Transportation Research Part F 27 (2014) 112123

Contents lists available at ScienceDirect

Transportation Research Part F

journal homepage: www.elsevier.com/locate/trf

How appraisals shape driver emotions: A study from discrete

and dimensional emotion perspectives
Tingru Zhang , Alan H.S. Chan
Department of Systems Engineering and Engineering Management, City University of Hong Kong, Hong Kong

a r t i c l e

i n f o

Article history:
Received 24 June 2014
Received in revised form 24 September 2014
Accepted 24 September 2014

Keywords:
Driver
Emotion
Appraisal
Dimensional model
Discrete model

a b s t r a c t
This study aimed to investigate how the emotional responses of drivers, from both dimensional and discrete perspectives, may be predicted by using the appraisal components of
goal relevance, blame party, and certainty. Trafc scenarios representing a combination
of the three appraisal components were designed and presented to participants. The emotional responses to each scenario were measured on an ArousalValence emotional space
and were assigned with discrete emotion labels by applying a cluster analysis. For the
dimensional model, the results showed that valence was signicantly associated with
the blame party and the goal relevance components. The arousal was, as hypothesised, predicted by the blame party and the certainty components. For the discrete model, it was
found that driving anger was most likely to be provoked when other drivers were responsible for the adverse driving outcome; driving fear was most commonly experienced in situations where driver safety was threatened by the driver himself/herself or by impersonal
circumstance; and driving anxiety was an outcome of uncertain arrival-blocking events
caused by driver himself/herself or impersonal circumstance. Findings from this study suggest the feasibility of predicting emotional dimensions on the basis of the appraisal process. Moreover, this study contributes to the research on driver emotion by
demonstrating that the certainty feature of trafc events plays an important role in determining the emotional responses of drivers.
2014 Elsevier Ltd. All rights reserved.

1. Introduction
Driving can provoke a large range of negative emotions such as anger, anxiety or fear. Using a two-week period driving
diary, Underwood, Chapman, Wright, and Crundall (1999) found that 85% of respondents surveyed reported experiencing
anger at least once while driving. Mesken, Hagenzieker, Rothengatter, and de Waard (2007) showed that within a one-hour
driving period, participants on average experienced anxiety 2.6 times and anger 1.5 times. These negative emotions can
adversely affect driving performance by interfering with cognitive activities such as attention (Schimmack & Derryberry,
2005), evaluation (Evans, 2008), and decision-making (Morgan, Jones, & Harris, 2013). It has been found that drivers are
more likely to violate speed limits when they feel angry (Stephens & Groeger, 2011) or anxious (Briggs, Hole, & Land,
2011). Moreover, aggressive behaviors such as tailgating or running amber lights have been demonstrated to be related
to anger (Abdu, Shinar, & Meiran, 2012; Beck, Daughters, & Ali, 2013), and driving errors and lapses are related to fear

Corresponding author. Tel.: +852 59832370.

E-mail addresses: trzhang3-c@my.cityu.edu.hk (T. Zhang), alan.chan@cityu.edu.hk (A.H.S. Chan).
http://dx.doi.org/10.1016/j.trf.2014.09.012
1369-8478/ 2014 Elsevier Ltd. All rights reserved.

T. Zhang, A.H.S. Chan / Transportation Research Part F 27 (2014) 112123

113

and anxiety (Shahar, 2009; Taylor, Deane, & Podd, 2007). There is also evidence that accident risk increases for angry or anxious drivers (Dula, Adams, Miesner, & Leonard, 2010; Underwood et al., 1999).
1.1. Determinants of emotions: appraisal theory
To decrease the adverse impacts associated with negative emotions while driving, it is important to understand why and
how drivers are provoked emotionally by certain trafc events. Appraisal theory provides a theoretical framework that
describes how emotions are determined by how a driver evaluates an event (Roseman, 1996). According to Roseman
(1996), ve appraisal components, namely, motivation consistency, blame party, certainty, situational state, and legitimacy,
are critical in the emotion generation process, with the rst three being the most signicant. The appraisal of motivation
consistency refers to the evaluation of whether an individuals goal is promoted or blocked in a given situation and such
an appraisal helps to differentiate positive from negative emotions. Positive emotion will be provoked in goal-promoting situations and negative emotion provoked in goal-blocking situations. The blame party refers to whether an outcome is seen to
be caused by impersonal circumstances, some other person, or the self. In general, sadness or shame are associated with selfblame events; anger or dislike tend to be the emotional responses to other-blame events; fear or anxiety are most likely to be
provoked when impersonal circumstances are held accountable (Roseman, 1996). The appraisal of certainty is the evaluation
of whether the outcomes of events can be predicted or not and it can further differentiate emotion such as fear and anxiety.
According to Sadock, Kaplan, and Sadock (2007), fear is an emotional response to certain or known threats while anxiety is a
response to uncertain or unknown threats. Rosemans appraisal theory has been supported by a lot of research and there is
evidence that appraisal outcomes are the key to predicting emotional response patterns (Conati & Zhou, 2002; Gratch,
Marsella, Wang, & Stankovic, 2009).
In recent transportation research, the emotional responses of drivers to trafc events have been investigated from the
perspective of appraisal theory. In a study where participants drove an instrumented vehicle and reported their emotions,
Mesken et al. (2007) found that negative emotions were associated with goal incongruent events while happiness was associated with goal congruent events. They also found that goal relevance, which referred to achieving the two most important
goals in trafc situations (i.e. arrival at the destination and safe driving), partially determined the type of emotions provoked.
Anger was mostly triggered by progress-impeding events while anxiety associated with safety-threatening events. Similar
ndings have also been reported by Roidl, Frehse, Oehl, and Hger (2013) in their questionnaire studies. The importance
of goal relevance in shaping the emotional responses of drivers has been further supported by Zhang and Chan (2014), where
it was found that drivers become angrier when their safety is threatened than when their progress is blocked. Regarding the
effects of blame party, it has been shown that driving anger is mostly associated with other-blame and anxiety with circumstance-blame (Mesken et al., 2007; Roidl et al., 2013).
It is evident that all previous driver emotion studies have only focused on goal relevance and the blame party. The effect
of the certainty appraisal component on the emotion generation process of drivers has not been explored much even though
the outcomes of many trafc events are unpredictable (Hault-Dubrulle, Robache, Pacaux, & Morvan, 2011). For instance,
drivers can usually predict how long they will be blocked by red trafc lights, but it is difcult for them to estimate how
long trafc congestion may last. It seems like that drivers will develop rather different emotional responses to these two situations, though both of which are arrival-blocking events caused by impersonal circumstances. Therefore, it is necessary to
take the certainty component of trafc events into consideration to better understand and more accurately predict the emotional experiences of drivers. One aim of this study was to investigate how the combinations of the three most relevant
appraisal components (goal relevance, blame party, certainty) determine the type of emotions experienced by drivers in
given trafc scenarios.
1.2. Emotion experience measures: discrete and dimensional approaches
Psychological emotion models developed by previous researchers can be categorized into two main groups; one with discrete emotion states and the other with emotions expressed in a dimensional continuous space. In the discrete emotion
models, human emotions can be classied into a prescribed list of words or adjective labels, e.g. happiness, anger, etc.
(Ekman, 1999; Thamm, 2006). The dimensional space approach contends that a few fundamental dimensions together specify emotional responses (Colibazzi et al., 2010) and the most recognised dimensions are valence and arousal. Valence refers
to a subjective feeling of pleasantness or unpleasantness, and arousal is dened as a subjective state of feeling activated or
deactivated. With the purpose of better accounting for differences in emotional experiences, new emotion dimensions, such
as the submissive-dominance dimension (Bradley & Lang, 1994) or the approach-avoidance dimension (Davidson, Ekman,
Saron, Senulis, & Friesen, 1990), have been proposed. However, these newly proposed dimensions seem to explain only a
small proportion of the variance in emotion and are less consistently accepted by emotion researchers than valence and
arousal (Scherer, Dan, & Flykt, 2006). By proposing that a discrete emotion represents a combination of several dimensions,
it is possible to reconcile the two perspectives to some extent (Mauss & Robinson, 2009). The locations of discrete emotions
within the dimensional continuous space have been investigated in a number of psychological studies aiming to integrate
the two emotion models (e.g. Christie & Friedman, 2004; Sun, Yu, Huang, & Hu, 2009). Their ndings suggest that dimensions
of valence and arousal can portray the structure of discrete emotions. For example, anger is found to be characterized by high
arousal and extremely negative valence, whereas anxiety characterized as low arousal and less negative valence.

114

T. Zhang, A.H.S. Chan / Transportation Research Part F 27 (2014) 112123

Although there is emerging evidence that the appraisal components could be important to the prediction of emotional
response patterns, almost all past driver emotion research has focused on the prediction of discrete emotions (Mesken
et al., 2007; Roidl et al., 2013) and little is known about how to link appraisal theory to dimensional emotion models for
drivers. To ll this research gap, this study aimed to investigate how self-reported emotions in the ValenceArousal (VA)
emotion space could be predicted by the three above-mentioned appraisal components (goal relevance, blame party and certainty) in the context of driving. To examine this question, a set of driving scenarios encompassing different combinations of
appraisal components were designed by referring to examples and cases used in previous related studies (Lu, Xie, & Zhang,
2013; OBrien, Shaw, Watson, & Lennon, 2012; Roidl et al., 2013). These trafc scenarios were presented to participants in the
form of text and participant emotional responses to the scenarios were measured in the VA emotional space. The mapping
relationships available between dimensional and discrete emotion models then make it possible to convert the recorded
dimensional emotions into basic discrete emotions. The present study will explore how such converted discrete emotions
can be predicted by the appraisal components. Results from this study should shed light on the underlying mechanism that
elicit driver emotions, which may further serve as guidance for the development of intervention strategies aimed at eliminating negative driving emotions and improving road safety.
1.3. Aims and hypotheses
The present study aimed to investigate the effects of goal relevance, blame party and certainty on emotional experiences
whilst driving. Also, it aimed to link appraisal theory to both dimensional and discrete emotion models. Two previous studies
have attempted to predict dimensional emotions with appraisal components. In the study by Scherer et al. (2006), pictures
extracted from the International Affective Picture System (IAPS), all with extreme valence and arousal values, were used to
induce participant emotions. One group of participants was required to rate their evaluation of the pictures on a prole of nine
appraisal factors (e.g. goal relevance, blame party, and coping potential) while another group rated their emotional reactions
on a 3-dimension (valance, arousal and coping ability) emotion model. The results showed that the valence dimension was
best explained by the goal-relevance component whereas the arousal dimension was signicantly correlated with blame
party and certainty of outcome. A recent study conducted on 40 Canadian individuals demonstrated that the extent to which
an event is congruent or incongruent with ones goal only inuenced ratings of valence but not arousal (Nicolle & Goel, 2013).
Based on the above review, it was hypothesized that
Hypothesis 1 (Goal relevance). The valence dimension of the evoked emotions on the road would be determined by the goalrelevance feature of the provocative trafc events.
Hypothesis 2 (Blame party). The arousal level of the evoked emotion is higher when other drivers are responsible for the
adverse outcome than when the individual himself/herself or impersonal circumstances are responsible.
Hypothesis 3 (Certainty). The provoked arousal level will be higher when drivers are uncertain about the outcome of a trafc
event than when they are certain.
How the appraisal components determine discrete emotions has been widely studied (Lu et al., 2013; Roidl et al., 2013;
Roseman, 1991). In the context of driving, it has been generally accepted that anger is mostly associated with goal-inconsistent events caused by other people (Roidl et al., 2013), while fear and anxiety are the most likely emotional responses to
safety-threatening events caused by impersonal circumstances or by the individual himself/herself (Roidl et al., 2013;
Roseman & Smith, 2001). These well-established associations can be used to evaluate the appropriateness of the method
used here to convert the dimensional emotions to discrete emotions. As mentioned above, the discrete emotions here were
not directly measured but converted from the valence and arousal ratings based on the mapping relationships between
dimensional and discrete emotion models. If similar associations between the converted discrete emotions and the appraisal
components are found, it will indicate that the converting method, which will be elaborated later, is valid. There is no direct
evidence regarding how appraisal of the outcome certainty shapes driver emotional responses, however, according to Sadock
et al. (2007), anxiety is a response to an imprecise or unknown threat while fear is an emotional response to a known or
denite threat. Based on this, it was hypothesized that:
Hypothesis 4 (Certainty). If identied as discrete emotions, driving fear is the emotional response associated with outcome
certainty and driving anxiety is associated with outcome uncertainty.

2. Method
2.1. Design of trafc scenarios
This study examined the three most important appraisal components in driver negative emotion generation: goal
relevance, blame party and certainty of the outcome. While drivers may drive for different reasons (e.g. fun seeking or risk

115

T. Zhang, A.H.S. Chan / Transportation Research Part F 27 (2014) 112123

seeking), the two main types of driving goal are to achieve the destination on time (arrival goal) and to drive safely (safety
goal). These two goals have been investigated in previous driver emotion studies (Mesken et al., 2007; Roidl et al., 2013). To
provoke negative emotions, the trafc scenarios used here will block either the arrival goal or the safety goal. Four possible
blame parties were dened: other drivers responsible for the adverse events (other-blame); impersonal trafc circumstance
responsible (circumstance-blame); and two dimensions of self-directed blame: participant inuences own goal (self-self)
and own self inuences the goal of others (self-other). The four levels of the blame party component used here are the same
as those used by Roidl et al. (2013). Certainty refers to whether a driver is certain or uncertain as to the outcome of trafc
events. By referring to examples and cases used in previous relevant studies (Lu et al., 2013; Roidl et al., 2013), a total of 16
trafc scenarios were designed to cover all combinations of the three appraisal factors: goal relevance (arrival/safety) 
blame party (other-blame/self-self/self-other/circumstance-blame)  certainty (certain/uncertain). Short descriptions of
the 16 trafc scenarios are presented in Table 1.
To guarantee that a scenario will be interpreted as expected, there are explicit descriptions on each appraisal component
to guide the appraisal process. For example, the text used to describe the arrival-blocking  other-blame  uncertain scenario is Today, you have an important meeting. When you drive on a narrow road, one car in front of you drives at a speed lower
than normal in good trafc conditions and your progress is blocked by the car in front. That car does not speed up even though you
pressed the horn many times. You do not know how long you need to wait to pass that car. The arrival-blocking feature of the
scenario is stressed by your progress is blocked, the other-blame feature is delivered through by the car in front and the
uncertain feature is reected by do not know how long.
2.2. Recording of emotional responses
The two dimensional ValanceArousal (VA) space, adapted from Cai and Lin (2011), was used in this study to record participant emotional responses for the trafc scenarios tested (Fig. 1). In the study by Cai and Lin (2011), the rating for both
valence and arousal ranged from 1 to 1, with a resolution of 0.1. However, the negative value of arousal seemed to be a
bit confusing so the rating scales were revised slightly to make them more reasonable. In this study, each dimension in
the VA space was an 11-point rating scale, with the valence dimension ranging from 5 (extremely negative) to +5 (extremely positive) and the arousal dimension ranging from 0 (low arousal) to 10 (high arousal). Participants were required to
place an X on the grid intersections to indicate their ratings on arousal and valence. An example of a possible emotional
response is shown in Fig. 1. Since people are more familiar with the discrete emotion labels than with the arousal and
valence concepts (Cai & Lin, 2011), several emotion labels were plotted on this VA space for their reference. However, participants were encouraged to locate their responses based on their assessment on valence and arousal.
2.3. Structure of questionnaire
The questionnaire consisted of two parts. In the rst part of the questionnaire, the 16 trafc scenarios expressed in text
were randomly presented. A VA emotional space was presented after each scenario for participants to indicate their
Table 1
Short descriptions of trafc scenarios used to induce drivers emotions.
Goal
relevance

Certainty

Arrival

Certain

Uncertain

Safety

Certain

Uncertain

Blame party
Other-blame

Self-other

Self-self

Circumstance-blame

1. At an intersection, a slow
moving car in front makes you
miss the green light and you
have to wait for the red light
2. Slow moving car in front
blocks your progress and does
not speed up though you have
honked the horn. You are unsure
how long you have to wait
9. Dangerous braking of front
car. You know that if you brake
quickly, the collision can be
avoided and if not, the collision
can not be avoided

3. You slow down to answer a

phone call and block vehicles
behind. The call would be
nished within a few minutes
4. Your car breaks down and
blocks vehicles behind. You do
not how long the tow truck
will take to come

5. Drive to a wrong route

but the police may help you

7. Encounter trafc jam

during rush hours but
the trafc polices are
directing the trafc
8. Encounter trafc jam
and unsure how long the
jam would last

10. Dangerous braking of front

car, and you are not sure
whether a collision can be
avoided

11. You change lane without

indicating signals. You know
that if the car behind brakes
quickly, a rear-end collison can
be avoided. Else, a collision
cannot be aovided
12. You change lane without
indicating signals and not sure
if a rear-end collision has been
caused by you

6. Driving to a wrong route

and not sure how long it
takes to nd the right way

13. Tired while driving and

when more alert, you
realize that you are very
close to the front car. A
collision can be avoided if
you brake quickly. Else, a
collision will happen
14. Tired while driving and
when more alert, you
realize that you are very
close to the front car and
you are not sure whether
the collision can be avoided

15. Malfunctioning trafc

lights in the intersection
ahead. Accident can be
avoided only if you have
carefully checked the
trafc in the intersection
16. Badly constructed
road may cause the loss
of control over your
vehicle

116

T. Zhang, A.H.S. Chan / Transportation Research Part F 27 (2014) 112123

Fig. 1. Valence-Arousal emotional space used to record participants emotional responses.

emotional response. The second part of the questionnaire asked about the demographic information of participants. Participants were required to give responses on their gender, age (<30 years old, 3040 years old, 4150 years old,
>50 years old), education level (primary, secondary, or higher education), driving experience (number of years
licensed), and driving frequency per week (0 time, 13 times, 46 times, >6 times).
2.4. Participants
The participants were recruited from social network and they were made sure that this study was voluntary and anonymous. A total of 50 (16 females and 34 males) Hong Kong Chinese drivers participated in the survey. About 48% of them
were younger than 30 years old, 26% of them belonged to the 3140 age group, 14% of them belonged to the 4150
age group, and the other 12% participants were older than 50 years old. The average driving experiences of the surveyed male
and female drivers were 10.49 (SD = 7.47) and 3.41 (SD = 2.59) years, respectively. The majority (88%) of the respondents
drove car more than once every week. About 10%, 36% and 54% of them received primary, secondary, and higher education,
respectively.
2.5. Procedure
The investigation was taken place in a classroom. The aim of the study and the method of indicating responses with the
VA emotional space were introduced to the participants before they responded to the questionnaire. They were then asked
to imagine encountering the 16 text-described trafc scenarios and respond as to how they would be provoked emotionally
by each scenario. After completing questions about the emotional responses to road events, the personal characteristics of
the participants were assessed. The questionnaire was presented in traditional Chinese, which is the native language for all
the 50 participants. The whole procedure took approximately 30 min.
2.6. Statistical analysis
To convert the emotions shaped by dimensions into discrete labels (e.g. anger or fear), the statistical technique of cluster
analysis was applied to data in the present study. Cluster analysis classies a set of observations into several groups, where
members of groups share properties in common. The clustering could be achieved with the use of different algorithms and
the procedures outlined by Milligan and Sokol (1980) were followed in this study. First the hierarchical clustering algorithm
was used to determine the number of clusters and then the K-means clustering algorithm was applied to generate the classied results. In hierarchical cluster analysis, the individual cases begin as individual clusters and step-by-step the two most
similar clusters (i.e. clusters with a shortest distance) would be jointed together, eventually resulting in one cluster containing all cases. In this study, the distance between two clusters was dened as the squared Euclidean distance generated with
Wards method. The distance between the two most similar clusters usually increases with the merging process and the optimal number of clusters corresponds to the step where further combination will lead to the largest increase in this distance
(Salvador & Chan, 2004). The cluster number determined in this way was used in the subsequent K-means algorithm.
The main effects of the three appraisal components on valance and arousal ratings and all possible two-way interactions
were tested using a repeated-measures ANOVA. The three-way interaction may have insufcient statistical power when the
sample size is small and therefore was excluded from analysis. To measure the strength of the associations, the estimated
proportion of the variance (g2) in a dependent variable explained by the corresponding independent variable, was calculated

117

T. Zhang, A.H.S. Chan / Transportation Research Part F 27 (2014) 112123

and reported. To investigate how the discrete emotions were predicted by the three appraisal components, the Multinomial
Logistic Regression (MLR) was used in this study. The MLR evaluated the net effects, in the form of odds ratios (ORs), of the
appraisal components on the provocation of discrete emotions. It also estimated the probability of generating a specic discrete emotion given a combination of the appraisal components.
3. Results
The mean ratings for valence and arousal dimensions were 1.98 and 6.93, respectively, indicating that the designed scenarios in general were effective in inducing negative emotions. The demographic variables showed no signicant inuences
on both arousal and valence ratings except for gender, where the arousal level of male drivers (mean = 7.16, SD = 1.68) was
signicantly higher than that of female drivers (mean = 6.82, SD = 1.74) (p < 0.05). The Spearman correlation test was applied
and no signicant correlation between the two dimensions was identied (p > 0.05).
3.1. Appraisal components and dimensional emotions
The mean arousal and valence ratings for each combination of the three appraisal components are presented in Fig. 2. In
order to assess the inuences of the three appraisal components on arousal and valence ratings, a repeated-measures ANOVA
was conducted and the results are presented in Table 2. The three independent variables were: goal relevance (two levels:

Fig. 2. The mean valence and arousal ratings for different combinations of appraisal components. C: certain; U: uncertain; O: other-blame; S-O: self-other;
S-S: self-self; Cir: circumstance-blame.

Table 2
The results of the repeated measure ANOVA.
Valence

Arousal

Appraisal

F-value (g2)

F-value (g2)

Goal relevance
Blame party
Certainty
Goal relevance  blame party
Goal relevance  certainty
Blame party  certainty

27.198
15.331
.149
3.059
.174
5.002

.000
.000
.701
.087
.679
.002

.938
19.131
9.020
2.018
28.096
2.443

.338
.000
.004
.114
.000
.066

(.357)
(.238)
(.003)
(.059)
(.004)
(.093)

g2: The proportion of the variance in a dependent variable explained by the corresponding independent variable.

(.019)
(.281)
(.155)
(.040)
(.364)
(.047)

118

T. Zhang, A.H.S. Chan / Transportation Research Part F 27 (2014) 112123

arrival/safety), blame party (four levels: other-blame/self-self/self-other/circumstance-blame), certainty (two levels: certain/
uncertain); the two dependent variables were arousal and valence ratings.
3.1.1. Effects of appraisal factors on valence ratings
Consistent with Hypothesis 1, the goal relevance was the most signicant predictor of valence (F(1, 49) = 27.198; p < 0.001;
g2 = 0.357), with valence rating being much lower in safety-threatening situations (meansafety-threatening = 2.208) than in
arrival-blocking situations (meanarrival-blocking = 1.753). Unexpectedly, the main effect of blame party was also signicant
(F(3, 147) = 15.331; p < 0.001; g2 = 0.238). The post hoc analysis showed that other-blame situations resulted in a signicantly
more negative valence rating than the other three blame parties (meanother-blame = 2.460; meanself-other = 1.850;
meanself-self = 1.840; meancircumstance-blame = 1.770). The main effect of certainty was not statistically signicant
(F(1, 49) = 0.149; p > 0.05; g2 = 0.003). There was a signicant certainty  blame party interaction effect on valence rating
(F(3, 147) = 5.002; p < 0.005; g2 = 0.093). So the effect of certainty at each level of the blame party and the effect of blame party
at certainty and uncertainty level were separately investigated. The results showed that the certainty component showed
signicant effects in other-blame and self-other situations but no signicant effects in the other two blame levels. In
the self-other situation, outcome uncertainty resulted in a more negative valence rating than outcome certainty
(meanuncertainty*self-other = 2.010; meancertainty*self-other = 1.690). However, in the other-blame situation, it was the certainty
of outcome that lead to a more negative valence rating (meancertainty*other-blame = 2.620; meanuncertainty*other-blame = 2.300).
In the present study, the self-blame party was further divided into self-self and self-other for testing the possible emotional
differences when either the driver himself/herself or other drivers suffered the results of the adverse events. The only
difference between the two blame parties was in situations where the outcome was uncertain, self-other situations
resulted in a signicantly more negative valance rating than self-self conditions (meanself-other*uncertainty = 2.010;
meanself-self*uncertainty = 1.730).
3.1.2. Effects of appraisal factors on arousal ratings
In general, the results showed that blame party and certainty had signicant main effects on arousal ratings while the
goal relevance did not (Table 2). The blame party component explained the greatest amount of variance in arousal ratings
(F(3, 147) = 19.131; p < 0.001; g2 = 0.281). The hypothesized effect of blame on arousal in Hypothesis 2 was therefore conrmed. Post hoc analysis showed that other-blame led to a signicantly higher arousal rating than the other three forms
of blame (meanother-blame = 7.705; meanself-other = 6.460; meanself-self = 6.720; meancircumstance-blame = 6.825). No signicant differences among the other three blame parties were identied. The main effect of certainty was also signicant
(F(1, 49) = 9.020; p < 0.005; g2 = 0.155). The provoked arousal was higher when drivers were uncertain of the adverse outcome
than when they were certain (meanuncertainty = 7.083; meancertainty = 6.773). This nding provided support for Hypothesis 3.
Since the 2-way interaction goal relevance  certainty was signicant (F(1, 49) = 28.096; p < 0.001; g2 = 0.364), the effect of
certainty was separately investigated at each level of goal relevance. The results showed that uncertainty of outcome
only provoked higher arousal rating when drivers safety goal was blocked (F(1, 49) = 43.688, p < 0.01; g2 = 0.471;
meanuncertainty*safety-threatening = 7.445; meancertainty*safety-blocking = 6.720). The role of certainty was not signicant in cases of
arrival-blocking events. The effect of goal relevance at either certainty or uncertainty level was tested and no signicant
result was found.
3.2. Appraisal components and discrete emotions
How the combination of the three appraisal components determined the discrete emotions will be explored in this
section. For this end, cluster analysis was rst applied to convert the emotions shaped by dimensions into discrete labels
(e.g. anger or fear). Then, Multinomial Logistic Regression was applied to explore how discrete emotions would be predicted
by the appraisal components.
3.2.1. Discrete emotion generation
Here, it was investigated how the 800 responses (50 participants  16 scenarios), measured by valence and arousal
dimensions, could be grouped into several discrete emotion categories. Prior to analysis, the data used to generate cluster
partitions were screened for multivariate outliers and no outliers were identied. Hierarchical clustering analysis (HCA)
was conducted to determine the optimal number of clusters for the data set. The 800 responses began as individual clusters
and every step two clusters with the smallest squared Euclidean distance were combined. The distance between the two
most similar clusters at each step was plotted against the number of clusters at that step (Fig. 3). An inspection of Fig. 3 suggested that the largest increase in the squared Euclidean distance appeared when the number of cluster changed from 4
(marked with a triangle marker) to 3, indicating that further merging was inappropriate when the cluster number was 4.
As a result, the number of clusters was determined as 4 in the subsequent K-means clustering analysis.
This rst cluster generated by the K-means method was characterized as medium level of negative valence
(mean = 2.65) and low arousal level (mean = 2.88). Based on the above introduced mapping relationship between dimensional and discrete emotions (Fig. 1), this cluster was most appropriate to be labeled as shame. The mean valence and arousal rating of the second cluster was 2.09 and 6.30 respectively, which corresponded to the discrete fear emotion. The
third cluster, with an average valence rating of 0.86 and arousal rating of 7.68, could be labeled as anxiety based on

T. Zhang, A.H.S. Chan / Transportation Research Part F 27 (2014) 112123

119

Fig. 3. The squared Euclidean distance between the two clusters combined at each step. The largest increase in squared Euclidean distance appears when
the number of cluster changes from 4 to 3 and therefore, the most appropriate number of cluster for the present data is 4, as indicated by the triangle
marker.

the mapping relationship. Finally, the fourth cluster was characterized as extremely high negative valance (mean = 2.99)
and high arousal (mean = 7.68) and could be labeled as anger emotion. Two separate ANOVA was performed to investigate
whether the four discrete emotions differed in the valance and arousal ratings. The results of ANOVA indicated that there
were signicant differences in both valence rating (F(3, 796) = 250.920; p < 0.001; g2 = 0.486) and arousal rating
(F(3, 796) = 1165.917; p < 0.001; g2 = 0.815) among the four discrete emotions.
3.2.2. Effects of appraisal factors on discrete emotions
Multinomial Logistic Regression (MLR) was conducted to evaluate how the discrete emotions could be predicted by the
three appraisal components. In general, the results of MLR (Table 3) showed that goal relevance (p < 0.005) and blame party
(p < 0.001) contributed signicantly to discrimination between different discrete emotions. Based on conventions in interpreting p-value (Yoccoz, 1991), the effect of the certainty component was marginally signicant (p < 0.1).
The net effects of the appraisal components, in the form of odds ratio (OR), are presented in Table 4. An OR > 1 indicates a
higher probability whereas an OR < 1 indicates a lower probability, of the outcome falling into the comparison emotion
group than falling into the baseline group, in the presence of the corresponding independent variable. It was found that
in other-blame situations, the anger emotion was more likely to be experienced than shame (OR(anger vs. shame) = 5.882)1,
fear (OR(anger vs. fear) = 3.289) or anxiety (OR(anger vs. anxiety) = 5.291). This was in line with the previous results in driver emotion
studies (Mesken et al., 2007; Roidl et al., 2013). Also, it was found in arrival-blocking situations, drivers were more likely to
experience anxiety than anger (OR(anxiety vs. anger) = 1.664)2, shame (OR(anxiety vs. shame) = 2.174) and fear (OR(anxiety vs.
fear) = 1.742). Moreover, when drivers were not sure regarding the outcome of trafc events, there was a signicantly higher
probability for them to experience anxiety rather than shame (OR(anxiety vs. shame) = 1.712) or fear (OR(anxiety vs. fear) = 1.431).
These results showed that fear and shame were associated with certainty of events while anxiety was associated with uncertainty of conditions, which support Hypothesis 4. The effects of the three appraisal components in differentiating shame and
fear were not signicant.
To summarize, the blame party component played an important role in separating anger from the other three emotions
while the certainty component was effective to differentiate anxiety from fear and shame.
3.2.3. The most likely provoked discrete emotion in each of the 16 scenarios
For each scenario, the probability of provoking each of the four types of emotions was calculated based on the MLR
results. These probabilities are summarized in Fig. 4. The most likely emotional responses for the 16 investigated scenarios
are presented in Table 5. Of the 16 scenarios, 4 scenarios (scenario No. 1, 2, 9 and 10) were most likely to provoke the anger
emotion. The common characteristics of the 4 scenarios were that they were all negative driving events that caused by other
drivers. Two scenarios (scenario No. 6 and 8) were most likely to provoke anxiety emotion and they both were uncertain
arrival-blocking events that caused either by the driver himself/herself or by the circumstance. The most probable emotional
responses for all the other scenarios were fear. The four with the highest probability in provoking fear were scenario No. 11,
12, 13 and 14. These four scenarios depicted safety-threaten events caused by the driver himself/herself. The probability of
experiencing shame was generally very low (less than 0.2) across all 16 scenarios, indicating that the shame emotion was not
1
2

The latter group is the baseline category and the former is the comparison group.
OR(anxiety vs. anger) equals to the reciprocal of OR(anger vs. anxiety).

120

T. Zhang, A.H.S. Chan / Transportation Research Part F 27 (2014) 112123

Table 3
Results of multinomial logistic regression predicting discrete emotions by appraisal components.
Appraisal

Chi-square

df

Sig.

Goal relevance
Blame party
Certainty

13.300
138.059
6.411

3
9
3

.004
.000
.093

Model t information: Likelihood ratio = 247.53, Chi-Square (15) = 157.359, p < 0.001; Cox and Snell Pseudo RSquare = 0.179.

Table 4
The effect sizes of the appraisal components, with different category as the baseline category.
Appraisal

Anger vs. shame

Anger vs. fear

Anger vs. anxiety

Anxiety vs. shame

Anxiety vs. fear

Shame vs. fear

Arrival-blocking
Other-blame
Self-other
Self-self
Uncertain

1.309
5.882*
0.267*
0.349*
1.574

1.047
3.289*
0.325*
0.305*
1.314

0.601*
5.291*
0.498*
0.408*
0.919

2.174*
1.115
0.536
0.856
1.712*

1.742*
0.623
0.652
0.747
1.431*

0.800
0.559
1.216
0.873
0.835

The latter category is the baseline category and the former is the comparison category. An OR > 1 indicates a higher probability whereas an OR < 1 indicates
a lower probability, of the outcome falling into the comparison emotion group than falling into the baseline group, in the presence of the corresponding
independent variable.
*
p < 0.05.

Fig. 4. The estimated probability of each discrete emotion in the 16 investigated scenarios.

Table 5
The most likely emotional response in the 16 investigated scenarios.
Goal relevance

Arrival-blocking
Safety-blocking

Certainty

Certain
Uncertain
Certain
Uncertain

Blame party
Other-blame

Self-other

Self-self

Circumstance-blame

Anger
Anger
Anger
Anger

Fear
Fear
Fear
Fear

Fear
Anxiety
Fear
Fear

Fear
Anxiety
Fear
Fear

commonly provoked in the context of driving. The two scenarios that resulted in the highest probability of experiencing
shame were scenario No. 11 and 12, where the participants threatened other drivers safety.
4. Discussion
This study investigated how the emotional responses of drivers, from both dimensional and discrete perspective, are
determined by the three most relevant appraisal components (goal relevance, blame party, and certainty). The dimensional
emotions were directly recorded using the Valence-Arousal emotion space whereas the discrete emotions were indirectly
derived from by applying cluster analysis to valence and arousal ratings. In general, the results supported all the four

T. Zhang, A.H.S. Chan / Transportation Research Part F 27 (2014) 112123

121

hypotheses proposed and showed that both dimensional and discrete emotion models could be predicted by driver appraisals of the trafc scenarios.
Inconsistent with the frequently reported V-shaped relation between valence and arousal (Kuppens, Van Mechelen,
Nezlek, Dossche, & Timmermans, 2007; Lang & Bradley, 2007), no signicant correlation between the two dimensions
was identied (p > 0.05). This is probably due to the cultural differences in the valence-arousal relation. Kuppens,
Tuerlinckx, Russell, and Barrett (2012) have investigated how such a relation differs across different populations. They found
that the slope of the V shape was steepest for western cultures (Canada, Spain) but less steep (Japan, Korea) to almost at
(Hong Kong) for eastern cultures.
The results regarding how arousal and valence dimensions are determined by the appraisal components are in line with
some of the ndings of two previous studies (Nicolle & Goel, 2013; Scherer et al., 2006). For instance, this study showed that
goal relevance was a signicant predictor of the valence dimension, a nding that has been reported by both Scherer et al.
(2006) and Nicolle and Goel (2013). Moreover, consistent with results reported by Scherer et al. (2006), it was found that the
certainty component was a signicant predictor of the arousal rating. Apart from these results, some new relationships have
been identied in the present study. First, Scherer et al. (2006) found that an event caused by a human agency provoked a
higher level of arousal than an event caused by impersonal circumstance. In the present study, the human agency has been
further divided into classes of other-blame, self-other blame and self-self blame. It was found that only when the human
agency was other drivers the arousal rating was signicantly higher than the circumstance-blame condition. The other
two types of human agency exhibited effects on arousal rating similar to those aroused by circumstance. Second, blame party
appraisal was identied to have signicant effects on valence rating in this study, a nding that has not been reported previously. This seems to be a reasonable nding considering that the valence and arousal dimensions are not completely independent from each other. For instance, Kuppens et al. (2012) found that there was a consistent V-shaped relation of arousal
as a function of valence. If the appraisals are sufcient causes of an emotional response (Siemer, Mauss, & Gross, 2007), then
some appraisal components are supposed to account for the common variance in the valence and arousal dimensions. The
nding in the present study, that valence and arousal are both predicted by the blame party component, suggest that the
previously identied V-shaped relationship between the two dimensions is a result of the effects of the blame party
component.
From the discrete perspective, it was found that driving anger was most likely to be provoked by adverse trafc events
caused by other drivers. Similar results have been reported in a number of appraisal theory studies (e.g. Roseman, 1991;
Smith & Kirby, 2009) as well as in many driver emotion studies (Lu et al., 2013; Mesken et al., 2007; Roidl et al., 2013).
The present study also found that in situations where driver safety was threatened, the most likely emotional response
was fear. This nding is consistent with the general notion that a fear response arises from driver perceptions of danger
or threatening stimuli (Schmidt-Daffy, 2013). Unexpectedly, the results showed that drivers would be fearful in four arrival-blocking scenarios (No. 3, 4, 5, and 7). This is probably a result of the currently imperfect mapping relationship between
discrete emotion and dimensional space. According to the mapping relation proposed by Onton and Makeig (2009), there is
some overlap between fear and frustration, which means it is possible that the most likely emotional response to the above
four situations is frustration. According to Lewis (1995), shame is a negative emotion that results from comparison of the
selfs action with the selfs standards. In the context of driving, it has been reported that shame occurs when the driver himself is responsible for blocking other drivers goal (Roidl et al., 2013). Consistent with this result, this study found that the
four scenarios (No. 11, 12, 13, and 3) showing the highest probability in eliciting shame were all situations where the driver
himself was the blame party. In the nal prediction result, however, no scenario was identied as most likely to provoke a
shame emotion probably because there were only a few cases in this cluster. This nding is not surprising as it is generally
accepted that compared to other three emotions, drivers experience less shame while driving (Mesken et al., 2007; Roidl
et al., 2013).
The most important nding from this study was concerned with provocation of driving anxiety. Previous studies have
only identied the effects of goal relevance and concluded that anxiety is mostly associated with events affecting driver
safety (Mesken et al., 2007; Roidl et al., 2013). In the present study, the certainty appraisal component, which has never been
considered in previous driver emotion studies, was proposed as being effective in the differentiation between driving fear
and anxiety. The results showed that arrival-blocking and uncertainty of outcome were critical in anxiety provocation. Such
ndings conrmed the notion that anxiety is a response to an imprecise or unknown threat (Sadock et al., 2007) in the
context of driving. It should be noted that apart from the certainty dimension, other distinctions between fear and anxiety
have been proposed in driving literature. For example, Schmidt-Daffy (2013) demonstrated that drivers felt fearful when current task demands exceeded their perceived capabilities and experienced anxiety when there was a conict between safety
and velocity while driving.
There may be some doubt about the validity of the method applied in this study to generate discrete emotions from the
dimensional space, because the mapping relationship has not been well established. However, it is suggested here that this
method is reasonable for two reasons. First, although an accurate mapping relationship is not available, some general associations, e.g. anger is associated with high arousal and very negative valence, have been commonly accepted. Since the
valence and arousal properties of the four discrete emotions in this study have been well recognized, it is believed that
the discrete emotions generated should be valid. Second, the discrete emotions reveal anticipated associations with the
appraisal components, which, at least partially, supported the validity of the generation method.

122

T. Zhang, A.H.S. Chan / Transportation Research Part F 27 (2014) 112123

This study has some limitations, one being related to whether the designed scenarios were representative of the desired
appraisal framework. Although there are explicit descriptions of the characteristics of three appraisal components in trafc
scenarios, it is still possible that participants would interpret a situation in a way inconsistent with expectations. One
approach to solve this problem might be to ask participants not only to indicate their emotions but also to rate their evaluations of the appraisal components. Such methods have been applied in Scherer et al. (2006) and Siemer et al. (2007). A
second limitation concerns the relatively small driver sample size, which may not represent the general driver population
and may have caused a relative low power in the analyses. Also, the low proportion of women to men and the large discrepancy in driving experience between men and women are potential problems because they prevent gender comparisons and
one gender might have unique emotional responses to some types of scenarios. However, it should be realized that this is
only a preliminary study exploring the feasibility of predicting dimensional driver emotions based on the appraisal components. The conrmation of the proposed hypotheses suggests that this is a promising direction for further research using representative samples to provide more generalized prediction models for emotional experience of both male and female
drivers. Finally, only the three appraisal components identied as determinant by Roseman (1991) have been investigated.
Appraisal components proposed in other appraisal models of emotion may also play important roles in driver emotion prediction. For example, driver evaluation of his/her coping ability to adverse trafc events, which could be measured by driving
skill or driving experience, may partially determine the emotion provoked (Lajunen, Parker, & Stradling, 1998). It is suggested that future research should take more appraisal components into consideration for a more comprehensive appraisal
model to predict driver emotions.
5. Conclusion
To the best of our knowledge, this is the rst study that has investigated how driver emotions, recorded in dimensional
form, may be predicted by appraisal components. It was demonstrated that both valence and arousal are signicantly associated with the blame characteristics of the trafc events, while valence is also predicted by goal relevance and arousal predicted by certainty. There are an increasing number of studies investigating the effects of driver emotions on driving
behaviors from the dimensional perspective. The ndings from the present study may benet such studies by shedding light
on the underlying mechanism of emotion elicitation. This study also contributes to the research on driver emotions by demonstrating that apart from the two commonly studied appraisal components i.e. goal relevance and blame party, the certainty of trafc events also plays an important role in determining how drivers respond emotionally while driving.
Finally, the unique discrete emotion generation method applied in this study may inspire other researchers to explore
new methodologies in driver emotion research.
Acknowledgement
The authors would like to thank LF Li for her assistance with the recruitment of participants and data collection.
References
Abdu, R., Shinar, D., & Meiran, N. (2012). Situational (state) anger and driving. Transportation Research Part F: Trafc Psychology and Behavior, 15, 575580.
Beck, K. H., Daughters, S. B., & Ali, B. (2013). Hurried driving: Relationship to distress tolerance, driver anger, aggressive and risky driving in college students.
Accident Analysis & Prevention, 51, 5155.
Bradley, M. M., & Lang, P. J. (1994). Measuring emotion: The self-assessment manikin and the semantic differential. Journal of Behavior Therapy and
Experimental Psychiatry, 25, 4959.
Briggs, G. F., Hole, G. J., & Land, M. F. (2011). Emotionally involving telephone conversations lead to driver error and visual tunnelling. Transportation
Research Part F: Trafc Psychology and Behavior, 14, 313323.
Cai, H., & Lin, Y. (2011). Modeling of operators emotion and task performance in a virtual driving environment. International Journal of Human-Computer
Studies, 69, 571586.
Christie, I. C., & Friedman, B. H. (2004). Autonomic specicity of discrete emotion and dimensions of affective space: A multivariate approach. International
Journal of Psychophysiology, 51, 143153.
Colibazzi, T., Posner, J., Wang, Z., Gorman, D., Gerber, A., Yu, S., et al (2010). Neural systems subserving valence and arousal during the experience of induced
emotions. Emotion, 10, 377389.
Conati, C., & Zhou, X. (2002). Modeling students emotions from cognitive appraisal in educational games. In Proceedings of the 6th international conference on
intelligent tutoring systems (pp. 944954). Springer-Verlag.
Davidson, R. J., Ekman, P., Saron, C. D., Senulis, J. A., & Friesen, W. V. (1990). Approach-withdrawal and cerebral asymmetry: Emotional expression and brain
physiology: I. Journal of Personality and Social Psychology, 58, 330.
Dula, C. S., Adams, C. L., Miesner, M. T., & Leonard, R. L. (2010). Examining relationships between anxiety and dangerous driving. Accident Analysis &
Prevention, 42, 20502056.
Ekman, P. (1999). Basic emotions. Handbook of Cognition and Emotion, 4, 560.
Evans, J. S. B. (2008). Dual-processing accounts of reasoning, judgment, and social cognition. Annual Review of Psychology, 59, 255278.
Gratch, J., Marsella, S., Wang, N., & Stankovic, B. (2009). Assessing the validity of appraisal-based models of emotion. In Affective computing and intelligent
interaction and workshops (pp. 18).
Hault-Dubrulle, A., Robache, F., Pacaux, M.-P., & Morvan, H. (2011). Determination of pre-impact occupant postures and analysis of consequences on injury
outcome. Part I: A driving simulator study. Accident Analysis & Prevention, 43, 6674.
Kuppens, P., Tuerlinckx, F., Russell, J. A., & Barrett, L. F. (2012). The relation between valence and arousal in subjective experience. Psychological Bulletin, 139,
917940.
Kuppens, P., Van Mechelen, I., Nezlek, J. B., Dossche, D., & Timmermans, T. (2007). Individual differences in core affect variability and their relationship to
personality and psychological adjustment. Emotion, 7, 262.

T. Zhang, A.H.S. Chan / Transportation Research Part F 27 (2014) 112123

123

Lajunen, T., Parker, D., & Stradling, S. G. (1998). Dimensions of driver anger, aggressive and highway code violations and their mediation by safety
orientation in UK drivers. Transportation Research Part F: Trafc Psychology and Behavior, 1, 107121.
Lang, P., & Bradley, M. M. (2007). The international affective picture system (IAPS) in the study of emotion and attention. In Handbook of emotion elicitation
and assessment (pp. 2946). New York and Oxford: Oxford University Press.
Lewis, M. (1995). Shame: The exposed self. Simon and Schuster.
Lu, J., Xie, X., & Zhang, R. (2013). Focusing on appraisals: How and why anger and fear inuence driving risk perception. Journal of Safety Research, 45, 6573.
Mauss, I. B., & Robinson, M. D. (2009). Measures of emotion: A review. Cognition and Emotion, 23, 209237.
Mesken, J., Hagenzieker, M. P., Rothengatter, T., & de Waard, D. (2007). Frequency, determinants, and consequences of different drivers emotions: An onthe-road study using self-reports,(observed) behavior, and physiology. Transportation Research Part F: Trafc Psychology and Behavior, 10, 458475.
Milligan, G. W., & Sokol, L. M. (1980). A two-stage clustering algorithm with robust recovery characteristics. Educational and Psychological Measurement, 40,
755759.
Morgan, J. I., Jones, F. A., & Harris, P. R. (2013). Direct and indirect effects of mood on risk decision making in safety-critical workers. Accident Analysis &
Prevention, 50, 472482.
Nicolle, A., & Goel, V. (2013). Differential impact of beliefs on valence and arousal. Cognition & Emotion, 27, 263272.
OBrien, S. R., Shaw, L., Watson, B. C., & Lennon, A. J. (2012). Do drivers respond emotionally and behaviorally differently to an intentionally anger provoking
driving situation than to an ambiguous, but potentially provocative, one? Results of a self-report survey. In Proceedings of Australasian road safety
research, policing and education conference 2012: Australasian College of Road Safety (ACRS).
Onton, J., & Makeig, S. (2009). High-frequency broadband modulations of electroencephalographic spectra. Frontiers in Human Neuroscience, 3.
Roidl, E., Frehse, B., Oehl, M., & Hger, R. (2013). The emotional spectrum in trafc situations: Results of two online-studies. Transportation Research Part F:
Trafc Psychology and Behavior, 18, 168188.
Roseman, I. J. (1991). Appraisal determinants of discrete emotions. Cognition & Emotion, 5, 161200.
Roseman, I. J. (1996). Appraisal determinants of emotions: Constructing a more accurate and comprehensive theory. Cognition & Emotion, 10, 241278.
Roseman, I. J., & Smith, C. A. (2001). Appraisal theory: Overview, assumptions, varieties, controversies. In K. R. Scherer, A. E. Schorr, & T. E. Johnstone (Eds.),
Appraisal processes in emotion: Theory, methods, research (pp. 319). New York and Oxford: Oxford University Press.
Sadock, B. J., Kaplan, H. I., & Sadock, V. A. (2007). Kaplan & Sadocks synopsis of psychiatry: Behavioral sciences/clinical psychiatry (10th ed.). Lippincott Williams
& Wilkins Press.
Salvador, S., & Chan, P. (2004). Determining the number of clusters/segments in hierarchical clustering/segmentation algorithms. In 16th IEEE international
conference on tools with articial intelligence, 2004. ICTAI 2004 (pp. 576584). IEEE.
Scherer, K., Dan, E., & Flykt, A. (2006). What determines a feelings position in affective space? A case for appraisal. Cognition & Emotion, 20, 92113.
Schimmack, U., & Derryberry, D. (2005). Attentional interference effects of emotional pictures: Threat, negativity, or arousal. Emotion, 5, 5566.
Schmidt-Daffy, M. (2013). Fear and anxiety while driving: Differential impact of task demands, speed and motivation. Transportation Research Part F: Trafc
Psychology and Behavior, 16, 1428.
Shahar, A. (2009). Self-reported driving behaviors as a function of trait anxiety. Accident Analysis & Prevention, 41, 241245.
Siemer, M., Mauss, I., & Gross, J. J. (2007). Same situationdifferent emotions: How appraisals shape our emotions. Emotion, 7, 592600.
Smith, C. A., & Kirby, L. D. (2009). Putting appraisal in context: Toward a relational model of appraisal and emotion. Cognition and Emotion, 23, 13521372.
Stephens, A. N., & Groeger, J. A. (2011). Anger-congruent behavior transfers across driving situations. Cognition & Emotion, 25, 14231438.
Sun, K., Yu, J., Huang, Y., & Hu, X. (2009). An improved valence-arousal emotion space for video affective content representation and recognition. In IEEE
International Conference on Multimedia and Expo, 2009. ICME 2009 (pp. 566569). IEEE.
Taylor, J. E., Deane, F. P., & Podd, J. V. (2007). Driving fear and driving skills: Comparison between fearful and control samples using standardised on-road
assessment. Behavior Research and Therapy, 45, 805818.
Thamm, R. A. (2006). The classication of emotions. In Handbook of the sociology of emotions (pp. 1137). Springer.
Underwood, G., Chapman, P., Wright, S., & Crundall, D. (1999). Anger while driving. Transportation Research Part F: Trafc Psychology and Behavior, 2, 5568.
Yoccoz, N. G. (1991). Use, overuse, and misuse of signicance tests in evolutionary biology and ecology. Bulletin of the Ecological Society of America, 72,
106111.
Zhang, T., & Chan, A. H. (2014). Factor structure for a short driving anger scale: A study based on Chinese drivers. Manuscript submitted to Journal of safety
research.