage in these various behaviors and the circumstances of this engagement, it is difficult to gauge their potential impact on driving safety. The current study was intended to address the need for real-world data on the occurrence and effects of driver distractions. The primary research questions we sought to address were: (1) How often do drivers engage in distracting behaviors? (2) Are there age and sex differences in the occurrences of driver distractions? (3) How do contextual variables such as vehicle movement affect driver distractions? and (4) What are some of the consequences of distractions on driving performance?

METHODS

VIDEO LOGGING METHODOLOGY

The system developed for continuous unobtrusive recording of in-vehicle driving behavior in subjects’ own cars consisted of a camera unit, a recording unit, and trigger and connecting cables. Figure 1 shows a schematic of the overall system. The camera unit measured approximately 18 cm × 6 cm × 5 cm, and was designed to be mounted on the vehicle’s front windshield just under the rear view mirror, using suction cups. It contained a microphone and three miniature video cameras – one directed to capture a closeup view of the driver’s face, another a broader view of the interior of the vehicle, and the third the roadway immediately ahead of the vehicle. The cameras were hidden from the driver’s view by near-infrared filters covering both sides of the camera box. An infrared light source was mounted underneath the camera box to facilitate recording under low light conditions.

Figure 1.

Schematic of video recording system

A trigger cable connected to the vehicle’s accessory fuse was used to power the cameras whenever the vehicle ignition was turned on. A locked box stored in the vehicle’s trunk contained a video recorder, quad processor, and battery pack. Cables connecting the camera unit, trigger, and recording units were discretely routed along the edges of the windshield and door frame on the driver’s side of the car, before passing through the back seat to the trunk of the car. More detailed information about the video logging methodology, including the User’s Manual developed for installing and operating the equipment, is included in Stutts, Feaganes, Rodgman, et al. (2003).

DATA COLLECTION

The video logging equipment was installed in the vehicles of 70 volunteer subjects. Subjects were recruited primarily from ads placed in local newspapers. Half resided in central North Carolina (near Chapel Hill) and half near Philadelphia, PA. There were 35 male and 35 female subjects, equally distributed among five age groups:18–29, 30–39, 40–49, 50–59, and 60+. This distribution of participants closely mimics the overall licensed driver population, except for a slight overrepresentation in the 50–59 year age group and underrepresentation in the 18–29 year age group.

Potential study participants were screened via a brief telephone interview to ensure that they had a valid driver’s license, drove at least six hours per week, and drove a vehicle that had rear seat access into the trunk (through which equipment cables could pass). They also had to be willing to come to the research offices to have the equipment installed in their vehicle and return a week later to have it removed. All participants were compensated $100 at the completion of the study.

The consent form that subjects were asked to read and sign when they came in to have the equipment installed identified the study only as an effort to learn “how traffic and roadway conditions affect driving behavior.” Although subjects were informed that their driving was being recorded, the true nature of the study was not revealed. Subjects remained inside the research center offices while the equipment was being installed. Installation of the equipment generally required 30 minutes or less, while removal required about 15 minutes. Data collection activities were initiated in late November, 2000 and extended into the following November, thus spanning a one-year driving period. The study protocol was reviewed and approved by the Institutional Review Board on Research Involving Human Subjects of the School of Public Health at the University of North Carolina at Chapel Hill.

DATA CODING AND REDUCTION

Data coding was based on a driver distraction taxonomy that evolved from the earlier CDS data analysis (Stutts et al., 2001). The taxonomy incorporated all of the major driver distraction categories appearing in the CDS datafile, with some further refinements. In addition to driver distractions, the coding taxonomy also incorporated a variety of contextual variables to describe the conditions under which drivers engage in various distracting activities, and three outcome or driver performance measures. Table 1 shows the overall data coding scheme. More detailed operational definitions used in the coding are available in the full report (Stutts et al., 2003).

Table 1.

Coded study variables

Driver Distractions:	Context Variables:
Phone/pager	Occupants in vehicle
Phone not in use1	No other occupants
Dialing phone	Front seat occupant(s) only
Answering ringing phone	Rear seat occupant(s) only
Talking/Listening	Both front and rear occupants
Eating/drinking	Travel lanes
Not eating or drinking	Neighborhood street
Preparing to eat/drink	2-lane road
Eating	3+ lane road
Drinking	Interstate
Spilled/dropped food	Other divided, multilane
Spilled/dropped drink	Other/unknown road type
Music/audio	Light/weather conditions
Music, radio, etc. not on	Light
Music, radio, etc. on	Dark
Manipulating music controls	Gray/dim light
	Dark - Unable to code
Smoking	Traffic level
Not smoking	Traffic light
Lighting cigarette, pipe, etc.	Traffic moderate
Finishing smoking	Traffic heavy
Smoking
Reading/writing or grooming	Vehicle movement
Not reading/writing/grooming	Vehicle moving
Reading/writing	Vehicle stopped
Grooming	Veh. passing through intersct*
Reading/writing & grooming	Veh. turning at intersection*
Occupant distraction	Outcome Measures:
No occupant distraction	Vehicle position
Baby distracting	Vehicle wandering in lane*
Child distracting	Veh. encroach across lane line*
Adult distracting	Sudden/sharp braking*
Conversing	Hand position
Not conversing	Both hands on steering wheel
Conversing	One hand on steering wheel
	Both hands off wheel
Internal distraction	Eyes/head
No internal distracting event	Eyes directed outside vehicle
Manipulating vehicle controls	Eyes directed inside vehicle
Falling object*
Insect*	Yawning *
Pet	Clear anger/aggressiveness *
Reaching/leaning/looking for/etc.	Clear drowsiness (head jerk, eyes drooping/closed) *
Other internal distraction
External distraction
No external distracting event
External distracting event

¹Initial value for each variable is the default setting, if a default is set.

^*Signifies an event having no time duration.

Actual coding was carried out using the Observer Video-Pro, a software-based system especially developed for coding, editing, and analyzing video data (Noldus Information Technology, Inc., 2001). The system allows up to 16 “channels” of data to be coded simultaneously. Within each channel, coding options (called “states”) are required to be mutually exclusive and exhaustive, i.e., at any one time, one state, and one state only, can be active. As an example, one of the 16 channels was designated for recording cell phone use; its four mutually exclusive and exhaustive states were (1) phone not in use, (2) phone in use (talking or listening), (3) dialing phone, and (4) ringing phone. Since each distraction category was recorded in a separate channel, multiple distractions could be coded simultaneously, e.g., a person could be both talking on the cell phone and manipulating the radio controls or eating and drinking. Separate channels were also used to record the various contextual circumstances (lanes, traffic level, light conditions, etc.) and to track the three outcome measures (hand position, direction of eyes, and vehicle wanderings and encroachments). “Events” could be coded independently of states, and had no associated duration. For example, passing through an intersection was an “event” that would only be coded while the vehicle was moving (a “state”).

Three staff members of the UNC Highway Safety Research Center were trained to use the Video-Pro system. Actual coding was carried out by simultaneously monitoring the three video screens on the quad-camera monitor display (the fourth quadrant contained a time stamp display) and entering a 2-letter variable code to record all changes in behavior as they appeared on the tape. Generally at least two complete passes of the videotape were required – the first pass to record eye direction and hand position, and the second pass to record all other behaviors. For very active drivers, a third pass was sometimes needed.

Due to the time required for coding the data and the varying amounts of available data per subject, a decision was made to code only three hours of total data per subject. For most of the 70 subjects, the three hours were evenly distributed across the total recorded time in one-half hour blocks, skipping the first one-half hour to allow drivers time to adjust to having the equipment in their vehicle. For two subjects there was less than three hours of usable data, and all of this was coded. Altogether a total of 207.2 hours of video data was coded.

Formal inter-observer reliability checks were made at several stages in the coding process using Video-Pro’s reliability test procedure. The procedure checks for coding matches within a preset window of time (e.g., ± 2 seconds). Although the overall percentage of agreement (number of matches/total number of matches plus errors) among the three coders only reached about 65% to 70%, this is very much a reflection of the subjective nature of many of the behaviors coded. With an almost infinite possibility of behaviors and conditions, it was not possible to develop a single set of objective coding criteria to cover all the behaviors and situations observed. Our approach to dealing with this issue primarily involved having the coders continuously review each other’s coding, and also having them working together in the same office so that if questions arose they could get a second opinion.

DATA ANALYSIS

As an initial step in the data analysis, basic descriptive results were generated for each of the coded variables on each of the 70 datafiles. This allowed us to check for inconsistencies in the data coding. If a questionable result was identified, it was checked and verified by reviewing the videotapes. The individual file summaries also allowed us to identify how many of the 70 subjects had displayed each of the various distracting behaviors; for example, how many used a cell phone, smoked, or transported young children. All subsequent analyses involved all 70 data files combined.

After the Video-Pro data was cleaned and finalized, the data were converted to a SAS datafile to allow for more in depth analyses, including statistical testing. Since the longitudinal nature of the data did not meet the assumptions for classic statistical analysis methods, confidence intervals for proportions and linear combinations of proportions (e.g., differences in the likelihood of eyes directed inward when dialing a cell phone versus not dialing a cell phone) were constructed using the bootstrap percentile method (Mooney and Duval, 1993). Bootstrapping is a computationally intensive nonparametric technique for constructing confidence intervals that employs large numbers of repetitive computations to estimate the shape of a statistic’s sampling distribution, rather than relying on strong distributional assumptions and analytic formulas. Details of the procedure can be found in Stutts et al. (2003), Appendix D.

RESULTS

FREQUENCY AND DURATION OF DISTRACTIONS

Information on the percentage of subjects engaging in each of the identified activities, while their vehicle was in motion, is presented in Table 2 (Column A). The data reflect any recorded incidence of a behavior, without considering the actual number or duration of the occurrences. Thus, a subject who tried one time to place a call while driving would be coded as using a cell phone, the same as a subject who placed calls and talked for most of the three hours of recorded driving.

Table 2.

Percentage of drivers engaging in potentially distracting activities when their vehicle was moving and duration of these activities.

Potential Distraction	(A) % of Subjects	(B) % of Total Driving Time1	(C) Adjusted % of Total Time2
Using cell phone (includes talking, dialing, answering)	34.3	1.30	3.8
Eating, drinking, spilling	71.4	1.45	2.0
Preparing to eat or drink	58.6	3.16	5.4
Manipulating audio controls	91.4	1.35	1.5
Smoking	7.1	1.55	21.1
Reading or writing	40.0	0.67	1.8
Grooming	45.7	0.28	0.6
Baby distracting	8.6	0.38	4.4
Child distracting	12.9	0.29	2.2
Adult distracting	22.9	0.27	1.2
Conversing	77.1	15.32	19.9
Reaching, leaning, etc.	97.1
Manipulating veh. controls	100.0	3.783	3.83
Other internal distraction	67.1
External distraction	85.7	1.62	1.9

¹Based on total sample of 70 drivers.

²Adjusted to reflect the percentage of drivers engaging in that activity, i.e., (% total time)/(proportion of drivers engaged in distraction).

³Combined categories.

During their three hours of coded driving time, nearly all subjects were observed manipulating vehicle controls (such as air conditioning or window controls) and reaching for objects inside their vehicle. Nearly as many were observed manipulating their vehicle’s music or audio controls, or had their attention drawn to something outside the vehicle (external distraction). Approximately three-fourths ate or drank while driving or conversed with a passenger. Reading/writing and grooming activities were less common, but were still observed in almost half the subjects. About a third of the subjects used a cell phone while driving, and nearly as many were distracted by passengers riding in their vehicle – either another adult, a child, or a baby.

Age differences in the likelihood of engaging in a particular distraction on at least one occasion were generally small, although sample sizes were not large enough for valid statistical testing. Compared to males, females were more likely to groom themselves (X²=8.467, df=1, p=0.004) and more likely to attend to things outside the vehicle (X²=4.456, df=1, p=0.04).

Altogether, excluding any time spent simply conversing with other passengers in the vehicle, drivers were engaged in some form of potentially distracting activity up to 16.1% of the total time that their vehicles were moving (Table 2, Column B total). Eating and drinking (including preparing to eat or drink and holding food in one’s hands) headed the list, followed by internal distractions, external distractions, and smoking. Less total time was devoted to manipulating audio controls, using a cell phone, other occupant distractions, reading or writing, and grooming. Although some of these activities may have occurred simultaneously, their totaled amount nevertheless represents a significant portion of overall time spent driving.

The percentages shown in Column B of Table 2 reflect the observed “population level” of exposure to the identified driving distractions; that is, the actual proportion of total driving time that the 70 subjects participating in our study were observed engaged in each of the identified activities, while their vehicles were moving. However, one might also be interested in a driver’s level of exposure, given that he or she engages in an activity at all. These are the adjusted percentages shown in Column C of the table. Thus, among those drivers who used a cell phone at all in their moving vehicle, the phone was in use 3.8% of the time spent driving. This latter percent also reflects the percent of total time exposed if all drivers use cell phones at the same level as did the 24 cell phone users in our study.

Following are descriptive highlights pertaining to the individual distractions. Due to the “naturalistic” nature of the data, these results include both moving and stopped vehicles.

Cell Phone

The 24 subjects in our sample of 70 who used a cell phone placed 122 calls, received 15 calls, and carried on 100 phone conversations. Assuming three hours of coded data per subject, this translates into 1.7 placed calls, 0.2 incoming calls, and 1.4 conversations per hour for those 24 subjects who used a cell phone. The average time required to place a call was 12.9 seconds, and to answer a call 7.9 seconds; the average conversation lasted 1.5 minutes, but ranged from only a second or two to over 20 minutes in length.

Eating and Drinking

Eating or drinking was coded whenever food or drink was brought to the mouth; it was not coded during chewing or swallowing, or if the food or drink was simply being held in the hand or lap. In the latter situation, “preparing to eat or drink” was coded. “Preparing to eat or drink” was also coded for activities such as removing bottle caps, unwrapping take-out food, etc. Eating and drinking were about equally common activities, and were generally short duration but very frequent events.

Music/Audio

Some form of audio, either music or talk, was playing in the vehicles 71.5% of the time – most often the radio. Only three individuals did not listen to the radio or music at all. Subjects adjusted their audio controls a total of 1,539 times, or an average of 7.4 times each per hour of driving (1,539/207.2 coded hours of driving). If the nine hours for the three individuals who did not record time with the radio on is subtracted from the total hours driving, the average number of audio control manipulations per hour of driving increases to 7.8. These manipulations averaged 5.5 seconds each.

Smoking

Only five of the 70 subjects smoked at all while driving. These five subjects lit 38 cigarettes, cigars, etc. and were recorded smoking on 45 occasions. (Note that since the data were coded in one-half hour segments, some of the coding sessions could include smoking, but not lighting or extinguishing.) This averages to 2.5 lightings and 3.0 periods of smoking per hour of driving, for the five subjects who smoked. The average time required to light a cigarette was 4.1 seconds, and the average smoking event lasted 3.4 minutes (the latter increased by a 17.4 minute interval in which a subject was smoking a cigar). There were only 17 recorded instances of finishing smoking, averaging 7.3 seconds each.

Reading/Writing

Reading or writing was observed on a total of 303 occasions. A review of the descriptive comments revealed about equal instances of each activity: sample comments included reading a map, reading a piece of paper, opening and reading mail, writing on an envelope, writing in a check book, reading the newspaper, writing in a notebook, etc. While the average duration of these events was relatively long, at 18.4 seconds, it should be emphasized that these results do not differentiate between whether the subject’s vehicle was moving or stopped at the time.

Grooming

Some form of grooming activity was noted on 229 occasions, included looking at self in mirror, combing hair, putting on lipstick, using a toothpick, putting in eye drops, putting on gloves, and a variety of other such activities. These events averaged 11.8 seconds, but again included many occurrences when the vehicle was stopped.

Other Occupant Distractions

There were a total of 243 recorded instances of drivers being distracted by other occupants in the vehicle, most often by babies (n=114) or children (n=81), but also by other adults in the vehicle (n=48). This information is best interpreted in light of the percentage of time these passengers in these age categories were being carried in the vehicles (recorded as a “context” variable). Altogether, babies were carried in drivers’ vehicles a total of 13.6 hours, children 18.1 hours, and other adults 43.9 hours. Thus, the “hourly rate” of driver distractions for infants was 8.4, for children 4.5, and for other adults 1.1.

Conversing

Talking or carrying on a conversation with another occupant in the vehicle was coded whenever it occurred, without attempting to judge whether it was distracting to the driver. Conversations were recorded 15.5% of the time overall. Since occupants were present in vehicles 30.3% of the time, this indicates active conversations about half the time another occupant was present in the vehicle.

Internal Distractions

The most frequently cited internal distraction was reaching, leaning, looking for, picking up, etc. something inside the vehicle – purse, sunglasses, sun visor, glove compartment, tissue, garage door opener, change for toll booth, etc. This behavior was noted on 2,246 occasions, or an average of 10.8 times (2,246/207.2 hours of coded data) per hour of driving per subject. Almost as frequent was manipulating vehicle controls other than the radio or music controls. These might include heat and air conditioning controls, window controls, cruise control, etc. (but not turn signals, horn, or other controls integral to the operation of the vehicle). Manipulating vehicle controls was recorded a total of 2,095 instances, or 10.0 times per hour per subject. Reaching events lasted an average of 7.6 seconds, while manipulating vehicle control events averaged 4.8 seconds. Distractions by pets, falling objects in the vehicle, and insects or bugs were quite infrequent, although it should be noted that only a few drivers carried pets in their vehicle. The larger category of “other internal distraction” captured such activities as opening one’s purse to get change, cleaning sunglasses, using a garage door opener, adjusting the sun visor, etc.

External Distractions

No attempt was made to identify a priori specific external distractions, since the potential list was so long and since, in many instances, the source or nature of the distraction might not be revealed by the outside-facing camera. Typical external distractions identified in the comment field included waving or talking to someone outside the vehicle, looking at houses or pretty scenery, toll booths, drive-through windows at banks or fast-food restaurants, work zone activity, simply looking out the side window at something, and bright sun glare. There were few recorded instances of being distracted by pedestrians, children, or animals outside the vehicle. At least one external distraction was coded for 90% of the participants. The overall count of 659 external distractions corresponds to an average of 3.2 external distractions per hour per driver, based on the full sample of 207.2 coded hours.

EFFECT OF VEHICLE MOVEMENT

Although a number of contextual variables were coded and analyzed, the variable most closely associated with distracting driving behaviors was whether the vehicle was stopped or moving at the time. Overall, vehicles were stopped 15.3% of the total recorded driving time. However, vehicles were more likely to be stopped when the driver was engaged in certain behaviors. These included reading or writing (69.5%), manipulating vehicle controls (43.3%), attending to events outside the vehicle or external distraction (41.4%), reaching/leaning/etc. (36.6%), other internal distractions (34.8%), grooming activities (34.1%), and distractions associated with other adults in the vehicle (22.2%).

These results suggests that, at least to some degree, drivers engage in these potentially distracting activities at “safer” times while driving – either at the start or end of a trip, or when their vehicle is stopped in traffic. On the other hand, answering a cell phone, eating and drinking, smoking, and distractions involving babies and children appear to occur independently of whether the vehicle is moving or stopped.

CONSEQUENCES ON DRIVING PERFORMANCE

Three measures of driving performance were identified and coded. These were (1) whether one hand, two hands, or neither hand was on the steering wheel, (2) whether the driver’s eyes were directed outside or inside the vehicle, and (3) whether the vehicle was swerving or wandering within the travel lane, crossing into another travel lane, or stopping from sudden braking. Overall, when their vehicles were moving, subjects had both hands on the steering wheel only 34.8% of the time; one hand was on the wheel 63.8% of the time, and no hand on the wheel 1.4% of the time. With respect to eye direction, eyes were directed outside the vehicle 97.2% of the time and inside the vehicle 2.8% of the time. Thus, in these first two instances where the outcomes of interest were coded as event occurrences over time, our analysis was directed at determining whether drivers spent a greater proportion of their driving time with no hands on the steering wheel or with their eyes directed inside the vehicle when engaged by a particular distraction.

The last group of driving performance measures had no associated duration, but instead were coded as single point-in-time events. Overall, there were 900 instances of recorded lane wanderings (4.3 per hour of recorded data), 444 instances of lane encroachments (2.1 per hour), and 22 instances of sudden braking (0.11 per hour). These events were summed to create an overall “adverse vehicle event” outcome, and the analysis examined whether drivers experienced higher rates of adverse vehicle events when engaged by a particular distraction.

Results of these analyses are summarized in Table 3. They represent 30 independent bootstrap analyses: one for each category of distraction with respect to each identified outcome measure. Within each distraction category, levels of that distraction (e.g., dialing or answering a cell phone, talking on a cell phone) were compared to the absence of the distraction (e.g., phone not in use). The table identifies those that were statistically significant at the .01 (**) and .05 (*) levels of confidence. (More detailed information, including 95% confidence intervals for the estimates, is included in Stutts et al., 2003.)

Table 3.

Results of bootstrap analyses for three measures of driving performance as a function of each potential distracting event, when vehicle was moving.¹

Potential Distraction	% No Hand on Wheel	% Eyes Directed In	# Vehicle Events/Hour
Cell phone/pager
Phone not in use (Ref.)	1.35	2.63	7.77
Dialing/answering	8.21 **	67.58 **	14.24
Talking/listening	6.97 *	1.35	6.24
Eating or drinking
Not eating/drinking (Ref.)	1.25	2.61	7.40
Preparing to eat/drink	4.40 **	5.52 *	18.20 *
Eating/drinking/spilling	5.32 **	6.24 *	9.02
Music/audio
Music/audio not on (Ref.)	1.00	2.85	7.98
Music/audio on	1.58	2.40	7.65
Manipulating controls	2.06 *	22.58 **	10.08
Smoking
Not smoking (Ref.)	1.43	2.76	7.83
Lighting or extinguishing	3.60	19.31 *	30.16
Smoking	0.82	1.57	3.02 *
Reading/writing
Not reading/writing (Ref.)	1.39	2.51	7.73
Reading/writing	15.10 **	91.50 **	20.93
Grooming
Not grooming (Ref.)	1.39	2.66	7.73
Grooming	12.44 *	34.62 **	20.18
Occupant distraction
No occupant distract. (Ref.)	1.42	2.60	7.65
Distracted by baby	2.75	21.93	24.21
Distracted by child	0.27	14.64	11.59
Distracted by adult	2.82	19.00	22.88
Conversing
Not conversing (Ref.)	1.41	2.53	7.54
Conversing	1.50	3.97	9.00
Internal distraction
No internal distaction (Ref.)	1.24	2.22	7.52
Manipul. veh. controls	9.79 **	15.42 **	11.30
Reach/lean/look for/etc.	3.80 **	20.10 **	18.37 **
Other internal distraction	6.97 **	12.17 **	9.95
External distraction
No external distraction (Ref.)	1.41	2.76	7.64
External distraction	2.30	2.40	15.45

¹Each variable level compared to reference (Ref.) level, e.g., talking/listening on cell phone compared to phone not in use.

^**Significant at p<.01

^*Significant at p<.05

In general, the models reveal fairly consistent trends of higher levels of no hands on the steering wheel and eyes directed inside the vehicle, along with higher rates of adverse vehicle events, associated with each of the identified driving distractions. Although in the anticipated direction, however, the results frequently do not attain statistical significance. This may be due to a number of factors, including small sample sizes for some of the categories, lack of precision in the coding, and the relative rarity of the outcomes as well as some of the measured behaviors.

Nevertheless, Table 3 results do show significantly higher proportions of no hands on the steering wheel while dialing/answering or talking on a cell phone, eating and drinking or preparing to eat or drink, manipulating audio controls, reading or writing, grooming, and the various internal distractions including manipulating vehicle controls and reaching/leaning/etc. Eyes were significantly more likely to be directed inward when dialing or answering a cell phone, eating and drinking or preparing to eat or drink, manipulating audio controls, lighting or extinguishing a cigarette, reading or writing, grooming, and when manipulating vehicle controls or reaching for something inside the vehicle. And lastly, higher incidence rates of adverse vehicle events were associated with preparing to eat or drink and reaching for something inside the vehicle.

There were also some notable exceptions to the trend of higher levels of potentially dangerous driving behaviors for a few of the identified distractions. In particular, smoking (but not lighting or extinguishing a cigarette) was found to be associated with a lower incidence of adverse vehicle events, and talking on a cell phone was associated with a lower proportion of time with the eyes directed inward. The latter result, however, was not significant statistically.

DISCUSSION

This study provides some of the first naturalistic data on drivers’ exposure to potential distracting events that have been related to crash involvement. Results show distractions to be a common component of everyday driving. Altogether, excluding any time spent simply conversing with other passengers in the vehicle, drivers were engaged in some form of potentially distracting activity up to 16.1% of the total time that their vehicles were moving. Eating and drinking (including preparing to eat or drink and holding food in one’s hands) headed the list, followed by internal distractions, external distractions, and smoking. Less total time was devoted to manipulating audio controls, using a cell phone, other occupant distractions, reading or writing, and grooming. Although some of these activities might have been conducted simultaneously, their total nevertheless represent a significant proportion of time spent driving.

The occurrence of driver distractions also varied according to whether the vehicle was stopped or moving at the time. This suggests that, at least to some degree and for some activities, drivers are choosing to engage in them at “safer” times on the roadway. Finally, the data provided some evidence that distractions can negatively affect driving performance, as measured by higher levels of drivers having no hands on the steering wheel, their eyes directed inside the vehicle, and their vehicles wandering in the travel lane or crossing into another travel lane.

There were a number of important limitations to these field data collection efforts. Foremost were problems in objectively defining all categories of driver distraction, as well as context and outcome variables. This made it difficult to achieve high levels of inter-rater reliability when coding the data. This might lead to biased estimates of the durations of our potentially distracting events, and to inflated error measurements. The latter would make our results conservative in the sense of being less likely to detect significant differences in our measured outcome effects.

We were also not able to distinguish between different levels of intensity of a distraction, which would have introduced additional subjectivity into the coding. Some potentially important variables, such as vehicle speed, vehicle deceleration, and following distances, could not be objectively coded from the video data at all. There was also the substantial time and effort entailed by the coding, which limited the number of subjects recruited and the amount of data coded per subject.

Another important limitation of the study is that the measures of driving performance we were able to code and analyze – hands on steering wheel, direction of eye gaze, and vehicle wanderings or encroachments across travel lanes – have not been directly linked to crash risk. While we may intuitively feel that drivers who engage in activities that require them to take their hands off the steering wheel or their eyes off the road for short periods of time have a higher risk of crashing, we do not know this to be true. Neither do we know that increased wandering in the travel lane is associated with higher crash risks in real world driving.

Most importantly, we were unable to capture any measure of cognitive distraction, which has been linked in the literature to poorer driving performance and increased likelihood of crashing. Such studies have typically been carried out in more controlled settings, using driving simulators or instrumented vehicles (or drivers) on test tracks (see, e.g., Strayer, Drews and Johnston, 2003). Other studies have suggested that drivers’ fixed gaze may be an indicator of cognitive distraction. These studies distinguish two types of eye movements that can indicate a driver is distracted: either short glances away from the driving task, or the longer fixed gazes that signify a cognitive distraction. In our less controlled naturalistic driving study, we could not differentiate fixed gazes from the desirable category of “eyes directed at the roadway,” for example, when subjects were talking on a cell phone. Consequently, our study is not able to provide a definitive answer as to which activities, or which driver distractions, carry the greatest risks of crash involvement.

What the study does provide is very detailed data, from a reasonably large sample of drivers, about the activities that people engage in while driving that affect some aspects of driving performance and that also might increase their risk or crashing. The project also had other strengths. Foremost was the development of the video logging methodology itself, and the demonstration of its feasibility and practicality for unobtrusively collecting real-world driving data. Another was the continued refinement of a driver distraction taxonomy, including more detailed levels of several distractions and identification of important contextual variables. Finally, much was learned about the practicalities of naturalistic data collection in this important research area, along with the reduction of multi-stream in-vehicle video data, that might be applied to future research efforts.

In the end, a better understanding of the role of driver distraction in traffic crashes is most likely to emerge from a combination of research approaches including naturalistic data studies like the current one, but also analyses of crash data and a variety of more controlled research studies in laboratory, simulation, or test track environments. Given the proliferation of new technologies anticipated in future vehicles, there is an urgent need to better manage all forms of driver distraction to maintain safety on our roadways.