driver distraction in long-haul truck drivers

www.elsevier.com/locate/trf

Transportation Research Part F 8 (2005) 441–458

Driver distraction in long-haul truck drivers

Richard J. Hanowski *, Miguel A. Perez, Thomas A. Dingus

Virginia Tech Transportation Institute, Virginia Tech, 3500 Transportation Research Plaza,

Blacksburg, VA, 24061, United States

Received 2 February 2005; received in revised form 25 July 2005; accepted 2 August 2005

Abstract

Research on driver distraction has typically been conducted by means of epidemiology or experimentaltesting. The study presented here uses a naturalistic approach, where real-world driving data were collectedfrom truck drivers as they worked their normal delivery runs. Crash, near-crash, and crash-relevant conflictdata from 41 long-haul truck drivers, driving approximately 140,000 miles, were examined. Of the 2737crashes, near-crashes, and crash-relevant conflicts (collectively termed ‘‘critical incidents’’) that wererecorded, 178 were attributed to ‘‘driver distraction’’. The 178 distraction-related critical incidents wereanalyzed and 34 unique distraction types were identified. Results showed that a small number of long-hauldrivers were involved in a disproportionate number of distraction-related critical incidents. For example,two of the drivers accounted for 43 of the 178 distraction incidents. Important insight was also gained intothe relative safety impacts of different distracting agents and behaviors. The frequency and duration of atask, along with the visual demand associated with performing the task, were found to contribute in com-bination to the prevalence of critical incidents. Finally, it was found that simply because a task does notnecessarily require visual attention does not mean that long-haul drivers will not look (sometimes often)away from the roadway. However, it is also clear that visually demanding tasks carry the highest degreeof risk, relative to other categories of tasks.� 2005 Elsevier Ltd. All rights reserved.

Keywords: Driver distraction; Commercial vehicle; Naturalistic; Cellular telephone

1369-8478/$ - see front matter � 2005 Elsevier Ltd. All rights reserved.doi:10.1016/j.trf.2005.08.001

* Corresponding author. Tel.: +1 5402311500.E-mail address: [email protected] (R.J. Hanowski).

mailto:[email protected]

442 R.J. Hanowski et al. / Transportation Research Part F 8 (2005) 441–458

1. Introduction

Driver inattention occurs whenever the operator of a vehicle diverts his or her attention awayfrom the driving task. Driver distraction, on the other hand, has been defined to occur when thisinattention leads to a delay in the recognition of information that is necessary to accomplish thedriving task safely (Stutts, Reinfurt, Staplin, & Rodgman, 2001b). Thus, distraction occurs wheninattention leads to a critical incident. This definition describes the construct of distraction on aquantifiable basis. It also accounts for the fact that drivers often gaze at areas that are irrelevantto the driving task without any undesirable consequences. By this definition, visual inattention(and many other types of inattention, including cognitive inattention) is considered harmless untilit results in a critical incident. Therefore, driver distraction can be represented as: inatten-tion + critical incident = distraction.

Using this model, studying driver distraction requires the identification of critical incidents.Critical incidents vary from high to low severity. Crashes are high severity critical incidents wherethere is an impact between the vehicle and another object. Low severity critical incidents includecrash-relevant conflicts which involve a safety risk but where a crash does not occur. An exampleof a crash-relevant conflict is an unintended lane deviation in which the vehicle drifts outside thevehicle�s lane of travel.

The study of distraction in the context of critical incidents has been described previously (e.g.,Hancock, Lesch, & Simmons, 2003). Traditional research studying driver distraction can be clas-sified into two broad methodological categories: epidemiology and empirical testing. Epidemiol-ogy involves looking at crashes (i.e., high severity critical incidents) after they have occurred.Researchers can use a variety of crash databases, such as the Fatal Accident Reporting System(FARS), in an attempt to assess crash causal factors. Conservative estimates based on epidemio-logical evidence suggest that driver distraction is a primary factor in 12.9% of all crashes (Stutts,Feaganes, Rodgman, Hamlett, Meadows, & Reinfurt, 2003a), although some of the estimates ofthis incidence are as high as 25–30% (Llaneras, 2000; Minter, 2000). Epidemiological research fo-cused on specific technologies also suggests an increased crash risk due to the distraction gener-ated by those technologies (Goodman et al., 1997; Redelmeier & Tibshirani, 1997; Violanti &Marshall, 1996). These data have prompted the National Highway Traffic Safety Administration(NHTSA) to study the driver distraction problem (Llaneras, 2000; Tijerina, Johnston, Parmer,Winterbottom, & Goodman, 2000).

However, these databases are not sufficiently detailed to assess driver behavior. For example,the crash report from a fatal crash seldom provides any information on the driver�s behaviorimmediately preceding the crash. To obtain crash information of interest to driver distraction,researchers would require a substantial restructuring of the current data collection system. Forexample, police accident report forms could include a check box to indicate that the driver wasusing an electronic or telematics device (e.g., cellular phone) at the time of the crash (assumingthat this could, in fact, be determined). Some of these restructuring efforts are already occurring(Model Minimum Uniform Crash Criteria, 2003).

Even these changes in the crash data collection process are unlikely to provide all the necessarydata for the complete assessment of driver distraction, such as eye-scanning behavior. Becausedriving is primarily a visual task, secondary tasks and in-vehicle devices should not significantlydivert the driver�s eyes away from the forward roadway. By quantifying driver inattention, mea-

R.J. Hanowski et al. / Transportation Research Part F 8 (2005) 441–458 443

sures such as task completion time and eyes-off-road time assess the potential for distraction asso-ciated with a task or device. However, neither of these measures can be obtained from an epide-miological, post-crash data collection analysis.

Empirical testing for driver distraction addresses many of the limitations associated withthe epidemiological approach. For example, in a controlled simulator or test-track environ-ment, driver behavior can be closely monitored. Typical measures of interest in driver distrac-tion research, such as task completion time and eyes-off-road time, can be precisely measured.However, as with epidemiological studies, there are limitations to studying driver distractionusing empirical methods. For example, it can be argued that the experimental situation mayalter driver behavior; thus, drivers may act differently in the real world as compared to acontrived experimental world. A related problem is that many experimental studies involvedata collection for a relatively short duration (e.g., up to a few hours of driving time), there-by limiting the investigation of behavior change that may occur over time. Drivers are usu-ally asked to employ an unfamiliar technology in an unfamiliar instrumented vehicle for arelatively short time period, which may not allow the driver to become familiar with the vehi-cle�s control characteristics (e.g., Green, Hoekstra, & Williams, 1993) or in the case of drivingsimulators, may not accurately reflect driving dynamics. Altogether, while these drawbacksmay be acceptable in the study of prototype devices used while driving, they are problematicin the study of everyday distractions that may serve as contributing factors for real-worldcrashes.

Naturalistic data collection may address some of the aforementioned shortcomings of epidemi-ological and experimental methods. In these studies, vehicles are instrumented with a variety ofdata collection equipment and driven by study participants for one or more weeks (Dinguset al., 2002; Hanowski, Wierwille, Garness, & Dingus, 2000) to a few months (Hanowski, Nakata,& Olson, 2004) to one year or more (Dingus et al., in press). The data collection equipment typ-ically consists of video cameras to record driver behaviors and sensors to record driver input andperformance. Naturalistic studies have been carried out in both light-vehicle (Dingus et al., inpress; Stutts, Feaganes, Rodgman, Hamlett, Reinfurt, & Gish, 2003b) and commercial-vehicle(Dingus et al., 2002; Hanowski et al., 2000) environments. With light vehicles, drivers use instru-mented vehicles in their daily drives. For commercial vehicles, operators use instrumented truckson their normal delivery, revenue-producing runs. Because data collection occurs in a real worlddriving environment, naturalistic driving studies tend to have limited control of independent vari-ables. However, because the data are collected in the real world, experimental results generallyhave high external validity.

Due to the expense, logistics, and effort associated with naturalistic data collection studies,few have been conducted. However, when naturalistic experiments are conducted, they canprovide very rich data that address a variety of research areas and questions. This paper de-scribes a naturalistic study where the data were used to investigate driver distraction in com-mercial vehicle operations. It was believed that the data gathered using this methodologywould provide information concerning the characteristics of various distractions, includingdurations and frequencies, which would allow comparisons of the relative risks associated withvarious distracters. In addition, it was believed that this methodology would allow forthe examination of relationships between the frequency of distraction incidents and driverfactors.


2. Method

Data from 41 long-haul truck drivers, driving approximately 140,000 miles, were collected andwere originally analyzed to investigate driver fatigue during long-haul trucking operations (‘‘Slee-per Berth’’, Dingus et al., 2002). More specifically, analyses on the data examined the relationshipof various fatigue-related factors, including driver sleep quantity and quality with sleeper berthdesign, environmental factors, team versus single operations, and length of trip. The methodsand equipment necessary to perform these analyses are described in Dingus et al.

Of the 41 drivers, 33 were involved in distraction-related critical incidents. The driver sampleincluded 28 males and 5 females, 16 were single drivers and 17 drove as part of a two-person team.All drivers who participated in this study were recruited from one of four for-hire commercialtrucking companies. Two of these companies primarily hauled perishable items and used refrig-erated trailers. The other two companies primarily hauled dry goods and used standard box trail-ers. Neither of the companies had a union affiliation. The average age of the drivers was 41.7 years(range: 28–63 years) with an average of 13 years of driving experience (range: 1–42 years of drivingexperience). All participants signed an informed consent form and all protocols of the study wereapproved by the Virginia Tech Institutional Review Board.

Two tractors, owned by the experimenters� research organization, were used in this study: a1997 Volvo L4 VN-series tractor and a 1995 Peterbilt 379. Functionally identical instrumentationpackages and data collection systems were installed in both trucks. The data acquisition systemfunctioned to record: (a) four camera views, including the drivers face; (b) driving performanceinformation including steering, lane departure, and braking; (c) sleeper berth environmental dataincluding noise, vibration, and temperature; (d) subjective alertness ratings; and (e) data from asleep analysis system. Because the data collection runs conducted in the Sleeper Berth project were6–10 days long (up to 240 h), recording data continuously was not feasible given the technologyused at the time. As such, a method of collecting only data of interest to the project was devised.Of primary interest were the occurrence of critical incidents; as such, the data collection systemwas programmed to only record and save data in which the on-board sensors detected a potentialcritical incident (e.g., a hard braking maneuver). Pre-determined thresholds were set for the sen-sors and the data were recorded and saved only when these thresholds were exceeded. Driverinteraction with the data collection system was minimal.

The naturalistically collected data from Dingus et al. (2002) contained a large number of inci-dents where the data were flagged due to sensor readings that exceeded preset thresholds. Thesethresholds were carefully selected to flag data that, with a high probability, surrounded a crash,near-crash, or crash-relevant conflict. For example, a situation where the truck exceeded a laneboundary but did not complete a lane-change maneuver (i.e., unintended lane deviation) wasflagged as a potential critical incident. Other sensor flags included unusually high (i.e., greaterthan 99th percentile) lateral acceleration, unusually high longitudinal deceleration, short time-to-collision, and high rate of closure relative to a lead vehicle. Once the flagged data were down-loaded from the trucks, a trained data analyst reviewed the video and electronic data to determinewhether or not the criteria had been met for a valid critical incident. Three types of critical inci-dents were considered valid for the purposes of this study. A ‘‘crash’’ was defined as an incidentwhere the truck came in contact with any other vehicle, fixed object, pedestrian, or animal. A‘‘near-crash’’ was operationally defined as a conflict with another vehicle, fixed object, pedestrian,


or animal that required a rapid evasive maneuver for crash avoidance. A ‘‘crash-relevant conflict’’was operationally defined as a conflict or driver error that resulted in an unsafe scenario but didnot require an evasive maneuver.

Each incident that was recorded was examined by trained data analysts. The data analysts�training process included written training on the classification system and practice with video al-ready reduced by expert data analysts. After the training process, data analysts were closely super-vised over a period of at least two weeks. After this probationary period, data analysts weresupervised by a senior researcher and encouraged to ask questions when unsure about the appro-priate classification for an event. These data analysts reviewed the video stream surrounding thevalid critical incidents and assigned an incident type (assumed cause) classification. Only one clas-sification category was selected by the data analyst as the primary cause for the incident. The threemain causes were judgment error, other vehicle, and driver distraction. ‘‘Judgment error’’ was thegeneral term used to identify any error by the truck driver, including errors in the decision-makingand execution stages of a maneuver. Judgment errors included factors such as following too clo-sely and changing lanes with insufficient gap. ‘‘Other vehicle’’ was used for situations where thedriver of another vehicle was judged to have initiated the incident. For a critical incident to beassessed with a ‘‘driver distraction’’ contributing factor, the driver had to be engaged in a tertiarytask immediately before the incident occurred. Tertiary tasks included using a cell phone, tuningthe radio, eating, looking away from the forward roadway, and similar tasks that did not directlyinvolve the driver�s primary task of safely operating the vehicle. The incidents in the Dingus et al.(2002) study where driver distraction was an assumed contributing factor were then further ana-lyzed as the events of interest for this study.

Three steps were taken to analyze the distraction events. The first step was to review the eventsand verify that distraction was the likely primary cause in the critical incident. The videotapes thatrecorded the 178 distraction events were carefully reviewed to identify the cause of the distraction.Events were also classified based on demographics and other between-driver factors. Frequencyhistograms and counts were created to summarize these classifications. Demographic comparisonsof distracted drivers versus the complete driver sample were also drawn to identify any sources ofbias in the sample. In addition, the frequency of distraction-related incidents was examined as afunction of whether the driver drove independently (i.e., single driver) or as part of a team (i.e.,team driver). The reader should note that the concept of team driving implies that two driversshare (take turns) in operating the tractor. In a team driving situation, one driver drives whilethe other driver rests.

The second step in the analysis involved determining exposure criteria to assess relative riskamong the various distractions. Methodologically, however, traditional exposure estimates aredifficult to obtain for event-triggered data; to determine traditional exposure estimates accurately,continuous data collection and substantial video analysis would be required. While baselineevents were collected as part of this study, these events were timed-triggers that required the driverto interact with a device in the truck in order to provide a self-rating of fatigue. These baselineevents did not allow for the generation of accurate base rates, as their sparseness and short dura-tion precluded the observation of long driving periods. Thus, the exposure due to total frequency ofthe presence of the distracting agents cannot be determined directly from these data; data on thetime exposure required by the task were considered instead as an estimate of exposure. This timeexposure was estimated separately via two different aspects of inattention: the average time it took


to finish the task (an estimate of the overall exposure) and the average amount of time that thedriver�s eyes were off the forward roadway while the task was performed (an estimate of the expo-sure relative to the risk presented by the task). One might expect tasks and activities with a largetime exposure to be more hazardous than those with a small time exposure; tasks that drivers en-gage in more frequently and for a longer period of time may serve to divert more of the driver�sattention away from operating the vehicle. Because data are available for only those distractionsthat resulted in a critical incident, it is already known that the task negatively affected driver per-formance. Thus, inferences can be made, on a relative time exposure basis, about the hazardous-ness of a particular distraction.

The third step in the analysis involved conducting eye-glance analysis on a 20-s epoch sur-rounding the event trigger (10 s prior to the start of the event and 10 s after the start of the event).Using custom software and the recorded video of the driver�s face, data analysts assessed driverglance locations and lengths during this 20-s distraction epoch. This window allowed for reason-able observation of driver behavior before and after the event while limiting the inclusion of nui-sance or confounding events in the data. These glance locations and lengths were then statisticallysummarized by distraction event. Measures included in the analysis were the proportion of timespent with eyes-off-road and the mean and maximum glance durations.

A cluster analysis was performed on the data, using event frequency, duration, proportion oftime spent with eyes-off-road, and associated mean glance duration away from the forward viewas dependent variables. Analysis of variance (ANOVA) and Chi-square tests were used where pos-sible to determine significant differences (a-value of 0.05) in various dependent variables. TheTukey honestly significant difference (HSD) procedure was performed to assess pair-wise differ-ences between clusters.

3. Results

3.1. Distribution of distraction incidents

A total of 2737 critical incidents were recorded in the Sleeper Berth study (Dingus et al., 2002),of which 178 (6.5%) were attributed to driver distraction. The most frequent cause of critical inci-dents was the broad category of ‘‘judgment error’’, which was the assessed cause for 2108 criticalincidents (77.0%). ‘‘Other vehicle’’ accounted for the second largest number of incidents (265 inci-dents or 9.7%). ‘‘Driver distraction’’ was the third highest assessed cause.

No crashes were recorded. Six of the 178 distraction events had no kinematics trigger event be-cause the trigger that resulted in the data collection was timed. One event was classified as a near-crash; the associated distraction was adjusting the Citizen Band (CB) radio, which is employed bytruck drivers as a two-way travel information system by using the knowledge of other drivers thatare further ahead on the route (among other uses). The event was classified as a near-crash be-cause the subject driver had to brake hard in order to avoid a crash with a lead vehicle that brakedsuddenly. The remaining events were classified as crash-relevant conflicts. Given this distributionand the lack of any pre-incident characteristics suggesting a difference between the near-crash andcrash-relevant conflicts, all of the 178 events were considered as a group (‘‘critical incidents’’), andno distinction is made in the remainder of this paper based on the severity of an event.


Driver distraction was associated with approximately 7% of the critical incidents observed inthe experimental sample. Within the causes over which the truck driver had direct control, onlyjudgment error, as previously described, was more frequent. Recall that distraction-related crit-ical incidents occurred among 33 different drivers out of 41 drivers that comprised the driverpool in the Dingus et al. (2002) study. The incident frequency of occurrence varied substantiallyas a function of driver. Two of the drivers accounted for 43 (24.2%) of the distraction incidentsrecorded. No age or gender effects were observed. Though there were nearly equal numbers ofsingle and team drivers represented in the distraction data set (16 single drivers and 17 teamdrivers), single drivers accounted for 115 of the 178 recorded distraction-related incidents(64.6%).

3.2. Critical incident analysis

Table 1 displays the results of the cluster analysis, which showed separation of the thirty-fourunique distraction types into seven distinct clusters. These clusters tended to separate incidentsbased on their duration and frequency. The frequency of occurrence of critical incidents for eachDistraction Type is shown in Fig. 1, with the two drivers exhibiting the largest number of inci-dents shown separately. The frequency of distraction incidents in each of the clusters shown inFig. 1 was statistically different (X 6

1 ¼ 243:9, p < 0.0001). While there were 121 incidents in the firstcluster (which contained most of the tasks), there were 34 instances of the second (looking right—outside and looking at Instrument Panel, IP) and seventh (looking left—outside) events. Theselooking outside distractions were by far the most frequent and were grouped separately. The ob-jects people were looking at, however, could not be identified by data analysts as camera place-ment did not support this type of observation.

Once these distractions were identified, further analysis of their occurrences was performed toidentify common factors that suggested reasons for the link between these events and a criticalincident. Fig. 2 displays the duration of the activity prior to the occurrence of the critical incidentand the proportion of time that the driver�s eyes were off the road for a 20-s period around theincident as a function of cluster and associated distracter. The one-way ANOVA showed thatthe average duration of a distracter, F(6,27) = 10.42, p < 0.0001, and the proportion of eyes-off-road time, F(6,27) = 19.32, p < 0.0001, were significantly different across clusters. Task durationclearly differentiated between two of the cell phone tasks, exhibiting the longest durations, andthe remaining tasks. Results of the Tukey HSD procedure indicated that the mean duration ofdistracters in Clusters 5 and 6 (cell phone tasks, M = 190.9 s) was significantly different thanthe duration of distracters in the remaining clusters (M = 15.61 s). However, Tukey HSD testsbased on the proportion of eyes-off-road time showed few clear-cut differences; distracters in Clus-ter 3 (M = 58.6%), which required more than a 50% proportion of eyes-off-road time, were signif-icantly different from distracters in Clusters 1, 4, and 5 (M = 23.9%). In general, the averagelength of the distractions varied greatly and did not correspond with the relative frequency ofthe occurrence of the critical events. It is important to note that the critical event frequencies varygreatly across these categories, and several have only a single data point. However, since this fig-ure represents only safety-critical event data in which a distracting agent was an apparent contrib-uting factor, it provides important insight into the characteristics of the pre-event behavior thatled to the event.

Table 1Description of distractions and associated definitions

Cluster Distraction Definition

1 Talking on CB Driver is holding CB to mouth and talking; usually looking forward;one hand off the wheel

Looking at CB Driver is looking up at CB receiver located on ceiling at the frontand center of cab; both hands on the wheel

Adjusting CB Driver is adjusting knobs, with right arm extended up, on CBreceiver located on ceiling at the front and center of cab;glancing at CB periodically; one hand off the wheel

Looking at radio Driver is looking at the music radio, down and to the right, on centerconsole of cab; both hands on the wheel

Adjusting radio Driver is reaching to the music radio, on center console of cab,adjusting station or volume; glancing at radio periodically; one handoff the wheel

Looking up Driver is looking up at the visor; both hands on the wheelLooking down Driver is looking down; either in lap at something unknown,

or at hands; may have one or both hands off the wheelLooking at floor Driver is looking at/for something on the floor (down and to the right);

both hands on the wheelTalking to passenger Driver is talking to another person in the cab; sometimes looking

to the right at passenger; both hands on the wheelEating/talking Driver is eating food and looking at passenger; one hand off the wheelToothpick/visor mirror Driver is looking up in the visor mirror, while picking teeth with a

toothpick; one hand off the wheelDrinking Driver is drinking out of a soda bottle or mug; usually

looking forward; one hand off the wheelGetting cigarette Driver is removing a cigarette from rest of pack; often looking at pack;

one hand off the wheelLighting cigarette Driver is lighting a cigarette; often looking at cigarette; one or

both hands off the wheelBlowing smoke Driver has head turned, blowing smoke out the window; usually

holding cigarette with one hand off the wheelAdjusting in seat Driver is adjusting himself/herself in driver seat; usually

looking forward; both hands on the wheelReaching to floor Driver is reaching for something either on the floor of the cab

(down and to the right) or somewhere in the cab; usuallylooking forward; takes one hand off the wheel

Rubbing face Driver is wiping face off or rubbing eyes; usually looking forward buteyes may close for a few moments; one hand off the wheel

Taking off jacket Driver is taking off jacket; usually looking forward; one handoff the wheel

Brushing hair Driver is using a hairbrush to brush hair; looking forward;one hand off the wheel

Wiping dash Driver is wiping off dash of cab with a cloth; usually looking at dash;one hand off the wheel

Release of steering wheel Driver is looking forward but does not have either hand on wheelwhile moving in seat; is not holding anything

Answering ringing cell phone/Looking at cell phone display

Driver is answering ringing cell phone; reaches to middle console,picks up phone, looks down at phone several times, but neverputs it to ear; one hand off the wheel


Table 1 (continued)

Cluster Distraction Definition

2 Looking right—outside Driver has head turned to the right, either looking in passengerside mirror, or out passenger window; usually both hands are onthe wheel

Looking at IP Driver is looking down, through steering wheel, at instrumentpanel containing speedometer and gauges; both hands on the wheel

3 Dialing cell phone Driver is looking down at cell phone in hands, dialing number;one hand off the wheel

Plugging in cell phone Driver is plugging in battery charger to bottom of cell phone;usually looking at the phone; one hand off the wheel

Getting food Driver is getting food out of a bag in their lap; often looking at bag/food; one or both hands off the wheel

Looking at paperwork Driver is holding paperwork on steering wheel and is looking downat it; one or both hands off the wheel

4 Reaching in pocket Driver is reaching for something in either front shirt pocket,or back pant pocket; usually looking forward but moving aroundin seat; one hand off the wheel

Looking outside Driver is looking at a road sign, something along side of the road,or another car, but is still looking out front window; both hands onthe wheel

5 Talking on cell phone Driver is holding cell phone up to ear and talking on it; usuallylooking forward; one hand off the wheel

6 Phone call/hanging up cell phone Driver makes phone call and is hanging up cell phone; reachesdown to floor to put phone back; usually looks down; one handoff the wheel

7 Looking left—outside Driver has head turned to the left, either looking in driver sidemirror, or out driver window; usually both hands are on the wheel

The definition for each distraction highlights the relevant task/activity and indicates where the driver tended tolook during the distraction (the associated visual demand) and the typical status of the driver�s hands on the steeringwheel during the distraction (the associated manual demand). Groupings obtained from the cluster analysis are alsoindicated.


Fig. 3 illustrates the mean glance durations for viewing locations away from the forward road-way. The one-way ANOVA showed a significant difference in the mean downward glance dura-tion between clusters, F(6,27) = 15.13, p < 0.0001. Results from the Tukey HSD procedureindicated that Cluster 4 (M = 0.00 s) had a significantly lower mean downward glance durationthan the remaining clusters (M = 1.00 s, 1.36 s, 1.34 s, 0.93 s, 1.23 s, and 1.08 s for Clusters 1–3and 5–7, respectively). This was expected given that the tasks in Cluster 4 had no downwardglances associated with them, thus, their mean downward glance duration was zero. The overallmean downward glance duration across clusters was 1.01 s (SD = 0.33 s). Maximum single glancedurations were generally shorter than four seconds, with two exceptions, Clusters 2 and 7. Theoverall maximum task duration, however, was contained within Cluster 3 and occurred while aparticipant was looking at paperwork.

0

5

10

15

20

25

30

35

40

Talk

ing

on C

B

Look

ing

at C

B

Adju

stin

g C

B

Look

ing

at ra

dio

Adj

ustin

g ra

dio

Look

ing

up

Look

ing

dow

n

Look

ing

at fl

oor

Talk

ing

to p

asse

nger

Eat

ing/

Talk

ing

Toot

hpic

k/V

isor

mirr

or

Drin

king

Get

ting

ciga

rette

Ligh

ting

ciga

rette

Blow

ing

smok

e

Adj

ustin

g in

sea

t

Rea

chin

g to

floo

r

Rub

bing

face

Taki

ng o

ff ja

cket

Bru

shin

g ha

ir

Wip

ing

dash

Rel

ease

of s

teer

ing

whe

el

Ans

wer

ing

cell

phon

e/Lo

okin

g at

cel

l pho

ne d

ispl

ay

Look

ing

right

- ou

tsid

e

Look

ing

at IP

Dia

ling

cell

phon

e

Plug

ging

in c

ell p

hone

Get

ting

food

Look

ing

at p

aper

wor

k

Rea

chin

g in

poc

ket

Look

ing

outs

ide

Talk

ing

on c

ell p

hone

Phon

e C

all/H

angi

ng u

p ce

ll ph

one

Look

ing

left

- out

side

Distraction Type

Num

ber o

f Inc

iden

t Occ

urre

nces

Other Drivers (N=31)Driver 22Driver 21

Clu

ster

1

Clu

ster

2

Clu

ster

3

Clu

ster

4

Clu

ster

5

Clu

ster

6

Clu

ster

7

Fig. 1. Frequency of critical incidents for each of the 34 distraction types. Totals for each of the two drivers with themost incidents are shown relative to the rest of the drivers. Vertical lines separate incidents in the different categoriesdefined by the cluster analysis.


4. Discussion

The current study analyzed critical incidents, collected in a naturalistic study with long-haultruck drivers, in which driver inattention was observed prior to the incident occurrence. As such,‘‘driver distraction’’ was assessed to have been a contributing factor in these critical incidents.

The examination of the distribution of overall incidents yielded some insights as to their causalfactors. For example, judgment error was associated with substantially more critical incidentsthan driver distraction. The difference in frequency between these two causal factors can be ex-pected based on findings from previous research (Wierwille, Kieliszewski, Hanowski, Keisler, &Olsen, 2000). The finding of few drivers accounting for a large number of incidents is consistentwith the findings from a study conducted with local/short haul drivers which found that a largenumber of recorded critical incidents were caused by a small number of drivers (Hanowski et al.,2000). However, in the current study the distribution of incidents among distraction types forhigh-incidence drivers seemed similar to the overall incident distribution, indicating that the highincidence observed for the drivers was not due to excessive involvement in a particular activity.

0.0

10.0

20.0

30.0

40.0

50.0

60.0

70.0

80.0

90.0

100.0

Talk

ing

on C

B

Look

ing

at C

B

Adju

stin

g C

B

Look

ing

at ra

dio

Adju

stin

g ra

dio

Look

ing

up

Look

ing

dow

n

Look

ing

at fl

oor

Talk

ing

to p

asse

nger

Eatin

g/Ta

lkin

g

Toot

hpic

k/Vi

sor m

irror

Drin

king

Get

ting

ciga

rette

Ligh

ting

ciga

rette

Blow

ing

smok

e

Adju

stin

g in

sea

t

Rea

chin

g to

floo

r

Rub

bing

face

Taki

ng o

ff ja

cket

Brus

hing

hai

r

Wip

ing

dash

Rel

ease

of s

teer

ing

whe

el

Answ

erin

g ce

ll ph

one/

Look

ing

at c

ell p

hone

dis

play

Look

ing

right

- ou

tsid

e

Look

ing

at IP

Dia

ling

cell

phon

e

Plu

ggin

g in

cel

l pho

ne

Get

ting

food

Look

ing

at p

aper

wor

k

Rea

chin

g in

poc

ket

Look

ing

outs

ide

Talk

ing

on c

ell p

hone

Phon

e C

all/H

angi

ng u

p ce

ll ph

one

Look

ing

left

- out

side

Distraction Type

Prop

ortio

n of

Tim

e (%

, Whi

te B

ars)

0

50

100

150

200

250

Tim

e D

urat

ion

(sec

, Bla

ck B

ars)Proportion of Task Time not Looking Forward

Average Event Duration

A AB C AC B ABCa a a a b b a

Clu

ster

1

Clu

ster

2

Clu

ster

3

Clu

ster

4

Clu

ster

5

Clu

ster

6

Clu

ster

7

B

Fig. 2. Mean duration and proportion of eyes-off-road time for each of the 34 distraction types. Vertical lines separateincidents in the different categories defined by the cluster analysis. Uppercase letters indicate differences between clustersbased on the proportion of eyes-off-road time (clusters with the same letter are not significantly different). Lowercaseletters indicate differences between clusters based on the distracter�s duration (clusters with the same letter are notsignificantly different).


Furthermore, single drivers accounted for more than 60% of the events collected, even thoughsimilar numbers of single and team drivers were included in the study. This finding is consistentwith the overall results of the Dingus et al. (2002) study, which indicated single drivers had manymore critical incidents than team drivers. More specifically, Dingus et al. found that single driversaccounted for approximately 77% of all recorded critical incidents, compared to approximately65% observed in this subset of the data. Based on results from a naturalistic driving study withlight vehicle drivers (Dingus et al., in press), it is possible that the presence of a second personin the cab in the team situation provided the driver with assistance to avoid potential hazards(i.e., a second set of eyes to monitor for unsafe situations). Additional research is needed to ex-plore this hypothesis.

The cluster analysis provided information on the groupings that best separate these events,based on a number of driving performance variables. While somewhat speculative, it is interestingto assign descriptions to each of the clusters based on their characteristics, which are presentedgraphically in Figs. 1–3. Distractions that could logically group together are instead contained

0

1

2

3

4

5

6

7

8

Talk

ing

on C

B

Look

ing

at C

B

Adju

stin

g C

B

Look

ing

at ra

dio

Adj

ustin

g ra

dio

Look

ing

up

Look

ing

dow

n

Look

ing

at fl

oor

Talk

ing

to p

asse

nger

Eat

ing/

Talk

ing

Toot

hpic

k/V

isor

mirr

or

Drin

king

Get

ting

ciga

rette

Ligh

ting

ciga

rette

Blow

ing

smok

e

Adj

ustin

g in

sea

t

Rea

chin

g to

floo

r

Rub

bing

face

Taki

ng o

ff ja

cket

Bru

shin

g ha

ir

Wip

ing

dash

Rel

ease

of s

teer

ing

whe

el

Answ

erin

g ce

ll ph

one/

Look

ing

at c

ell p

hone

dis

play

Look

ing

right

- ou

tsid

e

Look

ing

at IP

Dia

ling

cell

phon

e

Plug

ging

in c

ell p

hone

Get

ting

food

Look

ing

at p

aper

wor

k

Rea

chin

g in

poc

ket

Look

ing

outs

ide

Talk

ing

on c

ell p

hone

Phon

e C

all/H

angi

ng u

p ce

ll ph

one

Look

ing

left

- out

side

Distraction Type

Gla

nce

Dur

atio

n (s

ec)

Mean Downward Glance DurationMax Downward Glance Duration

AB AB B C AB AB AB

Clu

ster

1

Clu

ster

2

Clu

ster

3

Clu

ster

4

Clu

ster

5

Clu

ster

6

Clu

ster

7

Fig. 3. Mean and maximum downward glance duration for each of the 34 distraction types. Vertical lines separateincidents in the different categories defined by the cluster analysis. Uppercase letters indicate differences between clustersfor the mean downward glance duration (clusters with the same letter are not significantly different).


within different clusters (e.g. �looking right—outside� and �looking left—outside� could be groupedinto �looking outside�), and speculating on the reason for the separation also aids in pointing outseveral unique characteristics of the data. Cluster 1 can be characterized as an average frequency,average event duration grouping, and correspondingly encompasses the majority of the events(Figs. 1 and 2). Cluster 2 is a high frequency, high maximum downward glance group (see Figs.1 and 3). Cluster 3 is a group with events that exhibit a high proportion of time not looking for-ward (Fig. 2). Cluster 4 is characterized by the lack of downward glances; participants performingthese tasks spent the duration of the task time looking forward (Fig. 3). Cluster 5 is characterizedby long event duration (Fig. 2). Cluster 6 also has a long event duration, but is combined with alarge proportion of time spent not looking forward (Fig. 2). Finally, Cluster 7 is characterized bya very large frequency of occurrence (Fig. 1).

It is possible, even likely, that some of these clustering results may be due to noise in the data.This would cause a realignment of events in the case that another similar dataset was collected andanalyzed. However, it is reasonable to expect that similar groupings would be observed with re-spect to the task characteristics. Thus, disproportionately high-frequency events would begrouped separately from others based on their increased risk exposure. This risk could be hypoth-


esized to be different from the corresponding risk for events that might occur relatively infre-quently, but that have longer durations. It could be argued that the risk of the high-frequencyevents lies in the probability of a serious change in the driving environment within a short periodof time while the driver is being inattentive to the driving task. The increase in risk lies in the fre-quency with which drivers engage in tertiary non-driving tasks while driving. In the second case,where a long event duration occurs (albeit the event itself occurs infrequently), the risk is a func-tion of the probability of a serious change in the driving environment within a long period of time.The differential increase in risk as a function of frequency or time accounts for the differential riskbetween these observed clusters. Further research is needed to determine these risk functions, butthe current research shows how these particular dependent variables can differentiate betweenevents that would appear similar at first glance.

Incidents related to drivers looking at or reaching for something around the cabin were alsofrequent events. Given that each of these events resulted in a critical incident, these findings pro-vide insight into the relative risk associated with differing distracting agents. CB categories, forexample, accounted for approximately 11% of the total number of incidents. In contrast, eating,drinking, and grooming-related (e.g., combing hair) tasks, as well as the cell phone dialing/answering tasks, had lower frequencies of associated critical incidents. While baseline informationis not available that would match these figures with an exposure factor, this finding is still impor-tant because it indicates how often the events are causal factors in a critical incident. The issue ofhow often these actions are performed is secondary to the fact that, for the substantial time periodof driving observed for this investigation, these actions resulted in a hazardous driving condition.

The analysis of event duration and proportion of time spent not looking forward illustrate animportant point about driving-distraction-related safety surrogate measures that are time based,such as the time required to complete a tertiary task. While such measures may be an importantcomponent of the risk associated with a distracting agent, there are clearly other important fac-tors, including visual demand and task frequency, which contribute to the overall risk associatedwith distraction. For example, the category of ‘‘reaching to the floor’’ clearly constitutes a riskybehavior based on the number of critical incidents. However, the duration of the task was veryshort relative to other categories. Thus, it appears that the visual-manual aspects of reaching(including an element of bending to retrieve something on the floor) are important, risk-producingfactors. A second important consideration is that reaching to the floor was, in all likelihood, muchless frequent during normal driving than talking on the CB or looking at an object in the environ-ment. Thus, it is apparent that on a case-by-case basis, reaching to the floor is a very risky behav-ior, as empirical studies would no doubt show.

The concept of cognitive distraction also appears to be present for some of the incidents thatwere identified. For example, in the looking outside incidents, no glances were made to areas out-side the forward roadway. However, the driver was distracted for several seconds without anyapparent source of distraction within the forward view (which was visible to the data analysts),as is evident from the critical incident observed in the video. In this case, simply basing the degreeof distraction on glances outside the forward view would be inappropriate. Furthermore, simplybecause a task does not require visual attention does not imply that drivers will not look awayfrom the roadway (sometimes often), as was the case for some tasks consisting of primarily audi-tory stimuli. Overall, these findings indicate that the concept of distraction is multifaceted; thus,multiple constructs must be understood and quantified before distraction can be modeled.


These data indicate that there are three very important factors contributing to the prevalence ofsafety-related critical events. First, if a distraction agent is frequent, it has an associated risk evenwhen it is not time intensive or visually demanding. For example, looking at objects outside of thevehicle or glancing at the instrument panel seems to fit in this category. Note that while the datado not provide an indication of overall frequency, some of these events would logically be ex-pected to occur often. Second, some distracting agents are demanding tasks of short durationand lower frequency. Reaching to the floor to retrieve an object appears to fit in this category.Third, some tasks that are moderately high in time demand, visual demand, and frequency areassociated with the prevalence of safety-related critical events. Talking on or adjusting the CBor talking on the phone appears to fit this pattern.

It seems reasonable from these data that the combination of the frequency of a particular task,the time that task takes to complete, and the visual demand of the task are the pre-dominant fac-tors that affect driving risk. Other studies (e.g., Wierwille, 1995; Wierwille & Tijerina, 1998) havefound similar results. These results, combined with past epidemiological research on driver dis-traction (Redelmeier & Tibshirani, 1997), support this assertion.

Although exposure frequency data were not gathered for this study, precluding a direct test ofthe importance of task frequency on the occurrence of critical events, data from other studies canbe used to shed some light on the relative frequency of some of these behaviors. Wierwille andTijerina (1998) summarized data from several sources and estimated the frequency of usage perweek (for light vehicle drivers) for the following devices: (a) speedometer—300, (b) mirror—250, (c) radio controls—56, and (d) smoking/lighting—32. While these data might not be directlyapplicable to heavy vehicle drivers, when converted to relative frequencies (e.g., the speedometer isused more frequently than the radio), they are comparable to the relative frequencies with whichthese types of events appeared as critical incidents in the current study (i.e., for looking at theinstrument panel, adjusting the radio, and smoking). This suggests that frequent distracting eventscan also result in critical incidents, and it stresses the importance of considering frequency of per-formance in assessing the risk associated with a secondary task or tertiary behavior.

4.1. Comparing two studies of distraction hazardousness

While this was the first known study to consider distractions from naturalistically collected crit-ical incident data, previous research has used crash data to quantify the extent to which severaldistractions result in a crash (Stutts, Reinfurt, & Rodgman, 2001a). When data from Stuttset al. are compared with the current data, considerable qualitative differences in the frequency dis-tribution of the distraction categories are observed. Fig. 4 displays the percent of distracters as apercentage of the total frequency of distracters in the current study and Stutts et al. While the pro-portions of drivers distracted due to an outside person, object, or event are similar between studies(6.5% in the current study versus 8.3% in Stutts et al.), the percent of distracters as a percentage ofthe total frequency of distracters are considerably different.

One possible reason for the differences in frequency distributions is the inherent differences be-tween critical incidents (examined in this study) and crashes (examined in Stutts et al., 2001a).That is, the distractions that result in critical incidents might be different from those that resultin a crash. However, given the multitude of factors that must interact to cause a crash, simplyarriving at this conclusion is not appropriate. It is likely that the main difference between the crit-

0

5

10

15

20

25

30

35O

utsi

de p

eson

, obj

ect,

or e

vent

Adju

stin

gra

dio/

cass

ette

/CD

Oth

er o

ccup

ant

Mov

ing

obje

ct in

vehi

cle

Oth

er d

evic

e/ob

ject

Adju

stin

gve

hicl

e/cl

imat

e co

ntro

ls

Eatin

g an

d/or

drin

king

Usi

ng/d

ialin

g ce

llph

one/

CB

Smok

ing

rela

ted

Oth

er d

istra

ctio

ns

Unk

now

n di

stra

ctio

ns

Distraction Type

Perc

enta

ge o

f Tot

al O

bser

ved

Inci

dent

s

Stutts et al., 2001in situ data

Fig. 4. Comparison of data from Stutts et al. (2001a, 2001b) and current data.


ical incidents in the current study and the crashes in Stutts et al. simply lies in the absence of sev-eral other factors necessary to transform the critical incident into a crash (e.g., presence of othertraffic).

The difference in frequency distribution may also be due to an under-representation of crashesclassified as distraction-related in Stutts, Reinfurt, and Rodgman (2001a). These data are basedon the crashworthiness data system (CDS), which annually summarizes information from a sam-ple of all crashes in the United States. Two factors can affect the proportion of distraction-relatedcrashes in these data: (a) unless witnesses are present, drivers are unlikely to specify certain dis-tractions (e.g., talking on the cell phone) as the reason for the crash and are more likely to blameother causes (e.g., infrastructure) and (b) in many of the crashes, the reason is specified as un-known. Both of these factors are likely to underestimate the influence of certain distracters.

Another possible explanation for the difference in frequency distribution is the population beingused. While Stutts et al. (2001a) included all types of drivers in their study (e.g., light and heavyvehicles), the current data only considered heavy-vehicle commercial drivers. It is possible thatcertain distractions that affect light-vehicle drivers (which compose the majority of the samplein the CDS data) have a different impact on heavy vehicle drivers and vice versa. In otherwords, the frequency with which a distraction occurs may not be completely indicative of itshazardousness. Rather, driver experience with the distraction might also affect the influence of


that distraction on driver behavior. For example, ‘‘adjusting vehicle/climate controls’’ was classi-fied as a distraction type in the Stutts et al. study but was not an apparent pre-critical-incidentdistraction with heavy-vehicle drivers. One possible explanation for this finding is that heavy-vehi-cle drivers, who spend much of their workday in the cab of their truck, may be more knowledge-able and better practiced at performing secondary tasks while driving than light-vehicle drivers.This result was found in a study by Blanco, Biever, Gallagher, and Dingus (2003). Therefore,when comparing this particular task between the two driving groups, it may only have a negligibleimpact on safety for heavy-vehicle drivers.

Both the under-representation of distraction-related crashes in the Stutts et al. (2001a) data andthe differences in the driver groups may have played a role in the observed differences in the fre-quency distributions between the two data sets. The relative influence of each, however, is un-known and will likely remain so until long-term naturalistic driving studies are performed forboth light and heavy vehicles.

4.2. Assessment of naturalistic data collection methodology

Because the dataset from the Dingus et al. (2002) study included only critical incidents and thedata collection methodology did not use a continuous data collection approach, one limitation ofthe current analysis lies in the unavailability of accurate exposure information. Simply as a func-tion of exposure frequency, without considering causation, one would expect critical incidents andcrashes to occur more frequently with tasks and activities that drivers perform more frequently.Because only critical incidents were recorded in this research, the true extent to which a drivertalked on a CB, for example, and did not have an incident is unknown. Rather, only the tasksand activities that drivers were engaged in during an incident are known. Thus, we can determinethe relative risk of differing activities, but we can not accurately separate out the effects of taskfrequency. Efforts are underway to collect continuous driving data for extended periods of time,which would allow the estimation of exposures through random sampling or other statistical tech-niques (Dingus et al., in press).

From a methodological standpoint, this study has provided a glimpse of the type of data thatcan be derived from naturalistic data collection efforts. It is believed that, along with epidemiologyand empirical testing, naturalistic data collection can provide key data that can be used to answerquestions related to driver distraction and many other safety-related questions. It is suggested thatnaturalistic driving studies provide a unique approach to investigating driver behavior issues,including driver distraction, and address many of the limitations associated with epidemiologicaland experimental approaches to studying this topic. As with any methodological approach, thereare limitations associated with naturalistic data collection. However, it is believed that the datafrom naturalistic driving studies such as the one described here provide important data to assesssafety, going well beyond what can presently be gleaned from database inquiries or short-termempirical studies.

Finally, the results of this study were obtained from long-haul truck drivers, which form a un-ique subset of the driver population. Thus, care should be taken in extrapolating these results toother driver populations. However, the methodology presented here could be as effective as it wasin this study in researching the behaviors of other driver populations (e.g., older drivers, teendrivers).


Acknowledgements

The data used in this paper were collected under a Federal Motor Carrier Safety Administra-tion (FMCSA) project entitled, ‘‘The Impact of Sleeper Berth Usage on Driver Fatigue’’. TheContract Number was DTFH61-96-C-00068. The Contracting Officer�s Technical Representativewas Robert J. Carroll. The data mining and analysis work presented in the paper was sponsoredby the US Department of Transportation/Research and Special Programs Administration/VolpeCenter. Dr. Thomas Ranney served as the project�s technical advisor. Thanks to Jeff Hickman ofthe Virginia Tech Transportation Institute for his review of the manuscript.

References

Blanco, M., Biever, W. J., Gallagher, J. P., & Dingus, T. A. (2003). The impact of secondary task cognitive processing

demand on driving performance. Unpublished manuscript.Dingus, T. A., Klauer, S. G., Neale, V. L., Petersen, A., Lee, S. E., & Sudweeks, J., et al. (in press). The 100 car

naturalistic driving study, phase II—Results of the 100 car field experiment. Washington, DC: National HighwayTraffic Safety Administration.

Dingus, T. A., Neale, V. L., Garness, S. A., Hanowski, R. J., Keisler, A. S., & Lee, S. E., et al. (2002). Impact of sleeper

berth usage on driver fatigue (Contract No. DTFH61-96-C-00068). Blacksburg, VA: Virginia Tech TransportationInstitute.

Goodman, M., Bents, F. D., Tijerina, L., Wierwille, W., Lerner, N., & Benel, D. (1997). An investigation of the safety

implications of wireless communications in vehicles. Retrieved June 2002, from http://www.nhtsa.dot.gov/people/injury/research/wireless/.

Green, P., Hoekstra, E., & Williams, M. (1993). On-the-road tests of driver interfaces: examination of a route guidance

system and a car phone (Technical Report No. UMTRI-93-35). Ann Arbor, MI: The University of MichiganTransportation Research Institute.

Hancock, P. A., Lesch, M., & Simmons, L. (2003). The distraction effects of phone use during a crucial drivingmaneuver. Accident Analysis and Prevention, 35(4), 501–514.

Hanowski, R. J., Nakata, A., & Olson, R. L. (October 2004). Methodological overview of the drowsy driver warningsystem field operational test. In Safety Performance and Accident Free Driving, SP-1911 (pp. 103–108). (SAETechnical Paper Series 2004-01-2718). Warrendale, PA: Society of Automotive Engineers International.

Hanowski, R. J., Wierwille, W. W., Garness, S. A., & Dingus, T. A. (2000). Impact of local/short haul operations on

driver fatigue, final report (Report No. DOT-MC-00-203). Washington, DC: US Department of Transportation,Federal Motor Carriers Safety Administration.

Llaneras, R. E. (2000). NHTSA driver distraction Internet forum: summary and proceedings (Report). Washington, DC:National Highway Traffic Safety Administration.

Minter, S. G. (2000). Driver safety is no distraction. Occupational Hazards, 10(October).Model Minimum Uniform Crash Criteria (MMUC). (2003). Retrieved January 21, 2003, from http://www.nhtsa.

dot.gov/people/perform/trafrecords/pages/mmucc/mmucc.htm.Redelmeier, D. A., & Tibshirani, R. J. (1997). Association between cellular-telephone calls and motor vehicle collisions.

The New England Journal of Medicine, 336(7), 453–458.Stutts, J., Feaganes, J., Rodgman, E., Hamlett, C., Meadows, T., & Reinfurt, D., et al. (2003a). Distractions in everyday

driving (Report No. N/A). Washington, DC: AAA Foundation for Traffic Safety.Stutts, J. C., Feaganes, J., Rodgman, E., Hamlett, C., Reinfurt, D., & Gish, K., et al. (2003b). The causes and

consequences of distraction in everyday driving. Paper presented at the 47th annual proceedings for the association forthe advancement of automotive medicine.

Stutts, J. C., Reinfurt, D. W., & Rodgman, E.A. (2001a). The role of driver distraction in crashes: an analysis of 1995–

1999 crashworthiness data system data. Paper presented at the proceedings of the 45th annual conference of theassociation for the advancement of automotive medicine, San Antonio, TX.

http://www.nhtsa.dot.gov/people/injury/research/wireless/

http://www.nhtsa.dot.gov/people/injury/research/wireless/

http://www.nhtsa.dot.gov/people/perform/trafrecords/pages/mmucc/mmucc.htm

http://www.nhtsa.dot.gov/people/perform/trafrecords/pages/mmucc/mmucc.htm


Stutts, J. C., Reinfurt, D. W., Staplin, L., & Rodgman, E. A. (2001b). The role of driver distraction in traffic crashes.Retrieved May 2, 2005, from http://www.aaafoundation.org/projects/index.cfm?button=distraction.

Tijerina, L., Johnston, S., Parmer, E., Winterbottom, M. D., & Goodman, M. (2000). Driver distraction with wireless

telecommunications and route guidance systems (Report No. DOT HS 809-069). Washington, DC: National HighwayTraffic Safety Administration.

Violanti, J. M., & Marshall, J. R. (1996). Cellular phones and traffic accidents: An epidemiological approach. AccidentAnalysis and Prevention, 28, 265–270.

Wierwille, W. W. (1995). Development of an initial model relating in-vehicle visual demands to accident rate. Paperpresented at the proceedings of the 3rd annual mid-Atlantic human factors conference, Blacksburg, VA.

Wierwille, W. W., Kieliszewski, C. A., Hanowski, R. J., Keisler, A. S., & Olsen, E. C. B. (2000). Identification and

evaluation of driver errors: Task E report—Investigation of critical incidents (Interim Report for the Federal HighwayAdministration Contract No. DTHF-61-97-C-00051). Washington, DC: Federal Highway Administration.

Wierwille, W. W., & Tijerina, L. (1998). Modeling the relationship between driver in-vehicle visual demands andaccident occurrence. In A. G. Gale (Ed.), Vision in vehicles VI (pp. 233–243). Amsterdam: North Holland/ElsevierPress.

Richard J. Hanowski is the Director of the Center for Truck & Bus Safety at the Virginia Tech Transportation Institute.He received his Ph.D. in Industrial and Systems Engineering at Virginia Tech in 2000.

Miguel A. Perez is a Senior Research Associate at the Virginia Tech Transportation Institute. He received his Ph.D. inIndustrial and Systems Engineering at Virginia Tech in 2005.

Thomas A. Dingus is the Newport News Shipbuilding/Tenneco Professor of Civil and Environmental Engineering atVirginia Tech. In addition, he serves as Director of the Virginia Tech Transportation Institute. He received his Ph.D. inindustrial engineering and operations research from Virginia Tech in 1987.

http://www.aaafoundation.org/projects/index.cfm?button=distraction

driver distraction in long-haul truck drivers

Documents