Download Report

Laila M. Martinussen
Technical University of Denmark
[email protected]
Implicit attitudes towards risky and safe driving in a Danish sample
1
INTRODUCTION
Since the work of Greenwald and Banaji (1995), the literature on attitudes distinguishes
between implicit or automatic attitudes, and explicit or deliberate attitudes. Explicit attitudes
are conscious evaluative judgments that have their roots in propositional reasoning
(Gawronski and Bodenhausen, 2006). Implicit attitudes are attitudes that reflect
“introspectively unidentified (or inaccurately identified) traces of past experience”
(Greenwald and Banaji, 1995, p. 5). Measures of implicit cognition reveal associative
information that people are either unwilling to share or that they are not conscious of, and
therefore not able to share (Nosek, Greenwald, and Banaji, 2007). These associative
evaluations can be described as automatic affective reactions resulting from a particular
association to that stimulus.
Implicit cognition have been tested in a broad range of disciplines including
including social and cognitive psychology (Fazio and Olson, 2003), clinical psychology
(Teachman et al., 2001), developmental psychology (Baron and Banaji, 2006; Dunham et al.,
2004; Phelps et al., 2000), market research (Maison et al., 2001, health psychology
(Teachman et al., 2003), and recently also in traffic psychology (Harre and Sibley, 2007;
Hatfield et al., 2008; Sibley and Harre, 2009a; Sibley and Harre, 2009b). Implicit cognition
has been shown to predict behavior particularly well if the behavior is associated with social
desirability concerns and/or if a decision must be made spontaneously. Driving behavior is
characterized by frequent decisions made under time pressure that potentially can be costly
and/or dangerous both for the driver and other road users. Further, in our society, driving in a
safe manner is socially desirable as safety is largely promoted. Thus, safe driving is expected,
not only by car users, but by all road users. Because drivers have to make frequent decisions
under time pressure and self-reports of the intention to drive safely (or not) can be socially
sensitive, the motivation behind the present study was to measure implicit attitudes towards
safe and risky driving.
Measuring implicit driving-related cognition
Research applying implicit cognitive methods in traffic psychology is quite limited. To the
authors knowledge there are only four studies so far (Harre and Sibley, 2007; Hatfield et al.,
2008; Sibley and Harre, 2009a; Sibley and Harre, 2009b). Thus, there are no standard
methods available to test implicit attitudes towards safe and risky driving. The present study
applies the method called the “Go/No-go Association Task” (GNAT; Nosek and Banaji,
2001) to the traffic psychology settings for the first time (to the authors knowledge). The
GNAT is similar to its closest relative, the implicit association test (IAT), in that it assesses
strengths of associations between concepts in simple computer-administered categorizationtasks (Nosek and Banaji, 2001). In an IAT, the participants are typically asked to distinguish
between two contrasting pairs of target concepts like for example men versus women or
speeding versus keeping the speed limit, and contrasting pairs of attributes or evaluations, for
example good versus bad, or pleasant versus unpleasant. In critical trial blocks, the
participants distinguish between four concepts at the same time. For example, they may be
asked to press one response key when a word is shown that represents “speeding” or “bad”,
but to press a different key when the word represents “keeping the speed limit” or “good.”
The instructions are later changed such that in subsequent trial blocks, “speeding” and “good”
2
share a response key, whereas “keeping the speed limit” and “bad” share the other key.
Differences in the task performance between critical blocks are interpreted as indicators of the
relative preference of a person, that is, her/his implicit attitude.
The Implicit Association Test uses contrasting target concepts, or pairs of
attitude objects. However, having a positive attitude towards one attitude object does not
necessarily imply that one has a negative attitude towards the “opposite” attitude object.
Further, there might not always be a natural opposite concept to the concept under
investigation. For example, having a positive attitude towards talking on the mobile phone
while driving, does not automatically mean to have a negative attitude towards not talking on
the phone when driving. Thus, it would be useful to measure attitudes towards single attitude
objects. This is what the GNAT has to offer Different from the IAT, the GNAT does not need
contrasting pairs of attitude objects in order to test the attitudes, as with the GNAT one can
assess attitudes towards concepts without a clear opposite attitude object. Similar to the IAT,
the theory behind the GNAT is that it is easier for people, i.e., goes faster, to associate
concepts that are more strongly associated in the mind than concepts that are not (for further
reading about IAT and the GNAT see Nosek and Banaji, 2001; Nosek et al., 2007).
Measuring driving behavior
Driving behavior can be both costly, time consuming and unethical to measure directly.
Therefore, researchers often measure driving behavior through self-report measures. Two of
the most influential and most frequently applied instruments to measure driving behavior and
driving skills are respectively, the Driver Behavior Questionnaire (DBQ, Reason et al., 1990)
and the Driving Skill Inventory (DSI, Lajunen and Summala, 1995). The DBQ measures
aberrant driving behavior by asking drivers how frequent drivers perform intentional
violations, and unintentional errors and lapses while driving (for further reading see
Martinussen et al., 2013; Reason et al., 1990). The DSI measures driving skills by asking how
good drivers consider themselves to be in perceptual-motor skills (technical driving skills)
and safety skills (abiding rules and considering other road users) (for further reading see
Lajunen and Summala, 1995). Both the DBQ and the DSI have been shown to predict selfreported accidents (de Winter and Dodou, 2010).
The present research
The aims of the current study are (1) to develop a GNAT which can be used to assess implicit
attitudes towards risky and safe driving; (2) to explore the relationship between implicit
attitudes towards risky and safe driving, and self-reported driving behavior and driving skills
as measured with two established instruments: the DBQ, and the DSI.
METHOD
Design of the GNAT
The current GNAT consisted of two target categories, namely pictures of risky and safe
driving situations, and two attribute dimensions, namely good and bad words. The implicit
attitude is assessed by measuring the strength of association between the target situation
3
(risky versus safe) and the evaluative dimension (good versus bad). The GNAT works by
presenting the stimuli for a short time on the computer screen, one stimulus at a time. The
participants are asked to press a response button (“go”) if the stimulus on the screen belongs
to either a given target situation (for example risky driving) or a given evaluation (for
example good). If the stimulus does not belong to either of these categories, then the
participants are asked to do nothing (“no-go”). The participants are given a short deadline
(less than one second) for their decision, after which the computer proceeds automatically.
Measures of task performance under the various pairings (e.g., risky driving + good) are
computed from the proportions of correct and wrong responses. The difference in task
performance between situation/evaluation pairings (e.g., risky driving + good vs. risky driving
+ bad) reflects the association between that kind of situation and its implicit evaluation by the
participant. This association is taken to be a measure of the participants’ automatic or implicit
attitude.
The proportion of correct responses in the GNAT is a trade-off between
response speed and response accuracy: Increasing the speed of one’s responses increases the
potential for errors, and the other way around. To compensate for potential differences in the
participants’ response strategies (fast versus error-free), signal-detection theory’s d’ measure
of sensitivity was used as a measure of task performance. Further, two different response
strategies were induced for each participant in a controlled way, namely by repeating all parts
of the GNAT with two response deadlines: 750 milliseconds (ms) and 600 ms.
From a traffic-safety point of view, two performance patterns in the GNAT
indicate desirable attitudes: performance should be greater when the “go” response is required
for (a) pictures of risky situations and negative (rather than positive) words, and (b) pictures
of safe situations and positive (rather than negative) words.
Selection of GNAT materials
The stimuli used as the target categories were pictures of respectively risky and safe driving
situations, and the attributes were respectively “good” and “bad” words (see Figure 1 and
Table 1). The reason for using pictures as stimulus was that the link between attitude and
behavior is stronger if the measure used is similar to the actual behavior (Ajzen and Fishbein,
1977). The pictures used as stimuli in the current GNAT study were borrowed from the
Transport Educational Center (TUC, Fyn) which uses these pictures in the Danish driver
schools (totally 108 pictures). The pictures included many different safe and risky traffic
situations seen from the driver perspective. The reason the pictures was seen from the driver
perspective was that the participants should feel that he or she was driving, thus in the control
over the situation. The good and bad words were selected by the researchers to be both
familiar to, and unambiguously classifiable, by the potential participants (totally 68 words).
An online pre-test with a convenience sample of 80 drivers was performed in
order to identify which of the driving situations on the pictures was considered most
dangerous and least dangerous, and which of the attribute words were considered most
positive and most negative. The situations on the pictures was ranged on a 5-point Likert scale
from not dangerous to very dangerous (0 = not dangerous, 4 very dangerous), and the words
from very positive to very negative (2 = very positive, 1=positive, 0=neither positive nor
negative, -1=negative, -2 = very negative). The 12 pictures rated as most dangerous and the
12 pictures rated as least dangerous, and the 12 words rated as most positive and 12 words
rated as most negative were used in the main study.
4
Figure 1. Pictures used as stimulus in the GNAT. Left = risky situation. Right = safe
situation.
Table 1. Words used as stimulus in the GNAT.
Negative words
Catastrophe (Katastrofe)
Evil (Ondskab)
Hatred (Had)
Terrible (Forfærdeligt)
Nasty (Ækel)
Tragic (Tragisk)
Brutal (Brutal)
Evil (Onde)
Sickening (Kvalmende)
Nauseous (Væmmelig)
Painful (Smertefulde)
Anxiety (Angst)
Note. Danish translation in brackets.
Positive words
Laugh (Grine)
Smile (Smile)
Sweet (Sød)
Joy (Glæde)
Pleasure (Fornøjelse)
Lovely (Dejlig)
Friendly (Venlig)
Beautiful (Flotte)
Happy (Glad)
Comfortable (Behageligt)
Cosy (Hyggeligt)
Cheerful (Munter)
Self-report measures of driving behavior and skill
Two measures of driving behavior were included in the study, namely the DBQ and the DSI
(see Table 2). The DBQ measures aberrant driving behavior by asking drivers how frequently
they perform intentional violations as well as unintentional errors and lapses while driving a
six-point Likert scale (0 = never, 5 = nearly all the time) across different driver behaviors (see
Table 2; for a detailed description see Reason et al., 1990).
The DSI measures driving skills by asking drivers how good they consider
themselves to be in perceptual-motor skills (technical driving skills) and safety skills (abiding
rules and considering other road users) on a five-point scale (0 = well below average, 4 = well
5
above average) across different driving situations (see Table 2; for a detailed description see
Lajunen and Summala, 1995).
Table 2. Examples of DBQ and DSI items.
DSI items
Fluent driving (management of your car in
heavy traffic)
Conforming to the traffic rules
Performance in a critical situation
Driving carefully
Perceiving hazards in traffic
Paying attention to other road users
DBQ items
Unknowingly speeding
Turn right on to vehicle’s path
Drive as fast on dipped lights
Try to pass without using mirror
Overtake on the inside
Fail to see pedestrian waiting
Participants
In the main study, drivers with a type B driver license (Danish license for personal car) were
randomly selected from the Danish Driving License Register. The participants were contacted
by mail and asked to participate in the experiment online. Out of 600 contacted drivers, 77
letters came in return because the person either had moved out of the country or passed away.
Of the remaining 523 persons, 114 participated in the experiment leading to a total response
rate of 22 %. The participants had all finished the DBQ and the DSI on a previous occasion.
Statistical analysis
From the GNAT data, signal detection sensitivity scores (d’) were computed (see Nosek and
Banaji, 2001 for further information). This measure reveals how good the participants can
discriminate or distinguish between the foreground categories (for example risk and good)
from the noise or the background stimulus (for example safe and bad). Then implicit attitude
scores were computed for safe and risky driving by subtracting the sensitivity scores in
counter-normative blocks from the sensitivity scores in normative blocks (“risky driving is
bad” minus “risky driving is good”, and “safe driving is good” minus “safe driving is bad”
blocks). Greater values on the implicit attitude scores thus indicate more normative (or
socially desirable) implicit attitudes. Then, Pearsons correlations between the attitude scores
for safe and risky driving were computed. Finally, Pearsons correlations between the attitude
scores of safe and risky driving on the one hand, and the scores in the DBQ and the DSI on
the other were calculated.
RESULTS
Inter-correlations between implicit-attitude scores
As can be seen in Table 3, across the two response deadlines, implicit attitudes towards the
same attitude object (safe driving vs. risky driving) correlated positively and significantly;
that was observed both for implicit attitudes towards safe driving (r = .39, p < .01) and for
6
implicit attitudes towards risky driving (r = .46, p < .01). This finding shows that implicit
attitudes towards each driving style can be measured reliably, with repeatable results. In
contrast, within each response deadline, the correlation between implicit attitudes towards
different attitude objects (safe driving and risky driving) was smaller and non-significant; that
was observed both for the 750ms deadline (r = .27, p > .05) and for the 600ms deadline (r =
.25, p > .05). This findings show that implicit attitudes towards the two driving styles are
empirically separable constructs, rather than redundant with each other.
Table 3. Correlations between implicit attitudes towards risky and safe driving.
NoRisk600
Risk750
Risk600
**
+
NoRisk750
.39
.27
.28+
NoRisk600
.18
.25+
Risk750
.46**
Note. Cell entries are Pearson correlation coefficients. ** p < .01, + p < .10. N = 48-52
because not all participants finished all trials.
Implicit attitudes and self-reported driving behavior
To gain insight into the relation between implicit attitudes and self-reported driving style, we
first collapsed the GNAT scores across response deadlines. Then, we correlated the resulting
two implicit attitude scores (towards risky driving and towards safe driving) with the
participants’ DSI and DBQ scores. This was done separately for female and male participants.
Table 4 shows the relations between implicit attitudes and self-reported driving
style, separately for women and men. For the women, none of the correlation coefficients was
close to statistical significance. Thus, our female participants’ implicit attitudes were not
related to their self-reported driving style. For men, in contrast, two significant correlations
were observed. First, the DBQ scores and implicit attitudes towards risky driving correlated
significantly such that a greater number of own traffic violations and errors was associated
with more risk-aversive implicit attitudes towards risky driving, r = .45, p < .05. Second, the
DSI scores and implicit attitudes towards safe driving correlated significantly such that
lesser/lower self-reported driving abilities and skills were associated with more positive
implicit attitudes towards safe driving, r = -.63, p < .01.
Table 4. Correlations between implicit attitudes and self-reported driving style.
Gender
DSI
DBQ
Male
Implicit attitude risk
-.14
.45*
**
Implicit attitude no risk
-.63
.08
Female
Implicit attitude risk
.16
-.00
Implicit attitude no risk
.06
.10
Note. Cell entries are Pearson correlation coefficients. ** p < .01, * p < .05. Nmale = 23, Nfemale
= 30.
DISCUSSION
The current study is, to the authors’ knowledge, the first GNAT study performed in order to
test implicit attitudes towards risky and safe driving. The present GNAT results show great
promise as a measure to reveal implicit attitude towards both risky and safe driving.
7
Psychometric properties of the GNAT
Two response deadlines (600ms, 750ms) were used to measure implicit attitudes towards each
of two attitude objects (safe driving, risky driving). The inter-correlations between the
resulting four GNAT scores show that the instrument reveals similar implicit attitudes
towards the same attitude object, independent of the particular response deadline used. These
results speak to the reliability of the research instrument. Within each response deadline,
implicit attitudes towards different (though related) attitude objects were found to correlate
moderately, with the expected positive sign. The observation that the attitude scores
correlated positively may be interpreted as first evidence for the GNAT’s convergent validity:
when measuring attitudes towards related objects, the instrument reveals related attitudes. At
the same time, the positive correlation of implicit attitudes towards safe and risky driving was
only of moderate magnitude and just marginally significant. This may be interpreted as first
evidence for the GNAT’s discriminant validity: when measuring related attitudes, the GNAT
is sensitive enough to capture differences in the two attitudes.
The DSI, the DBQ, implicit attitudes, and gender
The results showed that self-reported judgments of own driving skills correlate with an
implicit pro-safety attitude, though not with an implicit anti-risk attitude. Conversely, selfreported violations and errors correlated with an implicit anti-risk attitude, though not with an
implicit pro-safety attitude. This pattern was observed in male, but not in female participants.
Neither the double-dissociation pattern nor the gender difference had been predicted. Are
these chance findings, or do they make sense?
To cover the gender difference first, some prior evidence for gender differences
in the effects of implicit cognition in relation to driving can be found in the literature. Harré
and Sibley (2007) found a stronger effect of the implicit driver self-image in men than in
women. It has been shown that men report greater gender-stereotypical ‘‘macho” driving
attitudes than women (Harré et al., 1996), and that such attitudes actually are linked to greater
driving aggression (Krahé and Fenske, 2002). From such findings it appears that when it
comes to driving, men tend to rely on their intuition, using stereotypical roles and gender
prototypes such as "machos" as the basis for their driving behavior. Implicit attitudes are one
of several psychological mechanisms that create "intuitions," and may here fore be more
important determinants of men’s (rather than women’s) driving behavior. The present data are
thus in line with prior findings as well as current theorizing in the field.
To cover the double-dissociation pattern next, evidence from the literature
suggests that drivers who think highly of their own driving skills also perceive a lesser risk
than others of becoming involved in an accident (DeJoy, 1989; Harré, Foster, and O’Neill,
2005; Harré and Sibley, 2007). Conversely, drivers who think lowly of their own driving
skills and abilities should then perceive a greater risk and value driving safety more than
others. This relation may explain the negative correlation between self-reported driving
abilities skills and implicit safety attitudes: drivers who estimate their own skills to be rather
low, have a greater implicit desire for safety in driving. The second correlation in the doubledissociation pattern indicates that drivers with more self-reported traffic violations have
greater risk-aversive implicit attitudes. Although counter-intuitive at first glance, that
correlation may be explained as the result of a learning process: to the degree that the own
past violations had unpleasant consequences, these drivers may have “learned their lesson” -which is not to like safety, but to dislike risk.
8
As a theoretical model that covers both of the observed correlations, we
tentatively suggest that pro-safety attitudes and anti-risk attitudes may have different
experiential sources: Pro-safety attitudes may essentially come from social comparison
processes that lead to low self-ratings of own skill, and thereafter to the insight that "in traffic,
I better play it safe." One could call this source of experience "soft variables" or
"psychological evidence." Anti-risk attitudes, in contrast, may come from external variables
such as one’s own past, rule-violating behavior that had unpleasant consequences, teaching
the driver that "taking risk is bad." One could call this source of experience "hard variables"
or "real-world evidence." We readily admit that our study does not demonstrate all steps in
the model. At the same time, however, the study illustrates the heuristic, theory-building
value of measuring implicit cognition with instruments such as the GNAT.
Practical applications
The results can give valuable practical value as it potentially can be applied in driver testing
both for transport companies employing new personnel and for the state driving license
administration. The results should, however, be replicated with a larger sample as the
limitations of the current study is the small sample size. Moreover, the link between explicit
attitudes, implicit attitudes and actual behavior should be explored. Self-report measures have
not always shown to be predictive of actual behavior (af Wåhlberg and de Winter, 2012),
thus, the implicit attitudes of drivers towards risky and safe driving might give valuable
information that can explain the relationship between self-reported behavior and actual
behavior to a greater extent, and also to help predict driving behavior better.
CONCLUSION
Pending replication in future research, the apparent dissociation between implicit attitudes
towards safe versus risky driving that we observed may contribute to a greater theoretical
understanding of the causes of unsafe and risky driving behavior. A practical advantage of
measuring implicit attitudes is a lesser susceptibility to social desirability biases, compared to
self-report methods. It is proposed that research on driving behavior may benefit from
routinely including measures of implicit cognition.
REFERENCES
Af Wåhlberg, A., & de Winter, J.C.F. (2012). Commentaries and responses to “The Driver
Behavior Questionnaire as a predictor of accidents: A meta-analysis”. Journal of Safety
Research, 43, 83-99.
Ajzen, I., and Fishbein, M. (1977). Attitude-behavior relations: a theoretical analysis and
review of empirical research. Psychological Bulletin, 84, 888-918.
Baron, A. S., and Banaji, M., R. (2006). The Development of Implicit Attitudes : Evidence of
Race Evaluations From Ages 6 and 10 and Adulthood. Psychological Science, 17, 53-58.
DeJoy, D. M. (1989). The optimism bias and traffic accident risk perception. Accident
Analysis and Prevention, 21, 333–340
Dunham, Y., Baron, A. S., and Banaji, M. R. (2004). From American city to Japanese
Village: A cross-cultural investigation of implicit race attitudes. Child Development,
77 (5), 1268-1281.
Fazio, R. H., and Olson, M. A. (2003). Implicit measures in social cognition research: Their
meaning and use. Annual Review of Psychology, 54, 297-327.
9
Greenwald, A. G. and Banaji, M. R. (1995). Implicit social cognition: Attitudes, self-esteem,
and stereotypes. Psychological Review, 102, 4-27.
Harré, N., Foster, S., & O’Neill, M. (2005). Self enhancement, crash-risk optimism and the
impact of safety advertisements on young drivers. British Journal of
Social Psychology, 96, 215–230.
Harre, N., and Sibley, C. G. (2007). Explicit and implicit self-enhancement biases in drivers
and their relationship to driving violations and crash-risk optimism. Accident Analysis
and Prevention, 39, 1155-1161.
Harré, N., Field, J., & Kirkwood, B. (1996). Gender differences and areas of common concern
in the driving behaviours and attitudes of adolescents. Journal of Safety Research, 27,
163–173.
Hatfield, J., Fernandes, R., Faunce, G., and Job, R. F. S. (2008). An implicit non-self-report
Analysis and Prevention, 40, 616-627.
Lajunen, T., & Summala, H. (1995). Driving experience, personality, and skill and safety
motive dimensions in drivers’ self-assessments. Personality and Individual Differences,
19 (3), 307-318.
Lewis, I., Watson, B., and Tay, R. (2007). Examining the effectiveness of physical threats in
road safety advertising: The role of the third-person effect, gender, and age.
Transportation Research Part F, 10 (1), 48-60.
Krahé, B., & Fenske, I. (2002). Predicting aggressive driving behaviour: The role of macho
personality, age and the power of the car. Aggressive Behaviour, 28, 21–29.
Maison, D., Greenwald, A. G., and Bruin, R. (2001). The implicit association test as a
measure of implicit costumer attitudes. Polish Psychological Bulletin, 32 (1), 61-69.
Martinussen, L.M., Hakamies-Blomqvist, L., Møller, M., Lajunen, T., and Özkan, T. (2013).
Age, gender, mileage and the DBQ: The validity of the Driver Behavior Questionnaire in
different driver groups. Accident Analysis and Prevention, 52, 228-236.
Nosek, B. A., and Banaji, M. R. (2001). The go/no-go association task. Social Cognition, 19
(6), 625-664.
Nosek, B. A., Greenwald, A. G., and Banaji, M. R. (2007). The implicit association test at
age7: A methodological and conceptual review (pp 265-292). In J. A. Bargh (Ed.),
Automatic processes in social thinking and behavior. Psychology Press.
Parker, D., Reason, J.T., Manstead, A., Stradling, S.G. (1995). Driver errors, driving,
violations, and accident involvement. Ergonomics, 38 (5), 1036–1048.
Petty, R. E., Tormala, Z. L., Brin˜ol, P., & Jarvis, W. B. G. (2006). Implicit ambivalence from
attitude change: An exploration of the PAST model. Journal of Personality and Social
Psychology, 90, 21–41.
Phelps, E. A., O’Connor, K. J., Cunningham, W. A., Funayama, E. S., Gatenby, J. C., Gore, J.
C., and Banaji, M. R. (2000). Performance on indirect measures of race evaluation
predicts amygdala activation. Journal of Cognitive Neuroscience, 12 (5), 729-738.
Siebley, C. G. and Harre, N. (2009a). The impact of different styles of traffic safety
advertisements on young drivers’ explicit and implicit self-enhancement biases.
Transportation Research Part F, 12, 159-167.
Siebley, C. G. and Harre, N. (2009b). A gender role socialization model of explicit anf
implicit biases in driving self-enhancement. Transport Research Part F, 12 (6), 452-462.
Teachman, B. A., Gapinski, K. D., Brownell, K. D., Rawlins, M., and Jeyaram, S. (2003).
Demonstration of implicit anti-fat bias: The impact of providing causal information and
evoking empathy. Health Psychology, 22 (1), 68-78.
Teachman, B.A., Gregg, A.P., and Woody, R. S. (2001). Implicit association for fear-relevant
stimuli among individuals with snake and spider fears. Journal of Abnormal Psychology,
110 (2), 226-235.
10
11