category: literaturenote citekey: verschuurmodelingsafeunsafedriving2008 title: Modeling safe and unsafe driving behaviour authors: "Verschuur, William L.G.; Hurts, Karel" year: 2008 date: 2008-00-00 2008 doi: 10.1016/j.aap.2007.09.001 publication: "Accident Analysis & Prevention" url: "https://doi.org/10.1016/j.aap.2007.09.001" zotero_key: IFPXIQQ9 zotero_storage: IQHHKFCE collections: transpordi_planeerimine folder: 001_artiklid firstAuthor: "Verschuur, William L.G."

status: converted

Accident Analysis and Prevention 40 (2008) 644–656

Modeling safe and unsafe driving behaviour

William L.G. Verschuur, Karel Hurts ∗

Department of Psychology, Leiden University, P.O. Box 9555, 2300 RB, Leiden, The Netherlands Received 21 February 2007; received in revised form 23 August 2007; accepted 3 September 2007

Abstract

A sample of 743 Dutch drivers was queried about driving-related errors and violations, and about the frequency of their accident involvement over the past 3 years. In addition, the following risky behaviours and characteristics related to driving were measured through self-report: strategic decisions made about driving before starting a trip, attitudes related to committing violations, psychological precursors of unsafe driving (such as being tired or stressed when driving), and physical precursors of unsafe driving (suffering physical or psychological handicaps when driving). The results show that several scales have acceptable reliability coefficients, though several others are in need of improvement. Path analysis showed that an acceptable structural model could be developed to describe all empirical relationships. Considerable parts of the variance in the violations scores and the psychological precursors scores could be explained by the other model variables. In terms of predictive power, "unsafe" attitudes, physical precursors, and psychological precursors had the most impact, each explaining up to 9% of the variance in violations scores, errors scores, and/or psychological precursors scores. The way these results can be utilized in future accident prevention programs and in driver training programs are indicated, as are their implications for future research.

Keywords: Psychological modeling; Human error; Traffic safety; Accident causation

1. Introduction

In this study we will describe an attempt to improve on Reason's (1990, 1997) accident causation model through a large-scale investigation into the driving-related behaviours of drivers. We felt Reason's model is not explicit about several factors that are important for predicting and understanding the behavioural aspects of accidents, such as type of unsafe acts preceding an accident, and the specific, more or less permanent, attitudes and behaviours that facilitate or inhibit safe driving behaviour. The consequence is that the model is somewhat underspecified and, therefore, difficult to use to generate and test hypotheses about the specific causes of unsafe driving behaviour under specific circumstances and/or for specific driver populations.

2. Study objective

We therefore set out to develop a driver behaviour model that compensates for the above-mentioned shortcomings by adding the following constructs to Reason's (1990, 1997) accident causation model, relating to:

1. The particular type of human failure preceding an accident. To this end, we borrowed constructs from the Driver Behaviour Questionnaire.
2. Drivers' attitudes, norms and control beliefs, using constructs derived from the Theory of Planned Behaviour. For the empirical part of the study, these constructs were "instantiated" to driving-related violations (i.e., the attitudes, norms and control beliefs that would be measured were about driving-related violations only).
3. The strategic, driving-related decisions drivers usually make before making a trip.
4. The physical state the driver is usually in when driving a car.

The empirical part of the study consisted of an extensive survey using a large sample of drivers (n = 743) that were queried about their (self-reported) driving-related errors and violations, their (self-reported) degree of accident involvement, and the attitudes and behaviours mentioned above under 2, 3 and 4. A construct called Psychological precursors, similarly defined as in Reason's accident causation model, was also included in the survey. Note that these attitudes and behaviours can all be seen as

^∗ Corresponding author. Tel.: +31 71 527 3489; fax: +31 71 527 3783. E-mail address: hurts@fsw.leidenuniv.nl (K. Hurts).

examples of risky behaviours and characteristics. The model to be developed and tested will be referred to as the safety-related driver behaviour model.

Demographic variables, like age and gender and other variables of this kind will be treated as background variables, moderating the relationships between the remaining (core) concepts of the model.

Two limitations and assumptions underlie our approach. These are explained below.

1. Emphasis on psychological causes or precursors of safetyrelated behaviour. In other words, we will not look at environmental or other causes of accidents which are external to the individual driver (e.g., road condition or the physical state of the vehicle).
2. Emphasis on individual variables. In other words, all variables are measured at the level of the individual driver. Interactions with other traffic participants, or with vehicle- or environment-related characteristics will not be considered.

In the next section all theories and frameworks are discussed in detail.

3. Existing frameworks and the safety-related driver behaviour model

3.1. Reason's accident causation model applied to driver behaviour

The idea that an accident is directly preceded by unsafe actions but is also preceded by causes more remote in time and place from the accident is advocated byReason (1990, 1997) and by Wagenaar and Reason (1990). Reason's accident causation model was primarily developed to understand organizational accidents.

In this model unsafe acts are also referred to as active failures, which are understood as suboptimal driving behaviours, such as errors (slips, mistakes) and violations. The more remote causes of accidents are called latent failures, usually understood as faulty management decisions and related deficiencies in the management system (e.g., poor road design).

Latent failures can only result in active failures if certain psychological preconditions exist, called psychological precursors. Examples of these preconditions are inattention and stress. Psychological precursors may be related to or caused by latent failures (e.g., lack of training), but may also be caused by other factors having nothing to do with the organization or system under consideration (e.g., fatigue due to lack of sleep or a temporary depression caused by personal problems).

In Reason's (1990, 1997) accident causation model active failures may eventually result in an accident if the last (material or procedural) defenses are broken. However, defenses, not being a typical individual or psychological variable, were not included in our model.

3.2. The driver behaviour questionnaire (DBQ)

The DBQ was originally proposed as a reactive tool, supplying information about the frequency of occurrence of aberrant behaviours for various subgroups of drivers. DBQ constructs expand the active failures component of Reason's (1990, 1997) accident causation model, as will be shown below.

The DBQ aims at counting and classifying unsafe driving behaviours typically experienced by drivers over the past year in the form of a self-report questionnaire (Lawton et al., 1997a,b; Parker et al., 1998, 2000). Behaviours are typically classified as errors or violations, which is also consistent with Reason's (1990) Generic Error Modeling System (GEMS).

Though the DBQ has been investigated intensively in the UK, it has also been used in many other countries inside and outside Europe, such as Sweden (Aberg and Rimm ˚ o, 1998 ¨ ), Australia (Blockey and Hartley, 1995), Greece (Kontogiannis et al., 2002), China (Xie and Parker, 2002), Finland (Mesken et al., 2002; Ozkan et al., 2006 ˝ ), The Netherlands (Verschuur, 2003a,b), France (Obriot-Claudel and Gabaude, 2004), and Brazil (Bianchi and Summala, 2002). The DBQ has also been used to investigate or differentiate specific groups of drivers, such as truck and other professional drivers (Sullman et al., 2002, 2004; Xie and Parker, 2002), older drivers, or male (versus female) drivers (Obriot-Claudel and Gabaude, 2004; Parker et al., 1998, 2000; Rimmo, 2000; Westerman and Haigney, 2000 ¨ ). In a cross-cultural study Lajunen et al. (2004) showed that the DBQ has similar psychometric properties in a number of European countries.

Several studies showed there is a relationship between violations and accidents (Parker et al., 1995a,b; Underwood, 1997) and, although less frequently, between errors and accidents (Mesken et al., 2002; Sullman et al., 2004; Verschuur, 2003a,b).

A study among 2002 Dutch drivers showed that the questionnaire's items could roughly be described in terms of four underlying factors: violations, dangerous errors, inattention errors, and a lapse factor, explaining 15.1%, 9.9%, 4.1%, and 3.3%, respectively, of the variance in the responses given. The version of the DBQ used covers items of the Manchester and the Swedish DBQ (Aberg and Rimm ˚ o, 1998; Rimm ¨ o,¨ 2000). The corresponding alphas of the factors were 0.88, 0.77, 0.70, and 0.72, respectively. In general, the factor structure was similar to the one found earlier in the Swedish study.

3.3. Theory of planned behaviour (TPB)

The TPB considers the origins of everyday behaviours that are largely, but not completely, under voluntary control (Ajzen, 1991). According to the theory, a behavioural intention is formed by a combination of the following factors: an attitude towards the intended behaviour (indicating the extent to which the behaviour is considered attractive and/or useful), a subjective or personal norm regarding the behaviour (indicating the extent to which the behaviour is considered socially or personally acceptable), and perceived behavioural control (indicating the extent to which one thinks one can actually carry out the behaviour).

Originally, we intended to add constructs borrowed from the TPB in an attempt to better predict violations (one of the constructs underlying the DBQ). In other words, the TPB was used as a proactive tool (Ajzen, 1991), i.e., to enable us to develop countermeasures or preventive measures with which to avoid violations in the future.

The theory has been used to explain driver behaviour, more specifically various types of traffic violation (Manstead and Parker, 1995; Parker et al., 1992a,b, 1995a,c, 1996; Rothengatter and Manstead, 1997; Stradling and Parker, 1997). Nevertheless, this application of TPB has been criticised, because of discrepancies in the contributions that have been reported for attitudes and social norms as predictors of driving intentions and behaviour. Possible reasons for these discrepancies are differences in the methodology used, in the specification of the target behaviour, and cultural differences (Rothengatter and Manstead, 1997).

3.4. The safety-related driver behaviour model

Fig. 1 contains the model we came up with for explaining (un)safe driving behaviours and which we will refer to as the safety-related driver behaviour model. In Section 4 we will discuss the instruments we used or developed for measuring the model variables.

Starting on the right-hand (outcome) side, we consider Accident involvement to be the final behaviour to be predicted by the model and to be avoided by means of to-be-developed future intervention programs.

Accident involvement is directly preceded by Unsafe acts, which are broken down in the following constituent behaviours (based on the DBQ studies summarized above): Violations, Dangerous errors, and Inattention errors.

Next, we added the following two types of precursors (behaviours or characteristics preceding unsafe behaviour, not

Fig. 1. The safety-related driver behaviour model. Arrows indicate hypothesized relationships to be tested in the field study. Age, gender, and mileage represent moderator variables.

being inherently related to driving):

1. Psychological precursors, referring to the (temporary) psychological state of the driver when driving, e.g., being under the influence of alcohol or being tired. Not all aspects of this state may be under control of the individual driver. However, we assume the driver is always aware of this state while driving. We hypothesize that Psychological precursors increase the likelihood of Inattention errors and Dangerous errors (mainly without awareness of the driver), but not that of Violations (which are more intentional and conscious in nature). They are themselves influenced by Strategic decisions (see below). Note that our definition of psychological precursors is similar to the one used by Reason (1990, 1997).
2. Physical precursors, referring to the (more or less) permanent physical and psychological state of the driver when driving, e.g., sickness, or the presence of physical handicaps. Again, not all aspects of this state may be under control of the driver, but we assume the driver is always aware of this state when driving. Just as we did for Psychological precursors, we predict that Physical precursors affect the occurrence of Inattention errors and Dangerous errors, but not that of Violations.

Finally, we added the following more remote causal behaviours (inherently related to driving):

1. Strategic decisions, defined as the quality of driving-related decisions made by the driver just before a trip is undertaken. Examples of such decisions would be: when to leave home in order to meet an appointment, and what road to choose for a trip. It is hypothesized that these decisions are interdependent with one or more of the TPB constructs (see below). However, in order to indicate our uncertainty about the precise nature if this relationship, we have marked the corresponding arrows in Fig. 1 with a question mark.
2. Attitudes, Subjective/personal norms, and Perceived behavioural control (borrowed from the Theory of Planned Behaviour, Ajzen, 1991). These constructs were only studied with respect to two specific types of violation, i.e., speeding and drunk driving. Obviously, we expect all these TPB constructs to affect the frequency of self-reported Violations (see Fig. 1). However, we do not expect the TPB constructs to have an effect on the frequency of self-reported Dangerous and Inattention errors, because the latter are less intentional in nature. As mentioned above, the question mark in Fig. 1 indicates we are not completely certain about the nature of the relationship between the TPB constructs and Strategic decisions.

These causal behaviours are similar to latent failures, as defined by Reason (1990, 1997). However, because we emphasize the behavioural, not the situational, aspects of these failures, we prefer to call them "causal behaviours", rather than "latent failures".

4. Method

4.1. Participants

A sample of 743 drivers was selected randomly from a large panel with more than 200,000 members. The sample was matched on two variables (age and gender) with mobility statistics of the Dutch Ministry of Transport (MON, 2005).

4.2. Instruments used in the field study

The following psychological variables and behaviours were measured:

Accident involvement was measured using the following semi-open question: "In the last three years, how frequently have you been involved in a traffic accident?" Alternatives: never; material damage only ... times; mild bodily dam-

age (no hospitalization) ... times; serious bodily damage (including hospitalization) or fatality ... times.

2. Active failures: from the DBQ questionnaire we borrowed items for measuring the three factors Dangerous errors, Violations, and Inattention errors, as identified in previous DBQ research (Verschuur, 2003a,b). For each factor we used the eight items from the original questionnaire with the highest factor loadings on the respective factors (see Table 1). Six-point Likert scales were used for measuring the items, running from 1 (never) to 6 (almost always).
3. Strategic decisions was measured by means of nine items on a 5- or 6-point Likert scale (1 = never to 6 = almost always, or: 1 = completely agree to 5 = completely disagree) (see Table 2).
4. Psychological precursors was measured using four items on a 5- or 6-point Likert scale (1 = never to 6 = almost always, or: 1 = completely agree to 5 = completely disagree) (see Table 2).

Table 1 DBQ items and their loadings on factor analysis components

| | DBQ item | Component loadings | | | | | |----|-------------------------------------------------------------------------------------------------------------------------------------------------|--------------------|--------|-------|--|--| | | | V | IE | DE | | | | 1 | Deliberately speeding on a not-busy road | 0.866 | | | | | | 2 | Exceeding the speed limit early in the morning or late at night | 0.848 | | | | | | 3 | Deliberately speeding in order to follow the traffic flow | 0.838 | | | | | | 4 | Deliberately speeding in order to overtake | 0.838 | | | | | | 5 | Pulling up when the traffic lights turn yellow | 0.719 | 0.195 | 0.126 | | | | 6 | Deliberately following a car at close distance in order to make it move
faster | 0.700 | | 0.203 | | | | 7 | Deliberately following a car at close distance in order to make it move to
the side | 0.697 | −0.127 | 0.234 | | | | 8 | Crossing an intersection knowing the traffic light has just tuned red | 0.385 | 0.261 | 0.115 | | | | 9 | Deliberately entering a road just in front of an upcoming car without
necessity | 0.308 | 0.141 | 0.282 | | | | 10 | Missing an exit on the highway forcing you to make a long detour | | 0.663 | | | | | 11 | Failing to notice the traffic light for right-turning traffic, allowing you to
make a right turn, is green | | 0.658 | 0.182 | | | | 12 | Failing to notice the traffic light has turned green | | 0.638 | | | | | 13 | Forgetting where you have parked the car in a parking lot | 0.112 | 0.579 | | | | | 14 | Misreading a traffic sign and ending up not knowing where you are | | 0.554 | 0.184 | | | | 15 | Finding yourself following the wrong road, probably caused by this road
being highly familiar | 0.256 | 0.524 | | | | | 16 | Misreading a traffic sign resulting in your taking a wrong exit on a
roundabout | | 0.487 | 0.130 | | | | 17 | Incorrectly lining up when approaching an intersection or roundabout | | 0.391 | 0.306 | | | | 18 | Misjudging the distance to a car coming from the opposite direction,
causing a near-collision when overtaking a car in front of you | | | 0.644 | | | | 19 | Underestimating the speed of a car coming from the opposite direction,
while overtaking the car in front of you | | | 0.611 | | | | 20 | Misjudging the distance to a car coming from the opposite direction when
you are making a left turn, forcing the other car to use his brakes | 0.157 | 0.198 | 0.607 | | | | 21 | Failing to notice a traffic sign telling you to yield to traffic coming from the
left or right, causing a near-collision with this traffic | 0.104 | 0.255 | 0.535 | | | | 22 | Failing to notice a bicyclist on your right, when making a right turn,
causing a near-collision | | 0.225 | 0.497 | | | | 23 | Failing to notice crossing pedestrians when making a right turn to a
secondary road | | 0.218 | 0.479 | | | | 24 | Failing to notice that the car in front of you, which you are about to
overtake, indicated he was going to the left | 0.112 | | 0.306 | | |

Note. Loadings represent rotated component matrix (N= 717). Extraction method: principal component analysis. Rotation method: varimax with Kaiser normalization. Rotation converged in 5 iterations. Labels of factors (components) based on Verschuur (2003a,b): V: Violations; IE: Inattention errors; DE: Dangerous errors.

Table 2 Items used for measuring strategic decisions, psychological precursors, physical precursors, and the TPB constructs

Strategic decisions

1. During short trips I use the seat belt
2. If I choose a certain route then I do that because I find that route safer
3. Security considerations play an important role when choosing a means of transport
4. Sometimes I make a detour in order to end up driving safer
5. If I make trips of at least 2 h I take 1 or more breaks
6. Passengers in the back of my car must always use a seat belt
7. When choosing a car I take security provisions into account (ABS, airbags, collision tests, etc.)
8. If it is foggy or skiddy I often consider not making my planned trip, or making it later or using another means of transport
9. If the traffic rules are modified, I want to know everything about it

Psychological precursors

1. Sometimes I swallow medicines which can negatively influence driving skill, even when I still have to drive my car
2. I sometimes drink more than 2 glasses of alcoholic beverage before driving my car
3. Even when I feel tired, I will make a trip in my car as planned
4. Even when I am mad at something or someone, I will make trip in my car as planned

Physical precursors

1. Do you think it is difficult once in a while to read traffic signs?
2. Even when I suffer physical complaints, I will make a planned trip in my car
3. Even when I think driving is scary, I will make a planned trip in my car

1. If I speed when driving, I do that to save time	Attitudes
2. If I speed when driving, I do that because it gives me a kick	Attitudes
3. If I speed when driving, it does not affect me or others negatively	Attitudes
4. I am a better driver after having drunk at least three glasses of alcoholic beverage	Attitudes
5. I feel better when driving after having drunk at least three glasses of alcoholic beverage	Attitudes
6. If I drink at least three glasses alcoholic beverage before driving my car, it does not affect me or others negatively	Attitudes
7. I often speed when driving, it simply happens	Attitudes
8. I often drink at least three glasses of alcoholic beverage before driving my car, it simply happens	Attitudes
9. I disapprove of speeding when driving	Subj./pers. norms
10. I disapprove of drinking at least three glasses of alcoholic beverage before driving your car	Subj./pers. norms
11. If others would persuade me not to drink at least three glasses of alcoholic beverage before driving my car, I would consider	Subj./pers. norms
following their advice
12. If others would persuade me not to speed when driving, I would consider following their advice	Subj./pers. norms
13. When I speed when driving, nothing will happen to me because I stay in control of the situation on the road and in my vehicle	Perc. behav. control
14. The presence of traffic police or cameras may keep me from speeding when driving	Perc. behav. control
15. After having drunk at least three glasses of alcoholic beverage, nothing will happen to me because I always know what I am doing	Perc. behav. control
in the car
16. The presence of traffic police or cameras may keep me from drinking more than two glasses of alcoholic beverage before driving	Perc. behav. control
my car

TPB items Construct

5. Physical precursors was measured using three items on a 5- or 6-point Likert scale (1 = never to 6 = almost always, or: 1 = completely agree to 5 = completely disagree) (see Table 2).
6. The TPB constructs (Attitudes, Subjective/personal norms, Perceived behavioural control) were measured by presenting respondents with two examples of traffic-related behavioural violations, i.e., speeding and driving under the influence of at least three alcoholic drinks. The constructs were measured using 16 items on a 5-point Likert scale running from 1 = completely agree to 5 = completely disagree (see Table 2).

4.3. Data analysis

The relationships of Fig. 1 can be thought of as forming a series of simultaneous regression equations, also called path diagram orstructural equations model. Therefore, we used EQS (version 6.1) to analyze and test the relationships between the variables shown in Fig. 1 (Dunn et al., 1993). Because it is assumed that all model variables are measured directly (without measurement error), this type of analysis is also called path analysis.

In structural modeling techniques like EQS a distinction is made between exogenous variables and endogenous variables. The first kind can only influence other variables, but cannot be influenced itself. They may be mutually correlated. Endogenous variables, on the other hand, can also be influenced by other variables (McDonald and Ho, 2002). They are not allowed to be correlated with each other.

According to this definition, all variables shown in Fig. 1 are endogenous, except Physical precursors. Because path analysis does not allow the specification of bi-directional causation, we decided to test the relationship between Attitudes and Strategic decisions (see Fig. 1) by means of the covariance among these two variables. This also had the effect of turning them into exogenous variables from the point of view of path analysis.

Table 3 Comparison of this study's sample with Dutch Mobility Statistics (MON, 2005)

| Age | | Our sample (n = 743) | | MON, 2005 sample (n = 66,481) | | | | |-------|------|----------------------|-------|-------------------------------|--------|-------|--| | | Male | Female | Total | Male | Female | Total | | | 18–34 | 14.8 | 16.1 | 30.9 | 14.2 | 13.7 | 27.9 | | | 35–54 | 21.4 | 22.7 | 44.1 | 22.9 | 20.6 | 43.5 | | | >55 | 13.8 | 11.2 | 25.0 | 16.6 | 12.0 | 28.6 | | | Total | 50.1 | 49.9 | 100.0 | 53.7 | 46.3 | 100.0 | |

Note. All table entries are percentages.

Similarly, we decided to leave the covariances among the TPB constructs free (estimable), as in this theory it is acknowledged that these constructs tend to covary.

5. Results

Before we present the outcomes of the EQS analyses, below we will first address sample characteristics, factor analysis of DBQ items, reliabilities of DBQ scales, and the distributions of the model variables.

5.1. Sample characteristics

Compared to the Dutch Mobility statistics (MON, 2005) younger drivers were slightly overrepresented in our sample (31% versus 28% in the MON sample), and, therefore, older drivers were slightly underrepresented (25% versus 29% in the MON sample) (see Table 3).

In addition, our sample contained more women (50%) than the MON sample did (46%).

Statistical analysis showed only the gender difference to be significant (p < 0.05). However, with respect to age, we can assume the results of our study will be representative of this population.

5.2. Factor analysis of DBQ items

A factor analysis carried on the data (using varimax rotation) results in six factors with an eigenvalue greater than (or equal to) 1.0. The first three factors had eigenvalues of 5.28, 3.07, and 1.62, explaining 22.9%, 13.3%, and 7.1%, respectively, of the overall variance in item responses (see Table 1). In combination, they explain 43.35% of this variance. By and large, these results confirm earlier findings (Verschuur, 2003a,b).

5.3. Reliabilities of scales

The internal consistencies (alphas) of the DBQ scales were as follows: Violations 0.88, Inattention errors 0.72, and Dangerous errors 0.62. In the original Dutch DBQ (Verschuur, 2003a) the alphas for the same scales were 0.88, 0.70, and 0.77, respectively, suggesting that the scale Dangerous errors could be improved if all original 14 (instead of only eight) items would have been included.

The TPB scales had the following alphas: Attitudes 0.71, Subjective/personal norms 0.59, and Perceived behavioural control 0.41.

Table 4 Descriptive statistics, reliability coefficients, and meanings of scales used for measuring model variables

Scale/variable	Meaning	#Items	Alpha	N	M	Std.	Skewness
Attitudes (v1)	A higher (+) score means one disapproves/a lower (−) score that one approves traffic violations	8	0.71	744	32.83	4.21	−0.42
Subj./pers. norms (v2)	A higher (+) score means a social setting that approves, a lower score (−) one that disapproves traffic violations	4	0.59	749	8.53	2.96	0.54
Perceived behav. control (v3)	A lower score (−) means feeling more in control of violation-inducing behaviour, a higher score (+) means the opposite	4	0.41	752	12.64	2.88	0.31
Strategic decisions (v4)	A lower score (−) means one takes less frequently safety-enhancing preventative measures before starting a trip, a higher score (+) means the opposite	9	0.75	751	35.50	8.26	−0.11
Psychological precursors (v5)	A lower score (−) means that one drives less frequently under conditions that might be unsafe (use of medicine, alcohol, fatigued or in rage), a higher score (+) means the opposite	4	0.52	752	5.81	2.32	2.07
Physical precursors (v6)	A lower score (−) means one drives less frequently with physical/mental complaints, whereas a higher score (+) means the opposite	3	0.35	752	3.56	2.53	1.11
Gender	1 = male; 2 = female		N/a	752	N/a	0.50	0.03
Age			N/a	752	43.02	13.54	−0.19
Violations (v7)	A lower score (−) means fewer self-reported violations, a higher score (+) means more violations	8	0.88	752	12.05	4.92	1.48
Inattention errors (v8)	A lower score (−) means fewer self-reported errors of attention, a higher score (+) means more such errors	8	0.72	752	6.65	1.66	0.92
Dangerous errors (v9)	A lower score (−) means fewer self-reported dangerous errors, a higher score (+) means more such errors	8	0.62	752	4.10	1.10	1.40
Accident involvement (past 3 years)	Runs from 0 to 5, number of accidents in this interval		N/a	752	0.23	0.57	3.33
Log Accident involvement (v10)	Log value of previous variable		N/a	752	0.14	0.32	2.19

Table 5 Covariance matrix for the ten model variables (based on 743 cases)

	v1	v2	v3	v4	v5	v6	v7	v8	v9	v10
v1 (Attitudes)	17.68
v2 (Subj./pers. norms)	−5.32	8.71
v3 (Perceived behav. control)	4.57	−1.16	7.95
v4 (Strategic decisions)	11.81	−7.28	5.01	67.35
v5 (Psychol. precursors)	−3.35	1.29	−0.13	−4.60	5.41
v6 (Physical precursors)	−0.30	0.01	−0.43	−1.37	1.55	6.42
v7 (Violations)	−11.32	6.06	−3.68	−16.10	4.49	0.43	24.18
v8 (Inattention errors)	−0.96	0.46	−0.25	−1.03	0.83	1.22	1.72	2.71
v9 (Dangerous errors)	−0.92	0.32	−0.001	−0.68	0.43	0.57	1.42	0.77	1.22
v10 (Log accident involvement)	−0.14	0.09	−0.01	−0.23	0.12	−0.01	0.19	0.02	−.002	0.10

Finally, the scales for Strategic decisions, Physical precursors and Psychological precursors had alphas of 0.75, 0.30 and 0.53, respectively.

5.4. Distributions of model variables

The descriptive statistics of the scales (see Table 4) show that the distributions of some variables are skewed, especially of Psychological precursors and Accident involvement. The values of Accident involvement were transformed by taking the log values, in an attempt to reduce the extreme amount of skewness of this variable. The resulting values are still skewed. As it turns out, accident statistics are generally Poisson-distributed, making it virtually impossible to reduce skewness completely. In addition, regression analysis is usually rather robust against violations of the assumption of having normal distributions.

5.5. Structural equations

5.5.1. Analysis approach

Table 5 contains the covariance matrix of the variables shown in Fig. 1. Initially, EQS was used for estimating the strengths of the relationships shown in Fig. 1. The Lagrange modification index was used in order to gradually (one parameter at a time) improve the overall model's fit, i.e., by adding free parameters of which the index indicates that they help increase the overall fit.

Fig. 2 and Table 6 show the results of the model optimization process where no further improvement was possible. Ten variables were analyzed, including 6 variables which were defined as dependent (endogenous) variables. Altogether, 37 free parameters and 6 fixed nonzero parameters were included.

The numbers indicated near the one-sided arrows of Fig. 2 represent the standardized regression (path) coefficients. The numbers indicated near the double-sided arrows represent correlation coefficients. It turned out that the Chi-square value for this model (45.29) was significant, d.f. = 18, p = 0.00038. Normally speaking, in structural modeling a significant Chi-square value indicates that the model does not fit well to the data. However, in this particular study finding a significant Chi-square value is not surprising, given the large sample size (N= 743) of the study. The literature suggests that for large samples a value of 2:1–5:1 for the ratio Chi-square/degrees of freedom indicates an acceptable fit (Marsh and Hocevar, 1988). In our case, this ratio is 2.5.

Our results show a RMSEA value of 0.045 (with a 90% confidence interval between 0.029 and 0.062) and other fit indices (NFI, NNFI, CFI, IFI, MFI, GFI, AGFI) are greater than or equal to 0.945. In general, for AGFI, NFI, CFI and NNFI, fit values greater than 0.90 indicate a good fit to the data. RMSEA indicates a good fit if its value is smaller than 0.05 (Knoke, 2004).

We summarize by concluding that the model shown in Fig. 2 gives an accurate representation of the covariances mentioned in Table 5.

5.5.2. Interpretation of structural equations

Below we summarize the main conclusions that are indicated by the structural equations shown in Table 6. Further discus-

Table 6 Best fitting structural equations, corresponding to arrows shown in Fig. 2

Parameter	Equation	R-squared
Psychological precursors (v5)	v5 = 0.25 × v6 − 0.29 × v1 × − 0.12 × v4 + E5	0.194
Physical precursors (v6)	v6 = −0.07 × v4 + E6	0.004
Violations (v7)	V7 = −0.32 × v1 + 0.18 × v2 − 0.04 × v3 − 0.18 × v4 + 0.20 × v5 + E7	0.412
Inattention errors (v8)	V8 = 0.12 × v5 + 0.26 × v6 − 0.09 × v1 + E8	0.114
Dangerous errors (v9)	V9 = 0.06 × v5 + 0.19 × v6 − 0.21 × v1 + 0.09 × v3 + E9	0.091
Log accident involvement (v10)	v10 = 0.13 × v5 + 0.08 × v7 + 0.02 × v8 − 0.05 × v9 + E10	0.032

Note. Variable labels (except E) are preceded by standardized regression weights. Variables labeled "Exx" denote disturbance terms (part of dependent variable which cannot be explained by independent variables). Their variance equals (1 − R-squared) times the overall dependent variable variance. Abbreviations: v1: Attitudes; v2: Subjective/personal norms; v3: Perceived behavioural control; v4: Strategic decisions.

Fig. 2. Outcomes of path analysis conducted on field study data. Numbers indicated near one-directional arrows represent standardized regression weights (path coefficients). Numbers indicated near bi-directional arrows are correlation coefficients. All coefficients are significant at 0.05, except the one marked with symbol (*).

sion of these findings will be postponed until Section 6. While reading these conclusions, keep in mind that for a proper interpretation of the regression coefficients we must take into account the polarity of the scales (i.e., sometimes a high scale score indicates a low value on the underlying dimension, see Table 4 for details).

5.5.2.1. Psychological precursors. As Fig. 2 shows, Psychological precursors (v5) has three incoming arrows: one from Physical precursors (v6), one from Attitudes (v1), and one from Strategic decisions (v4). Taking the signs of the coefficients into account, this shows that driving under unsafe conditions (as measured by Psychological precursors) is predicted by: more physical and psychological handicaps (measured by Physical precursors), Attitudes approving of traffic violations (v1), and by fewer Strategic decisions made before starting a trip (v4).

5.5.2.2. Physical precursors. Driving under physical and mental constraints (as measured by Physical precursors (v6)) is predicted by a smaller amount of safety-inducing Strategic decisions (v4).

5.5.2.3. Violations. The frequency of Violations (v9) is predicted by: driving under unsafety-provoking conditions (as measured by Psychological precursors (v5)), Attitudes approving of traffic violations (v1), Subjective and personal norms that approve of traffic violations (v2), feeling in control of behaviours provoking violations, as measured by Perceived behavioural control (v3), and by fewer safety-enhancing Strategic decisions (v4).

5.5.2.4. Inattention errors. The number of Inattention errors (v10) is predicted by: more driving under unsafe conditions (as measured by Psychological precursors (v5)), driving under physical and mental constraints (as measured by Physical precursors (v6)), and Attitudes approving of traffic violations (v1).

5.5.2.5. Dangerous errors. The frequency of self-reported Dangerous errors (v11) is predicted by: more driving under unsafety-provoking conditions (as measured by Psychological precursors (v5)), driving under physical and mental constraints (as measured by Physical precursors (v6)), Attitudes approving of traffic violations (v1), and by lower control over violationprovoking behaviours, as measured by Perceived behavioural control (v3).

5.5.2.6. Accident involvement. Accident involvement (v13) is predicted by: driving under unsafety-enhancing conditions (as measured by Psychological precursors (v5)), Violations (v9), Inattention errors (v10), and by fewer Dangerous errors (v11).

5.6. Other results

Fig. 2 also shows that the following covariances among exogenous variables were significant (p < 0.05):

1. For all combinations of the TPB constructs (Attitudes, Perceived behavioural control, Subjective/Personal Norms).
2. Between Strategic decisions on the one hand and each of the TPB constructs on the other hand.

6. Conclusions and discussion

6.1. Predictability and comparison with theoretical expectations

It can be seen in Table 6 that of all constructs Violations are predicted best (41% explained variance). Least well predicted is Accident involvement (3%). Of its predictors, Psychological precursors had a (small) effect which was not hypothesized. Explanations for the low predictability of Accident involvement are addressed in Section 6.4.3.

The predictability of Psychological precursors was second best (19% explained variance). Though the effects of Physical precursors and Attitudes on Psychological precursors were not hypothesized in Fig. 1, they can nonetheless be understood in terms of the relationships that are known to exist between physical well-being, psychological well-being, and attitudes.

Psychological and Physical precursors can be distinguished empirically in the sense that the former hardly (or do not) help predict Dangerous errors (though this was expected), but do help predict Violations (unexpected), whereas the opposite holds for Physical precursors. Both types of precursor help predict Inattention errors to some degree. Notice that both effects associated with Physical precursors were as expected (see Fig. 1).

As can be seen in Fig. 2, Strategic decisions was related to all TPB constructs (significant correlations), confirming our expectation that these variables are somehow related (see question mark Fig. 1). In addition, Strategic decisions contributes somewhat to the direct prediction of Violations (expected) and, to a smaller extent, to that of Psychological precursors (also expected).

It should be noted that in Reason's (1990, 1997) model psychological precursors are considered a kind of psychological vulnerability which, if present to a large extent, increases the likelihood that latent failures affect unsafe acts and/or accident involvement. This means that the precise relationships holding between Strategic decisions and Psychological precursors may be more complex than we were able to study. This remains an issue to be investigated in future research.

6.2. Conclusions with respect to attitudes (TPB)

The results show that of the unsafe acts not only Violations were affected by Attitudes (which was expected, see Fig. 1), but also Dangerous errors. However, Inattention errors were not affected by Attitudes (see Fig. 2). The following explanation is offered here for the effect on Dangerous errors and the lack of one on Inattention errors.

Attitudes reflect evaluative aspects, as well as knowledgebased aspects of behaviour. Hence, the knowledge-based aspects of Attitudes (e.g., "speeding is fully acceptable") may propagate to Violations and Dangerous errors, both of which may be said to reflect erroneous (partly intentional) thinking related to driving (e.g., making a wrong estimation of the speed of a car approaching from the other direction, while overtaking the car one is following). On the other hand, errors of attention (slips, lapses) are more skill-based and unintentional and, therefore, less likely to be influenced by erroneous knowledge.

Fig. 2 shows that Attitudes not only predict Violations and Dangerous errors, but also predict one's temporary state of psychological well-being, as measured by Psychological precursors. Looking at all these effects of Attitudes, one cannot help but conclude that the attitudes one has about speeding and drunk driving have substantial implications for the types of driving failures one commits and for a person's temporary state of psychological well-being.

Also, notice that the TPB constructs covary among each other in the expected direction (see Fig. 2). This confirms earlier observations done by Ajzen (1991).

Finally, the effects associated with Perceived behavioural control (though both significant), were only small in size (see Fig. 2). Of these, the direct effect on Violations was expected and shows that there was a small tendency for respondents who thought they were in control of the occurrence of violations to commit violations more frequently.

The latter finding reminds one of the notion of illusion of control as a basis for non-compliance to rules (Reason et al., 1994). This illusion implies that violators overestimate the extent to which they can govern the outcome of risky situations. It also implies a conflict between expected behavioural outcomes and personal control beliefs (McKenna and Lewis, 1991).

6.3. Overall model assessment

In summary, the approach we have taken to extend Reason's accident causation model has resulted in several interesting, new findings. For example, adding Physical precursors to the model has resulted in (moderately sized) effects on other drivingrelated behaviours (e.g., an expected effect on Inattention errors and an unexpected effect on Psychological precursors) that have not been reported yet elsewhere.

Moreover, adding the TPB constructs to the model has allowed us to explain additional, non-trivial, portions of the variance in Violations scores (expected), Dangerous errors scores (unexpected), and Psychological precursors scores (unexpected).

Of all 26 relationships shown in Fig. 2, only six were not hypothesized explicitly in the introduction to the safety-related driver behaviour model, indicating the model's theoretical basis is rather robust. Obviously, the unexpected relationships (even though most of them are significant) should be treated with caution until we have replicated, analyzed, and understood them through future studies. The same is true for the ill-understood relationships between Strategic decisions and the TPB constructs. Limitations of the model are addressed in Sections 6.4 and 6.6.

Section 6.5 addresses the way the model can be used for segmentation of various subgroups of drivers, e.g., male versus female drivers and younger versus older drivers.

6.4. Methodological issues

6.4.1. Improvement of scales

Some of the scales constructed in our model would certainly need improvement in terms of consistency. For example the TPB scale Perceived behavioural control had an alpha of only 0.41. The other TPB scales (Attitudes and Subjective/personal norms) and all DBQ scales (Violations, Inattention error, Dangerous errors) had alphas greater than 0.50, though the DBQ alphas were a bit lower than the ones found in previous studies.

Most studies accept Nunnaly's rule of thumb that alphas greater than 0.70 are acceptable (Spector, 1992). Therefore, various scales had unsatisfactory consistencies in our study. In some cases, this was caused by the small number of items.

For example, computations using the Spearman–Brown prophesy formula (Spector, 1992) show that expanding the *Dangerous errors*scale with the additional 6 items used in the longer version of the DBQ (Verschuur, 2003a) would increase the alpha from 0.62 to 0.77.

Of the remaining scales, those that are in most need of improvement are Perceived behavioural control (alpha = 0.41) and Physical precursors (alpha = 0.35).

6.4.2. Self-reports

As many driver behaviour studies, including DBQ studies, our study is based on self-reported data. The use of questionnaires in traffic behaviour studies has apparent advantages (time, costs) and is often preferred over other more costly methods like collecting data on participants' actual violations and/or accidents from public police records. However, self-reports may be vulnerable to biases. Lajunen and Summala (2003)tested the effects of social desirability in reporting violations, using a quasiexperimental research design in which the degree of anonymity of the DBQ-based self-reports was varied. The results showed only small effects. One DBQ item involving driving under the influence of alcohol formed an exception, so the results collected in our study on this type of behaviour must be interpreted with caution.

Other possible biases not covered by the former study are forgetting and over-estimating of skills as possible sources of under-reporting. The latter source is likely to influence the accuracy of self-reports about lapses. To report how frequently one had a lapse one must (paradoxally) be able to recall these lapses. However, precise assessment of these influences has not been conducted yet, as far as we know, and awaits future research.

6.4.3. Accident measurement

af Wahlberg (2003) ˚ mentions three problems associated with the use of short time periods such as the 3 years interval, employed by our study and most other DBQ studies, for measuring accident frequencies: low variance, Poisson distributions, and the instability of a measure based on accident frequencies. He recommends to distinguish different types of accidents because they probably have different causes, rather than treating all accidents as belonging to the same group. He also recommends to use the concept of culpability as a way of elaborating the unelaborated concept of "an accident". However, this "solution" may generate other measurement problems, like knowing how to distinguish "guilty" of an accident from "not guilty" of an accident.

Another problem associated with the measurement of accidents is that accidents often are influenced by many situational (e.g., traffic-related, road system-related) factors, not all of which may be known or measurable. No doubt, this is one of the explanations for the low predictability of Accident involvement, as measured in this study (see Table 6).

A final explanation for the low predictability may be the skewed distribution of*Accident involvement*scores (seeTable 4). Even after we had applied a log transformation, this variable remained highly skewed.

6.5. Applications: using the overall model for discriminating between genders and age groups

Below we will give two illustrations of how the general model can be applied in order to differentiate between different demographic groups. The first example takes the distinction between male drivers and female drivers as a starting point. The second example looks at the way the model parameters are dependent on the particular age group the driver is in. Specifically, we will look at the distinction between drivers older than 50 years of age, and drivers younger than 50 years of age.

We decided to contrast the groups not in terms of the means, but in terms of the path coefficients for the various structural equations. Technically speaking, we used the multi-group option of EQS in order to see if the best fitting models are significantly different for the two genders (indicating a gender effect) or for the two age groups (indicating an age effect). This is done (manually) as follows.

First of all, the Chi-squared value is computed belonging to the overall fit of the two models with maximum allowed variation among the parameters for the two models (all parameters free). Next, we had EQS compute the Chi-squared value for the overall fit with no variation among the models allowed (parameters constrained to be identical). Next, the Chi-squared values are subtracted, as are the numbers of degrees of freedom. Finally, the difference Chi-squared value is tested for significance, using the difference number of degrees of freedom.

In addition to conducting an overall significance test, using the above-mentioned procedure one can determine for each parameter whether it is the same for each group (statistically speaking) or whether it is significantly different for the two groups.

6.5.1. Gender effects

Table 7 shows the best fitting equation found for predicting Violations, this equation being computed separately for men and women. The corresponding equation describing the overall sample is also shown (copied from Table 6).

It can be seen that Subjective and personal norms (Norms in Table 7) discriminates significantly between men and women, the coefficient for men taking on a higher value than for women. Nonetheless, the R-squared values for men and women are roughly the same (0.40). The interpretation of this contrast is

Table 7 Gender contrasts for equation predicting Violations

	Best fitting model	R-squared value
Overall sample	Violations = −0.32 × Attitudes + 0.18 × Norms − 0.04 × Perc. behav. control − 0.18 × Strateg. dec. + 0.20 × Psych. prec.	0.41
Men	Violations = 0.19 × Psych. prec. − 0.29 × Attitudes + 0.24 × Norms − 0.05 × Perc. behav. control − 0.15 × Strateg. dec.	0.41
Women	Violations = 0.18 × Psych. prec. − 0.34 × Attitudes + 0.13 × Norms − 0.06 × Perc. behav. control − 0.18 × Strateg. dec.	0.38

Note. Bold printed coefficients discriminate significantly between male and female drivers (p < 0.05). Overall CFI = 0.98. Difference Chi-squared = 33.71, d.f. = 4, p < 0.05. All coefficients are standardized regression coefficients. Generally speaking, coefficients higher than 0.10 are significantly different from0(p < 0.05).

as follows. If a driver indicates (s)he holds "unsafe" norms, and if the driver is male, he is more likely to report committing violations, as compared to women. Both coefficients turned out to be significantly different from 0 (p < 0.05). This result confirms similar findings reported elsewhere (Parker et al., 1992a,b; Reason et al., 1990; Rimmo, 2000; Verschuur, 2003a ¨ ).

Because standardized coefficients may be interpreted as partial correlation coefficients, taking their squared values yields the amount of variance in dependent variable scores explained by the corresponding model variable (expressed as fraction). Comparing the coefficients for men and women in this way, we find that for men Norms explains an additional 4% of the variance in the Violations scores, as compared to women. In summary, unsafe norms are more predictive of reported violations for men than for women.

6.5.2. Age effects

Table 8 shows, again, the best fitting equation found for predicting Violations, but this time the equation was computed separately for drivers younger than 50 and drivers of age 50 or older. The corresponding equation describing the overall sample is also shown (again, copied from Table 6).

It can be seen that predicting violations for older and younger drivers is best done with different weights (coefficients) for Psychological precursors ("Psych. prec." in Table 8) and Subjective and personal norms (Norms in Table 8). This is despite the fact that the overall predictability of Violations is comparable for older and younger drivers (see R-squared values in Table 8).

The interpretation of this contrast is as follows. Mirroring the finding mentioned above for the gender contrast, one can say that if a driver indicates (s)he holds "unsafe" norms, and if the driver is younger than 50, (s)he is more likely to report committing violations, as compared to older drivers. In addition, if a driver indicates (s)he occasionally drives under unfavourable psychological conditions, and if the driver is younger than 50, (s)he is more likely to report committing violations, as compared to older drivers. Again, all coefficients turned out to be significantly different from 0 (p < 0.05) and evidence for these age-related finding has also been reported elsewhere (Parker et al., 1992a,b; Reason et al., 1990; Rimmo, 2000; Rimm ¨ o and ¨ Hakamies-Blomqvist, 2002; Verschuur, 2003a).

In terms of percentages of variance explained, this analysis shows that for younger drivers Norms explains an additional 4% of the variance in the Violations scores, and Psychological precursors explains an additional 4.5% of this variance, as compared to older drivers. In summary, unsafe norms and Psychological precursors are more predictive of reported violations for younger drivers than for older drivers.

These results show the advantage of the generic model shown schematically in Fig. 1: it offers us an opportunity to distinguish subgroups of drivers in terms of model parameters (e.g., types of variables involved, strengths of relationships between variables, direction of causality). Information resulting from such analyses (e.g., distinguishing male from female drivers, or younger from older drivers) may be of use to designers of traffic-related educational programs or to drivers' licensing agencies, to mention a few examples.

6.6. Limitations of the model

Despite the strengths of the safety-related driver behaviour model that were discussed above, we also realize that the fitted model is still incomplete in the sense that important (e.g., trafficrelated, situational) variables are still missing and that many variables are only weakly related to each other. This has several explanations, some of which were already indicated in Section 6.4. This incompleteness also explains the low R-squared values

Table 8 Age contrasts for equation predicting Violations

	Best fitting model	R-squared value
Overall sample	Violations = −0.32 × Attitudes + 0.18 × Norms − 0.04 × Perc. behav. control − 0.18 × Strateg. dec. + 0.20 × Psych. prec.	0.41
Younger than 50 years	Violations = 0.25 × Psych. prec. − 0.25 × Attitudes + 0.22 × Norms − 0.05 × Perc. behav. control − 0.15 × Strateg. dec.	0.39
50 years or older	Violations = 0.13 × Psych. prec. − 0.38 × Attitudes + 0.09 × Norms − 0.09 × Perc. behav. control − 0.23 × Strateg. dec.	0.40

Note. Bold printed coefficients discriminate significantly between younger and older drivers (p < 0.05). Overall CFI = 0.99. Difference Chi-squared = 29.70, d.f. = 3, p < 0.05. All coefficients are standardized regression coefficients. Generally speaking, coefficients higher than 0.10 are significantly different from0(p < 0.05).

mentioned in Table 6 for several model equations (for example, for the equation predicting Accident involvement).

Even so, we remind the reader that the overall fit of the model is adequate, as indicated by the fit indices discussed in Section 5.5.1. In other words, the fitted model gives an accurate description of the "real" relationships holding among the model variables. Moreover, despite the low predictability of several model variables, other variables have relatively high R-squared values. For example, the R-squared value for Violations equals 0.412, indicating that 41.2% of the variance in the Violations scores is explained by the predictors that are shown in Fig. 2 and Table 6.

Acknowledgements

The authors wish to thank Prof. Dr. A. Mooijaart of the Methodology section, Department of Psychology, Leiden University, for his helpful comments on the methods section of this article.

This study is based on data derived from an internal report on the development of diagnostic tools for measuring risky driver behaviours, written for the Dutch Ministry of Transport, Public Works, and Water Management, Advisory group for Traffic and Transport (AVV).

References

Aberg, L., Rimm ˚ o, P.A., 1998. Dimensions of aberrant driving behaviour. ¨ Ergonomics 41, 39–56.
af Wahlberg, A.E., 2003. Some methodological defiences on traffic accident ˚ predictors. Acci. Anal. Prevent. 35, 473–486.
Ajzen, I., 1991. The theory of planned behaviour. Organ. Behav. Hum. Dec. Process. 50, 179–211.
Bianchi, A., Summala, H., 2002. Moral judgment and drivers' behaviour among Brazilian students. Psychol. Rep. 91, 759–766.
Blockey, P.N., Hartley, L.R., 1995. Aberrant driving behaviour: errors and violations. Ergonomics 38, 1759–1771.
Dunn, G., Everitt, B., Pickles, A., 1993. Modelling Covariances and Latent Variables using EQS. Chapman Hill, London.
Knoke, D., 2004. Structural equation models. In: Kepf-Leonard, K. (Ed.), Encyclopedia of Social Measurement. Academic Press, San Diego, CA.
Kontogiannis, T., Kossiavelou, Z., Marmara, N., 2002. Self-reports of aberrant behaviour on the roads: errors and violations in a sample of Greek drivers. Acci. Anal. Prevent. 34, 381–399.
Lajunen, T., Summala, H., 2003. Can we trust self-reports of driving? Effects of impression management on driver behaviour questionnaire responses. Transport. Res. Part F 6, 97–107.
Lajunen, T., Parker, D., Summala, H., 2004. The Manchester driving behaviour questionnaire: a cross-cultural study. Acci. Anal. Prevent. 36 (2), 231–238.
Lawton, R., Parker, D., Stradling, S.G., Manstead, A.S.R., 1997a. Predicting road traffic accidents: the role of social deviance and violations. Br. J. Psychol. 88 (2), 49–262.
Lawton, R., Parker, D., Stradling, S.G., Manstead, A.S.R., 1997b. Self-reported attitude towards speeding and its possible consequences in five different road contexts. J. Commun. Appl. Soc. Psychol. 7 (2), 153–165.
Manstead, A.S.R., Parker, D., 1995. In: Miles, Hewstone, Wolfgang, Stroebe (Eds.), Evaluating and extending the theory of planned behaviour. Eur. Rev. Soc. Psychol. 6, 69–95.
Marsh, H.W., Hocevar, D., 1988. A new, more powerful approach to multitraitmultimethod analyses. J. Appl. Psychol. 73, 107–117.
McDonald, R.P., Ho, Moon-Ho Ringo, 2002. Principles and practice in reporting structural equation analyses. Psychol. Methods 7 (1), 64–82.
McKenna, F.P., Lewis, C., 1991. Illusory judgements of driving skill and safety. In: Grayson, G. (Ed.), Behavioural Research in Road Safety. Transport Research Laboratory, Crowthorne, Berkshire, UK, pp. 24–130.
Mesken, J., Lajunen, T., Summala, H., 2002. Interpersonal violations, speeding violations and their relation to accident involvement in Finland. Ergonomics 45 (7), 469–483.
MON, 2005. (in Dutch) Mobiliteitsonderzoek Nederland 2004. Conducted by the MON project team, Dutch Ministry of Traffic, Public Works and Water Management, Advisory Group on Traffic and Transport (AVV).
Obriot-Claudel, F., Gabaude, C., 2004. The Driver Behaviour Questionnaire: A French Study Applied to Elderly Drivers. French National Institute for Transport & Safety Research, Laboratory of Ergonomics and Cognitive Science applied to Transport, France.
Ozkan, T., Lajunen, T., Summala, H., 2006. Driver behaviour questionnaire: a ˝ follow-up study. Acci. Anal. Prevent. 38, 386–395.
Parker, D., Manstead, S.R., Stradling, S.G., Reason, J.T., 1992a. Determinants of intention to commit driving violations. Acci. Anal. Prevent. 24 (2), 117–131.
Parker, D., Manstead, S.R., Stradling, S.G., Reason, J.T., Baxter, J.T., 1992b. Intention to commit driving violations: an application of the theory of planned behaviour. J. Appl. Psychol. 77 (1), 94–101.
Parker, D., Manstead, S.R., Stradling, S.G., 1995a. Extending the theory of planned behaviour: the role of personal norm. Br. J. Soc. Psychol. 34, 127–137.
Parker, D., Manstead, S.R., West, R., Stradling, S.G., Manstead, A.S.R., 1995b. Behaviour characteristics and involvement in different types of traffic accident. Acci. Anal. Prevent. 27 (4), 571–581.
Parker, D., Reason, J.T., Manstead, S.R., Stradling, S., 1995c. Driving errors, driving violations and accident involvement. Ergonomics 39, 1036– 1048.
Parker, D., Stradling, S.G., Manstead, S.R., 1996. Modifying beliefs and attitudes to exceeding the speed limit: an intervention study based on the theory of planned behaviour. J. Appl. Soc. Psychol. 26 (1), 1–19.
Parker, D., Lajunen, T., Stradling, S.G., 1998. Attitudinal predictors of interpersonally aggressive violations on the road. Transport. Res. Part F: Traffic Psychol. Behav. 1, 1–14.
Parker, D., McDonald, L., Rabbitt, P., Sutcliffe, P., 2000. Elderly drivers and their accidents: the aging driving questionnaire. Acci. Anal. Prevent. 32, 751–759.
Reason, J.T., 1990. Human Error. Cambridge University Press.
Reason, J., 1997. Managing the Risk of Organizational Accidents, Organisational Accidents. Ashgate Publishing Limited, Aldershot, England.
Reason, J., Manstead, A., Stradling, S., Baxter, J., Campbell, K., 1990. Errors and violations on the roads: a real distinction? Ergonomics 33 (10/11), 1315–1332.
Reason, J., Parker, D., Free, R., 1994. Bending the Rules: The Varieties, Origins and Management of Safety Violations. Department of Psychology, University of Manchester.
Rimmo, P.A., 2000. A four factor model of self-reported aberrant driving ¨ behaviour. Paper Presented at the ICTTP 2000, September 3–7, Bern, Switzerland, 2000.
Rimmo, P.A., Hakamies-Blomqvist, L., 2002. Older drivers' aberrant driving ¨ behaviour, impaired activity, and health as reasons for self-imposed driving limitations. Transport. Res. Part F 5, 47–62.
Rothengatter, T., Manstead, A.S.R., 1997. The role of subjective norm in predicting the intention to commit traffic violations. In: Talib, Rothengatter, Enrique, Carbonell Vaya (Eds.), Traffic and Transport Psychology. Pergamon, Amsterdam, pp. 389–394.
Spector, P., 1992. Summated rating scale construction: an introduction. Quantitative Applications in the Social Sciences, vol. 82, A Sage University Paper.
Stradling, S.G., Parker, D., 1997. Extending the theory of planned behaviour: the role of personal norm, instrumental beleifs and affective beliefs in predictiong driving violations. In: Talib, Rothengatter, Enrique, Carbonell Vaya (Eds.), Traffic and Transport Psychology. Pergamon, Amsterdam, pp. 367–374.
Sullman, M.J.M., Meadows, M.L., Pajo, K.B., 2002. Aberrant driving behaviours amongst New Zealand truck drivers. Transport. Res. Part F 5, 217–232.
Sullman, M.J.M., Meadows, M.L., Pajo, K., 2004. Errors, lapses and violations in the drivers of heavy vehicles. In: Rothengatter, T., Huguenin, R.D. (Eds.), Traffic & Transportation Psychology. Elsevier, Amsterdam, pp. 147–154.
Underwood, G., 1997. Estimating accident liability. In: Rothengatter, T., Carbonell Vaya, E. (Eds.), Traffic Transport Psychology. Theory and Application. Pergamon, Amsterdam, pp. 247–258.
Verschuur, W.L.G., 2003a. (in Dutch) Onderzoek uitgevoerd met de 'Driver Behaviour Questionnaire'. Project Gedragsdeterminanten van Onveilig Verkeersgedrag, eindrapportage. Faculteit der Sociale Wetenschappen, Universiteit Leiden, Werkgroep Veiligheid/R-03/68 (Also available at the website of the Dutch Ministry of Transport, Public Works, and Water Management: http://www.rws-avv.nl/pls/portal30/docs/9456.PDF).
Verschuur, W.L.G., 2003b. The driver behaviour questionnaire' in the Netherlands. In: Proceedings of the 16th ICTCT Workshop, Soesterberg, 29th October–1st November.
Wagenaar, W.A., Reason, J.T., 1990. Types and tokens in road accident causation. Ergonomics 33 (10/11), 1365–1375.
Westerman, S.J., Haigney, D., 2000. Individual differences in driver stress, error and violation. Person. Ind. Differ. 29, 981–998.
Xie, C., Parker, D., 2002. A social psychological approach to driving violations in two Chinese cities. Transport. Res. Part F: Traffic Psychol. Behav. 5 (4), 293–308.

verschuurmodelingsafeunsafedriving2008.md 74 KB Geçmiş Ham