Theoretical Issues in Ergonomics Science, Volume 12, Issue 1, 2011, Pages 66-86

The Manchester Driver Behaviour Questionnaire as a

Predictor of Road Traffic Accidents

A.E. af Wåhlberg* (a)

L. Dorn (b)
and
T. Kline (c)

a. Department of Psychology
Uppsala University
Uppsala, Sweden

b. Department of Systems Engineering and Human Factors,

Cranfield University
Cranfield, United Kingdom

c. University of Calgary
Calgary, Canada

Received: 14 February 2008

* e-mail: [email protected]
Homepage: http://www.psyk.uu.se/hemsidor/busdriver

1
 Abstract

The Driver Behaviour Questionnaire (DBQ) has mainly been used as predictor of self-

reported road traffic accidents. The associations between crashes and the violation and

error factors of the DBQ however, may be spuriously high due to reporting bias. In the

present study, the DBQ was tested as predictor of self-reported and recorded accidents in

four samples of private and professional drivers. The findings show that the DBQ scale

only predicts self-reported accidents, not recorded crashes, despite the higher validity of

company data, and the higher means of the recorded data across these samples. The

results can be explained by a common method variance bias. In our review of the DBQ

research, the use of the instrument was found to be heterogeneous concerning the number

of items, scales used, and factor analytic methods applied. Thus, the DBQ may not be as

homogenous and as successful in predicting accidents as is often claimed.

Keywords: DBQ, questionnaire, traffic accident, crash, driver, common method variance

2
Introduction

Road traffic accidents are a major health and economical problem throughout the

world, and research within traffic psychology has largely centered on this problem.

One of the main approaches has been the use of individual differences to predict their

later accident involvement (for reviews, see Signori & Bowman, 1974; McGuire,

1976; Golding, 1983; Hansen, 1988; Lester, 1991; Arthur, Barrett & Alexander, 1991;

Elander, West & French, 1993; Peck, 1993; af Wåhlberg, 2003). If this could be done

with an acceptable degree of accuracy, it would be possible to identify drivers for

training or other interventions. However, for practical use, the measurement technique

must be valid and easy to use on a large scale.

The simplest way of measuring behaviour is to ask people how they

typically behave, effectively using individuals as their own monitoring devices. In this

way, any behaviour, thought, attitude and experience can, in principle, be canvassed

very cheaply, for any time period. Other methods, which have been tried for traffic

accident prediction, do not share these advantages. Medical examinations, for

example, are mainly about the current somatic state of the individual (unless

supplemented with self-reports or archive data). Objective driving behaviour

measurements are similarly restricted to the, usually very short, period of data

gathering.

Possibly due to these factors, driver behaviour questionnaires have been

very popular for many decades within traffic research. However, few of the many

suggested variants have reached wider acceptance. On the contrary, many inventories

seem to have been constructed specifically for a single study, or possibly used a few

times by a research group. This would seem to be the case for the Driving Behavior

3
Questionnaire (Arthur, Strong & Williams, 1994), the Driving Habits Questionnaire

(Schwebel, Severson, Ball & Rizzo, in press), the Driving Behavior Rating Scale

(Barkley, Murphy, Du Paul & Bush, 2002), the Driving History Survey (Knouse,

Bagwell, Barkley & Murphy, 2005) as well as a number of others without specific

names (e.g. Gumpper & Smith, 1968; McGuire, 1972; Smith & Kirkham, 1982;

Larson & Merritt, 1991; Chou, Miyazaki & Maruyama, 1994).

However, the Manchester Driver Behaviour Questionnaire (DBQ)

(Reason, Manstead, Stradling, Baxter & Campbell, 1990) has gained wide acceptance

(e.g. Blockey & Hartley, 1995; Parker, Reason, Manstead & Stradling, 1995; Rimmö

& Åberg, 1999; Westerman & Haigney, 2000; Sullman, Meadows & Pajo, 2002). So

far, at least fifty-four published studies have used at least parts of this instrument in

various ways (list available upon request from first author). The DBQ has its origins

in Reason‟s error theory (Reason, 1987; Reason, Manstead, Stradling, Baxter &

Campbell, 1990) and measures what is called aberrant driving behaviours (errors1,

lapses and violations). One of its basic intended uses is the prediction of (individual

differences in) traffic accidents. Although a number of studies have tested this (thirty-

one of the above-mentioned fifty-four), and reported some success, there is one

feature about the DBQ reports that should be noted; they have almost exclusively

used self-reported accidents as the dependent variable. The only known exception in

the use of recorded accidents (Caird & Kline, 2004) found no significant effects.

This is notable because self-reports of traffic accidents have repeatedly

been shown to have dubious validity (e.g. Arthur, 1991; Szlyk, Seiple & Viana, 1995;

McGwin, Owsley & Ball, 1998) due to memory loss (Maycock, Lockwood & Lester,

1991; Maycock & Lester, 1995; Chapman & Underwood, 2000) and possibly other

1
The terminology has changed somewhat; other terms used include mistakes, slips, aggressive
violations etc.

4
biases, like over-reporting (af Wåhlberg, 2002a) and selective under-reporting (Brown

& Beardie, 1960). Furthermore, archival sources of accidents (state and company

records) seem to have different associations, as compared to self-reported ones, with

various predictors (e.g. Owsley et al, 1991; Arthur et al, 2001; af Wåhlberg, 2002a).

Given that the DBQ uses self-reports to predict other self-reports, it is

likely that at least some part of the significant associations reported between one or

two of the factors and accidents is due to the problem of common method variance

(Podsakoff, MacKenzie, Lee & Podsakoff, 2003). Common method variance refers to

(in part) people‟s tendency to want to respond consistently across related measures. In

other words, they demonstrate a tendency to locate their responses in particular

regions of a scale such as at one end, or the middle. When these responses are

analysed they tend to create a correlation that is artificially inflated. Although the

widespread existence of this problem has been questioned, even a critic such as

Spector (2006) acknowledges that there are many situations where this type of

problem might occur. Other researchers have shown effects that can be traced to

common methods of gathering data within studies (e.g. Williams, Cote & Buckley,

1989; Moorman & Podsakoff, 1992).

With regards to traffic psychology, it could be that people report in a

self-congruent manner about their accidents and behaviours, when the latter are

perceived as relevant for traffic safety. This mechanism is a slight extension of the

„consistency motif‟ suggested for cognitions and attitudes (Osgood & Tannenbaum,

1955), in which respondents search for similarities between items and respond in

ways that are more consistent than is actually the case in real life. For traffic

behaviour and its link with accidents, it could be that there is an artifactual

consistency in reporting between the number of accidents respondents have been

5
involved in and the behaviours they believe may have had an impact on their accident

record. If this hypothesis is applied to the DBQ, it can, for example, be predicted that

associations between these factors and accidents will be stronger if the accidents are

self-reported, than if they are retrieved from actual records.

The present study was undertaken to test the specific hypothesis that the

DBQ can predict self-reported accidents, but not recorded crashes. The two main

sources for objective accident data are state records and transport company data.

There are two methodological problems that emanate from this hypothesis. First, state

records typically have lower means (and variability) than self-reported accident data

for the same drivers over short time periods (Schuster & Guilford, 1962; McGuire,

1973; Sobel & Underhill, 1976), and any comparison of predictive power between

these sources is therefore automatically confounded by this difference. However, for

the present study, it was possible to use self-reported and transportation company

data. Although the evidence regarding the validity of such data is sparse, it has been

shown for one of the companies used here that the records contain more accidents

than the drivers themselves reported for a three-year period (af Wåhlberg, 2002a).

The second problem to consider is that the DBQ was not developed for

bus drivers or other professional driver groups, and it could be argued that the DBQ is

not applicable for such samples. However, it is possible to adapt the DBQ for bus

driving, deleting items that are not suitable (car specific), and replacing them with

similar ones that are pertinent to bus driving. If the general principle of aberrant

driving behaviour predicting accident involvement is robust, then the DBQ should

withstand minor alterations to the surface content of some items and still be able to

predict accidents. It should also be remembered that a few studies have made these

kinds of alterations for specific populations other than normal car drivers (e.g. older

6
people (Parker, 1999), truck drivers (Sullman, Meadows & Pajo, 2002) and various

other professional drivers (Sümer, 2003)), yet managed to predict their self-reported

accident involvement.

Turning to accidents as a dependent variable, this is an area fraught with

difficulties, both statistical and methodological. The basic problem is the

uncommonness of traffic incidents of such severity as to be called accidents, which

creates a skewed distribution in most populations for the time periods used for

calculations. This means that statistical power will be low, and the range of suitable

methods somewhat restricted. At the same time, it should be remembered that many

different statistical analyses have been used by various researchers, and there is some

virtue in using certain tests even when they have sub-optimal power, given the

importance of being able to compare results.

One methodological problem of some importance in the present study

concerns what kind of accidents should be used as the dependent variable. Most

researchers take all accident data recorded, without further consideration (af

Wåhlberg, 2003), while some deliberate over culpability, responsibility, active-

passive and similar concepts. The differences between the first two terms (and several

other, such as „avoidable‟) would seem to be slight, although it is most often hard to

know exactly what is meant by any of these terms, as they have seldom been defined

(af Wåhlberg, 2003). They all concern whether a driver has been partly responsible

for the cause of an accident, and mainly seem to differ in their degree of severity of

their criterion; the percentage of collisions that are ascribed as at least partly a driver‟s

fault differ rather strongly between studies (from 35 percent (Brandaleone & Flamm,

1955) to 85 percent (Drummer et al, 2004)). The active-passive notion (West, 1997),

however, is different from the various culpability terms, and mainly concerns which

7
party was moving faster before the collision. Exactly what active-passive means is

hard to determine (af Wåhlberg, 2002b), but there is probably a good degree of

overlap with between this term and culpability. The problem is whether there is such a

thing as an accident involvement that is unrelated to the driver's behaviour. This is a

daunting issue that has so far not been resolved. It is not the purpose of this study,

however, to resolve this issue, merely to bring it to the reader‟s attention.

The originators of the DBQ did not state explicitly which factors may

relate to the concept of culpability, but the notion of active/passive accidents has at

times been used in DBQ studies (e.g. Parker, McDonald, Rabbitt & Sutcliffe, 2000;

Fergusson, Horwood & Boden, in press). As the whole concept of aberrant driving

would seem to imply that the drivers with these behaviours cause accidents, it would

be reasonable to assume that only culpable accidents should be included as the

dependent variable. Yet most DBQ studies have not considered, or even discussed the

issue. For the present study then, different variants of the dependent variable were

used to address this methodological problem.

Summing up, the present study compared the predictive power of the

DBQ using traffic accidents from different sources as dependent variables, with the

main consideration being self-reported versus recorded ones. To make the results

interpretable and comparable across samples, several statistical methods were

employed, as well as different accident variables.

Method

General

Data from four different research projects were analyzed in this study. The first was

undertaken in the United States, the second in the United Kingdom, the third in

Sweden, and the fourth in Canada. For the first two samples, a DBQ for car drivers

8
was distributed, in the third, respondents completed a DBQ modified for bus driving,

and in the fourth a DBQ for professional truck drivers was used. Each study had its

own unique combination of sample, materials and dependent variables, but all

contributed a facet concerning the use of self-reported accidents and the

administration of the DBQ. Descriptive data for the four samples are shown in Table

1.

Questionnaires

The DBQ exists in many different versions, where each researcher, or even study,

seems to have used a unique combination of number and type of items (see further

information in the discussion). In the present report, the original 50-item version was

used (Reason, Manstead, Stradling, Baxter & Campbell, 1990; Blockey & Hartley,

1995) in the U.S. and U.K. studies (see Appendices 1 and 2), but with a six-point

Likert-type response scale instead of the original five (see further information in the

results section). The U.S. study DBQ also had a code for 'Not applicable'.

The Swedish Bus-DBQ questionnaire administered for the present study

consisted of thirty-two items with a six-point Likert-type response scale (see

Appendix 1), where fourteen were newly created to be bus driver specific, and the

others were taken from the Swedish version of the DBQ (Åberg & Rimmö, 1998;

Rimmö & Åberg, 1999). The aim was to include similar numbers of violations, errors,

slips and lapses (see comments in Appendix 3).

In the Canadian sample, 15 items from the DBQ (Parker, Reason,

Manstead & Stradling, 1995) and an additional item on using cellular phones were

used to assess general driving errors and violations (see Appendix 4). The items were

responded to on a five-point Likert-type scale.

Samples and procedures

9
The U.S. sample consisted of older adults without diagnosis of dementia, Parkinson‟s

disease, psychosis, eye conditions other than refractive error or cataract, or any

illness, enrolled in a prospective study on the impact of cataracts on mobility (ICOM)

as described in (Owsley et al., 2002; Owsley, Stalvey, Wells, & Sloane, 1999;

Owsley, Stalvey, Wells, Sloane, & McGwin, 2001). All were licensed to drive in

Alabama and were active drivers. Drivers had been recruited through eye care clinics

in the Birmingham, Alabama area. The questionnaire was distributed at the third

annual follow-up visit for the ICOM study. They were asked to complete the DBQ

using the reference of their behaviours over the past 10 years.

All U.K. drivers were employees of a major U.K. bus company. Two

hundred and forty-eight drivers returned the questionnaire, but ten of them had to be

discarded due to missing data, and the final sample consisted of 238 respondents. The

questionnaire contained items about name and payroll number, so all respondents

could be identified.

In Sweden, the Bus-DBQ and a pre-paid envelope were distributed to

about 350 drivers2 at the bus company Gamla Uppsalabuss (GUB) through their

personal mailboxes at the bus garage. Each questionnaire was marked with the

driver's employment number for identification purposes. A reward consisting of a free

ticket to the movies was sent to each respondent. After three weeks, a reminder was

sent to drivers that had not responded. Finally, a few drivers were contacted in person

and asked to complete the questionnaire. A total of 127 questionnaires were returned

with complete data for the analyses undertaken here (a somewhat larger number was

used for factor analysis and some accident calculations using all available data).

2
A higher number of questionnaires were distributed to mail slots, but several of these were probably
vacant, due to the employment numbers not having been updated on slots of drivers who had quit or
were on long-term sick leave.

10
 In Canada, all participants were professional drivers for a single oil and

gas distribution company. Surveys were distributed to drivers through a company

contact during January and February of 1998. To ensure confidentiality of the data

collection procedure, surveys were sent directly to the researchers. A total of 330

surveys were distributed and 190 of them were returned (58% response rate).

However, only 115 cases had complete data for all variables used in the present study.

These data were used in Caird and Kline (2004), but the present analysis was more

extensive.

Table 1 about here

Accident data

Questions concerning the number of traffic accidents the driver had been involved in

during the last year and how many of these instances the police had been called to

scene were included in the U.S. DBQ study, while recorded accident data was

supplied by the Alabama Department of Public Safety for a period of ten years (before

and after the collection of the self-report data). The recorded accident data included

information about culpability for each crash, assigned by the reporting police officer.

In a few cases (nine), culpability had not been recorded. These cases were treated as

culpable accidents, as it was more probable that some blame could have been allotted,

than no blame given the findings of af Wåhlberg and Dorn (2007). Therefore, four

different dependent variables were available for this sample, two self-reported

(number of crashes and number of police attended crashes) and two recorded

11
(culpable and All) with the latter variables differing as to culpability, while the first

two variables probably measured different degrees of crash severity.

A major British bus company supplied recorded traffic accident data for

the U.K. study. This company‟s reporting system records incidents occurring during

work after each shift according to a strict vehicle check policy before handing over

the vehicle to the driver on the next shift. Data recorded included all damage and

injuries involving the bus and due to bus driver behaviour or other road users.

Culpability for traffic accidents (including falls in the bus) was assigned by the

driver's manager in collaboration with an insurance team, incorporating such

information as the driver's report, photographs, witness reports etc. Three levels of

responsibility for the accident were used; none, some and sole. Therefore, in the U.K.

sample, drivers' accident record was represented by three culpability variables;

number of All accidents, All responsible accidents, and Solely responsible accidents.

No self-report data were gathered for this sample.

In the Swedish study, bus accident data were gathered from the archives

of GUB (af Wåhlberg, 2002b; 2004). For the present study, number of accidents for

the three years before the questionnaire was distributed was used as the dependent

variable, regardless of culpability. This was for two reasons. First, DBQ studies have

paid little attention to culpability, and instead canvass all accidents (e.g. Blockey &

Hartley, 1995; Rimmö & Åberg, 1999; Gras et al, 2006). Possible exceptions to this

include Dobson, Brown, Ball, Powers and McFadden (1999), and Parker, West,

Stradling and Manstead (1995). Second, the project that generated the data was partly

designed to investigate whether there was any difference between number of self-

reported accidents and archival data (af Wåhlberg, 2002a); adding culpability could

have confounded this issue. Separate questions on number of bus and car accidents,

12
regardless of culpability, for the last three years were included in the questionnaire.

Therefore, the Swedish Bus-DBQ could be tested against three accident variables (self

reported bus crashes, company recorded bus crashes and self reported car crashes).

In the Canadian Study, self-report information was gathered on the 1)

number of accidents, 2) number of moving violations and 3) accident free kilometers

driven in the past two years on the job. In addition, participants also reported the

number of „near misses‟ they experienced over the past month. Not all of these

variables can be said to be valid measures of safety, but they are commonly used

within traffic safety research. Drivers indicated whether or not the researchers could

see their personnel records for research purposes. The names of these individuals

were faxed to the company contact person who provided the information. Of the 190

respondents, 153 (80.5%) allowed access their driving records for the number of

accidents and the number of kilometers driven accident free (total-to-date with the

company). No more information about their driving records, such as the number of

accidents or the number of kilometers driven accident free over just the past two

years, nor any information about the accidents (e.g., severity, culpability), could be

obtained. Although the information about accident-free kilometers driven is an

unusual criterion measure, the company was tracking this information in particular so

as to encourage safe driving amongst its employees.

Statistical methods

Different researchers have used different statistical methods from which

to derive DBQ factors; for example principal axis factor or principal components

analysis with direct oblimin (Özkan & Lajunen, 2005; Özkan, Lajunen & Summala,

2006) or varimax (Özkan, Lajunen, Chliaoutakis, Parker & Summala, in press)

rotation, direct summing of scores (Dobson, Brown, Ball, Powers & McFadden, 1999;

13
Lajunen & Summala, 2003), or no report of the analysis used (Parker, West, Stradling

& Manstead, 1995; Parker, 1999). However, principal components analysis with

varimax (e.g. Reason, Manstead, Stradling, Baxter & Campbell, 1990; Blockey &

Hartley, 1995; Parker, Reason, Manstead & Stradling, 1995; Rimmö & Åberg, 1999;

Sullman, Meadows & Pajo, 2002; Gras et al, 2006) or oblimin (Parker, Lajunen &

Stradling, 1998; Parker, McDonald, Rabbitt & Sutcliffe, 2000) rotation seems to be

the most common choice. Thus, in the present study, principal components analysis

was selected for data reduction, with varimax rotation if more than one relevant

component was extracted.

Given the somewhat small samples used, component solutions might not

always be dependable. Therefore, as a complementary method, it was decided to

simply unit-weight the items on the pre-defined factors (e.g., violation) creating a

„forced‟ scale. It should be remembered that violations of the kind included in the

DBQ do not really need to form a reliable scale (i.e., a person is high or low on most

items) to be able to predict accident involvement. A person who only engages in

speeding violations is still a dangerous driver and could accumulate accidents by this

behaviour alone. Therefore, summing the violation items was deemed a valid method

under the present circumstances. This approach has also been used by other authors

(e.g. Lajunen & Summala, 2003; Schwebel, Severson, Ball & Rizzo, in press; King &

Parker, in press).

Measures of association (Pearson correlation and Spearman‟s rho) were

used in this study. Use of the Pearson correlation with accident data as one of the

variables has been questioned, due to the restriction of range on this variable. This

argument is particularly problematic if accident data are collected over short time

periods and with and low-accident risk populations. In most of the present samples

14
range restriction was not problematic for the recorded data as is evinced by the high

mean values (see Tables 2, 4-6). The main reason for using correlational analyses is

that it allows comparison across studies. To supplement the measures of association,

however, t-tests were used for analyses of differences between accident-involved and

accident-free drivers.

Results

The U.S. study

When analysing the U.S. DBQ data, two items concerning gear shifting were deleted

due to very low response rates, as most U.S. cars have automatic gearboxes. All item

responses were strongly skewed, with very low means. This was probably due to the

homogeneous sample of elderly drivers. The remaining U.S. data of forty-eight items

was analysed using a principal components analysis (various rotated versions yielded

very similar results to the unrotated solution). There were fifteen components with

eigenvalues above 1, but the screeplot indicated that a solution of only one component

was most appropriate.

The fourteen items (all were errors) that loaded above .50 on this first

factor were included in a second principal components analysis resulting in a single

component with an eigenvalue of 4.8 explaining 34.5 percent of the variance. Twelve

of the 14 items loaded at .50 or higher. These items then were summed and used as an

“error” predictor. In addition, a violation scale was created by summing across the

seventeen violation items. These variables were then correlated with crashes, with the

results presented in Table 2. It can be seen that the self-reported variables had

stronger effects for the DBQ error variable, while the situation was reversed for the

DBQ violation scale.

15
 The associations between the criterion variables were then investigated.

As there were two variants of each of these two data sources, a total of six

correlations could be computed. Spearman‟s rho was selected for these tests, due to

the low means for these variables. These associations are shown in Table 3. It can be

seen that within each source the correlations were positive and strong. Between

sources, however, all associations were negative.

Tables 2-3 about here

The U.K. study

The fifty items of the U.K. DBQ were analysed using principal components with a

varimax rotation3. Although there were fifteen factors with an eigenvalue above 1, the

screeplot indicated that a three-factor solution was acceptable in this data set.

The factor-based scores for these three components were created and a violation scale

based on the 16 violation items was generated. These were correlated using the

Pearson product-moment coefficient with all U.K. accident variables. These results

are shown in Table 4. T-tests of the differences in means on the DBQ factors and

scale were conducted with accidents/no accidents as the grouping variable, and are

also shown in Table 4. As can be seen, no significant results were found. We re-ran

the principal components analysis resulting in a three-factor solution, and left the

components unrotated. Component-based scales created from this analysis showed

weak associated with accidents.

It might be argued that this DBQ version (an early model) was not

optimal for accident prediction. However, exactly what version/which items should be

3
In an unrotated solution, forty-three items loaded most strongly on the first factor.

16
used cannot be ascertained, as there are so many of them (see further information

presented in the Discussion). Therefore, to provide the most liberal test of the DBQ in

predicting accidents, we conducted yet another components analysis. Starting with the

previous components analyses results, all items with a loading below .50 were

deleted, leaving twenty-five items. A principal components analysis with varimax raw

rotation yielded five components with eigenvalues above 1, but, again, the screeplot

indicated a three-component solution to be most appropriate. The first component had

nine slips, three mistakes and two violations (31.4 percent explained variance, alpha

.88), the second seven violations and one slip (9.6%), and the third two violations and

a slip (7.3%). Only one item loaded below .50. The correlations between these

component-based and the original ones were >.89 for each, and the associations for

the new ones with accidents very similar to those presented in Table 4, (i.e. weak and

not significant). Thus, regardless of how these items were pulled together, they were

not able to predict accidents in this sample.

Table 4 about here

The Swedish study

The thirty-two items of the Bus-DBQ were analysed (N=146), using principal

components with varimax rotation. The screeplot indicated that a two-component

solution was most appropriate, where the first explained 33.4 percent of the variance,

and the second accounted for 8.1 percent of the variance. The component loadings are

shown in Appendix 3. The results suggest that the first component was a general

“error” component, while the second consisted mainly of violations. Component-

based scales constructed using these results were uncorrelated. The alphas were

17
somewhat higher than most in DBQ research (e.g. Sullman, Meadows & Pajo, 2002).

The two component-based scores and a violation scale created by summing across all

violation items were correlated with the three available accident variables and the

results are presented in Table 5.

Table 5 about here

The Canadian study

The fifteen items of the DBQ and the additional cellular telephone use item were

analysed, using principal components with varimax rotation. Although six

components had eigenvalues greater than 1, the first accounted for most of the

variance (19%) and the screeplot indicated that a one-component solution was most

appropriate. Of the 16 items, eight had loadings above 0.40 on this component (see

Appendix 4). Thus, a component-based DBQ scale was created using the eight items

that loaded on the first component (alpha = .63). This scale was a mix of items that

were mistakes, slips and violations. In addition, a violation scale was created by

summing across all six items of this type. The associations between the accident

variables and the DBQ variables are shown in Table 6.

Tables 6 about here

Discussion

The findings across this four-part study can be summarised as thus: if a

significant association between any DBQ factor or scale and accidents was found, it

was with a self-reported variable, not objectively recorded data, despite the higher

18
means and validities of the latter in all samples. It must be pointed out that there were

values for the recorded accident variables that approached significance at times, and

in a few instances, the effects were slightly stronger for the recorded variables than

for the self-report data. However, this is to be expected, as recorded data had higher

means than self-reported data.

These findings lead to the need to take a closer look at the DBQ, this

very popular questionnaire with its often-claimed success as a psychometric

instrument and accident predictor. Given the statements of various researchers in this

area (e.g. "...small differences in factor structures have been reported." (Lajunen,

Parker & Summala, 2004, p. 232)), it could be expected that the DBQ should be

homogenous in its content and results. However, the studies using the DBQ actually

show a bewildering array of different combinations of items, factors, statistical

methods and results.

Starting with items, these differ both in terms of quantity and content.

For example, the number included by Özkan, Lajunen, Chliaoutakis, Parker and

Summala (in press) was 19, Lajunen and Summala (2003) included 27, Sümer (2003)

used 28, Rimmö and Åberg (1999) had 32, Åberg and Rimmö (1998) 104, while

Kontogiannis, Kossiavelou and Marmaras (2002) included 112 items. Similarly, the

number of factors and their labels differ wildly between studies. The original three

factors of violations, errors and lapses (Reason, Manstead, Stradling, Baxter &

Campbell, 1990) have been split further into, for example, highway code violations

and aggressive violations (Parker, Lajunen & Stradling, 1998 mistakes, inattention

and inexperience (Åberg & Rimmö, 1998), while Blockey and Hartley (1995) found

two error factors; general and dangerous. Parker, McDonald, Rabbitt and Sutcliffe

(2000) reported five factors, and Kontogiannis, Kossiavelou and Marmaras (2002)

19
seven. See further the discussion about this in Özkan et al (in press).

It is also interesting to note that researchers have used many different

ways of factoring the DBQ, but none of them have clearly stated why a certain type of

analysis was used. As the various methods differ somewhat in their results on a given

set of data, these unexplained differences are troublesome, as it is possible that results

would not have been the same if a single type of analysis had been used. Still, it is

often claimed that the DBQ factor structure has been replicated many times (e.g.

Rimmö & Åberg, 1999; Parker, Lajunen & Stradling, 1998; Gras et al, 2006). Part of

this replication phenomenon may be due to the use of methods suited to the occasion.

Another result which is often stated by DBQ researchers is that

"...mainly violations not errors or lapses Summala, 2004, p. 231; see also Sullman,

Meadows & Pajo, 2002 for the same type of statement). However, counting the

number of studies with different results indicate that errors and lapses, taken together,

have been significant predictors of accidents about as many times as the various

violation factors (Parker, Reason, Manstead & Stradling, 1995; Parker, West,

Stradling & Manstead, 1995; Lawton, Parker, Stradling & Manstead, 1997; Meadows,

Stradling & Lawson, 1998; Dobson, Brown, Ball, Powers & McFadden, 1999 Rimmö

& Åberg, 1999; Parker, McDonald, Rabbitt & Sutcliffe, 2000; Kontogiannis,

Kossiavelou & Marmaras, 2002; Mesken, Lajunen & Summala, 2002; Sümer, 2003;

Özkan & Lajunen, 2005; Gras et al, 2006; Özkan et al, in press; Bener, Özkan &

Lajunen, in press). Only Blockey and Hartley (1995) did not find an effect for any of

the factors. On the other hand, about one third of all DBQ studies have not reported

on accident associations.

It is apparent that the DBQ literature is characterized by much less

coherence than is usually implied by the researchers active in this field, including the

20
power to predict accidents. Specific concerns with the DBQ include questions such as,

“What is the impact of items about hitting things while driving?” (e.g. Chapman,

Roberts & Underwood, 2000, see also items 12 and 31 in Appendix 2, and item 12 in

Appendix 4). Including an item in the independent variable that seems to measure

partly the same thing as the dependent variable would load the investigation in favour

of finding an association.

One limitation of the present study was the small n in the samples used.

However, it should also be remembered that the mean number of accidents and the

variation in professional driver populations used were much higher than those usually

used for DBQ studies, and any effect should therefore be easier to find.

It could also be claimed that the UK study was flawed, because it used a

DBQ version for car drivers on a population which it was not intended for, and the

dependent variable was bus accidents. However, Sullman, Meadows and Pajo (2002)

did find a significant association between truck drivers' self-reported accidents and the

violation factor of a DBQ-version which was not specifically adapted to truck driving.

Also, this possible problem could not explain the findings in the Swedish sample of

the present study, where self-reported bus driving behaviours were predictive of both

self-reported bus and car accidents.

Based on the results of the analyses reported here, it may be concluded

that the items of the DBQ can weakly predict self-reported accidents, but not

company – or state-recorded (i.e. objective) data. These results are in line with the

suggested hypothesis of a common method variance effect, specifically the

consistency motif. This explanation also applies to other driver behaviour

questionnaires, where self-reported attitudes, behaviour, personality, etc. are

correlated with self-reported accidents, and could explain why the DBQ has yielded

21
different accident predictors in different studies. Further, the present findings suggest

that using self-reported data describing behaviours, attitudes, personality, etc. is rather

overstated as a means of understanding the link between behaviour and actual

accident involvement.

In general, the main arguments of those who use self-reported crash data

are that state sources are even worse concerning their validity as a reflection of the

total number, and that significant findings using self-reported accidents are proof that

the real associations must be even stronger, as the dependent variable contains some

error. The problem is that none of the DBQ studies referenced herein has tested this

assumption, and that the present study would seem to indicate that it may be

erroneous. It can also be questioned whether sheer number of accidents is a good

indicator of validity, in terms of what is actually important to predict. Given that state

records tend to contain the most serious accidents, and those are the ones that are most

important to prevent, a method that is not able to predict this kind of traffic accident

involvement is of little use.

The main method employed for testing whether self-reports of crashes

lead to inflated results is to do what has been done here: testing to see if other sources

of the dependent variable yield different results. Unfortunately, few researchers seem

to have used such precaution, with Arthur et al (2001) and Arthur, Bell, Edwards,

Day, Tubre and Tubre (2005) being two of the exceptions. Also, it could be useful to

try to ascertain whether self-reported accidents are actually the same as the ones

found within company/state databases. This too, is very uncommon in the literature,

and so far the evidence points to substantial discrepancies between these sources

(McGuire, 1973; Smith, 1976; af Wåhlberg, 2002a). This leads to an important

question: If the accidents that people report are often not the ones that are found in the

22
company records, then what accidents are they reporting about? No research

concerning this question has been located.

Finally, it is clear from the present study that the DBQ (and other driver

behaviour questionnaires) should be used with caution by both researchers and

practitioners. Despite its popularity, the evidence for its accident predictive power

appears to be slight. It is a common rule of social interaction and logic that those who

make a claim are those who carry the weight of providing proof in favour of it.

Within science, it is equally common, it would seem, that it is those who dispute a

claim who need to do the research to show the error. The challenge now is for those

who rely on self-reported accident data to provide evidence that validates this measure

against objective records. By extension, the validity of self-reports of other data

crucial to traffic safety research, such as self-reported exposure measures, may also

deserve scrutiny.

Acknowledgements

The data used in this paper had been generously made available by a major UK bus company and

Gamla Uppsalabuss in Sweden, and by Cynthia Owsley and Gerald McGwin (University of Alabama)

in the US.

Author notes:
Anders af Wåhlberg graduated with a B. A. in history (1994) and an M. A. in
psychology (1996), and thereafter worked as a research assistant at Center for Risk
Research, Stockholm School of Economics 1996-1998. He received his PhD in
psychology 2006, at the Department of Psychology, Uppsala University, Sweden.
Main research interests are driver behavior, traffic accidents and research
methodology, most often together.

Lisa Dorn graduated with a BSc in Human Psychology in 1987 and was awarded a
PhD on 'Individual and Group Difference in Driving Behaviour' in 1992, both from
Aston University, UK. After several post-doctoral positions, Lisa was appointed as
Director of the Driving Research Group at Cranfield University in 2001. Her research

23
interests are driver training and education especially in the use of self-assessment
methods.

Theresa J.B. Kline is a professor in the Department of Psychology at the University of

Calgary. She has an active research program in the area of team performance,
psychometrics, organizational effectiveness, and work attitudes. Theresa has
published two books on teams, Teams that Lead (2003) and Remaking Teams (1999),
and one on psychometrics, Psychological Testing (2005), and over 50 peer-reviewed
articles. Theresa teaches statistics/methods and organizational psychology at both the
undergraduate and graduate levels. She has an active organizational consulting
practice with projects ranging from individual and organizational assessment to teams
and leadership to strategic alignment.

References

ARTHUR, W.Jr. 1991, Individual differences in the Prediction and Training of

Complex Perceptual-Motor Skill Tasks: The Development and Validation of The

Computer-Administered Test of Visual Selective Attention. Technical report. Texas

A&M University.

ARTHUR, W.Jr., BARRETT, G.V., and ALEXANDER, R.A. 1991, Prediction of

vehicular accident involvement: A meta-analysis. Human Performance, 4, pp. 89-105.

ARTHUR, W.Jr., TUBRE, T.C., DAY, E., SHEEHAN, M.K., SANCHEZ-KU, M.L.,

PAUL, D.S., PAULUS, L.E., and ARCHULETA, K.D. 2001, Motor vehicle crash

involvement and moving violations: Convergence of self-report and archival data.

Human Factors, 43, pp. 1-11.

ARTHUR, W.Jr., BELL, S.T., EDWARDS, B.D., DAY, E.A., TUBRE, T.C., and

TUBRE, A.H. 2005, Convergence of self-report and archival motor vehicle crash

24
involvement data: A two-year longitudinal follow up. Human Factors, 47, pp. 303-

313.

BARKLEY, R.A., MURPHY, K.R., DU PAUL, G. J., and Bush, T. 2002, Driving in

young adults with attention deficit hyperactivity disorder: Knowledge, performance,

adverse outcomes, and the role of executive functioning. Journal of the International

Neuropsychological Society, 8, pp. 655-672.

BLOCKEY, P.N., and HARTLEY, L.R. 1995, Aberrant driving behaviour: errors and

violations. Ergonomics, 38, pp. 1759-1771.

BROWN, P.L., and BERDIE, R.F. 1960, Driver behavior and scores on the MMPI.

Journal of Applied Psychology, 44, pp. 18-21.

CHAPMAN, P., ROBERTS, K., and UNDERWOOD, G. 2000, A study of the

accidents and behaviours of company car drivers. In G. B. Grayson (Ed.),

Behavioural Research in Road Safety X. Crowthorne: Transport Research Laboratory.

CHAPMAN, P., and UNDERWOOD, G. 2000, Forgetting near-accidents: The roles

of severity, culpability and experience in the poor recall of dangerous driving

situations. Applied Cognitive Psychology, 14, pp. 31-44.

CAIRD, J.K., and KLINE, T.J. 2004, The relationships between organizational and

individual variables to on-the-job driver accidents and accident-free kilometres.

Ergonomics, 47, pp. 1598-1613.

25
CHOU, M., MIYAZAKI, A., and MARUYAMA, K. 1994, Causal analysis of

aptitude, driving behavior, and accidents through the use of covariance structure

analysis: Based on self-reported data. Tohoku Psychologica Folia, 53, pp. 21-26.

DOBSON, A., BROWN, W., BALL, J., POWERS, J., and MCFADDEN, M. 1999,

Women drivers‟ behaviour, socio-demographic characteristics and accidents. Accident

Analysis and Prevention, 31, pp. 525-535.

ELANDER, J., WEST, R.J., and FRENCH, D.J. 1993, Behavioral correlates of

individual differences in road-traffic crash risk: An examination of methods and

findings. Psychological Bulletin, 113, pp. 279-294.

Fergusson, D. M., Horwood, L. J., & Boden, J. M. (in press). Is driving under the

influence of cannabis becoming a greater risk to driver safety than drink driving?

Findings from a longitudinal study. Accident Analysis and Prevention, , -.

GOLDING, A.P. 1983, Differential accident involvement: A literature survey.

Council for Scientific and Industrial Research Reports, PERS 356: 44.

GRAS, M.E., SULLMAN, M.J., CUNILL, M., PLANES, M., AYMERICH, M., and

FONT-MAYOLAS, S. 2006, Spanish drivers and their aberrant driving behaviours.

Transportation Research Part F, 9, pp. 129-137.

26
HANSEN, C.P. 1988, Personality characteristics of the accident involved employee.

Journal of Business and Psychology, 2, pp. 346-365.

HARTLEY, L.R., and EL HASSANI, J. 1994, Stress, violations and accidents.

Applied Ergonomics, 25, pp. 221-230.

KING, Y., and PARKER, D. in press, Driving violations, aggression and perceived
consensus. Revue Européenne de Psychologie Appliquée,

KNOUSE, L.E., BAGWELL, C.L., BARKLEY, R.A., and MURPHY, K.R. 2005,

Accuracy of self-evaluation in adults with ADHD: evidence from a driving study.

Journal of Attention Disorders, 8, pp. 221-234.

KONTOGIANNIS, T., KOSSIAVELOU, Z., and MARMARAS, N. 2002, Self

reports of aberrant behaviour on the roads: Errors and violations in a sample of Greek

drivers. Accident Analysis and Prevention, 34, pp. 381-399.

LAJUNEN, T., and SUMMALA, H. 2003, Can we trust self-reports of driving?

Effects of impression management on driver behaviour questionnaire responses.

Transportation Research Part F, 6, pp. 97-107.

LARSON, G.E., and MERRITT, C.R. 1991, Can accidents be predicted? An

empirical test of the Cognitive Failures Questionnaire. Applied Psychology: An

International Review, 40, pp. 37-45.

27
LESTER, J. 1991, Individual differences in accident liability: review of the literature.

Research Report 306. Crowthorne: Transport and Road Research Laboratory.

MAYCOCK, G. and LESTER, J. 1995, Accident liability of car drivers: follow-up

study. In G. B. Grayson (Ed., Behavioural Research in Road Safety V, pp. 106-120.

Transport Research Laboratory, Crowthorne.

MAYCOCK, J., LOCKWOOD, C., and LESTER, J.F. 1991, The accident liability of

car drivers. TRRL Research Report No. 315. Crowthorne: Transport and Road

Research Laboratory.

MCGUIRE, F.L. 1972, Smoking, driver education, and other correlates of accidents

among young males. Journal of Safety Research, 4, pp. 5-11.

MCGUIRE, F.L. 1973, The nature of bias in official accident and violation records.

Journal of Applied Psychology, 57, pp. 300-305.

MCGUIRE, F.L. 1976, Personality factors in highway accidents. Human Factors, 18,

pp. 433-442.

MCGWIN, G., OWSLEY, C., and BALL, K. 1998, Identifying crash involvement

among older drivers: Agreement between self-report and state records. Accident

Analysis and Prevention, 30, pp. 781-791.

28
MOORMAN, R.H., and PODSAKOFF, P.M. 1992, A meta-analytic review and

empirical test of the potential confounding effects of social desirability response sets

in organizational behaviour research. Journal of Occupational and Organizational

Psychology, 65, pp. 131-149.

OSGOOD, C. E., and TANNENBAUM, P. H. 1955, The principle of congruity in the

prediction of attitude change. Psychological Review, 62, pp. 42-55.

OWSLEY, C., BALL, K., SLOANE, M.E., ROENKER, D.L., and BRUNI, J.R. 1991,

Visual/cognitive correlates of vehicle accidents in older drivers. Psychology and

Aging, 6, pp. 403-415.

OWSLEY, C., MCGWIN, G.Jr., and MCNEAL, S.F. 2003, Impact of impulsiveness,

venturesomeness, and empathy on driving in older adults. Journal of Safety Research,

34, pp. 353-359.

OWSLEY, C., MCGWIN, G.J., SLOANE, M.E., WELLS, J., STALVEY, B.T., and

GAUTHREAUX, S. 2002, Impact of cataract surgery on motor vehicle crash

involvement by older adults. JAMA, 288, pp. 841-849.

OWSLEY, C., STALVEY, B., WELLS, J., and SLOANE, M.E. 1999, Older drivers

and cataract: Driving habits and crash risk. Journals of Gerontology Series A,

Biological Sciences and Medical Sciences, 54A. pp. M203–M211.

29
OWSLEY, C., STALVEY, B.T., WELLS, J., SLOANE, M.E., and MCGWIN, J.G.

2001, Visual risk factors for crash involvement in older drivers with cataract. Archives

of Ophthalmology, 119, pp. 881-887.

PARKER, D. 1999, Elderly drivers and their accidents: The Ageing Driver

Questionnaire. In G. Grayson Behavioural Research in Road Safety I, pp. 169-178.

Crowthorne: Transport Research Laboratory.

PARKER, D., MCDONALD, L., RABBITT, P., and SUTCLIFFE, P. 2000, Elderly

drivers and their accidents: the Aging Driver Questionnaire. Accident Analysis and

Prevention, 32, pp. 751-759.

PARKER, D., LAJUNEN, T., and STRADLING, S.G. 1998, Attitudinal predictors of

interpersonally aggressive violations on the road. Transportation Research Part F, 1,

pp. 11-24.

PARKER, D., REASON, J.T., MANSTEAD, A.S., and STRADLING, S. G. 1995,

Driving errors, driving violations and accident involvement. Ergonomics, 38, pp.

1036-1048.

PECK, R.C. 1993, The identification of multiple accident correlates in high risk

drivers with specific emphasis on the role of age, experience and prior traffic violation

frequency. Alcohol, Drugs and Driving, 9, pp. 145-166.

30
PODSAKOFF, P. M., MACKENZIE, S. B., LEE, J. Y., & PODSAKOFF, N. P.

2003, Common method biases in behavioral research: A critical review of the

literature and recommended remedies. Journal of Applied Psychology, 88, pp. 879-

903.

REASON, J. 1987, Generic Error-Modelling System (GEMS): a cognitive framework

for locating common human error forms. In J. Rasmussen, K. Duncan, and J. Leplat,

New technology and human error. Chichester: Wiley.

REASON, J., MANSTEAD, A., STRADLING, S., BAXTER, J., and CAMPBELL,

K. 1990, Errors and violations on the roads: a real distinction? Ergonomics, 33, pp.

1315-1332.

RIMMÖ, P.-A. and ÅBERG, L. 1999, On the distinction between violations and

errors: Sensation seeking associations. Transportation Research Part F, 2, pp. 151-

166.

SCHUSTER, D.H., and GUILFORD, J.P. 1962, The psychometric prediction of

problem drivers. Traffic Safety Research Review, 6, 16-20.

SCHWEBEL, D.C., SEVERSON, J., BALL, K.K., and RIZZO, M. in press,.

Individual differences factors in risky driving: The roles of anger/hostility,

conscientiousness, and sensation seeking. Accident Analysis and Prevention

31
SIGNORI, E.I., and BOWMAN, R.G. 1974, On the study of personality factors in

research on driving behavior. Perceptual and Motor Skills, 38, 1067-1076.

SMITH, D.I. 1976, Official driver records and self-reports as sources of accident and

conviction data for research purposes. Accident Analysis and Prevention, 14, 439-442.

SMITH, D.I., and KIRKHAM, R.W. 1982, Relationship between intelligence and

driving record. Accident Analysis and Prevention, 14, 439-442.

SOBEL, R., and UNDERHILL, R. 1976, Family disorganization and teenage auto

accidents. Journal of Safety Research, 8, pp. 8–18.

SPECTOR, P.E. 2006, Method variance in organizational research: Truth or urban

legend? Organizational Research Methods, 9, pp. 221-232.

SULLMAN, M..J., MEADOWS, M.L., and PAJO, K.B. 2002, Aberrant driving

behaviours amongst New Zealand truck drivers. Transportation Research Part F, 5,

pp. 217-232.

SŰMER, N. 2003, Personality and behavioral predictors of traffic accidents: testing a

contextual mediated model. Accident Analysis and Prevention, 35, pp. 949-964.

SZLYK, J.P., SEIPLE, W., and VIANA, M. 1995, Relative effects of age and

compromised vision on driving performance. Human Factors, 37, pp. 430-436.

32
WESTERMAN, S.J., and HAIGNEY, D. 2000, Individual differences in driver stress,

error and violation. Personality and Individual Differences, 29, pp. 981-998.

AF WÅHLBERG, A.E. 2002a, On the validity of self-reported traffic accident data.

E140 Proceedings of Soric'02. Available at www.psyk.uu.se/hemsidor/busdriver

AF WÅHLBERG, A.E. 2002b, Characteristics of low speed accidents with buses in

public transport. Accident Analysis and Prevention, 34, pp. 637-647.

AF WÅHLBERG, A.E. 2003, Some methodological deficiencies in studies on traffic

accident predictors. Accident Analysis and Prevention, 35, pp. 473-486.

AF WÅHLBERG, A.E. 2004, Characteristics of low speed accidents with buses in

public transport. Part II. Accident Analysis and Prevention, 36, pp. 63-71.

AF WÅHLBERG, A.E., and DORN, L. 2007, Culpable versus non-culpable traffic

accidents; what is wrong with this picture? Journal of Safety Research, 38, pp. 453-

459.

ÅBERG, L., and RIMMÖ, P.-A. 1998, Dimensions of aberrant driver behaviour.

Ergonomics, 41, pp. 39-56.

ÖZKAN, T., and LAJUNEN, T. 2005, A new addition to DBQ: Positive Driver

Behaviours Scale. Transportation Research Part F, 8, pp. 355-368.

33
ÖZKAN, T., LAJUNEN, T., and SUMMALA, H. 2006, Driver Behaviour

Questionnaire: A follow-up study. Accident Analysis and Prevention, 38, pp. 386-395.

ÖZKAN, T., LAJUNEN, T., CHLIAOUTAKIS, J.E., PARKER, D., and

SUMMALA, H. in press, Cross-cultural differences in driving behaviours: A

comparison of six countries. Transportation Research Part F,

34
Appendix 1: The items included in the U.S. Study Error factor.
Component loadings from a principal components analysis with varimax rotation.
Item (numbers from original) Loading
14. Miss your exit on an interstate and have to make a lengthy detour -0.52
20. Try to change lanes without first checking your mirror, and then be honked at by the car -0.47
behind you which is already in the lane
24. In a line of vehicles turning left on to a main road, pay such close attention to the -0.54
oncoming traffic that you nearly hit the car in front of you
27. Lost in thought or distracted, you fail to notice someone on the side waiting to cross in a -0.53
pedestrian crossing and you drive through the crosswalk
29. Misjudge the speed of the oncoming vehicle when passing a car -0.65
31. Fail to notice someone stepping out from behind a bus or parked car until it is nearly too -0.56
late
33. Pass a single line of stationary or slow-moving vehicles, only to discover that they were -0.49
in a line to get through one-lane construction zone
35. Cut the corner on a left-hand turn and have to swerve to avoid an oncoming vehicle -0.62
37. Fail to read the signs correctly at an intersection and turn onto the wrong road -0.57
39. Ignore the yield signs and narrowly avoid colliding with traffic that has the right of way -0.64
41. Attempt to pass a vehicle that you hadn‟t noticed was signaling its‟ intention to turn left -0.62
45. Fail to notice pedestrians crossing when turning into a side-street from a main road -0.67
48. Brake too quickly on a slippery road and skid -0.56
49. Misjudge your turning space when turning left and narrowly miss collision -0.72

35
Appendix 2: The items of the UK sample
Component loadings from a principal components analysis with varimax rotation. N=238.
Item Component Component Component
1 loadings 2 loadings 3 loadings
1. Attempt to drive away from traffic lights in third gear 0.48
2. Check your speedometer and discover that you are 0.33
unknowingly travelling faster that the legal limit
3. Lock yourself out of your car with the keys still inside 0.37
4. Become impatient with a slow driver in the outer lane and 0.72
overtake on the inside
5. Drive as fast along country roads at night on dipped lights 0.48
as on full beam
6. Attempt to drive away without first having switched on the 0.50
ignition
7. Drive especially close or „flash‟ the car in front as a signal 0.68
for that driver to go faster or get out of your way
8. Forget where you left your car in a multi-level car park 0.34
9. Distracted or preoccupied, realise belatedly that the vehicle 0.53
ahead has slowed, and have to slam on the brakes to avoid a
collision
10. Intend to switch on the windscreen wipers, but switch on 0.49
the lights instead, or vice versa
11. Turn left on to a main road into the path of an oncoming 0.45
vehicle that you hadn‟t seen, or whose speed you had
misjudged
12. Misjudge your gap in a car park and nearly (or actually) 0.46
hit adjoining vehicle
13.‟Wake up‟ to realise that you have no clear recollection of 0.54
the road along which you have just travelled
14. Miss your exit on a motorway and have to make a lengthy 0.44
detour
15. Forget which gear you are currently in and have to check 0.59
with your hand
16. Stuck behind a slow-moving vehicle on a two-lane 0.51
highway, you are driven by frustration to try to overtake in
risky circumstances
17. Intending to drive to destination A, you „wake up‟ to find 0.50
yourself en route to B, where the latter is the more usual
journey
18. Take a chance and cross on lights that have turned red 0.54
19. Angered by another driver's behaviour, you give chase 0.40
with the intention of giving him/her a piece of your mind
20. Try to overtake without first checking your mirror, and 0.57
then get hooted at by the car behind which has already begun
it‟s overtaking manoeuvre
21. Deliberately disregard the speed limits late at night or 0.49
very early in the morning
22. Forget when your road tax/insurance expires and discover 0.40
that you are driving illegally
23. Lost in thought, you forget that your lights are on full 0.56
beam until „flashed‟ by other motorists
24. On turning left, nearly hit a cyclist who has come up on 0.56
your inside
25. In a queue of vehicles turning left on to a main road, pay 0.60
such close attention to the traffic approaching from the right
that you nearly hit the car in front
26. Drive back from a party, restaurant, or pub, even though 0.36
you realise that you may be over the legal blood-alcohol limit
27. Have an aversion to a particular class of road user, and 0.41

36
indicate your hostility by whatever means you can
28. Lost in thought or distracted, you fail to notice someone 0.51
waiting at a zebra crossing, or a pelican crossing light that has
just turned red
29. Park on a double-yellow line and risk a fine 0.42
30. Misjudge speed of oncoming vehicle when overtaking 0.59
31. Hit something when reversing that you had not previously 0.37
seen
32. Fail to notice someone stepping out from behind a bus or 0.31
parked vehicle until it is nearly too late
33. Plan your route badly, so that you meet traffic congestion 0.33
you could have avoided
34. Overtake a single line of stationary or slow-moving 0.35
vehicles, only to discover that they were queuing to get
through a one-lane gap or roadwork lights
35. Overtake a slow-moving vehicle on the inside lane or hard 0.28
shoulder of a motorway
36. Cut the corner on a right-hand turn and have to swerve 0.62
violently to avoid an oncoming vehicle
37. Get into the wrong lane at a roundabout or approaching a 0.61
road junction
38. Fail to read the signs correctly, and exit from a 0.56
roundabout on the wrong road
39. Fail to give way when a bus is signalling its intention to 0.36
pull out
40. Ignore „give way‟ signs, and narrowly avoid colliding 0.63
with traffic having right of way
41. Fail to check your mirror before pulling out, 0.67
changing lanes, turning, etc
42. Attempt to overtake a vehicle that you hadn't noticed was 0.51
signalling its intention to turn right
43. Deliberately drive the wrong way down a deserted one- 0.57
way street
44. Disregard red lights when driving late at night along 0.56
empty roads
45. Drive with only „half-an-eye‟ on the road while looking at 0.47
a map, changing a cassette or radio channel, etc
46. Fail to notice pedestrians crossing when turning into a 0.46
side-street from a main road
47. Get involved in unofficial „races‟ with other drivers 0.57
48. „Race‟ oncoming vehicles for a one-car gap on a narrow 0.54
or obstructed road
49. Brake too quickly on a slippery road and/or steer the 0.56
wrong way in a skid
50. Misjudge your crossing interval when turning right and 0.64
narrowly miss collision

37
 Appendix 3: The Swedish sample Bus Driver Behaviour Questionnaire
New, bus-specific, items are marked with *.
Component loadings from principal components analysis with varimax rotation.

Instruction:
Estimate how often you do the following when driving a bus, on a six-point scale from 'Never' to
'Very often'.
C: When I drive a bus, it happens that I... Type of behavior Component Component
1 2
loadings loadings
1. Deliberately exceed speed limit on main roads during low traffic Violation 0.76
2. Queuing, you pay such close attention to the mainstream of traffic Slip 0.56
that you nearly hit the car in front.
3. Exceed the speed limit to catch up or avoid being late * Violation 0.76
4. Fail to notice "green arrow" at traffic signal allowing you to turn Slip 0.68
5. Fail to check your rear-view mirror before pulling out from a bus stop Mistake 0.58
*
6. Fail to see passengers at a stop and have to brake strongly * Slip? 0.59
7. Deliberately turn onto a main road just in front of oncoming vehicle Violation 0.64
although no other traffic is approaching
8. Turning just in front of pedestrian although he's got "walk now" Violation? 0.54
signal

D: When I drive a bus, it happens that I... Type of behavior Component 1 Component 2
loadings loadings
1. Deliberately disregard speed limit to follow traffic flow Violation 0.72
2. Take more passengers than allowed * Violation 0.49
3. Forget to loosen the parking brake when driving off Lapse 0.64
4. Forget to stop at a bus stop when a passenger want to leave * Lapse 0.56
5. Talk to passengers or colleagues while driving * Violation 0.47
6. Sell tickets while the bus is moving * Violation 0.42
7. Fail to notice when a traffic-signal turns green Slip 0.61
8. Drive over the curb with your rear wheels * Slip 0.63
9. Drive to the wrong terminus in the morning * Lapse 0.44
10. Take a wrong turn en route * Lapse 0.67
11. Try to drive away despite the back doors being open * Lapse 0.68
12. Read the time table wrong and start at the wrong time * Mistake 0.65
13. Feel unsure about the route you are driving * Lapse 0.58
14. Force a cyclist, who has come up along the bus, sideways * Violation? 0.67
15. Drive especially close to the car in front as a signal to its Violation 0.63
driver to go faster or get out of the way
16. Forget to dip the lights when driving during dark hours and is Lapse 0.49
reminded by other drivers flashing their lights
17. Speed up at traffic lights at a green /yellow phase Violation 0.68
18. Turn left onto a main road into the path of an oncoming Mistake 0.54
vehicle that you hadn't seen, or whose speed you misjudged
19. Intend to reverse and find that the bus is moving forward Slip 0.39
because it's in the wrong gear.
20. Misjudge the road surface and when braking find the distance Slip 0.45
to stop to be longer than you expected
21. Switch on e.g. wipers, when you meant to switch on Slip 0.50
something else, such as headlights
22. Fail to the read the day communication and drive the wrong Slip 0.58
way at a road work area *
23. Turn left onto a main road into the path of an oncoming Slip 0.50
vehicle that you hadn't seen, or whose speed you misjudged
24. Cut the bends and occasionally drive in your right lane when Violation 0.34

38
driving in rural areas even though your sight is short

Comments on some of the bus-specific items:

C5: This item was only slightly changed, from pulling out and changing lanes in general, to the more
specific situation of the bus stop.
D8: When driving a long vehicle, like a bus, it is very important not to start turning before the rear
wheels have passed all obstacles on the side that you are turning to, like the corner of the curb, or you
will hit it.
D5: This is not allowed, but very common.
D11: In Sweden, the brakes on buses are automatically on when any door is open. The back doors are
normally on automatic and will close when someone has passed through. However, if, for example,
someone is unloading something, like a baby carrier, the driver will open the doors manually. After
doing this, the driver will often have to attend to other passengers entering through the front door, and
forget that the back door is on manual and therefore still open when he has closed the front door and
tries to move off.
D14: A slight alteration of the original "On turning right, nearly hit a cyclist who come up on your
inside". Also, this item is particularly pertinent in Uppsala, a university town with many cyclists.
D22: The daily communication consists of messages that the drivers are supposed to read before they
start their day. These messages often concern temporary changes in the routes.

39
Appendix 4: The Canadian sample DBQ (items and factor loadings)
Component loadings from principal components analysis with varimax rotation.
Item Loading
1. Become impatient with a slow driver and pass on the right .239
2. Drive too closely to the car in front as a signal to its driver to go faster or get out of .368
the way
3. Attempt to pass someone on an undivided road that you hadn't noticed to be signaling .435
a left turn (E)
4. Run yellow lights (V) .510
5. Fail to notice pedestrians are crossing .380
6. Angered by another driver's behaviour, you yell or gesture at them (V) .406
7. Choose to exceed the speed limits when the police are not around (V) .420
8. On turning right, nearly hit a cyclist who has come up on your inside .381
9. When merging, you pay such close attention to the main stream of traffic that you .501
nearly hit the car in front of you (E)
10. Drive even though you consumed alcohol .373
11. Underestimate the speed of an oncoming vehicle when passing on an undivided road .613
(E)
12. Hit something while backing up that you had not previously seen .374
13. Miss "Yield" or "Stop" signs; narrowly avoiding a collision (E) .598
14. Fail to check your rearview mirror before pulling out, changing lanes, etc. .230
15. Brake too quickly on a slippery road, or steer the wrong way into a skid (E) .647
16. I use the cellular phone while driving .304
Note: Italicised items were used to create the DBQ scale, V = violation, E = error

40
 Table 1: Descriptive data for the four samples; number of drivers, percentage male drivers, percentage
ethnic mix, mean /std for age and experience.
Sample Number of Sex (percent Ethnicity Age Experience (years of holding a PCV
drivers male drivers) license)
U.S. 307 49.3% 85.3% (white) 69.6/6.4 -
U.K. 238 90.8% 95.8% (white) 46.7/10.7 12.7/11.7
Sweden 141 84.8% 65.2% (Swedish 45.6/9.9 11.8/8.9
names)
Canada 153 97.0% - 42.6/6.9 --

41
Table 2: The means and standard deviations of the accident variables and effects on these for the error
component (see Appendix 1) in the U.S. sample. Correlations between variables, and t-values (mean on
the factor scores categorized by accidents dichotomised into none/some).
Variable Self-reported, one year State recorded, ten years
All Crashes, police present All Culpable
crashes crashes crashes
N=291 Mean 0.13/0.55 0.07/0.32 0.29/0.59 0.20/0.49
Error Correlation .07 .07 .03 .01
Component
N=291 t -2.38* -1.61 1.16 -0.01
Violation scale Correlation .04 -.04 .08 .06
N=222 t -0.87 0.57 -1.96 -1.29
„ p<.05, one-tailed test

42
Table 3: The intercorrelations (Spearman) between the accidents variables of the U.S. study. N=300.
Variable All recorded crashes, Culpable recorded crashes, All self-reported, 1
10 years 10 years year
Culpable recorded crashes, .83*** -
10 years
All self-reported, 1 year -.17** -.14* -
Self-reported police -.14* -.11 .79***
recorded, 1 year
*p<.05, **p<.01, ***p<.001, one-tailed test

43
Table 4: The means and standard deviations of the (company-recorded) accident variables and effects
on these for the DBQ components in the U.K. sample. Correlations between variables and t-values
(mean on the component scores categorized by accidents dichotomised into none/some). Bus accidents
for the total time period of employment of each driver, as shown in Table 1. No effects significant at
p<.05 (one-tailed tests).
Variable N=238 All accidents All responsible accidents Solely responsible
accidents
Mean/std 2.84/2.62 1.43/1.55 1.27/1.45
Component 1 Correlation -.04 -.02 -.02
t 1.03 -0.95 -0.58
Component 2 Correlation .07 .04 .06
t -0.34 -0.31 -0.58
Component 3 Correlation .04 .00 .01
t -1.27 1.23 1.11
Violation Correlation .06 .03 .05
scale t -0.95 0.34 0.25
Variable N=221 All All responsible Solely responsible
accidents/year accidents/year accidents/year
Mean/std 0.87/.125 0.48/0.90 0.44/0.85
Component 1 Correlation .00 .00 .01
Component 2 Correlation .11 .05 .06
Component 3 Correlation .04 -.02 -.02
Violation Correlation .10 .03 .04
scale

44
Table 5: The means and standard deviations of the accident variables and effects on these for the DBQ
components in the Swedish sample. Correlations between variables and t-values (mean on the factor
scores categorized by accidents dichotomised into none/some). All accident variables cover three years.
Variable Self-reported bus Recorded bus Self-reported car
accidents (N=141) accidents (N=141) accidents (N=130)
Mean/std 0.71/0.91 0.79/1.06 0.14/0.39
Component 1 Correlation .21* -.05 .13
t -2.47* 0.79 -2.05*
Component 2 Correlation -.07 -.02 -.06
t 0.25 0.90 1.11
Violation Correlation .00 -.01 .01
scale t -1.06 0.44 0.08
* p<.05, one-tailed test

45
Table 6: The means and standard deviations of the accident variables and effects on these for the DBQ
component and violation scale in the Canadian sample. Correlations between variables and t-values
(mean on the factor score/scale categorized by accidents/violations dichotomised into none/some).
N=115.
Self-reported Recorded
Variable Accidents, Near Violations, Accident- Accidents, Accident-
two years accidents, two years free km, total free km
one two years
month
Mean/std 0.09/0.34 0.31/0.74 0.28/0.54 75 0.55/0.70 491 308/251
388/32 551
722
Component Correlation .10 .11 .18 -.03 -.02 -.12
1 t -1.26 -1.72 -1.86 - 0.44 -
Violation Correlation .16 .04 .19* -.13 .02 -.02
scale t -2.19* -0.63 -1.99* - -0.29 -
* p<.05, one-tailed test

46