Accident Analysis and Prevention 40 (2008) 1667–1673

Contents lists available at ScienceDirect

Accident Analysis and Prevention

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

A perceptual map for understanding concern about unsafe

driving behaviours
Ward Vanlaar ∗ , Herb Simpson, Robyn Robertson
Traffic Injury Research Foundation, 171 Nepean Street, Suite 200, Ottawa, Ontario, Canada K2P 0B4

a r t i c l e i n f o a b s t r a c t

Article history: The objective of this paper is to develop a model that can help explain the public’s level of concern associ-
Received 28 September 2007 ated with six dangerous driving behaviours (drinking and driving, speeding, distracted driving, using a cell
Received in revised form 21 May 2008 phone while driving, fatigued or drowsy driving, and using illegal drugs while driving). Understanding the
Accepted 28 May 2008
genesis of concern can be useful in addressing it and leveraging it to improve safe driving. Building on a
risk perception model that was developed previously, the study investigated the relationship between the
Keywords:
level of concern about the unsafe driving behaviours and the perceived level of concern of others about the
Concern
dangerous driving behaviours, the perception of the prevalence of the dangerous driving behaviours, the
Perceptual map
Multidimensional scaling
perception of the level of risk imposed by these dangerous driving behaviours, and the perception of the
Bandwagon effect severity of injuries that can result from them. Data from two independent samples were modeled using
Risk perception multidimensional scaling and logistic regression analysis. Both samples come from telephone surveys;
one was administered to a random sample of 750 drivers in the province of Ontario, Canada in November
2006, the other to a random sample of 1201 drivers across Canada in September 2006. Two dimensions
in particular were found to fit the data well: perceived risk and the perceived level of concern of others.
The results from these analyses are summarized using a perceptual map. The relevance of such a map is
illustrated by explaining the factors that impact levels of concern regarding several of the unsafe driving
behaviours.
© 2008 Elsevier Ltd. All rights reserved.

1. Introduction
Much has been written in the literature about risk perception
and concern for safety hazards (see Renn, 1998; Boholm, 1998, for
an overview). Concern should be distinguished from fear; fear is
a term used for reactions to immediate threats, while concern (or
anxiety) is used for reactions to future or past events (Jackson, 2006;
Warr, 2000). Concern or fear must not be equated with risk percep-
tion as it has been demonstrated that measures of the former do not
measure the same phenomenon as the latter (Rountree and Land,
1996; Ferraro, 1996). “In short, fear[/concern] is not perceived risk;
by all indications, it is its consequence.” (Warr, 2000, p. 454) It may
also be instructive to distinguish between personal concern/fear
for oneself and altruistic concern/fear for others since both may be
prevalent and have distinct consequences (Warr, 2000).
Several theoretical models of risk perception have been devel-
oped including the psychometric model (Fischhoff et al., 1978), the
Basic Risk Perception Model (Sjöberg and Drottz-Sjöberg, 1994),
the social amplification of risk (Kasperson et al., 1988; Burns et al.,
∗ Corresponding author. Tel.: +1 613 238 5235; fax: +1 613 238 5292.
E-mail address: wardv@trafficinjuryresearch.com (W. Vanlaar).
1993) and the Worry about Crime Model (Jackson, 2006). Wåhlberg
(2001) investigated these models (except the Worry about Crime
Model) and concludes that – despite a large body of literature – not
much theory is available in the field of risk perception research.
Nevertheless, several dimensions have been identified in these
models as having an influence on levels of concern; they include
perceived likelihood of occurring and seriousness of the conse-
quences of the offence or risky behaviour (Jackson, 2004; Nilsson,
discussed in Elvik and Vaa, 2004; Vanlaar and Yannis, 2006; Warr
and Stafford, 1983; Wolfgang et al., 1985); perceived prevalence
of the offence or risky behaviour (Jackson, 2004; Nilsson, dis-
cussed in Elvik and Vaa, 2004; Vanlaar and Yannis, 2006); locus
of control—being in control of the situation mitigates feelings of
concern while lack of control exacerbates such feelings (Jackson,
2006; Wåhlberg, 2001); and social amplification—referring to peo-
ple and organizations that can amplify fear or concern through
various ways of communication (Kasperson et al., 1988; Burns et
al., 1993).
Of particular concern regarding the social amplification of con-
cern is the bandwagon effect (McAllister and Studlar, 1991). It
refers to the theory positing the reinforcement of a person’s beliefs
through communal dynamics and means that one’s own beliefs will
be strengthened if one is convinced that others share that belief.

0001-4575/$ – see front matter © 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.aap.2008.05.009
1668 W. Vanlaar et al. / Accident Analysis and Prevention 40 (2008) 1667–1673
Traditionally, criminological research into concern/fear for
crime has investigated these issues primarily as a reaction to irra-
tionally high levels of concern/fear for offences that exist among the
public, and which can have negative consequences on a society as a
whole, such as avoidance behaviour, which may lead to social isola-
tion of neighborhoods (e.g., Conklin, 1975; Skogan, 1990). However,
low levels of public concern may be irrational as well and can be
equally undesirable for a society if it leads to dangerous behaviour
and to lack of public involvement and action. Conversely, pub-
lic concern can have positive effects because it “can motivate an
individual to take personal action by changing their behaviour or
incite them to demand that action is taken by government or other
responsible agencies.” (Vanlaar et al., 2007, p. 15) Concern can
therefore serve as a lever that can be used to influence people’s
behaviour.
This paper addresses the issue of what influences people’s level
of concern. The objective is to develop a model that can help identify
the factors that influence drivers’ level of concern about a variety of
dangerous driving behaviours—in this case, concern about drinking
and driving, speeding excessively, distracted driving, using a cell
phone while driving, fatigued or drowsy driving and using illegal
drugs while driving. Another objective is to obtain information that
provides insight into how similar or dissimilar people think these
dangerous driving behaviours are with regard to risk perception
and its various dimensions as identified in previous paragraphs.
A risk perception model previously developed by Vanlaar and
Yannis (2006) found that the perception of the level of risk imposed
by dangerous driving behaviours and the perception of the preva-
lence of dangerous driving behaviours influenced levels of concern
that drivers had about safety issues. This risk perception model was
built using multidimensional scaling (MDS) and proved particularly
useful in gauging whether or not people discern differences in risks
between dangerous driving behaviours that may seem very similar
but that are really different, at least to some extent.
MDS is suitable to investigate such perceived similarities and
dissimilarities because its prime objective is to build a perceptual
map in which distances between items in this map truly corre-
spond to differences in perceptions. For example, in Vanlaar and
Yannis’ perceptional map (2006) driving while impaired by alco-
hol was located very closely to driving while impaired by illegal
drugs, which means that the public thinks both are very similar
with regard to their prevalence and risk for an accident. However,
it can be concluded from the literature that this is not necessar-
ily the case. Such belief systems that exist among the public may
impede dealing efficiently with both issues, for example if there
would be pressure from the public to invest too many resources
than can be justified based on the evidence into one issue (e.g.,
driving while impaired from illegal drugs) to the detriment of the
other (e.g., driving while impaired from alcohol).
Building on this previously developed model, two more dimen-
sions were investigated in this paper: the level of concern of others
about those dangerous driving behaviours (i.e., the bandwagon
effect) and the perception of the severity of injuries that can result
from the dangerous driving behaviours. More precisely, the study
was designed to test the following hypotheses:
(1) Level of concern of others: if a person believes others are con-
cerned about dangerous driving behaviours, this will heighten
an individual’s level of concern about those behaviours.
(2) Prevalence or magnitude: if a person believes that a dangerous
driving behaviour is prevalent, his/her level of concern about it
will increase.
(3) Level of risk: if a person believes that the dangerous driving
will increase his/her chances of a crash, he/she will be more
concerned about those behaviours.
(4) Severity of injuries: if a person is convinced that the dangerous
driving behaviour typically leads to crashes with more severe
injuries then his/her level of concern about it will increase.
Ultimately, understanding the dynamics between these dimen-
sions and their role in the genesis of public concern about unsafe
driving behaviours can be useful for ensuring that concern is con-
sistent with the level of danger and risk posed by them. Appropriate
levels of concern might motivate the public to behave appropriately
and/or incite them to demand action to deal with them consistently.
2. Method
2.1. Sampling procedure and participants
Data from two independent random samples were used. Both
samples come from telephone surveys; one was administered to a
random sample of 750 drivers in the province of Ontario, Canada
in November 2006, the other to a random sample of 1201 Canadian
drivers in September 2006. Both surveys required an average of
approximately 15 min to complete.
Criteria for inclusion in the Ontario survey were: having a valid
driver’s license, residing in Ontario, and having driven in the past
30 days. Among the 3131 households contacted in Ontario in which
a person was asked to participate, 2052 (65.5%) refused, 270 (8.6%)
were not qualified, and 59 (1.9%) terminated the interview before
it was completed. The Ontario sample was weighted according to
gender and age to avoid bias and ensure it was representative of
the Ontario population. The final group of respondents for the study
included 750 Ontario drivers. Their age ranged from 16 to 93 (mean
of 48; median of 47). Forty four percent of all participants were
male.
Criteria for inclusion in the Canadian survey were: having a valid
driver’s license, residing in Canada, and having driven in the past
30 days. Among the 6076 households contacted in which a person
was asked to participate, 4418 (73%) refused and 457 (7.5%) were
not qualified. The Canadian sample was stratified by province and
also weighted according to gender and age to avoid bias. The final
group of respondents for the study included 1201 Canadian drivers.
The median age category was 45–49. Forty-three percent of the
participants were male.
2.2. Questionnaire
To determine the level of concern associated with the danger-
ous driving behaviours, survey respondents were asked to rate each
of them using a six-point Likert-type scale, where one means they
think the behaviour is “not a problem at all” and six means they
think it is “an extremely serious problem”. The behaviours probed
in the Canadian sample included drinking and driving, speeding
excessively, distracted driving, using a cell phone while driving
(both hand-held and hands-free) and fatigued or drowsy driving.
The Ontario survey also included the issue of using illegal drugs
while driving.
To determine perceptions about these behaviours, respondents
in the Ontario sample were also asked to answer the following ques-
tions about each of the listed behaviours (including drivers using
illegal drugs). First, they were asked how concerned they thought
others were about each of the behaviours, using the same six-point
scale as before (i.e., dimension 1: level of concern of others). Second,
respondents were asked to indicate what percent of drivers they
believe engage in each of the behaviours (i.e., dimension 2: preva-
lence or magnitude). Third, they had to indicate, on a scale from
one to six, how likely they think it is for drivers engaging in each of
the behaviours to cause a crash where one means “very unlikely”
 W. Vanlaar et al. / Accident Analysis and Prevention 40 (2008) 1667–1673
and six means “very likely” (i.e., dimension 3: level of risk). Finally,
respondents were asked whether they thought crashes caused by
these behaviours lead to more severe crashes than those caused
by a reference category (i.e., dimension 4: severity of injuries)—the
reference category was “fatigued or drowsy driving” as the Ontario
sample comes from a study dedicated to this issue (see Vanlaar et
al., 2007, 2008).
2.3. Data analysis
The data were analyzed using a multivariate approach and tak-
ing account of the stratified and weighted sampling design in the
case of the Canadian sample and the weighted sampling design
in case of the Ontario sample. More precisely, MDS and logistic
regression analysis in Stata, release 9 (StataCorp, 2005) were used
to examine the relationships between the different dimensions
and the dependent variable (i.e., the individual’s own level of con-
cern about the dangerous driving behaviours) for each of the six
behaviours.
According to Coxon (2004) MDS refers to a family of models
where the structure in a dataset is represented graphically by the
relationships between different points in a space. Results of MDS
can be presented in a perceptual map that visualizes the solution
in a two-dimensional or three-dimensional space. The dimensions
are then interpreted using additional information (see Kruskal and
Wish, 1978; de Leeuw and Heiser, 1982; Wish and Carroll, 1982;
Young, 1985; Coxon, 2004). While there are empirical ways to deter-
mine the number of dimensions, a qualitative assessment can be
valuable as well (Vanlaar and Yannis, 2006).
In this paper, results from the dependent variable that mea-
sured the respondents’ own level of concern for each of the six
probed behaviours (i.e., drinking and driving, speeding excessively,
distracted driving, using a cell phone while driving, fatigued or
drowsy driving and using illegal drugs while driving) in both the
Ontario sample and the Canadian sample were analyzed using MDS.
The multidimensional space that was obtained using this approach
was then interpreted with the aid of the variables containing the
information about the respondents’ perception of each of the four
Fig. 1. Model fit for six dangerous driving behaviours using Shepard diagrams.
1669
dimensions (i.e., level of concern of others, prevalence, risk, and
severity of consequences).
This process of modeling and interpreting the data is explained
in more detail in the paragraph entitled “Model fit” in Section 3,
while the results from the data analysis itself are presented in the
paragraph entitled “Perceptual map” in Section 3.
3. Results
3.1. Model fit
Developing and interpreting the perceptual map involved three
steps. First, a two-dimensional and a three-dimensional model
were fitted using a correlation matrix of the respondents’ level of
concern about the six behaviours in both samples. The solution
from the Canadian sample was almost identical to the one from
the Ontario sample, indicative of the robustness of the findings.
From then on, only data from the Ontario sample were further
investigated because, unlike the Canadian sample, this sample
also contained information about the four dimensions that can
possibly influence the respondent’s level of concern (see Section
2.2).
Logically, the three-dimensional model obtained with the
Ontario data fit the data better than the two-dimensional one
according to two model fit statistics: 84% of the variability is
explained by the three-dimensional model according to the Mardia
fit measure 2 (Mardia et al., 1979) versus only 71% in the two-
dimensional model; and, the overall value for Kruskal’s Stress 1 in
the three-dimensional model is 0.12, while this value is 0.15 with
only two dimensions, indicative of a model that fits the data slightly
less well.
However, this is not to say that the two-dimensional model does
not fit the data. This is illustrated in Fig. 1 which displays a Shepard
diagram for each of the six behaviours in a two-dimensional space
(“cell” refers to using a cell phone, either hands-free or hand-held,
“distract” to distracted driving, “drink” to drinking and driving,
“drugs” to using illegal drugs while driving, “fatigue” to fatigued
or drowsy driving, and “speed” to excessive speeding). Points that
1670 W. Vanlaar et al. / Accident Analysis and Prevention 40 (2008) 1667–1673

Table 1
Results from the logistic regression analysis for six dangerous driving behaviours (models only contain significant variables)

Variable Odds ratio Linearized S.E. p-Value

Model 1: Drinking and driving

1—others not concerned Reference category N/A N/A
2 0.75 0.68 0.751
3 3.04 2.27 0.137
4 9.09 7.26 0.006
5 20.76 15.17 0.000
6—others very concerned 83.26 66.04 0.000
Prevalence 1.03 0.01 0.009

Model 2: Speeding excessively

1—others not concerned Reference category N/A N/A
2 1.06 0.84 0.944
3 1.70 1.27 0.481
4 1.99 1.48 0.355
5 6.07 4.54 0.016
6—others very concerned 7.96 5.82 0.005
1—not risky Reference category N/A N/A
2 9.14 9.41 0.032
3 20.17 19.56 0.002
4 44.85 43.80 0.000
5 54.53 52.14 0.000
6—very risky 170.96 169.52 0.000

Model 3: Cell phones

1—others not concerned Reference category N/A N/A
2 1.21 0.64 0.714
3 0.93 0.44 0.877
4 2.16 1.05 0.111
5 2.67 1.45 0.069
6—others very concerned 3.93 2.24 0.017
1—not risky Reference category N/A N/A
2 4.17 2.75 0.031
3 9.95 6.47 0.000
4 15.61 10.09 0.000
5 52.31 37.16 0.000
6—very risky 58.92 45.37 0.000

Model 4: Driver distraction

1—not risky Reference category N/A N/A
2 1.81 1.75 0.542
3 5.24 4.90 0.077
4 5.02 4.55 0.076
5 10.08 9.50 0.014
6—very risky 8.38 8.53 0.037

Model 5: Fatigued or drowsy driving

1—others not concerned Reference category N/A N/A
2 1.00 0.58 0.995
3 1.58 0.85 0.396
4 4.73 2.66 0.006
5 8.96 5.29 0.000
6—others very concerned 8.84 5.95 0.001
Prevalence 1.02 0.01 0.005

Model 6: Illegal drugs

1—others not concerned Reference category N/A N/A
2 0.76 0.42 0.611
3 1.26 0.63 0.649
4 3.34 1.69 0.018
5 6.47 3.57 0.001
6—others very concerned 35.22 21.37 0.000
1—not risky Reference category N/A N/A
2 4.23 3.65 0.096
3 3.79 3.17 0.112
4 8.80 7.31 0.009
5 6.69 5.45 0.020
6—very risky 4.56 3.71 0.062
Prevalence 1.02 0.01 0.013

deviate from the diagonal indicate lack of fit (StataCorp, 2005). As
can be seen, with the exception of speed and fatigue, the model
fit is close to perfect according to this test. Overall, both the three-
dimensional model and the two-dimensional model obtained with
the Ontario sample fit the data well and are more generally applica-
ble beyond the province of Ontario, as illustrated with the Canadian
data.
In a second step, logistic regression was used to investigate
the relationship between the first three dimensions (dimen-
sion 1: level of concern of others; dimension 2: prevalence;
 W. Vanlaar et al. / Accident Analysis and Prevention 40 (2008) 1667–1673 1671

dimension 3: risk) and the respondents’ level of concern to

gain insight into factors that could possibly influence people’s
level of concern about dangerous driving behaviours. Six models
were fitted, one for each of the dangerous behaviours. In each
model, the respondent’s own level of concern about a particular
behaviour served as the dependent variable; for this purpose, the
six-point scale was recoded into a dummy variable with values
one (not a problem at all), two and three as the reference cate-
gory and values four, five and six (an extremely serious problem)
as the other category. The independent variables in these analy-
ses were dimensions one, two and three. Because of the format
of dimension 4 (perception of the severity of injuries), designed
to probe fatigued or drowsy driving in particular by means of
a comparison with a reference category, this dimension was not
included in the logistic regression. Instead, univariate distribu-
tions were used in a third step to help interpret the perceptual
map.
It was found that in five of the six logistic regression models,
Fig. 2. Percent of Ontario drivers who think the six dangerous driving behaviours
“concern of others” influenced significantly the respondent’s own are serious or extremely serious.
level of concern (see Table 1; note that each model in this table
only contains significant variables); i.e., this variable was significant
in the model for drinking and driving, speeding excessively, using Fig. 2 shows the percent of respondents who think the listed
a cell phone while driving, fatigued or drowsy driving, and using problems are serious or extremely serious (ranking of five or six on
illegal drugs while driving. This means that for all the dangerous the six-point answering scale). A series of univariate distributions
behaviours included in this research, with the exception of driver like this can be used to interpret the two-dimensional perceptual
distraction, people’s level of concern about these behaviours was map. Such an interpretation can be seen in Fig. 3. In this solution,
influenced by how concerned they think others are about them. the most promising dimensions according to the available data (i.e.,
Simply stated, the more concerned you believe others are about a combination of the results from the logistic regression analyses
some dangerous driving behaviour, then your level of concern about and the univariate distributions) were used to interpret the map;
it will increase as well. This result can be interpreted through the these dimensions are “how concerned respondents think others are
theoretical framework of the bandwagon effect and suggests that about the six dangerous driving behaviours” (i.e., dimension 1) and
such an effect exists among the public with regards to traffic safety “how risky they think these behaviours are” (i.e., dimension 3).
issues.
In three models – those related to drinking and driving, fatigued 3.2. Perceptual map
or drowsy driving, and using illegal drugs while driving – the per-
ception of the prevalence of these dangerous driving behaviours Using a combination of model fit measures, logistic regres-
was found to affect significantly the respondent’s own level of con- sion analysis and an interpretation of univariate distributions, as
cern (see Table 1). In other words, a person’s level of concern is explained in previous paragraphs, thus culminated in a percep-
positively related to their perception of the magnitude of the prob- tual map (cf. Fig. 3) that combines two dimensions to explain
lem. dis/similarities in the public perception of six dangerous driving
Finally, the perception of the level of risk associated with cer- behaviours according to those dimensions. These dimensions are
tain dangerous driving behaviours – speeding excessively, using a “how risky people think these dangerous driving behaviours are”
cell phone while driving, distracted driving and using illegal drugs and “how concerned people think others are about these dangerous
while driving – was found to be related to the level of concern (see driving behaviours”.
Table 1). Someone who believes these behaviours are risky is more
likely to be concerned about them.
Modeling the data using logistic regression analysis thus lead to
the identification of three possible candidates to interpret a two-
dimensional or three-dimensional space, namely the respondent’s
perception of how concerned others are about the behaviours in
question, the respondent’s perception of the prevalence of the
behaviours, and the respondent’s perception of how risky the
behaviours are.
In a third and last step of the analytic approach, univariate
frequencies of the four dimensions were used to complete the inter-
pretation process. While it was found that, from an empirical point
of view, a three-dimensional model fit the data better (see above), a
two-dimensional model was logically more consistent when giving
meaning to the dimensions, based on these univariate frequencies.
As stated previously, overall, the two-dimensional model fit the
data well (71% of the variability is explained). It warrants mention-
ing, however, that the item “fatigued or drowsy driving” fit the data
poorly compared to the overall fit of the two-dimensional model.
This will turn out to be important when interpreting the perceptual Fig. 3. Two-dimensional perceptual map of concern about six dangerous driving
map coming from the MDS. behaviours.
1672 W. Vanlaar et al. / Accident Analysis and Prevention 40 (2008) 1667–1673
Such a perceptual map can best be understood by reference to an
example involving the dangerous behaviour of drinking and driv-
ing. A respondent’s own level of concern about drinking and driving
is the result of how risky he/she thinks drinking and driving is and
how concerned he/she thinks others are about it. As such, it is sit-
uated in the ‘riskier/more concern’ quadrant. Generally speaking,
respondents seem to think that drinking and driving is very risky
(position on the vertical axis on the ‘riskier’ side) and that others
are more concerned about it as well (position on the horizontal axis
on the ‘more concern’ side). These two assessments combined lead
to a high level of concern among drivers as can also be seen in Fig. 2,
which shows that over 82% of the respondents think that drinking
and driving is a serious or extremely serious problem.
Drivers using cell phones, on the other hand, caused consider-
ably less concern among respondents; relative to other behaviours,
it was at the bottom of the list in Fig. 2. According to the perceptual
map this can be explained by its position in the ‘less risky/less con-
cern’ quadrant, corresponding to a perceived lack of concern among
others about drivers using cell phones (position on the horizontal
axis on the ‘less concern’ side) and a low perceived risk imposed
by it (position on the vertical axis on the ‘less risky’ side). Gen-
erally speaking, respondents believe others typically are not that
concerned about drivers using cell phones and they also think it is
not that risky, leading to a lower level of concern about it.
The level of concern about the other dangerous driving
behaviours can be explained in the same way. However, with
respect to fatigued and drowsy driving, the same rationale cannot
wholly be adopted based on this particular perceptual map. Given
this item’s bottom place in Fig. 2, one would expect to see it in
the ‘less risky/less concern’ quadrant at the top right hand side in
Fig. 3, rather than in the bottom right hand side ‘riskier/less concern’
quadrant. This can be explained by a poorer model fit for this partic-
ular item according to the MDS solution, which is corroborated by
the results of the logistic regression analysis—risk perception was
not found to have a significant influence on the individual’s level
of concern in the logistic regression model for fatigued or drowsy
driving.
4. Discussion
The objective of this paper was to develop a tool that can pro-
vide insights into the genesis of people’s concern about dangerous
driving behaviour. The impetus for gaining such an understanding
was motivated by the understanding that concern about dangerous
behaviour does not necessarily only have negative consequences for
a society but can also have positive reverberations. It was argued
that, overall, people will more likely refrain from engaging in dan-
gerous driving practices if they are concerned about it or that they
will demand action to do something about it. Concern thus serves
as a lever that can be used by social engineers to increase the level
of traffic safety.
The influence of four dimensions on an individual’s level of con-
cern about six dangerous driving behaviours was investigated using
two different multivariate analytic techniques (logistic regression
analysis and MDS). Using logistic regression analysis, it was found
that an individual’s own level of concern is determined in part
by their perception of other people’s level of concern as well as
their perception of the prevalence and risks associated with those
behaviours. These findings corroborate what we know from the
literature. Two dimensions in particular – concern of others and
risk – were also found to fit the MDS model particularly well and
facilitated its interpretation based on a two-dimensional percep-
tual map. Despite a poorer fit for fatigued or drowsy driving, overall,
the two-dimensional model coming from the MDS of the Ontario
sample fit the data well. It was also argued that the findings can
be generalized beyond the Ontario population as they were repli-
cated using Canadian data—an almost identical perceptual map was
found using the Canadian sample.
While this model can be regarded as an extension of a previously
developed one (see Vanlaar and Yannis, 2006), it is still limited.
For example, it does not address such factors as ‘locus of con-
trol’, which is also known to influence levels of concern (Wåhlberg,
2001). Believing that the hazards imposed by dangerous driving
can be controlled by one’s own behaviour tends to reduce the level
of concern. This may create a false sense of security and lead to
displaying dangerous behaviour. While this dimension could not
be investigated for each of the behaviours included in this study,
it was possible to evaluate its influence with respect to fatigued
or drowsy driving because the Ontario survey included pertinent
questions. Indeed, it was found that respondents in the Ontario
study believed there were several effective tactics that they could
use to overcome this problem, which may have reduced their level
of concern about it (Vanlaar et al., 2007, 2008).
Other models include dimensions such as perceptions of infor-
mal social control that can also influence a person’s level of
concern (Jackson, 2004). It was also suggested researchers distin-
guish explicitly between concern (or fear) for oneself and altruistic
concern (Warr, 2000). This was a limitation in this study as such
a distinction was not made. Another useful clarification may be
to distinguish more explicitly between being concerned for a crash
caused by another driver who engages in one of the dangerous driv-
ing behaviours versus concern for crashing yourself if you engage
in one of those behaviours. More research is needed to elaborate
on the perceptual map that was developed in this paper.
In the interim, this perceptual map provides a heuristic model
for gauging where dangerous driving behaviours sit on the public
agenda and where efforts might best be placed to raise awareness
and concern among the public about these behaviours. For example,
the limited evidence that is available about distracted driving does
suggest that it is an important traffic safety issue (Hedlund et al.,
2006; Wang et al., 1996). Given the dangers imposed by distracted
driving, the public’s level of concern about it seems incongruous.
Focusing on both dimensions that were found to be significant can
facilitate raising public awareness and concern. For example, cap-
italizing on the fact that people are more concerned about unsafe
driving practices when they believe others are, may serve as a lever
to create a bandwagon effect (McAllister and Studlar, 1991) based
on communal reinforcement dynamics; or, raising awareness about
the risks associated with distracted driving may incite drivers to
behave accordingly.
While it is acknowledged that such an awareness-raising
approach will have little or no effect with persistent offenders, it
may be a promising means to influence levels of concern among
the majority of the public about distracted driving and other unsafe
driving practices, which may, in turn, influence their behaviour.
The perceptual map also reveals nuances on the public’s per-
ception of several dangerous driving behaviours that cannot be
obtained simply by applying logistic regression analysis or investi-
gating univariate frequencies. For example, it is apparent from the
relatively small distance between drinking and driving and using
illegal drugs and driving in the perceptual map that people perceive
these issues as very similar with regard to the risks they imposed
and the perceived level of concern that others have about them. This
was also found in the previously developed perceptual map using
data from 23 European countries (see Vanlaar and Yannis, 2006).
The dimensions in this perceptual map were perceived risk and per-
ceived prevalence. However, drinking and driving and using illegal
drugs and driving are anything but similar in terms of prevalence
and risk (Vanlaar, 2007). There are many different illegal drugs
 W. Vanlaar et al. / Accident Analysis and Prevention 40 (2008) 1667–1673
and these are highly complex chemicals. The pharmacodynamics
of these chemicals vary, as do their users and prevalence levels. The
public may overestimate risk and prevalence associated with drugs
other than alcohol and as a consequence attribute less concern to
other important road safety issues. The challenge is to communi-
cate balanced messages to the public about these issues. A tool such
as a perceptual map can facilitate this.
Acknowledgements
Financial support to carry out the Canadian survey was provided
by Transport Canada, the Brewers Association of Canada and Toyota
Canada, Inc. The Ontario survey was made possible by a donation
from TD Meloche Monnex Group. The authors acknowledge with
appreciation the financial support from these sponsors.
References
Boholm, Å., 1998. Comparative studies of risk perception: a review of twenty years
of research. Journal of Risk Research 1 (2), 135–163.
Burns, W.J., Slovic, P., Kasperson, R.E., Kasperson, J.X., Renn, O., Emani, S., 1993.
Incorporating structural models into research on the social amplification of
risk: Implications for theory construction and decision making. Risk Analysis
13, 611–623.
Conklin, J., 1975. The Impact of Crime. Macmillan, New York.
Coxon, A.P.M., 2004. Multidimensional scaling. In: Lewis-Beck, M.S., Bryman, A.,
Liao, T.F. (Eds.), The Sage Encyclopedia of Social Science Research Methods. Sage,
Thousand Oaks.
de Leeuw, J., Heiser, W., 1982. Theory of multidimensional scaling. In: Krishnaiah,
P.R., Kanal, L.N. (Eds.), Handbook of Statistics, vol. 2. North-Holland Publishing
Company, Amsterdam, pp. 285–316.
Elvik, R., Vaa, T., 2004. The Handbook of Road Safety Measures. Elsevier, Oxford, UK.
Ferraro, K.F., 1996. Women’s fear of victimization: shadow of sexual assault? Social
Forces 75, 667–690.
Fischhoff, B., Slovic, P., Lichtenstein, S., Read, S., Combs, B., 1978. How safe is safe
enough? A psychometric study of attitudes towards technological risks and
benefits. Policy Sciences 9, 127–152.
Hedlund, J., Simpson, H.M., Mayhew, D.R., 2006. International Conference on Dis-
tracted Driving: Summary of Proceeding and Recommendations, Ottawa: TIRF
and CAA.
Jackson, J., 2004. Experience and expression. Social and cultural significance in the
fear of crime. British Journal of Criminology 44, 946–966.
Jackson, J., 2006. Introducing fear of crime to risk research. Risk Analysis 26 (1),
253–264.
1673
Kasperson, R.E., Renn, O., Slovic, P., Brown, H.S., Emel, J., Goble, R., Kasperson, J.X.,
Ratick, S., 1988. The social amplification of risk: a conceptual framework. Risk
Analysis 8, 177–188.
Kruskal, J.B., Wish, M., 1978. Multidimensional scaling. In: Sage University Paper
Series on Quantitative Applications in the Social Sciences, number 07-011. Sage
Publications, Newbury Park, CA.
Mardia, K.V., Kent, J.T., Bibby, J.M., 1979. Multivariate Analysis. Academic Press, New
York.
McAllister, I., Studlar, D.T., 1991. Bandwagon, Underdog, or Projection? Opinion Polls
and Electoral Choice in Britain, 1979–1987. The Journal of Politics 53, 720–740.
Renn, O., 1998. The role of risk perception for risk management. Reliability Engineer-
ing and System Safety 59, 49–62.
Rountree, P.W., Land, K.C., 1996. Perceived risk versus fear of crime: empirical
evidence of conceptually distinct reactions in survey data. Social Forces 74,
1353–1376.
Sjöberg, L., Drottz-Sjöberg, B.M., 1994. Risk perception of nuclear waste: experts
and the public. Rhizikon: Risk Research Report No. 16. Centre for Risk Research,
Stockholm School of Economics, Stockholm.
Skogan, W.G., 1990. Disorder and Decline: Crime and the Spiral of Decay in American
Neighborhoods. Free Press, New York.
StataCorp, 2005. Stata Statistical Software. Release 9. StataCorp LP, College Station,
TX.
Vanlaar, W., 2007. Substance use and driving—the state of knowledge. Smartrisk
Learning Series, Teleconference, January 29th 2007.
Vanlaar, W., Yannis, G., 2006. Perception of road accident causes. Accident Analysis
and Prevention 38, 155–161.
Vanlaar, W., Simpson, H., Mayhew, D., Robertson, R., 2007. Fatigued and Drowsy Driv-
ing. Attitudes, Concern and Practices of Ontario Drivers. Traffic Injury Research
Foundation, Ottawa.
Vanlaar, W., Simpson, H., Mayhew, D., Robertson, R., 2008. Fatigued and drowsy
driving: a survey of attitudes, opinions and behaviors. Journal of Safety Research
39, 303–309.
Wåhlberg, A.E.A.F., 2001. The theoretical features of some current approaches to risk
perception. Journal of Risk Research 4 (3), 237–250.
Wang, J., Knipling, R.R., Goodman, M.J., 1996. The role of driver inattention in Crashes:
new statistics from the 1995 crashworthiness data system (CDS). In: 40th Annual
Proceedings: Association for the Advancement of Automobile Medicine, AAAM,
Des Plaines, IL, pp. 377–392.
Warr, M., 2000. Fear of crime in the United States: avenues for research and policy.
Criminal Justice 4, 451–489.
Warr, M., Stafford, M.C., 1983. Fear of victimization: a look at the proximate causes.
Social Forces 61, 1033–1043.
Wish, M., Carroll, J.D., 1982. Multidimensional scaling and its applications. In:
Krishnaiah, P.R., Kanal, L.N. (Eds.), Handbook of Statistics, vol. 2. North-Holland
Publishing Company, Amsterdam, 317–345.
Wolfgang, M.E., Figlio, R.M., Tracy, P.E., Singer, S.I., 1985. The National Survey of Crime
Severity. U.S. Government Printing Office, Washington, DC.
Young, F.W., 1985. Multidimensional scaling. In: Kotz, S., Johnson, N.L., Read, C.B.
(Eds.), Encyclopedia of Statistical Sciences, 5. Wiley, New York, pp. 649–659.