AD

Ma

b

a

ARRA

KRVADDPN

1

fwwc2dba(iRtp2le

s

0h

Accident Analysis and Prevention 58 (2013) 88– 96

Contents lists available at ScienceDirect

Accident Analysis and Prevention

j ourna l h om epage: www.elsev ier .com/ locate /aap

comparison of the Driving Anger Scale and the Propensity for Angryriving Scale

ark J.M. Sullmana,∗, Amanda N. Stephensb

School of Engineering, Cranfield University, UKCollege of Arts, Psychology, Victoria University, Australia

r t i c l e i n f o

rticle history:eceived 8 April 2013eceived in revised form 3 May 2013ccepted 5 May 2013

eywords:isky drivingiolations

a b s t r a c t

The present study investigated the factor structures of the 14-item version of the DAS (Driving AngerScale) and the Propensity for Angry Driving Scale (PADS) using a sample of New Zealand drivers drawnfrom the general population. The two scales were also investigated with regards to their relationshipswith general trait anger, risky driving behaviour, along with crash involvement and a variety of crash-related conditions. Confirmatory Factor Analysis supported both scales as unidimensional, although thePADS was reduced from a 19-item to an 18-item scale. Both the PADS and DAS were significantly relatedto trait anger, risky driving behaviour and near-misses. However, once the influence of the demographic

ngry drivingriver angerASADSew Zealand

variables and trait anger had been partialled out, the addition of the PADS and DAS made a significantcontribution to predicting violations, but it was only the PADS which was significant. In contrast, after thedemographic variables and trait anger had been partialled out, the addition of the DAS and PADS againmade a significant contribution to the prediction of near-misses, but this time it was only the DAS whichmade a significant contribution. The present study clearly shows that both scales are robust measures,measuring similar, but slightly different aspects of driving anger.

© 2013 Elsevier Ltd. All rights reserved.

. Introduction

Driving evokes a wide range of emotions in people, including joy,rustration, anxiety, fear and anger. Anger is one of the emotionshich has become increasingly researched over the last ten years,ith one of the reasons for this being the fact that it is relatively

ommon to experience anger while driving (Deffenbacher et al.,002b). Furthermore, a number of studies have found that angryrivers engage more often in aggressive and dangerous drivingehaviours (Dahlen et al., 2005; Deffenbacher et al., 1994; Stephensnd Groeger, 2011; Sullman et al., 2013). In fact, Dahlen and Ragan2004) went so far as to state that driving anger is one of the mostnfluential predictors of aggressive and risky driving behaviour.esearch has also found driving anger to be significantly relatedo near-misses (Underwood et al., 1999), slower reaction times tootential hazards (Stephens and Groeger, 2011; Stephens et al.,013) and crash related conditions, such as loss of concentration,

osing control of the vehicle and crash involvement (Deffenbacher

t al., 2001, 2003; Sullman et al., 2007).

There are a number of ways in which driving anger can be mea-ured, with two such scales being the Driving Anger Scale (DAS;

∗ Corresponding author. Tel.: +44 7771892297.E-mail address: M.Sullman@cranfield.ac.uk (M.J.M. Sullman).

001-4575/$ – see front matter © 2013 Elsevier Ltd. All rights reserved.ttp://dx.doi.org/10.1016/j.aap.2013.05.002

Deffenbacher et al., 1994) and the Propensity for Angry DrivingScale (PADS; DePasquale et al., 2001). In addition there are twoversions of the DAS, a short fourteen item unidimensional measureand a longer thirty three item multidimensional measure. The shortversion of the DAS presents fourteen different situations and asksthe responding driver to report the degree of anger that each situ-ation makes them feel. In contrast, the PADS, which is a nineteenitem scale, presents situations that are likely to evoke anger andthen asks the respondent to indicate how they would respond byselecting one of four potential responses. These range from mildreactions, such as slowing down, to more extreme responses, suchas ramming the other car.

The DAS and PADS have both been found to have good psy-chometric properties. Research has shown the DAS to have goodinternal reliability, with alpha coefficients ranging from 0.80 to0.92 (Deffenbacher et al., 1994, 2002a). The alpha coefficients forthe PADS have also been good, ranging from 0.85 to 0.89 (Dahlenand Ragan, 2004; DePasquale et al., 2001). Convergent validity anddiscriminant validity for the PADS has been displayed through rela-tionships with trait anger and hostility (DePasquale et al., 2001),while the validity of the DAS has also been shown through cor-

relations with the Trait Anger Scale (Deffenbacher et al., 1994;Villieux and Delhomme, 2007). Moreover, the test–retest reliabil-ity of both scales has also been shown to be high. The PADShas been found to have four-week test–retest reliability of 0.91

ent An

(t

taivaehtg1

vaeenp(sr

r2cDaeftcdrrcFna

eMDFfiavrAb2st1ath1bwdsoot

M.J.M. Sullman, A.N. Stephens / Accid

DePasquale et al., 2001), while the DAS has been shown to haveen-week test–retest reliability of 0.84 (Deffenbacher et al., 2002a).

As would be expected, both scales seem to have similar rela-ionships with descriptive variables (e.g., age and gender), as wells driving behaviours and crash related conditions. For examplen the two studies which have used the original nineteen itemersion of the PADS neither reported any age differences (Dahlennd Ragan, 2004; DePasquale et al., 2001) and only DePasqualet al. (2001) reported a gender difference. Although some researchas found females score more highly on the shortened version ofhe DAS (Dahlen and Ragan, 2004) most research has found noender differences (e.g. Dahlen et al., 2005; Deffenbacher et al.,994).

Also in contrast to the research using the multidimensionalersion of the DAS (e.g. Lajunen et al., 1998; Sullman, 2006) noge differences were reported in the studies using the short-ned version of the scale (Dahlen and Ragan, 2004; Dahlent al., 2005; Deffenbacher et al., 1994). However, it should beoted that the studies using the short DAS have all used sam-les with very narrow age ranges, whereas the two studiesmentioned above) using the longer version of the scale usedamples from the general population with much broader ageanges.

The DAS has also been found to be related to aggressive andisky driving behaviour (Dahlen et al., 2005; Deffenbacher et al.,001, 2002b) and other crash related conditions, such as; loss ofoncentration, loss of control and near-misses (Dahlen et al., 2005;effenbacher et al., 2001). In addition, although one study found

relationship between the DAS and major crashes (Deffenbachert al., 2002b), this has not been a common finding. Similarly con-using findings have arisen from the PADS. In an American studyhe PADS was found to be correlated with both major and minorrashes (Dahlen and Ragan, 2004) and other crash-related con-itions, such as loss of control and receiving tickets for violatingoad rules (Dahlen and Ragan, 2004). In contrast, more recentesearch using an Australian version of the PADS found no signifi-ant relationship with crash involvement (Leal and Pachana, 2008).urthermore, like the DAS, the PADS has also been found to be sig-ificantly related to aggressive and risky driving behaviour (Dahlennd Ragan, 2004).

Although the PADS has been validated five times (DePasqualet al., 2001; Dahlen and Ragan, 2004; Leal and Pachana, 2008, 2009;axwell et al., 2005), only two of these studies have used both theAS and the PADS (Dahlen and Ragan, 2004; Maxwell et al., 2005).urthermore, one of these two studies (Maxwell et al., 2005) modi-ed the PADS by dropping four of the nineteen items and also used

twenty one item version of the DAS, rather than the fourteen itemersion. Thus, the findings generated by that study were not compa-able. Moreover, as with the British study (Maxwell et al., 2005), theustralian studies also modified the PADS by dropping four items,ased upon the results of a factor analysis (Leal and Pachana, 2008,009). The only remaining study to compare the two scales reliedolely upon psychology undergraduates as participants. This meanshat the participants were from a very restricted age range (median9) and were mainly female (75%), calling into question the gener-lisability of these findings. This concern is highlighted further byhe fact that in samples from the general population driving angeras been found to be related to both gender and age (Lajunen et al.,998; Sullman, 2006). Therefore, it seems important that the PADSe investigated in a broader sample of drivers. The present studyas the first to not only investigate the PADS in a broader sample ofrivers, but to use the PADS on drivers in New Zealand. The present

tudy also compared the PADS and DAS to test whether the previ-usly found relationships could be generalised to a broader samplef drivers. The scales were also subject to CFA in order to confirmheir factor structure.

alysis and Prevention 58 (2013) 88– 96 89

2. Methods

2.1. Participants

A total of 213 licenced drivers recruited from three cities inNew Zealand participated in the study. Participants (males = 92;43.7%) ranged in age from 17 to 80 years (M = 43.96; SD = 15.95),had been licenced between 1 and 65 years (M = 25.73; SD = 14.68)and reported driving between approximately 100 and 78,000 km(M = 16,242; SD = 10,603) per annum.

2.2. Materials

2.2.1. Propensity for Angry Driving Scale (PADS)The PADS (DePasquale et al., 2001) contains nineteen short

vignettes describing potentially anger-inducing situations driversmay encounter. Participants are asked to read each vignette andthen respond by circling the most appropriate of four responses.These responses range on a continuum from mildly aggressive toextremely aggressive. For example:

“You are driving on a city street. Without warning, a pedestriansuddenly runs in front of your car nearly causing you to hit him/her.How do you respond?”(a) Do nothing except feel grateful no-one was injured(b) Actually stop your car and get out to yell at the pedestrian for

being careless and stupid(c) Yell at the pedestrian out your window telling them to watch

where they are going(d) Curse loudly at the pedestrian out your window telling them

the next time you are not going to stop

Responses to the 19-items are rescored according to the proce-dure outlined by DePasquale et al. (2001) where item responses arereplaced by weighted mean response values ranging from 1 (repre-senting mild responses) to 7 (severe aggressive responses). Minorwording adaptations were made to make the terms suitable for aNew Zealand sample. For example, the measurement system waschanged to metric, reference to the left and right side of the roadswere adjusted and a number of American terms changed (e.g. carpark was substituted for parking lot).

2.2.2. Driving Anger Scale (short)Driving Anger was also measured using the short Driving Anger

Scale (DAS; Deffenbacher et al., 1994). The DAS (short) contains 14-items depicting anger-provoking situations. For example, “Someonebeeps at you about your driving”. Participants are asked to imagineeach of the situations happening to them and to rate the amount ofanger evoked by each on a five-point scale (1 = not at all; 3 = someanger; 5 = very much anger). Responses are then tallied to form oneoverall driving anger score.

2.2.3. Trait Anger ScaleThe Trait Anger Scale (TAS; Spielberger, 1988,1999) is a 10-item

scale designed to measure trait propensities for anger. Partici-pants are presented with 10 statements and asked to indicate howapplicable each one generally is to them. For example, “I am ahot-headed person”. Responses are on a 4-point scale (1 = almostnever; 4 = almost always). Responses are scored to provide an over-all measure of propensity to become angered. The TAS exhibits goodinternal consistency with ̨ ranging from 0.81 to 0.91 (Spielberger,1988, 1999).

2.2.4. Driving violationsDriving violations were obtained using the eight violation items

from the Driving Behaviour Questionnaire (DBQ; Reason et al.,

9 ent An

1htR5r

2

d(hmlvts

amwe

2

toudm

2

PNptptahctlwn

TD

0 M.J.M. Sullman, A.N. Stephens / Accid

990). Participants are asked to rate how often in the past year theyave committed each of the eight different types of driving viola-ions; for example, “disregarded the speed limit on a residential road”.atings are made on a 6-point scale (0 = never; 3 = occasionally;

= all the time). The violation subscale of the DBQ has good internaleliability ( ̨ = 0.81; Parker et al., 1995, 2002).

.2.5. Driving Survey and Speed PreferencesParticipants also provided information on accidents and acci-

ent related history using questions taken from Deffenbacher et al.2000) Driving Survey. Six questions were provided which askedow often in the past three months participants had: been in ainor accident; been in a major accident; received a traffic ticket;

ost concentration while driving; a minor loss of control of theehicle while driving; nearly had an accident (near-misses). Par-icipants provide one value indicating the frequency of each. Theseix questions were analysed as individual variables.

Participants also reported the speed they would generally drivet on five different types of roads. These were: an open road; a busyain street; a road through a residential area; a motorway; and, ainding country road. The posted speed limits were provided in

ach instance, for example “the open road (100 km/h)”.

.2.6. Demographic questionnaireDemographic information such as age, gender, annual mileage,

ype of licence held and information regarding the vehicle mostften driven (e.g. motorbike, van or car, engine size, year of man-facturing) was provided. Participants also rated how often theyrove on congested roads (almost never, biannually, bimonthly,onthly, weekly, everyday).

.3. Procedure

The study was approved by the University ethics committee.articipants were selected using updated electoral rolls from threeew Zealand cities. Random number generators determined whichages of each roll to target, as well as five addresses from each pageo send questionnaire packs to. Prospective participants were thenosted a questionnaire pack containing a cover letter, the ques-ionnaire and a reply paid envelope. Although the cover letter wasddressed to each randomly selected individual, the questionnairesad no identifiable features and therefore their responses wereompletely anonymous. Instructions were also provided outlining

hat participants were required to hold a valid New Zealand driver’sicence and to have driven at least once in the past six months. Those

ho did not meet these criteria were asked to pass the question-aire pack to an eligible household member whose birthday was

able 1escriptive variables.

Preferred speed Avera

Preferred speed on open road (100 km/h) 101.3Preferred speed on busy main street (50 km/h) 49.4Preferred speed in residential areas (50 km/h) 51.5Preferred speed on motorways (100 km/h) 102.9Preferred speed on winding country roads (100 km/h) 81.8Mean preferred driving speed 77.0

Frequency in congested traffic Percentage

Everyday 32.90

Weekly 31.90

Twice a month 8.00

Once a month 9.40

Twice a year 6.10

Almost never 11.70

alysis and Prevention 58 (2013) 88– 96

the closest to theirs. A total of 600 questionnaires were distributedand 225 returned, with a response rate of 37.5%.

Although a total of 225 drivers completed the questionnaires,twelve cases with 10% or more missing data from the two driv-ing scales were excluded from further analysis, resulting in a finalsample containing 213 participants.

Further missing data were handled in two ways. For the CFA,Maximum Likelihood (Robust) Estimations were calculated mean-ing missing values were not imputed, but parameter estimatesassumed complete data were available (Enders, 2001). For subse-quent analysis, cases with one missing data point were subjectedto trimmed mean (by gender) imputation. During the correlationaland regression analyses, two further cases were excluded becauseresponses for more than 8 of the 10 Trait Anger Scale items werenot provided. This will be reflected in lower degrees of freedom onthis measure. As the missing data constituted less than 5% of theoverall data and appeared random, these methods of data handlingare deemed satisfactory (Tabachnick and Fidell, 2007).

3. Results

3.1. Descriptive variables

Table 1 shows participants’ reported preferred speed across fivedifferent road scenarios: open road, busy main street, residentialareas, motorways and winding roads. The five speed questionswere standardised and combined to form a single measure of pre-ferred driving speed (Sullman et al., 2002). Overall, participantsreported preferring speeds that closely matched the posted speedlimits, suggesting a generally compliant sample. For example, theaverage preferred speed on an open road (posted limit 100 km/h)was 101 km/h. The national survey by the Ministry of Transport(2013) of 67 open road locations in New Zealand (between 2006and 2012) reported drivers have an average speed similar to this(95–96 km/h). For urban areas, drivers in the current study reportedan average preferred speed of 51 km/h (in a 50 km/h zone). Whilethe actual measured speed at 64 locations throughout New Zealandwas comparable (between 51 and 53 km/h; Ministry for Transport,2013).

Table 1 also shows how frequently participants experiencecongested traffic. Over half of the participants (64.8%) reportedexperiencing congestion on a daily or weekly basis.

3.2. Mean scores of the driving anger scales

Responses to the nineteen PADS items were re-scored accordingto the procedure outlined by DePasquale et al. (2001) and reportedabove in Section 2. Table 2 shows that average weighted scores

ge km/h (SD) Range

5 (8.77) 70–1204 (7.44) 20–705 (7.01) 15–1005 (6.98) 75–1200 (14.63) 75–1202 (7.23) 45–94

Cumulative

32.9064.8072.8081.2087.30

100.00

M.J.M. Sullman, A.N. Stephens / Accident Analysis and Prevention 58 (2013) 88– 96 91

Table 2Mean scores for PADS item responses re-calculated to the 7-point scale.

Item M (SD)

1 You are driving your car down a two-lane road. Without warning, another car pulls out in front of you from a parking lot. You had tobrake suddenly to avoid hitting it.

2.65 (1.06)

5 You are driving your vehicle in a traffic jam in the far left hand lane. Out of nowhere, a car comes up from behind on the shoulder andattempts to squeeze in front of you.

2.45 (1.42)

4 You are in a full parking lot. You see a driver leaving and you put on your indicator to signal you intend to take the parking space. Asthe other driver pulls out, a second driver cuts in front of you from the other side and takes the parking space.

2.40 (1.36)

9 You have been sitting in your car in a traffic jam for over 20 min. Suddenly, a car lightly bumps you from behind. 2.38 (1.61)2 You are driving your car down the motorway in the passing lane. You come up to a car driving much slower than you are in the

passing lane. Even though you flash your lights as a signal for the other car to move over, it does not.2.34 (1.50)

14 You are driving on the motorway. The driver in the car in front of you throws a cup of coffee out of his/her window. The cup hits yourwindshield.

2.32 (1.19)

19 You are travelling on a single-lane road late at night and the vehicle coming towards you has their lights on high beam. You flash yourlights, but the bright lights of the other vehicle do not change.

2.07 (1.15)

12 You are trying to exit the motorway. However, a car coming on to the motorway has failed to acknowledge a give way sign and theirbehaviour has caused you to miss the exit.

2.01 (0.68)

18 You are in the right-hand lane behind another vehicle. When the right turn light is given, the vehicle does not move because the driveris not paying attention. You tap on your horn to get their attention and they give you the middle finger in their rear view mirror.

1.95 (1.22)

11 You are driving on a city street. Without warning, a pedestrian suddenly runs in front of your car nearly causing you to hit him/her. 1.95 (1.47)10 You are driving on the motorway. One of the cars in front of you keeps switching lanes preventing other cars from passing efficiently.

Thus traffic is being slowed.1.65 (1.00)

15 While making a left-hand turn you accidentally cut-off another car. In response, the other driver follows you to the next intersectionat which point he/she pulls up to your car and proceeds to yell obscenities at you until the light turns green. When the light turnsgreen the other driver takes off in a hurry.

1.60 (0.77)

8 You are driving on the motorway when another vehicle pulls up alongside your car. You look over and see a total stranger makingobscene gestures at you.

1.59 (0.73)

3 You are driving on a single lane road. For no apparent reason the car in front of you is constantly braking and accelerating causing youto drive in the same manner.

1.58 (0.88)

17 You are driving on the motorway in the passing lane. You come up behind another car in the passing lane. You flash your headlights asan indicator for the other car to move over. Instead of moving over, you see the driver in the other car give you the finger and remainin the passing lane.

1.55 (0.94)

13 Your off ramp is quickly approaching. The driver next to you is driving in a manner that is preventing you from changing lanes. Youmay miss your exit.

1.52 (0.77)

16 You have been stuck in a traffic jam for nearly 40 min. While not paying attention you accidentally bump the car in front of you. Thedriver in the car in front of you leans out the window and swears at you very loudly.

1.52 (0.75)

7 You are driving in the passing lane at 120 km/h. The speed limit is 100 km/h. A car comes up behind you very quickly. Soon the other and h

1.32 (0.83)

you w

frfyohcha“yoo

ist(raat2

3d

a

vehicle is right on your bumper and the driver flashes his headlights6 You are sitting in your car at the traffic lights. A car pulls up next to

way too loud.

or the current study remained relatively low. The three scenariosated as receiving the most severe reactions were: a reversing carorcing the driver to make an emergency brake, “You are drivingour car down a two-lane road. Without warning, another car pullsut in front of you from a car park”. You had to brake suddenly to avoiditting it” (M = 2.65); and a car merging in front of the driver duringongestion “You are driving your vehicle in a traffic jam in the far leftand lane. Out of nowhere, a car comes up from behind on the shouldernd attempts to squeeze in front of you” (M = 2.45), or in a parking lotYou are in a full parking lot. You see a driver leaving and you put onour indicator to signal you intend to take the parking space. As thether driver pulls out, a second driver cuts in front of you from thether side and takes the parking space” (M = 2.40).

Mean responses for the Driving Anger Scale (short) are listedn Table 3. In agreement with the PADS, the most angering DAScenario also involved a vehicle merging in to traffic in front ofhe driver, “Someone backs out right in front of you without looking”M = 3.68). The next two most angering situations were “Someoneunning a red light or a stop sign” (M = 3.58) and “A slow vehicle on

winding road will not pull over and let people pass” (M = 3.34). Theverage total DAS score was 2.72 (SD = 0.68), which is comparableo previous published samples (for example; Deffenbacher et al.,004, M = 3.31; SD = 0.77).

.3. Confirmatory factor analysis and internal consistency of the

riving anger scales

The PADS and the DAS were subjected to independent Confirm-tory Factor Analyses (CFA) using EQS v6.1 for Windows (Bentler,

onks the horn.ith its windows rolled down and the stereo playing music 1.17 (0.64)

2005). The Robust Maximum Likelihood (ML) method was used inboth instances as Mardia’s normalised estimate was above 5, sug-gesting the data were non-normally distributed (Bentler, 2005).The results will be presented separately for each scale, howeverfor both CFA models, goodness of fit indices were evaluated withthe Santorra-Bentler Scaled Chi-Squared (S-B�2), S-B�2/df index,adjusted Comparative Fit Index (CFI) and Root Mean Square Error ofApproximation (RMSEA). Acceptable model fit is traditionally indi-cated by df index < 5, an adjusted CFI of 0.90 or greater, an RMSEA of0.60 or less (Hu and Bentler, 1995) and a confidence interval (C.I.)reporting a 90% interval surrounding the RMSEA acceptable level<5 (Browne and Cudeck, 1993). Further, in both cases the numberof free parameters fit well within the 5:1 (cases to free parame-ters) ratio suggested by Bentler and Chou (1987) indicating thatthe sample sizes were sufficient for each model.

3.3.1. Propensity for Angry Driving ScaleTable 4 outlines the model fit for the 19-item unidimensional

PADS. Lagrange Multiplier (LM) test identified misfitting param-eters and so the model was re-specified with one pair of errorsco-varied. These were Items 15 and 16, which are both concernedwith being yelled at by other motorists. The Wald Test on the finalmodel also suggested item 14, “You are driving on the motorway. Thedriver in the car in front of you throws a cup of coffee out of their win-dow. The cup hits your windshield” did not contribute to the model

and when the factor loadings were examined, the factor regres-sion co-efficient was not statistically significant. The model wastherefore re-specified excluding item 14. The final 18-factor modelshowed good fit.

92 M.J.M. Sullman, A.N. Stephens / Accident Analysis and Prevention 58 (2013) 88– 96

Table 3Item means for the 14-item Driving Anger Scale.

Item M (SD)

3 Someone backs our right in front of you without looking 3.68 (1.08)4 Someone runs a red light or stop sign 3.58 (1.27)2 A slow vehicle on a winding road will not pull over and let people pass 3.34 (1.15)6 Someone speeds up when you try to pass them 3.32 (1.11)9 Someone makes an obscene gesture towards you about your driving 3.08 (1.19)

10 Someone beeps at you about your driving 2.83 (1.15)1 Someone is weaving in and out of traffic 2.80 (1.05)

11 A cyclist is riding in the middle of the lane and slowing traffic 2.77 (1.21)13 A truck kicks up gravel on the car you are driving 2.47 (1.20)

8 You are stuck in a traffic jam 2.38 (1.07)14 You are driving behind a large truck and you cannot see around it 2.17 (1.10)

7 Someone is slow in parking and holds up traffic 2.08 (1.02)12 A police officer pulls you over 1.92 (1.12)

5 You pass a speed camera 1.75 (1.09)

Total 2.72 (0.68)

Table 4Summary of the CFA goodness-of-fit statistics for the unidimensional PADS.

Model fitted S-B�2 df S-B�2/df index CFI RMSEA C.I. 90%

First model on 19 items 183.50*** 152 1.21 0.89 0.03 0.007–0.050Second model on 19 items with 2 errors covaried 172.04*** 151 1.14 0.92 0.03 0.001–0.046

4

10rcrC

3

oiwsei

o

Final model on 18 items with 2 errors covaried 146.37*** 13

*** p < 0.001.

The standardised factor loadings are included in Fig. 1. The8-item model showed good internal consistency with a Rho of.85. Within-convergent validity was also met as the remaining 18egression coefficients for each item onto the factor were statisti-ally significant. The scale also had good reliability with a compositeeliability (CR) score of 0.80. Reliability is deemed good when theR score is >0.70 (Fornell and Larcker, 1981).

.3.2. Driving Anger ScaleTable 5 outlines the model fit for the Driving Anger Scale. The

riginal 14-items produced a poor fit. Lagrange Multiplier (LM) testdentified misfitting parameters and so the model was re-specified

ith four pairs of errors co-varied. These error correlations can beeen in Fig. 2. Of particular interest is the high correlation between

rrors for Item 9 and Item 10, which perhaps suggests redundancyn these questions.

The 14-item DAS showed good internal consistency with a Rhof 0.86. Within-convergent validity was met with all regression

Fig. 1. CFA model for Propensity for Angry Driving Scale (PADS)

1.09 0.95 0.02 0.000–0.043

coefficients being significant. Further, Fig. 2, shows that 85% offactor loadings were >0.50. Composite reliability was 0.86 also indi-cating good reliability of the 14-item DAS.

Before subsequent analyses were performed on the new scales,the distributions were examined. The distributions were withinnormal range demonstrating good values for skewness (<2) andkurtosis (<7).

3.4. Gender comparisons between the driving scales, trait angerscale and driving survey

Given that previous research has suggested gender differenceson self-reported driving anger propensities (Dahlen and Ragan,2004; DePasquale et al., 2001) gender comparisons were made for

the PADS, DAS and TAS means (see Table 6). There were no reliabledifferences on PADS, DAS or TAS scores between males and females.However, males reported a reliably higher number of violations(M = 2.00; SD = 0.53) than females (M = 1.77; SD = 0.45).

with standardised solution factor loadings (R2). *p < 0.001.

M.J.M. Sullman, A.N. Stephens / Accident Analysis and Prevention 58 (2013) 88– 96 93

Table 5Summary of the CFA goodness-of-fit statistics for the unidimensional DAS.

Model fitted S-B�2 df S-B�2/df index CFI RMSEA C.I. 90%

First model on 14 items 252.36*** 77 3.27 0.79 0.10 0.091–0.120Second model on 14 items with 8 errors covaried 117.99*** 73 1.62 0.95 0.05 0.036–0.073

*** p < 0.001.

ith sta

3

yIpsrttdP

umwsrtmnrt

TA

Fig. 2. CFA model for 14-item Driving Anger Scale (DAS) w

.5. Intercorrelations among variables

Scores on both the PADS and DAS were correlated with age,ears licenced, annual mileage and overall preferred driving speed.nterestingly, although the PADS was negatively related to age andositively related to preferred faster speeds, the DAS shared noignificant relationships with any background variable. The cor-elation between the DAS and the PADS was moderate, suggestinghat although these constructs are related (as would be expected),hey remain independent of each other. Previous research has alsoemonstrated only a moderate relationship between the DAS andADS (Dahlen and Ragan, 2004; r = 0.50, p < 0.01).

The intercorrelations between the PADS, DAS and other meas-res are presented in Table 7. Both the PADS and DAS wereoderately related to Trait Anger and all three anger measuresere positively related to tendency for driving violations. Higher

cores on both the PADS and DAS were reliably related to moreeported near-misses. The PADS also shared a weak positive rela-ionship with loss of control while driving. None of the three anger

easures shared any relationships with number of traffic ticketsor reported accident involvement (minor or major). No reliableelationships emerged when annual mileage was correlated withhe PADS or the DAS.

able 6lpha Coefficients and Means and Standard Deviations by gender.

Measure ̨ Males

M (SD)

Propensity for Angry Driving Scale 0.79 34.23 (10.00)

Driving Anger Scale (Short) 0.87 37.42 (9.83)

Loss of concentration 1.76 (3.05)Loss of control 0.20 (0.69)

Near-misses 1.07 (2.50)

Moving tickets 0.11 (0.58)

Minor accidents 0.11 (0.40)

Major accidents 0.00 (0.00)

Violations 0.73 2.00 (0.54)

Trait Anger Scale 0.83 16.15 (3.37)

* p < 0.01.

ndardised solution factor loadings (r-squared). *p < 0.001.

3.6. The PADS and DAS as predictors of detrimental drivingbehaviours

Hierarchical multiple regression analyses were conducted toexamine the extent to which the PADS and DAS predict scores fordriving violations as well as near-misses (see Table 8). For eachanalysis, the PADS and DAS were entered as the third of three steps.Step 1 entered background variables of age, gender and annualmileage, Step 2 was Trait Anger scores and then Step 3 includedboth PADS and DAS. By doing this, a direct comparison could bemade between the two driving anger scales while accounting fordemographic variables and trait levels of anger.

When predictors of driving violation scores were examined, age,gender and mileage were all found to significantly contribute to themodel at each step. Trait anger scores were also a reliable predictor,however they ceased to be reliable after Step 3 when PADS and DASwere entered. Of particular note, is that DAS scores were not a reli-able predictor of violations. Thus, the final model, which accountedfor 28% of the variance in violation scores, included age, gender,

mileage and PADS scores, but not DAS nor TAS. To explore this fur-ther, the regression was re-run without the PADS. Exclusion of thePADS led to a final model, accounting for 25% of the variance ofviolation scores, for which the DAS made a significant contribution

Females t (df)N M (SD) N

92 33.38 (8.41) 121 (211), <192 38.74 (9.25) 121 (211), −1.0088 2.18 (8.11) 115 (201), <190 0.16 (0.50) 118 (206), <189 0.77 (1.52) 119 (206), 1.0892 0.03 (0.18) 119 (104), 1.1992 0.08 (0.33) 119 (209), <191 0.02 (0.13) 119 (118), −1.4292 1.77 (0.45) 121 (211), 3.37*

90 17.11 (4.43) 120 (209), −1.74

94 M.J.M. Sullman, A.N. Stephens / Accident Analysis and Prevention 58 (2013) 88– 96

Table 7Intercorrelations among variables.

1 2 3 4 5 6 7 8 9

1. PADS –2. DAS 0.47*** –3. TAS 0.53*** 0.40*** –4. Violations 0.35** 0.22** 0.30** –5. Loss of Concentration 0.01 0.05 0.14 0.21** –6. Loss of Control (vehicle) 0.14* 0.01 0.06 0.13 0.22** –7. Near-miss 0.16* 0.25** 0.08 0.26** 0.21** 0.12 –8. Tickets −0.02 −0.02 0.07 0.15* 0.02 −0.03 0.02 –9. Minor crashes −0.10 −0.05 −0.04 −0.02 −0.03 0.04 −0.02 0.50*** –10. Major crashes 0.01 0.12 0.06 −0.05 −0.01 −0.03 −0.04 −0.01 0.25**

PADS = Propensity for Angry Driving Scale, DAS = Driving Anger Scale; TAS = Trait Anger Scale.

tpd

tiagaeoawim

4

v(mPiowt

TP

T

* p < 0.05.** p < 0.01.

*** p < 0.001.

o the change in R2 (0.02, p < 0.05). Both the DAS ( ̌ = 0.13, t = 1.99, < 0.05) and TAS ( ̌ = 0.22, t = 3.06, p < 0.01) were also reliable pre-ictors.

In contrast to violations, when near-misses were considered,he PADS did not emerge as a reliable predictor. As can be seenn the lower panel of Table 8, age, gender, mileage and TAS scoreslso failed to make a significant contribution to the model. Age,ender and mileage did not contribute at any Step. Entering PADSnd DAS significantly improved the model, although overall it onlyxplained 8% of variance of near-miss scores. However, inspectionf the individual predictors showed that only DAS served as a reli-ble predictor. To explore this further, the regression was re-runith the DAS excluded. When this was done, although PADS signif-

cantly contributed to the change in R2 (0.02, p < 0.05), the overallodel was not reliable (F = 1.29, p = 0.27).

. Discussion

The present study investigated the factor structure of the shortersion of the DAS and the PADS using Confirmatory Factor AnalysisCFA). The results showed that the fourteen item unidimensional

odel of the DAS proved good fit to the data. Further, the one factorADS model also proved a good fit to the data after the removal of

tem 14. Both the PADS and the DAS were related to crash-relatedutcomes, however the PADS reliably predicted driving violationshile the DAS reliably predicted near misses. Thus, highlighting

hese scales as complementary, rather than competing in nature.

able 8ropensity for Angry Driving Scale and Driving Anger Scale (short) as predictors of traffic

Step Variable ̌

ViolationsStep 1

Gender −0.17

Age −0.14

Mileage 0.21

Step 2 TAS 0.15

Step 3 PADS 0.22

DAS 0.06

Near-miss for accidentStep 1

Gender −0.07

Age −0.04

Mileage −0.03

Step 2 TAS −0.06

Step 3 PADS 0.08

DAS 0.25

AS = Trait Anger Scale; PADS = Propensity for Angry Driving Scale; DAS = Driving Anger S* p < 0.05.

** p < 0.01.*** p < 0.001.

Both the DAS and the 18-item New Zealand PADS showed goodinternal reliability. Previous research has also demonstrated goodreliability with different versions of the PADS (DePasquale et al.,2001; Leal and Pachana, 2008; Maxwell et al., 2005). However, thisis the first study to use CFA to factor analyse the PADS and also thefirst to use the PADS in a sample of New Zealand drivers. Further-more, this research joins the small number of studies in the areaof driving anger that have not relied primarily (or exclusively) onstudent participants.

In the present study, both the PADS and DAS were significantlycorrelated with violations. This is consistent with previous researchthat has found both scales to be significantly related to risky driv-ing behaviour (Dahlen and Ragan, 2004; Deffenbacher et al., 2001,2002b). However, it should be mentioned that when both drivinganger scales were entered together and following the partiallingout of the descriptive variables and trait anger, it was only the PADSthat was significant. This may be explained by that fact that thereis some shared variance between the DAS and PADS. However, thePADS has a stronger relationship with violations than the DAS.

Also in agreement with previous research, was the finding thatthe DAS was predictive of near-misses (or close calls) (Dahlenet al., 2005). Only one other study has compared both PADSand DAS in relation to near-misses (Dahlen and Ragan, 2004).

These researchers found the DAS scores from drivers in the USshared reliable, albeit rather weak (r = 0.15), relationships withself-reported near-misses, whilst the PADS did not. In the presentstudy of drivers in New Zealand, both anger scales were positively

violations.

R2 �R2 t

0.17 0.17***

−2.71**

−2.01*

3.36**

0.24 0.06*** 1.910.28 0.05** 2.85**

0.85

0.01 0.01<1<1<1

0.01 0.01 <10.08 0.07** <1

3.09**

cale (short).

ent An

re

2ntsnp2

DrtFehrArislawa

tiiwyyeaisfrd

PsD(flfaeswTt

dnswba2rc

a

M.J.M. Sullman, A.N. Stephens / Accid

elated to self-reported near-misses, although, only the DASmerged as a reliable predictor.

In further agreement with previous research (Dahlen et al.,005; Dahlen and Ragan, 2004; Deffenbacher et al., 1994), no sig-ificant gender differences emerged for the DAS or PADS scores inhe sample of New Zealand drivers. Male drivers however, reportedignificantly more driving violations than female drivers. The sig-ificant gender difference for violations is also consistent withrevious research (e.g. Parker et al., 2002; see de Winter and Dodou,010 for a DBQ meta-analysis).

In contrast to previous research (Dahlen et al., 2005;effenbacher et al., 2001), the DAS was not found to be significantly

elated to crash related conditions such as loss of concentration,ickets, loss of control, minor crash involvement or major crashes.urthermore, the PADS was not reliably related to involvement inither minor or major crashes, tickets or loss of concentration. Per-aps one reason contributing to the reduced number of significantelationships was the relatively modest number of respondents.s crashes are relatively rare events, a large sample of drivers isequired to adequately investigate the relationship between crashnvolvement and driving anger. If the number of participants werelightly larger, in the present study, the number of significant corre-ations may also have been larger. For example, the DAS correlatedt r = 0.12, p = 0.08 with major crashes and the PADS r = 0.10, p = 0.19ith minor crashes, both of which may have been significant with

larger sample.Also along these lines, the fewer number of factors related to

he PADS and DAS may have been in part due to the differencesn the demographic compositions of the respective studies. Mostmportantly, the mean age of the New Zealand drivers (44.1 years)

as much higher than the medium age of drivers in the US (19ears) reported by Dahlen and Ragan (2004). As research has shownounger drivers are more likely to engage in violations (Lajunent al., 1998; Sullman et al., 2002), report higher levels of drivingnger (Lajunen et al., 1998; Sullman, 2006) and are more frequentlynvolved in crashes (e.g. Parker et al., 2002; Sullman et al., 2002), ithould not be surprising that more significant relationships wereound with predominantly young participants. Therefore, futureesearch needs to investigate this relationship with a larger samplerawn from the general population.

The level of anger or hostile reactions reported on theADS (males = 34.2, females = 33.4) by participants in the presenttudy appeared to be considerably lower than that reported byePasquale et al. (2001) (overall = 50) and Dahlen and Ragan (2004)

males = 47.2, females = 43.2). This also appeared to be the caseor the DAS (males = 37.4, females = 38.7), which appeared to beower than was reported by Dahlen and Ragan (2004) (males = 43.9,emales = 47.2), Dahlen et al. (2005) (males = 45.6, females = 46.2),nd Deffenbacher et al. (2004) (males = 46.9, females = 45.8). Nev-rtheless, it should be noted that in all the previously mentionedtudies the participants were university students, who on averageere substantially younger than the sample in the present study.

herefore, it is not surprising that the present study would reporthe lowest levels of anger for both the PADS and DAS.

Construct validity for both the DAS and the PADS was clearlyemonstrated in the present study due to the moderate, but sig-ificant, correlations both scales had with trait anger. This findingupports previous research using drivers from other countries,hich has shown the original versions of the PADS and DAS to

e significantly related to measures of general anger (e.g. Dahlennd Ragan, 2004; Deffenbacher et al., 1994; DePasquale et al.,001). Construct validity was further demonstrated through the

egression analyses that showed driving anger trait scales predictedrash-related behaviour when trait anger was accounted for.

The present study provides further evidence that the DASnd PADS are measuring similar, but slightly different aspects of

alysis and Prevention 58 (2013) 88– 96 95

driving anger. The differences are evident firstly in the fact thatthey were moderately, but not strongly correlated (r = 0.47). Thiswas also demonstrated by the fact that they had different rela-tionships with other variables in the study. This may be explainedin part, by the nature of scales with the PADS measuring angryresponses and the DAS measuring reported anger. Nevertheless,both scales were correlated with similar crash related variables andboth were correlated with near-misses and violations. The PADSwas also correlated with losing control of the vehicle. Furthermore,although both scales contributed to the prediction of near-missesand violations, once the demographic variables and TAS had beenpartialled out, it was only the PADS that was a significant predictorof violations and only the DAS that was a significant predictor ofnear-misses.

In the present study, the PADS was more strongly correlated tothe TAS than the DAS. This was also mildly supported by Dahlen andRagan (2004) and is probably due to the fact that what the PADSmeasures is more similar to the TAS than what the DAS measures.The DAS measures how strongly each of the fourteen situationscause the driver to feel angry, while both the PADS and TAS measurereactions to anger. For example, the TAS responses may include: hitsomeone who angered you, which the PADS responses may include:use your vehicle to hit another vehicle whose driver has angeredyou.

The Driving Survey measures the accident related conditionsover the previous three months. Although collecting the data overthis period would have helped by reducing “recall bias”, it alsomeans very few individuals would have experienced minor ormajor accidents (8.1% and 0.8%, respectively), which are relativelyrare events. Although previous research using this time period hasfound relationships between the PADS and accidents, it shouldbe remembered that the previous research had very young sam-ples and young drivers are over-represented in accident statistics.Future research in this area should also collect the number of acci-dents over a longer period of time.

A potential limitation of the present study is the fact that theresearch relied solely upon self-reported data, which is subjectto self-report and social desirability bias. However, as nothingwas contingent upon the drivers’ answers, their names were notrecorded and they were all assured of complete confidentiality andanonymity, the impact of social desirability bias is not likely tohave been large. In support of this, research has found that theeffect of social desirability bias on self-reported risky behaviours isnot necessarily substantial (e.g. Sullman and Taylor, 2010; Lajunenand Summala, 2003). Furthermore, while acknowledging that self-report is not flawless, there is a considerable body of research thatclearly supports the accuracy of data gathered in this manner (e.g.Rolls et al., 1991; Walton, 1999; West et al., 1993).

5. Summary and practical implications

In summary, the present study has confirmed the unidimen-sional structure of the PADS and DAS in a sample of New Zealanddrivers. Further, a number of the previously reported relationshipsbetween these driving anger scales and crash related conditionswere replicated in the current sample, which was from the generalpopulation of drivers, rather than university students. However, anumber of previously reported relationships were not confirmed.The results have also demonstrated the independence of the twodriving-anger measures. Thus, although the PADS and DAS are sim-ilar in nature, they should be thought of as complementary rather

than competing scales.

The findings have practical implications for driving research.First, the measurements gained allowed greater understandingof the relationship between anger and crash-related outcomes,

9 ent An

bFfdard

A

RhYi

R

B

B

B

D

D

d

D

D

D

D

D

D

D

D

E

6 M.J.M. Sullman, A.N. Stephens / Accid

eyond that which could be obtained through accident databases.urther, the results validated both the PADS and the DAS as use-ul tools in understanding propensities for driving anger and angryriving. It is important to identify which situations lead to anger,nd in turn crash-related conditions as these findings can informoad-safety campaigns aimed at reducing anger-related dangerousriving behaviours.

cknowledgements

The authors would like to thank Beth Mackie (now Tootell),obyn Mason, Stephanie Schmitt and Joanna Young for their kindelp during the data collection. We would also like to thank Joannaoung for her time in carefully transferring the questionnaire data

nto SPSS.

eferences

entler, P.M., 2005. EQS 6 Structural Equations Program Manual. Multivariate Soft-ware, Encino, CA.

entler, P.M., Chou, C.P., 1987. Practical issues in structural modelling. SociologicalMethods & Research 16, 78–117.

rowne, M.W., Cudeck, R., 1993. Alternative ways of assessing model fit. In: Bollen,K.A., Long, J.S. (Eds.), Testing Structural Equation Models. Sage, Beverly Hills, CA,pp. 136–162.

ahlen, E.R., Ragan, K.M., 2004. Validation of the propensity for angry driving scale.Journal of Safety Research 35, 557–563.

ahlen, E.R., Martin, R.C., Ragan, K., Kuhlman, M.M., 2005. Driving anger, sensa-tion seeking, impulsiveness and boredom proneness in the prediction of unsafedriving. Accident Analysis and Prevention 37, 341–348.

e Winter, J.C.F., Dodou, D., 2010. The driver behaviour questionnaire as a pre-dictor of accidents: a meta-analysis. Journal of Safety Research 41, 463–470.

effenbacher, J.L., Deffenbacher, D.M., Lynch, R.S., Richards, T.L., 2003. Anger,aggression and risky behaviour. A comparison of high and low anger drivers.Behaviour Research and Therapy 41, 701–718.

effenbacher, J.L., Huff, M.E., Lynch, R.S., Oetting, E.R., Salvatore, N.F., 2000. Charac-teristics and treatment of high-anger drivers. Journal of Counseling Psychology43, 131–148.

effenbacher, J.L., Filetti, L.B., Lynch, R.S., Dahlen, E.R., Oetting, 2002a. Cognitive-behavior treatment of high anger drivers. Behaviour Research and Therapy 40,895–910.

effenbacher, J.L., Lynch, R.S., Oetting, E.R., Yingling, D.A., 2001. Further evidence ofreliability and validity for the Driving Anger Expression Inventory. PsychologicalReports 89, 535–540.

effenbacher, J.L., Lynch, R.S., Oetting, E.R., Swaim, R.C., 2002b. The driving angerexpression inventory: a measure of how people express their anger on the road.Behaviour Research and Therapy 40, 717–737.

effenbacher, J.L., Oetting, E.R., Lynch, R.S., 1994. Development of a driver angerscale. Psychological Reports 74, 83–91.

effenbacher, J.L., White, G.S., Lynch, R.S., 2004. Evaluation of two new scalesassessing driving anger: the driving anger expression inventory and the driver’sangry thoughts questionnaire. Journal of Psychopathology and BehaviouralAssessment 26, 87–99.

ePasquale, J.P., Geller, E.S., Clarke, S.W., Littleton, L.C., 2001. Measuring road rage:

development of the propensity for angry driving scale. Journal of Safety Research32, 1–16.

nders, C.K., 2001. A primer on maximum likelihood algorithms available for usewith missing data. Structural Equation Modeling: A Multidisciplinary Journal 8,128–141.

alysis and Prevention 58 (2013) 88– 96

Fornell, C., Larcker, D.F., 1981. Evaluating structural equation models with unob-served variables and measurement error. Journal of Marketing Research 18,39–50.

Hu, L.-T., Bentler, P.M., 1995. Evaluating model fit. In: Hoyle, R.H. (Ed.), StructuralEquation Modelling: Concepts, Issues and Applications. Sage, Thousand Oaks,CA, pp. 76–99.

Lajunen, T., Parker, D., Stradling, S.G., 1998. Dimensions of driver anger, aggressiveand Highway Code violations and their mediation by safety orientation in UKdrivers. Transportation Research Part F 1, 107–121.

Lajunen, T., Summala, H., 2003. Can we trust self-reports of driving? Effects ofimpression management on driver behaviour questionnaire responses. Trans-portation Research Part F 6, 97–107.

Leal, N.L., Pachana, N.A., 2008. Adapting the Propensity for Angry DrivingScale for use in Australian research. Accident Analysis and Prevention 40,2008–2014.

Leal, N.L., Pachana, N.A., 2009. Validating the Australian Propensity for Angry DrivingScale (Aus-PADS). Accident Analysis and Prevention 41, 1112–1117.

Maxwell, J.P., Grant, S., Lipkin, S., 2005. Further validation of the propensity forangry driving scale in British drivers. Personality and Individual Differences 38,213–224.

Ministry of Transport, 2013. 2012 Speed Survey results – car speeds, Retrievedfrom: http://www.transport.govt.nz/research/roadsafetysurveys/Pages/2012speedsurveyresultscarspeeds. aspx

Parker, D., Reason, J.T., Manstead, A.S., Stradling, S.G., 1995. Driving errors, drivingviolations and accident involvement. Ergonomics 38, 1036–1048.

Parker, D., Lajunen, T., Summala, H., 2002. Anger and aggression among drivers inthree European countries. Accident Analysis and Prevention 34, 229–235.

Reason, J., Manstead, A., Stradling, S., Baxter, J., Campbell, K., 1990. Errors andviolations on the roads: a real distinction? Ergonomics 33, 1315–1332.

Rolls, G.W.P., Hall, R.D., Ingham, R., McDonald, M., 1991. Accident Risk andBehavioural Patterns of Younger Drivers. AA Foundation for Road SafetyResearch, Southampton.

Spielberger, C.D., 1999. State-Trait Anger Expression Inventory, 2nd ed. Psycholog-ical Assessment Resources, Odessa, FL.

Spielberger, C.D., 1988. State-Trait Anger Expression Inventory, 1st ed. PsychologicalAssessment Resources, Odessa, FL.

Stephens, A.N., Groeger, J.A., 2011. Anger-congruent behaviour transfers acrossdriving situations. Cognition and Emotion 25, 1423–1438.

Stephens, A.N., Trawley, S.L., Madigan, R., Groeger, J.A., 2013. Drivers display anger-congruent attention to potential traffic hazards. Applied Cognitive Psychology27, 178–189.

Sullman, M.J.M., 2006. Driving anger amongst New Zealand drivers. TransportationResearch Part F 9, 173–184.

Sullman, M.J.M., Gras, M.E., Cunill, M., Planes, M., Font-Mayolas, S., 2007. Drivinganger in Spain. Personality and Individual Differences 42, 701–713.

Sullman, M.J.M., Meadows, M.L., Pajo, K., 2002. Aberrant driving behaviours amongstNew Zealand truck drivers. Transportation Research Part F 5, 217–232.

Sullman, M.J.M., Stephens, A.N., Kuzu, D., 2013. The expression of driving angeramongst Turkish taxi drivers. Accident Analysis and Prevention 56, 42–50,http://dx.doi.org/10.1016/j.aap.2013.03.013.

Sullman, M.J.M., Taylor, J.E., 2010. Social desirability and self-reported driv-ing behaviours: should we be worried? Transportation Research Part F 13,215–221.

Tabachnick, B.G., Fidell, L.S., 2007. Using Multivariate Statistics. Allyn and Bacon,Boston, MA.

Underwood, G., Chapman, P., Wright, S., Crundall, D., 1999. Anger while driving.Transportation Research Part F 2, 55–68.

Villieux, A., Delhomme, P., 2007. Driving Anger Scale, French adaptation: furtherevidence of reliability and validity. Perceptual and motor skills 104, 947–957.

Walton, D., 1999. Examining the self-enhancement bias: professional truck drivers’

perceptions of speed, safety, skill and consideration. Transportation ResearchPart F 2, 91–113.

West, R., French, D., Kemp, R., Elander, J., 1993. Direct observation of drivingself reports of driving behaviour and accident involvement. Ergonomics 36,557–567.