Road accidents caused by drivers falling asleep

Accident Analysis and Prevention 31 (1999) 639–649

Road accidents caused by drivers falling asleep

Fridulv Sagberg *Institute of Transport Economics, PO Box 6110, Etterstad, N-0602 Oslo, Norway

Received 8 October 1998; received in revised form 23 December 1998

Abstract

About 29 600 Norwegian accident-involved drivers received a questionnaire about the last accident reported to their insurancecompany. About 9200 drivers (31%) returned the questionnaire. The questionnaire contained questions about sleep or fatigue ascontributing factors to the accident. In addition, the drivers reported whether or not they had fallen asleep some time whilstdriving, and what the consequences had been. Sleep or drowsiness was a contributing factor in 3.9% of all accidents, as reportedby drivers who were at fault for the accident. This factor was strongly over-represented in night-time accidents (18.6%), inrunning-off-the-road accidents (8.3%), accidents after driving more than 150 km on one trip (8.1%), and personal injury accidents(7.3%). A logistic regression analysis showed that the following additional factors made significant and independent contributionsto increasing the odds of sleep involvement in an accident: dry road, high speed limit, driving one’s own car, not driving the cardaily, high education, and few years of driving experience. More male than female drivers were involved in sleep-related accidents,but this seems largely to be explained by males driving relatively more than females on roads with high speed limits. A total of10% of male drivers and 4% of females reported to have fallen asleep while driving during the last 12 months. A total of 4% ofthese events resulted in an accident. The most frequent consequence of falling asleep—amounting to more than 40% of thereported incidents—was crossing of the right edge-line before awaking, whereas crossing of the centreline was reported by 16%.Drivers’ lack of awareness of important precursors of falling asleep—like highway hypnosis, driving without awareness, andsimilar phenomena—as well as a reluctance to discontinue driving despite feeling tired are pointed out as likely contributors tosleep-related accidents. More knowledge about the drivers’ experiences immediately preceding such accidents may give a betterbackground for implementing effective driver warning systems and other countermeasures. © 1999 Elsevier Science Ltd. All rightsreserved.

Keywords: Driver behaviour; Sleep; Fatigue; Inattention; Road accidents

www.elsevier.com/locate/aap

1. Introduction

Fatigued or drowsy drivers have long been acknowl-edged to constitute a potential traffic safety hazard, andseveral research studies have addressed various aspectsof the problem. Brown (1994) has presented a compre-hensive review and discussion of the research literatureon the nature of fatigue and its effects on driver be-haviour and traffic accidents, on the basis of which heconcluded that ‘fatigue is insufficiently recognised andreported as a cause of road accidents’.

In discussing the effects of fatigue, Brown furtherpoints out that the main effect is ‘a progressive with-drawal of attention from road and traffic demands’. A

most extreme form of withdrawal of attention is obvi-ously the closing of eyes due to sleepiness.

To prevent accidents related to drowsiness and sleep-ing behind the wheel, it is important to acquire preciseknowledge about the extent of the problem as well asits preconditions and consequences. The following is-sues, all of which are investigated empirically in thepresent study, are considered relevant for a better un-derstanding of sleep-related accidents.1. What is the proportion of accidents caused by fa-

tigue or sleep behind the wheel, and what are themost likely types of accidents to occur under theseconditions?

2. How prevalent is the problem of actually fallingasleep while driving?

3. What are the most frequent consequences of fallingasleep while driving?

* Tel.: +47-2257-3800; fax: +47-2257-0290.E-mail address: [email protected] (F. Sagberg)

0001-4575/99/$ - see front matter © 1999 Elsevier Science Ltd. All rights reserved.

PII: S 0 0 0 1 -4575 (99 )00023 -8

F. Sagberg / Accident Analysis and Pre6ention 31 (1999) 639–649640

4. To what extent is sleeping behind the wheel relatedto characteristics of the driver, the road and trafficconditions, and the trip?

Quite a few studies have estimated the proportion ofaccidents caused by drivers who were fatigued or actu-ally were dozing off behind the wheel. Several studiesbefore the mid 1980s were reviewed by Nairn (1987),showing estimated proportions of crashes related tosleep or fatigue varying from 1% for all accidents to48% of fatalities.

Most typically, these studies were based on officialcrash report statistics. This holds true also for laterstudies. For example, a Norwegian study of policereports from frontal crashes (Amundsen and Chris-tensen, 1986) showed that falling asleep at the wheelwas noted as causal factor in 16 out of 286 accidents(5.6%).

In the US, analyses of larger databases have beenreported. Thus, Pack et al. (1995) found that the driverwas judged to be asleep in less than 1% of the crashesreported in North Carolina over the years 1991–1992.Lyznicki et al. (1998) report on the basis of a literaturereview that driver sleepiness is involved in 1–3% of allUS motor vehicle crashes. They point out, however,that surveys of the prevalence of sleepy behaviourindicate that sleepiness is a more common causativefactor that shown by these estimates. Similar estimateshave been reported also from New Foundland (seeTaoka, 1998), and from Switzerland (Laube et al.,1998).

The underestimation of the involvement of sleep andfatigue in police reports has been pointed out also byFell (1994). He reports on the development in newSouth Wales, Australia, of a set of criteria includingcrash trajectories and conditions of occurrence to deter-mine whether a crash is related to sleep. On the basis ofthese criteria, which however have not been validated,it was estimated that sleepiness accounts for as much as6% of all crashes. For fatal crashes the estimate was15%, and for fatal crashes on rural roads as high as30%. Knipling and Wang (1995) have pointed out thatthese criteria may be over-inclusive and capture somecrashes not necessarily related to sleep or drowsiness.Their estimate for all US police-reported crashes, basedon a series of criteria to ascertain drowsiness involve-ment in cases where this was not explicitly stated in thereport, was that between 1.2 and 1.6% of all crashes arerelated to drowsiness or fatigue.

As indicated by some of the estimates above, sleeptends to be more prevalent as contributing factor inhigh-severity crashes. According to Pack et al. (1995)data from the US Fatal Accident Reporting Systemshows that falling asleep accounts for 3–4% of fatalcrashes, an estimate that is higher than those givenabove for all accidents.

Probably more valid evidence on sleepiness involve-ment in fatal accidents comes from Finland, wherespecial investigation teams routinely collect data fromall fatal accidents. Based on the database from theseinvestigations, Summala and Mikkola (1994) foundthat about 7% of the fatal accidents were due to adriver falling asleep, and an additional 3% were causedby fatigue.

The higher involvement of sleep and fatigue in fatalaccidents is most likely related to a higher prevalence ofdriver sleepiness on major roads, especially in ruralareas, where driving speeds are higher. It is notable thatHorne and Reyner (1995) found that sleepiness wasinvolved in as much as 16% of police-reported accidentson major roads in England and over 20% onmotorways.

Even if the large discrepancies between the variousestimates of driver sleepiness involvement in accidentscited above can be explained by differences regardingthe accident categories included, there are good reasonsfor suspecting that this problem is underestimated inpolice reports, partly because of verification problems.An alternative source of information is therefore thedrivers themselves. There have been published a fewreports of self-reported sleep involvement in accidents,which generally tend to result in somewhat higherestimates than the police reports.

For example, in a British study, Storie (1984) col-lected reports from about 15 000 drivers who had beeninvolved in an accident, and found that at ‘the time justbefore the accident happened’ 3% reported to havebeen sleepy and 8% bored (not mutually exclusivecategories). It cannot, however, be concluded from thatstudy that the mentioned conditions were causally re-lated to the accident.

More recently, Maycock (1997) reported that out ofsome 1000 accidents reported by a sample of 4621 maleUK drivers over a 3-year period, 9–10% were associ-ated with tiredness.

An additional difference between police reports andself-reports of accidents is that the latter are likely toinclude more accidents, especially in the low-severitycategory. And since there seems to be a positive rela-tionship between severity and sleepiness involvement,this may tend to deflate the proportion of sleep-relatedaccidents in self-report data as compared to data basedon official reporting systems. This implies that theunderreporting of sleep as accident cause in officialdata is even larger than suggested by a direct compari-son in terms of proportion of reported crashes.

In the present study, a self-reported approach wasused for collecting data about sleep and fatigue ascausal factors in accidents. A sample of Norwegiandrivers who had reported a crash to their insurancecompany and were judged by the insurance company tobe at fault for the event, were asked in a questionnaire

F. Sagberg / Accident Analysis and Pre6ention 31 (1999) 639–649 641

whether they considered sleep or fatigue (among severalfactors listed) to have contributed to the accident. Inaddition, information was collected regarding crashtype and severity, making it possible to consider therole of sleep/fatigue for different kinds of accidents.

In addition to accidents, which represent the worst-case outcome of the driver falling asleep, there arelikely to be numerous instances of near-miss situations,where the driver wakes up soon enough to avoid aserious situation. To get information about the preva-lence of falling asleep while driving, Garder and Alex-ander (1995) carried out a survey among a randomsample of motorists as well as a sample of students inthe US. They found that 52% reported to have fallenasleep at least once while driving a motor vehicle, andthat 29% had fallen asleep during the last 12 months.As to the consequences, 7% reported an accident as aresult of having fallen asleep. Garder and Alexander(1995) also asked their drivers who had fallen asleep tostate their lateral position at the time when they awoke,and found that 62% were off their driving lane with atleast one pair of wheels. It was also shown that devia-tions to the right side of the lane are much more likelythan to the left. Apart from these results, knowledgeabout other consequences than accidents is extremelyscarce.

Somewhat lower prevalence estimates were reportedby McCartt et al. (1996) based on a telephone surveyamong 1000 New York State license holders. A total of23% reported to have fallen asleep some time whiledriving, and 2.8% had crashed when fallen asleep.

Information about the consequences of near-misssituations where the driver has fallen asleep can provideuseful indications about accident risk. Since the numberof near-miss incidents is large compared to accidents,use of such data is a good alternative to accident datafor driver samples that are too small for meaningfulanalyses of accidents. For instance, the relative preva-lence of crossing the right edge-line when falling asleepcan be assumed to be related to the relative risk ofrunning off the road, whereas crossing the centrelinebears a similar relationship to the relative risk of head-on collision. One purpose of the present study was tocollect information both regarding the proportion ofdrivers who had experienced episodes of falling asleepwhile driving, and what had been the most seriousconsequence of the incident.

A further objective was to investigate to what extentsleep-related accidents as well as the prevalence offalling asleep while driving can be predicted on thebasis of background information about the driver, suchas sex, age, and driving experience. For the accidentsadditional background factors to be investigated in-clude characteristics of the road, the car and the tripwhere the accident occurred.

It was also of interest to see if the findings of Garderand Alexander (1995) on US drivers could be replicatedon a sample of Norwegian drivers. The risk of fallingasleep during driving is probably related both to thetype of road and to the traffic density, with higher riskon straight, monotonous, dual carriageway roads withlittle traffic, where boredom is likely to occur. Suchroads make up a relatively large proportion of USroads, as compared to many other countries. It istherefore likely that the prevalence of falling asleepwhile driving is higher in the US than e.g. in Norway,where roads on the average are more hilly and curved,and where dual carriageway roads make up a relativelysmall part of the road network.

2. Method

2.1. The dri6er sample

From the files of the two largest insurance companiesin Norway a random sample of the most recent acci-dent reports was drawn. The sample consisted of 29 600accident reports received within the last few months.The car owners who had reported these accidents,received a postal questionnaire, to be filled in by theperson who had been driving the car during theaccident.

2.2. The questionnaire

Questions about sleeping behind the wheel were in-cluded as part of a more comprehensive questionnaireto accident-involved drivers, with the primary aim ofinvestigating accident risk associated with the use ofmobile telephones as well as various other in-car androadside sources of distraction. The results concerningdistraction and accidents are published elsewhere (Sag-berg, 1999).

The following questions were asked regarding fatigueand sleep:

‘Did fatigue or falling asleep while driving contribute tothe accident you were involved in?’

‘Have you ever fallen asleep (or dozed off for a shortmoment) while driving a car?’

If the answer was yes to the second question above, thefollowing two additional questions should be answered:

‘Has this happened to you during the last 12 months?’

‘What was the consequence of your dozing off whiledriving? (crossed centreline, crossed right edge-line,crossed left edge-line, ran off the road, crashed withother vehicle, other consequence)’


Further, the questionnaire contained questions aboutthe last accident (time, place, type and severity, etc.) Inaddition, there were questions about background fac-tors of the driver. For classifying drivers into responsi-ble (guilty) and not responsible (innocent) party of thereported accident, the questionnaire contained a ques-tion about who was the responsible party, as judged bythe insurance company. The questionnaire was anony-mous, and it contained no reference number or otherinformation enabling anyone to identify therespondent.

The insurance companies distributed the question-naires, together with pre-paid postage return envelopesaddressed to the Institute of Transport Economics.

2.3. Returned questionnaires

Of the 29 600 questionnaires sent, about 9200 werereturned. This implies a response rate of about 31%.This low response rate is partly explained by the inclu-sion in the original sample of several irrelevant cases(e.g. non-traffic damage or damage to parked cars), aproblem which was due to limitations in the insurancecompanies’ coding of accident reports. Car ownershaving had such accidents were asked in the accompa-nying letter not to return the questionnaire. Since thenumber of irrelevant cases in the original sample is notknown, it is not possible to estimate the true responserate.

Accidents with shared responsibility as well as acci-dents where accident type was not specified, were ex-cluded from the data analyses. The remaining sampleconsisted of 6161 reports, of which 3239 were fromguilty drivers and 2922 from innocent drivers.

2.4. Responsible 6ersus not responsible dri6er groups

The incidence of sleep or fatigue as a causal factor inthe last reported accident is a meaningful variable forthe responsible driver group only, and this relationshipwas therefore analysed only for this subgroup. Theremaining questions about sleep during driving wererelevant to all drivers. It would, however, have beendesirable to have such data for a representative sampleof drivers, to be able to make inferences to the generaldriver population. Our sample was selected on the basisof having reported accidents, and therefore accident-prone drivers can be assumed to be over-represented. Itcan be argued, however, that drivers who get involvedin accidents for which they are not at fault, are moresimilar to the general driving population than are thegroup of at-fault drivers (Thorpe, 1967; Carr, 1969;Haight, 1973). It was therefore decided to analyse theprevalence of driver sleep for not-at-fault drivers only.However, due to temporal and spatial variations inrisk, drivers who drive more in risky traffic environ-

ments (darkness, night, rush hours, etc.), tend to beover-represented among not-at-fault drivers as well. Inaddition, even though judged by the insurance com-pany not to be at fault, these drivers may on theaverage be less able to prevent negative consequences ofother drivers’ acts compared to the general driver popu-lation. Thus, neither of the two subgroups in our driversample can be considered to be quite representative ofthe driver population at large. This should be kept inmind when interpreting the present results.

3. Results

3.1. Accidents and sleep/fatigue

Among the 3239 responsible drivers 125 (3.9%), re-ported that fatigue or sleep contributed to the accident.In Fig. 1 it is shown how the proportion of accidentsrelated to fatigue or sleep varies with type of accident.Most notably, the highest proportion of sleeping orfatigued drivers is found for running-off-the-roadaccidents.

To find out how the proportion of accidents wherefatigue or sleep contribute varies with characteristics ofthe accident, the traffic situation and the driver, wecarried out a logistic regression analysis with severalbackground factors from the questionnaire as indepen-dent variables. The dependent variable was whether ornot fatigue or sleep had contributed to the accident.

One of the independent variables of interest was thedistance driven at the moment of the crash, assumingthat the probability of sleep involvement increases withtime at the wheel, and consequently with trip length.For the purpose of investigating the relationship of triplength to probability of sleep involvement, the driversample was subdivided by trip length into six approxi-mately equally sized groups. Then the proportion ofsleep-related crashes was computed for each group andplotted against the average trip length for the groups,as shown in Fig. 2. Since the relationship is clearlynon-linear and appears to be described fairly well by alogarithmic function, it was decided to perform a loga-rithmic transformation of the trip length variable be-fore including it in the logistic regression model. Itshould be noted that the relationship shown in Fig. 2does not necessarily reflect the effect of trip length onthe risk of a sleep-related crash as such, but only on theproportion of crashes caused by sleep or fatigue. Anincrease in the proportion of sleep-related crashes maybe due to either increased risk of such crashes, de-creased risk of other types of crashes, or both.

The regression model shown in Table 1 shows thestrongest predictor variable to be the hour of theaccident. The odds of fatigue or sleep being involvedincreases by a factor of about 6 if the accident happensbetween midnight and 06:00 h.


Concerning factors not retained in the model, it isnotable that neither age, gender nor annual drivingdistance yielded any additional significant effect. As togender, however, a bi-variate comparison showed malesto have a significantly higher proportion of sleep-re-lated accidents than females. In the logistic regressionmodel, the effect of gender disappeared when speedlimit was entered into the equation. Since speed limithas higher face validity than gender as a predictor ofsleeping while driving, it seems more correct to includethe former variable in the regression model. And theeffect of age was captured by the variable years withdriver’s licence, which showed a stronger relationshipwith sleep involvement and was therefore included inthe model instead of age.

The bi-variate relationships with sleeping/fatigue foreach of the variables in the logistic regression model areshown in Table 2. We see that fatigue or sleep play apart in as much as 18.6% of the accidents betweenmidnight and 06:00 h, in 8.3% of ran-off-the-roadaccidents, and in 7.3% of personal injury accidents.This illustrates clearly the over-representation of fatigueor sleep as causal factors in these accidents. We also seethat the percentage of sleep-related accidents increaseswith the length of the trip, amounting to 8.1% for tripsover 150 km.

It should be noted that the variable car ownership isnot significant when considered in isolation, eventhough clearly significant in the regression model, and‘frequency of driving…’ is only marginally significant.This can be explained by the correlation between thesetwo variables, which implies that drivers who weredriving their own car during the accident, are more

likely to be driving that car daily, than drivers whowere not driving their own car. And since these twovariables bear opposite relationships to sleep or fatigue(driving daily reduces the odds, whereas driving one’sown car increases the odds of sleep/fatigue as causalfactor), they are not significant when entered separatelyin the model, but only when both are included.

3.2. Pre6alence of falling asleep while dri6ing

Of the total sample of not responsible drivers, 27%reported that they had fallen asleep some time whiledriving, as shown in Table 3. For 8.3% this had oc-curred during the last 12 months. It can further be seenthat the proportion of drivers who have fallen asleep isabout 2.5 times higher for male than for female drivers.

3.3. Consequences of falling asleep while dri6ing

The most frequent consequence of falling asleepwhile driving is crossing of the edge-line to the right,which occurs in more than 40% of the occasions ofsleep, as shown in Fig. 3.

Crossing the centre-line is the second most frequentconsequence specified, but much less frequent than thefirst one. A total of 4% of the incidents result in anaccident, either running off the road or collision withother vehicle.

3.4. Relationship between in6ol6ement in sleep-relatedcrashes and pre6ious history of dri6er sleep

It was investigated whether the previous prevalence

Fig. 1. Accidents related to sleep or fatigue, by accident type (%).


Fig. 2. Relationship between trip length and proportion of accidents related to sleep or fatigue. The data value labels indicate number ofsleep-related/total crashes represented by each data point.

of falling asleep when driving differed between at-faultdrivers involved in sleep-related crashes and driversinvolved in other crashes. Among the 125 drivers atfault for a sleep-related crash, 26 drivers (20.8%) re-ported to have fallen asleep while driving some time inthe past. This was not significantly different from thecorresponding proportions among drivers at fault forcrashes with other causes (28.3%) or among the not-at-fault drivers (27.0%). It would have been informativealso to investigate differences between those drivergroups regarding the consequences of falling asleepwhile driving. However, since as few as 26 of the driversinvolved in a sleep-related crash reported previousepisodes of driver sleep, a breakdown into smallercategories was not considered meaningful.

3.5. Characteristics associated with falling asleepduring dri6ing

Logistic regression analyses of data for the not re-sponsible drivers were carried out to find out whetherany of the background variables included in the ques-tionnaire could predict the probability of having fallenasleep whilst driving. Two analyses were done, with‘falling asleep some time’ and ‘falling asleep last 12months’ as respective dependent variables. Table 4shows the regression models including independentvariables that were statistically significant in one orboth of the analyses.

It is clear that the strongest predictors of the proba-bility of falling asleep during driving are annual drivingdistance and being a male driver. This implies that thesex difference in proportion of drivers falling asleep,

which was shown in Table 3, remains even when correc-tion is made for the difference between male and femaledrivers in length of driving.

The age effect implies that the odds of having fallenasleep during driving, is reduced by about 2% per year.The effect of education means that higher education isassociated with higher odds of having fallen asleep.

4. Discussion and conclusions

The present study provides two kinds of evidencethat automobile drivers who are drowsy or who actu-

Table 1Logistic regression model for predicting the involvement of sleep orfatigue as contributing factor in crashesa

Independent variable Statistical sig-Odds rationificance

6.34 PB0.0001Accident happened between mid-night and 06:00 h

3.85 P=0.007Running-off-the-road accidentDry and bare road (vs. wet or 2.60 PB0.0001

slippery)P=0.0091.73Speed limit 60 km/h or higher

1.89 P=0.045Personal injury accident1.85 P=0.011Driving one’s own car

P=0.0201.68Not driving the car dailyP=0.0031.57Education (3 levels)

Distance driven since start of the 1.35 PB0.0001trip [ln(km)]

P=0.0100.98Years with driver’s licence

a Responsible drivers only (n=3061).


Table 2Number and percentage of crashes related to sleep or fatigue, for different levels of the significant variables in the logistic regression model

Levels Sleep-related crashes Statistical significanceVariables

PercentageNumber

8.338Running off the road x2=26.6; PB0.001Accident type87 3.1Other accident

13 7.3Accident severity x2=4.2; P=0.042PI accident112 3.6Property damage

x2=31.5; PB0.00118.611Midnight–06:00 hTime interval112 3.606:00 h–midnight

69 6.3Speed limit x2=21.6; PB0.00160 km/h or higher2.85350 km/h or lower

90 4.8Road condition Dry x2=10.2; P=0.0012.635Wet, snowy or icy

8.1 x2=6.4; P=0.011Kilometres since start of trip 150 km or more 12110 3.6Less than 150 km

67 x2=8.0; P=0.0055.1College/universityLevel of education57 3.1Below college level

6.3295 years or less x2=6.5; P=0.011Years with driver’s license3.7More than 5 years 96

5.4 x2=3.8; P=0.051Frequency of driving the actual car Less than every day 3688 3.7Daily

91 x2=1.68; n.s.Driving own car 4.4Car ownership3.4Driving other car 34

ally fall asleep while driving constitute a very significantroad safety hazard. First, the results regarding sleep orfatigue involvement in accidents showed that these fac-tors were causally related to about one in every 14personal injury accidents, and about one in 25 accidentsin total. Second, the results on the prevalence of fallingasleep among drivers in general showed that one in 12drivers had actually fallen asleep during the last 12months, and that about 4% of these events resulted inan accident. Applied on the general driver populationthe latter finding suggests that about one in 250 Norwe-gian drivers per year is involved in an accident becauseof falling asleep.

Concerning age and sex differences in the prevalenceof falling asleep while driving, our results largely repli-cate the findings for US drivers reported by Garder andAlexander (1995). As expected, however, the propor-tion of drivers who reported that they had fallen asleepis generally lower in our study, a finding probablyrelated to differences in road geometry, design andenvironment as well as differences in average annualdriving distance between the two countries. There mayalso be differences in lifestyle (sleep habits, work sched-ules, etc.) between the two countries, which may con-tribute to explaining the observed differences.

More male than female drivers reported having fallenasleep, even when correction was made for driving

distance. It may therefore appear somewhat surprisingthat sex did not come out as a significant factor in themultivariate analysis of sleep involvement in accidents.Closer scrutiny of the data showed that there actuallywas a significant bi-variate relationship in the expecteddirection between sex and the proportion of accidentscaused by fatigue or sleep. In the multivariate analysis,however, the effect of sex was considerably reduced(and was far from significant) when the variable ‘speedlimit’ was entered in the logistic regression equation.This indicates that the higher relative involvement ofmales in sleep-related accidents is partly explained bytheir driving relatively more on high-speed roads.

The results further showed that the percentage ofsleep-related accidents was higher for personal injuryaccidents than for less serious accidents. This is most

Table 3Drivers who have fallen asleep during drivinga

Have fallen asleepHave fallen asleepduring driving lastsome time during12 monthsdriving

Males (n=1889) 34.5 10.4Females (n=995) 13.0 4.3

27.1 8.3Total (n=2884)

a Percent of all not-responsible drivers.


Fig. 3. Consequences of falling asleep during driving (n=1061).

the sole explanation of the relationship between driversleep and road departure accidents. Irrespective of theexplanation, the strong association between driver sleepand running off the road explains why shoulder rum-bling strips have been shown to reduce the frequency ofaccidents (Garder and Alexander, 1995).

As expected, the relative involvement of sleep ordrowsiness as contributing factor to accidents increaseswith trip length. This may suggest that more emphasisshould be put on long compared to short trips whendesigning driver-targeted measures for preventing sleep-related accidents. On the other hand, 90% of thecrashes in the present study occurred to drivers whohad been driving 150 km or less, which means that therisk of a sleep-related crash is considerably increasedeven after driving for only a couple of hours.

It is also not surprising that driving one’s own car isassociated with increased relative involvement of sleepand drowsiness, since monotony and boredom proba-bly is more likely to occur when driving a car that oneis used to driving. The level of arousal can be assumedto be higher when driving an unknown car. On thisbackground, it is not easy to explain why ‘not drivingthe car daily’ is associated with higher odds of sleepinvolvement. It may be that regular and less regulardrivers drive in different road and traffic conditions,regarding dimensions that are not captured by thepresent analysis.

The relationship between educational level and rela-tive sleep involvement in accidents does not have anyobvious explanation. It is notable, though, that Mc-Cartt et al. (1996) also reported higher levels of educa-tion to be related to increased frequency of drowsydriving. Concerning traffic accident risk generally, pre-vious research has indicated that high-educated drivershave a lower-than-average risk (Murray, 1998); thepresent results, however, show that this apparently doesnot hold true for sleep-related crashes.

Some sources of bias may limit the possibility ofgeneralising from the present findings, at least as far asthe absolute magnitude of the effects is concerned. Forexample, the sample of crashes reported to the in-

likely explained by the fact that sleep-related accidentstend to occur under conditions where the speed can beassumed to be higher than average, which implies moreserious consequences of an accident. The percentage ofsleep-related accidents was higher on roads with highspeed limits, on dry roads as compared to wet or icyroads, and particularly during night driving.

The finding that deviations to the right side of thelane are much more likely than to the left, is probablyrelated to the typical transverse section characteristicsof most roads. The elevation is reduced towards theverge, causing vehicles on straight sections to drift offto the right if no steering-wheel corrections are made.This result is consistent with the over-representation ofsleep and fatigue in road departure accidents. The latterfinding could alternatively be explained by the fact thatsleep during driving is more frequent among young orinexperienced drivers, who also constitute a group ofdrivers particularly prone to be involved in road depar-ture accidents (Sagberg, 1998). However, since ‘yearswith drivers licence’ and ‘road departure accident’ makeindependent contributions to the odds of sleep as con-tributing factor, driving experience is not likely to be

Table 4Logistic regression models for the relationship between background factors and having fallen asleep during driving, for not-responsible drivers(n=2828)

Independent variable Have fallen asleep during driving some time or Have fallen asleep during driving last 12 monthsother

Odds ratio Statistical significance Odds ratio Statistical significance

PB0.00011.14Annual driving distance (km× PB0.0001 1.1710000)

PB0.0001 PB0.00013.49Male driver 2.410.98 PB0.0001Age 0.97 PB0.00011.32Education (3 levels) P=0.0001 1.05 n.s.


surance companies does probably not represent a trulyrandom sample of all crashes, and the rate of sleep orfatigue involvement may differ between reported andnot reported crashes. Further, the rather low responserate in the present study might possibly be invoked asan argument against the general validity of the findings.We do, however, not consider it very likely that therelative involvement of sleep or drowsiness in acci-dents—or the relationship of this involvement to back-ground factors—varies systematically betweenresponders and non-responders in a study like this. Adefinite possibility is that drivers who were at fault forthe accident could be more reluctant to answer. Such abias, however, would be a problem only if the dropoutrate among the drivers at fault differed systematicallybetween sleep-related accidents and other accidents,which seems hard to imagine. And concerning theprevalence of falling asleep, the analyses were basedonly on not-at-fault drivers. For this group it is difficultto find reasons for systematic differences in prevalenceof sleep during driving between responders and non-re-sponders. The possibility of underreporting of certaincauses of accidents among those who responded is notvery likely either, since the questionnaire was anony-mous. Biased results in such a study can, however,never be completely ruled out. For example, the re-sponse rate may be correlated with age, gender, orother social or demographic characteristics. The abso-lute size of the effects reported here may thereforepossibly differ somewhat from what we could be ex-pected in a truly representative driver sample. Since thequestionnaires were addressed to the car owners andwere supposed to be filled in by the person who wasdriving when the accident occurred, the distribution ofbackground data for non-responders is not known. Itwas therefore not possible to check whether the respon-dents differed from the non-respondents.

Another source of bias—resulting in underestimationof sleep or fatigue involvement in crashes—is related tothe fact that some of the most severe crashes areobviously not included, viz., driver fatalities, which arehigher for sleep-related crashes. The effect of this bias ishowever rather small. Assuming that the involvementof sleep and fatigue in fatal crashes is about three timesthat of personal injury crashes generally, as can beinferred from the studies cited in the introduction, thisbias can be estimated to imply less than 1% underesti-mation of the involvement.

Although drowsiness and sleep thus seem to con-tribute to a substantial number of traffic accidents, itmay be difficult to find effective measures to preventsleep-related accidents. It is of course important to tryand make drivers more aware of the problem of sleep-related accidents. Increased publicity about the problemmay possibly influence some drivers to refrain fromstarting on a trip—at least without a co-pilot—with

insufficient sleep in advance, during the night or atother times when one ordinarily feels drowsy.

It is notable, however, that in spite of sleep-relatedaccidents being over-represented at night, in terms ofabsolute numbers the majority of these accidents (likemost accident types) in the present data material hap-pened during daytime. An interesting question deserv-ing further attention is to what extent sleep-relatedcrashes during the day may occur without previoussleep deprivation. Apart from daytime hypersomno-lence due to medical conditions like narcolepsy andchronic sleep apnoe syndrome, which have been shownto be related to accidents risk (Taoka, 1998), there arereasons to believe that even normally rested driversunder certain conditions may experience sleep-relatedstates, possibly without a preceding phase of subjec-tively experienced fatigue. An example is the phe-nomenon of so-called highway hypnosis, which hasbeen defined as ‘a tendency to become drowsy and tofall asleep when driving an automobile’ (Williams andShor, 1970). It has been pointed out that this sleep-likeor trance-like state can occur with the driver sitting inan upright position and staring ahead, and that it isoften accompanied by ‘hallucinatory experiences anddistortions of thought and judgement’ (Williams, 1963).These experiences seem to be examples of so-calledhypnagogic imagery (Schachter, 1976), which refers tothe mental contents characterising the transition statefrom wakefulness to sleep.

Wertheim (1991) has suggested that highway hypno-sis is the result of the driving task becoming increas-ingly automatic due to a highly predictable roadenvironment, with a reduced level of arousal as aconsequence. What seems to be essentially the samephenomenon has been described by Kerr (1991) as thedriving without awareness mode (DWAM). This refersto the rather common experience of having been driv-ing a considerable distance without remembering any-thing from the trip.

Conceivably, in addition to a monotonous road envi-ronment, the high degree of comfort provided by mod-ern cars—comfortable seats, easy manoeuvring, and alow level of noise and vibration—is likely to produce astate of low arousal that is conducive to the develop-ment of the mentioned sleep-related states ofconsciousness.

The transition from wakefulness to sleep may bemore or less insidious and difficult to perceive by theperson, because the early precursors are not recognisedas danger signals. A paradox in this is that the moredrowsy one is, and consequently the more important itis to take action, the less the ability to react to the signsof drowsiness. There may be no natural warnings be-fore actual sleep sets in and the eyes close, at whichstage the process is not likely to reverse without exter-nal stimulation. For example, the difference between


ordinary daydreaming and the more bizarre andstrange images and thoughts characterising the hypna-gogic state is not very sharp. But if drivers knew thatstrange thoughts and images might be the first signs ofthe transition to sleep, it could possibly enable them toavoid a dangerous situation.

Correct recognition of danger signals indicating riskof falling asleep is, however, no guarantee that thedriver will react appropriately to the signals. Althoughit could be possible to teach drivers to become moreaware of the early signs of sleepiness, it is probably verydifficult to make them take the consequence of detect-ing these signs and take a break from driving. A strongmotivation to reach their destination in time will veryoften make drivers try various ways of combating thetiredness while continuing the trip—a combat that thedriver is very likely to lose.

In spite of this rather pessimistic scenario, everyeffort should of course be taken to inform drivers aboutthe early signs that may indicate a risk of falling asleep.But possibly there may be more to be gained by mea-sures applied to the road and the vehicles for waking upsleeping or drowsy drivers. As mentioned above, theprovision of rumbling strips along the edge-line seemsto be a very promising measure. In some places this isnow being applied also to the centrelines. The effects ofcentreline rumble strips have as far as we know notbeen evaluated yet. In addition to waking up drowsydrivers, these lines will have the effect of alerting driverswho deviate from the lane due to other kinds of inat-tention, caused by various kinds of internal or externaldistractions. The effectiveness of rumble strips probablydepends on their width, which determines the durationof the sound when the line is crossed. It is thereforeimportant to find the optimal width, which is probablyrelated to the road geometry. There is further a needfor continued research to find the optimal profile ofrumble strips to provide a sufficiently strong auditoryand/or tactile stimulus to wake up sleeping drivers. Foredge-lines, the effectiveness of rumbling strips will alsodepend on the availability of a shoulder of sufficientwidth for drivers to correct their bearing in time whenbeing awakened by the rumbling.

Concerning vehicles, the development of driver moni-toring systems that reliably can wake up drowsy driverswithout providing too many false alarms, seem still tobe some time away. Even if the systems will work asexpected from a technical point of view, we still do notknow enough about their possible effects on safety.There is a definite possibility that such systems mayresult in driver behaviour adaptation, by making tireddrivers continue driving for a longer time than theywould have done without this support. Such an effectmight easily offset any safety increase of these systems.

Hopefully, more knowledge about the conditionspreceding the occurrence of accidents related to fatigue

or sleep could contribute to the development and imple-mentation of adequate precautions. Questions needingto be answered include, for example, what proportionof drivers involved in sleep-related accidents have expe-rienced some of the known precursors? To what extentare such accidents systematically related to prolongedperiods without sleep? And are there other events thanthe known precursors of sleep that regularly precedesleep-related accidents?

In summary, then, it seems that although sleep anddrowsiness cause a high number of accidents, the onlymeans available at the present time to reduce theiroccurrence seem to be rumble strips and driver infor-mation, apart from measures to reduce the conse-quences of any vehicle accidents. If effective driverinformation measures are available, they should partic-ularly emphasise the risks associated with driving acomfortable car on dry, rural high-speed roads, anddriving under conditions of low sensory input or lowmental workload in general.

Acknowledgements

This study was supported by grants from TelenorMobil AS and The Ministry of Transport andCommunications.

References

Amundsen, F.H., Christensen, P., 1986. Tiltak mot møteulykker-analyse av politiets saksdokumenter og forslag til mottiltak(Countermeasures regarding head-on vehicle accidents—an anal-ysis of accident records. Norwegian, with summary in English).TØI Project Report. Institute of Transport Economics, Oslo.

Brown, I.D., 1994. Driver fatigue. Hum. Factors 36 (2), 298–314.Carr, B.R., 1969. A statistical analysis of rural Ontario traffic acci-

dents using induced exposure data. Accid. Anal. Prev. 1, 343–357.Fell, 1994. Safety update: Problem Definition and Counter measure.

Summary: Fatigue. New South Wales Road Safety Bureau, RUSNo. 5.

Garder, P., Alexander, J., 1995. Fatigue related accidents and contin-uous shoulder rumble strips (CSRS). Transportation ResearchBoard 74th Annual Meeting.

Haight, F., 1973. Induced exposure. Accid. Anal. Prev. 5, 111–126.Horne, J.A., Reyner, L.A., 1995. Sleep-related vehicle accidents. Br.

Med. J. 310 (6979), 565–567.Kerr, J.S., 1991. Driving without attention mode (DWAM): a formal-

isation of inattentive states in driving. In: Gale, A.G., et al. (Eds.),Vision in Vehicles, vol. III. North-Holland, Amsterdam, pp.473–479.

Knipling, R.R., Wang, J.-S., 1995. Revised estimates of the USdrowsy driver crash problem size based on general estimatessystem case reviews. Association for the Advancement of Auto-motive Medicine, 39th Annual Proceedings.

Laube, I., Seeger, R., Russi, E.W., Bloch, K.E., 1998. Accidentsrelated to sleepiness: review of medical causes and prevention withspecial reference to Switzerland. Schweiz. Med. Wochenschr. 128(40), 1487–1499.


Lyznicki, J.M., Doege, T.C., Davis, R.M., Williams, M.A., 1998.Sleepiness, driving, and motor vehicle crashes. J. Am. Med.Assoc. 279 (23), 1908–1913.

Maycock, G., 1997. Sleepiness and driving: the experience of UK cardrivers. Accid. Anal. Prev. 29 (4), 453–462.

McCartt, A.T., Ribner, S.A., Pack, A.I., Hammer, M.C., 1996. Thescope and nature of the drowsy driving problem in New YorkState. Accid. Anal. Prev. 28 (4), 511–517.

Murray, A, ., 1998. The home and school background of young driversinvolved in traffic accidents. Accid. Anal. Prev. 30 (2), 169–182.

Nairn, R.J., 1987. The role of fatigue in rural accidents. ReportGR/87/6. Road Traffic Authority, Hawthorne, Victoria.

Pack, A.I., Pack, A.M., Rodgman, E., Cucchiara, A., Dinges, D.F.,Schwab, C.W., 1995. Characteristics of crashes attributed to thedriver having fallen asleep. Accid. Anal. Prev. 27 (6), 769–775.

Sagberg, F., 1998. Month-by-month changes in accident risk amongnovice drivers. 24th International Conference of Applied Pscyhol-ogy, San Francisco.

Sagberg, F., 1999. Accident risk of car drivers during mobile tele-phone use. Int. J. Veh. Des. (in press).

Schachter, D.L., 1976. The hypnagogic state: a critical review of theliterature. Psychol. Bull. 83 (3), 452–481.

Storie, V.J., 1984. Involvement of goods vehicles and public servicevehicles in motorway accidents. Report 113. Transport and RoadResearch Laboratory.

Summala, H., Mikkola, T., 1994. Fatal accidents among car andtruck drivers: effects of fatigue, age, and alcohol consumption.Hum. Factors 36 (2), 315–326.

Taoka, 1998. Asleep at the wheel: a review of research related todriver fatigue. Transp. Quarterly 52 (1), 85–92.

Thorpe, J.D., 1967. Calculating relative involvement rates in acci-dents without determining exposure. Traffic Saf. Res. Rev. 11 (1),3–8.

Wertheim, A.H., 1991. Highway hypnosis: A theoretical analysis. In:Gale, A.G., et al. (Eds.), Vision in Vehicles, vol. III. North-Hol-land, Amsterdam, pp. 467–472.

Williams, G.W., 1963. Highway hypnosis: an hypothesis. Int. J. Clin.Exp. Hypn. 11 (3), 143–151.

Williams, G.W., Shor, R.E., 1970. An historical note on highwayhypnosis. Accid. Anal. Prev. 2, 223–225.

.

Documents

Road accidents caused by drivers falling asleep