Accident Analysis and Prevention 73 (2014) 296–304

Contents lists available at ScienceDirect

Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap

The influence of image valence on visual attention and perception of risk in drivers Jones M.P. *, Chapman P., Bailey K School of Psychology, University of Nottingham, Nottingham, England, United Kingdom

A R T I C L E I N F O

A B S T R A C T

Article history: Received 27 January 2014 Received in revised form 4 April 2014 Accepted 15 September 2014 Available online xxx

Currently there is little research into the relationship between emotion and driving in the context of advertising and distraction. Research that has looked into this also has methodological limitations that could be affecting the results rather than emotional processing (Trick et al., 2012). The current study investigated the relationship between image valence and risk perception, eye movements and physiological reactions. Participants watched hazard perception clips which had emotional images from the international affective picture system overlaid onto them. They rated how hazardous or safe they felt, whilst eye movements, galvanic skin response and heart rate were recorded. Results suggested that participants were more aware of potential hazards when a neutral image had been shown, in comparison to positive and negative valenced images; that is, participants showed higher subjective ratings of risk, larger physiological responses and marginally longer fixation durations when viewing a hazard after a neutral image, but this effect was attenuated after emotional images. It appears that emotional images reduce sensitivity to potential hazards, and we suggest that future studies could apply these findings to higher fidelity paradigms such as driving simulators. ã 2014 Elsevier Ltd. All rights reserved.

Keywords: Distraction Emotion Hazard perception Valence Attention

1. Introduction One of the main contributors of on-road collisions is inattention from drivers. Naturalistic studies of car crashes have found distraction to be the cause in 78% of cases (Neale et al., 2005). What such studies suggest is that many people do not necessarily look at the focus of expansion all of the time, despite the fact that approximately 90% of fixations are supposed to focus at this point (Lansdown, 1997). Drivers may be distracted by internal or external factors, which can take up to two seconds to significantly increase the risk of a crash (Zwahlen et al., 1988). It is therefore important to analyse why drivers decide to divert their attention away from the road for the purposes of road safety. One significant type of distracter to consider, which is not necessarily in the control of the driver, is road-side advertising. Typically advertising is large, bright and positioned within the driver’s central field of view. Self-report measures have also suggested that around 30–50% of driver’s attention is given to aspects unrelated to driving, including advertising, whereas only around 20% is given to road signs (Hughes and Cole, 1986). A significant proportion of such drivers can be distracted by advertising, according to a 2006 privilege insurance survey

* Corresponding author. Tel.: +44 7453328823. http://dx.doi.org/10.1016/j.aap.2014.09.019 0001-4575/ ã 2014 Elsevier Ltd. All rights reserved.

(Lansdown, 2012). Considering that it has previously been suggested that the risk of a crash can significantly increase within two seconds (Zwahlen et al., 1988), advertising can therefore be seen as a significant distracter, and whilst self-report studies do not necessarily indicate real-world data, they do emphasise the need to investigate advertising as a significant distracter. Laboratory studies, using joysticks to point to on-screen arrows, have found that reaction times are significantly slower when advertising is also shown (Johnston and Cole, 1976). Real-world studies have found similar results. Previous research has found that there are more crashes at junctions where there are advertisements (McMonagle, 1952), and correlations have been found between advert frequency, advert size and crashes (Holahan, 1977), suggesting that bottom-up processing is used to analyse the features of the advertisement. However, it may be that there are more crashes at junctions with advertisements simply because the junction itself creates road complexity, and correlation between advert frequency and crashes does not necessarily imply causation by the advertisements. In such cases it could be the physical features of the road, such as road type, weather conditions and even junctions. Contemporary research within simulators has demonstrated that adverts placed on rural, urban and motorway roads can have negative effects on lateral control and subjective mental workload as well as encouraging a short fixation sampling strategy,

M.P. Jones et al. / Accident Analysis and Prevention 73 (2014) 296–304

indicative of the complex environment created by in-car distracters such as mobile phones and iPods (Young et al., 2007). However, research has also suggested that placing advertisements a few feet above the ground, which is out of the way from the typical horizontal search strategy used by drivers, can have beneficial effects (Crundall et al., 2006). So street level adverts, which are typically placed within the driver’s horizontal field of view, can be seen as detrimental. Drivers may be fixating on the advertisement instead of any localised hazards, which could result in a near-miss or collision. However, it may not just be the sensory aspects of advertisements that capture attention. Advertisements often seek to evoke an emotional reaction from the viewer, which may then divert attention away from other real-world stimuli. Emotion has been demonstrated to have an effect on driver behaviour. For example, emotional abilities have been linked to self-reported risky attitudes without becoming correlated with age or driving experience (Arnau-Sabatés et al., 2012), and positive emotion priming has been associated with self-reported reckless driving (Ben-Ari, 2012). Again, this self-report study highlights the importance of studying the effects of emotion. Anxiety, as opposed to other emotions such as happiness or anger has been associated with an increase in perceived risk, as well as an increased heart rate (Mesken et al., 2007). One example of investigating the relationship between emotion and driving within the context of anxiety and its relationship to driving has been conducted by Briggs et al. (2011). In their study, they found that when participants with arachnophobia ere in a simulator and held what is described as emotionally involving conversations regarding spiders, their cognitive mental workload increased as demonstrated by heart rate, and their spread of eye movements indicated what is known as cognitive tunnelling (Easterbrook, 1959), or a reduction in visual spotlight (Driver and Baylis, 1989). This means that when an anxious participant was placed in what they felt was a threatening situation, it resulted in a case of hyper distractability limiting other sensory processing. In order to safely process the road ahead, the spread of eye movements reduced in order to focus on the central task. Such results are reflective in previous driving literature, where hazards on the road have been shown to capture the attention of a driver and create a cognitive tunnelling effect (Chapman and Underwood, 1998). Briggs et al. (2011) suggest that the driver is distracted due to the emotional involvement of the task. Whilst this may be true, discussions with a phobic person may result in the development of negative mental imagery and associated memories, thus creating a secondary task which may be the real cause of cognitive tunnelling rather than actual emotional processing. Furthermore, the use of mobile phones as distracters themselves could result in any effects on the road. In car distracters can affect steering precision (Reed and Green, 1999), and result in a greater likelihood of near-misses and collisions (Chisholm et al., 2008). So whilst the proposed effects of emotion are certainly plausible, conclusions such as this need to be investigated further before emotion can actually be considered as a significant on-road distracter. A good example of studying the effects of emotion on driving behaviour is in the context of music within the car. Pêcher et al. studied the effects of the valence of music and its effects on in-car driving (Pêcher et al., 2009). They found that happy music resulted in a decrease in speed; however, drivers also deteriorated in terms of control and tended to steer towards the hard shoulder. This could have been due to broadening their attention to global aspects of the driving environment, which can happen when using positive valence stimuli (Rowe et al., 2007). It may also have been due to participants reacting to the music and thus creating an unintentional secondary task. On the other hand, sad music resulted in a

297

decrease in driving speed, and an increase in control with a tendency to keep their vehicle in the middle of the lane. This may have been due to the negative valence stimuli resulting in greater control of the vehicle in the, which can be advantageous to the driver in the same way that cognitive tunnelling could be advantageous for spread of search in a hazardous situation (Miyazawa and Iwasaki, 2009). However, whilst differences were found between the two valence levels, the study confounded valence and arousal by not considering the effects of arousal on the speed at which participants chose to drive. They tended to change their speed according to different types of music. Speed is a factor associated with arousal levels of music, which in itself can result in more on-road collisions (Brodsky, 2002). The previous studies also highlight the fact that emotion and driving attention has not been widely studied within a visual stimuli context. This is important, because it is possible that two tasks using the same sensory modality, for example driving whilst looking at an emotional advert, could result in even greater detrimental consequences. Recent research by Trick et al. has attempted to look at the relationship between emotion and attention whilst driving by using visual stimuli (Trick et al., 2012). Based on previous research and the fact that hazard perception requires the use of search strategies employed by focal vision as opposed to ambient (Previc, 1998) it was hypothesised that positive valence images would create a broadening of attention whilst negative valence images would centralise attention. They used pictures taken from the international affective picture system (IAPS) whilst asking participants to drive around in a simulator. Whilst driving, an image would appear on a GPS device next to the steering wheel. The participants had to indicate whether this image was a positive or negative image whilst driving. It was concluded that steering control was affected by image valence rather than arousal, with negative images producing poorer steering quality. This is complimentary to other contemporary research investigating the effects of emotional visual stimuli and distraction in the context of IAPS stimuli, which suggests negative implications for its effects on increased risk taking within the car and rear-end accidents (Megías et al., 2011). However, the effect sizes found were small, and there are some methodological concerns. Firstly, the participants had to look down at the device and indicate whether the image was positive or negative. It may have been the act of looking down at the device and pressing a button, rather than whether the image had positive or negative valence, that had an effect on steering quality. Whilst previous research has suggested that in-car devices may not necessarily be disadvantageous to the driver in contexts such as hazard perception (Reed-Jones et al., 2008), other studies have shown that the further away a driver looks from the road, the less they are able to steer (Summala et al., 1996). So in the context of this study, it may have been the in-car distracter affecting the results; this demonstrates the need to fully investigate the role of emotion before making conclusions on in-car distracters, such as the case of Briggs et al.’s spider phobia study. The study also had problems controlling for arousal and valence levels of images, in a similar fashion to Pêcher et al., which once again could be confounding for the experimental results. Currently there are few studies that observe the relationship between emotion, attention and driving by using visual stimuli. Those that have done so have suffered from various methodological limitations. This study observes the relationship between image valence and driving attention whilst at the same time controlling for arousal levels. In accordance to recommendations from Trick’s research (2012), perception of hazard, eye movements and physiological data such as heart rate and galvanic skin response were measured. Such physiological measures have also been recommended as useful in the context of driving by other

298

M.P. Jones et al. / Accident Analysis and Prevention 73 (2014) 296–304

researchers (Johnson et al., 2011), and have been reflective of cognitive mental workload in previous research (Briggs et al., 2011). The confounding effects of looking down at an in-car device such as a GPS were removed by placing emotional images, taken from the IAPS (Lang et al., 2008), within the driver’s field of view. The effects of emotional images were observed in the context of a hazard perception test. Hazard perception is an important driving skill based on situational awareness; it has been argued that across studies, it is the only factor that can be related to the likelihood of being involved in an accident (Horswill and McKenna, 2004). It is also a useful tool to the extent that it has been introduced as a compulsory part of the UK Driving Standards Agency theory test since 2002, and its use in psychological research has also been useful in eliciting emotional responses (Kinnear et al., 2013). So whilst the previously mentioned studies have used dependent measures such as lane control and braking times, this is one of the first of its kind to take measures of what is considered to be biggest predictor of on-road accidents. If significant results are obtained, then it will result in real-world implications for the use of emotional advertising on the roads and its ability to prevent drivers from becoming aware of an oncoming hazard. It will be particularly interesting to see if and how perception of risk, physiological responses and eye movements interact with the valence of emotional image conveyed, according to, for example, the cognitive tunnelling effect for negative valence thoughts and images, or the broaden-and build hypothesis (Rowe et al., 2007; Frederickson, 2001) in relation to positive affect. 2. Method 2.1. Participants Thirty-six participants were recruited for this study, ranging between 18–26 years, with an average age of 20.2. Twenty-seven participants were female and nine were male. They were students from the University of Nottingham and took part for course credit. All participants were in possession of a full driver’s licence. Possession of a full driver's licence indicates that participants have previously passed a hazard perception test and have had experience of detecting hazards in a natural driving environment. 2.2. Design Each hazard perception clip was provided by the Driving Standards Agency (DSA); these clips typically act as practise clips for the actual hazard perception test. After using editing software, each clip lasted for 49 s, was taken from the driver’s point of view and contained only one hazard. Details of what each hazard perception clip contained can be found in Appendix A. Clips were selected according to how early or late they presented their hazard,

so six clips presented the hazard during the first half of the clip and the remaining six presented the hazard during the second half. Windows of measurement were defined in each clip as beginning 2 s before the hazard began, according to the DSA, until the presented hazard was no longer considered a danger. These windows ranged from 4.24 to 13.48 s, with an average duration of 9.3 s. A second window of measurement, a ‘safe’ window, was defined as symmetrical to its corresponding ‘dangerous’ window and was recorded for the same amount of time as the corresponding dangerous window within the clip. For example, if a dangerous window began 25% of the way into a clip and ended 30% of the way into a clip, the corresponding safe window would begin 75% of the way into the clip and end 80% of the way into the clip. Timings calculated according to frame rate, at a rate of 25 frames per second, are presented in Appendix A. Prior to each window, an emotional image was overlaid at 50% transparency onto the hazard perception video, thus presenting itself in full screen, for 2 s, with the three types of emotional image balanced across clips (see Fig. 1 for an example of this). The overall experiment included two within subject factors, which were the valence rating for each image (negative, neutral and positive) and the window in which the clip was presented (before a hazard or before an uneventful moment in the clip). Physiological measures included mean skin conductance in MicroSiemens and heart rate. For the purposes of the research topic, mean eye fixation durations in milliseconds and the horizontal and vertical spread of search in pixels were measured. The mean response on a variable response transducer, response rate per minute and hazard perception score acted as measurements for perception of risk. The variable response transducer was a scale that continuously measured the participant’s perception of a hazard on a scale of 0–9, with 0 indicating feeling safe in the driving environment and 9 indicating the immediate need to brake or take evasive action in the environment. 2.3. Materials/stimuli Stimuli consisted of twelve hazard perception practise clips taken from the Driving Standards Agency and twenty-four images taken from the international affective picture system (Lang et al., 2008). The twenty-four images were divided into three categories according to level of valence; eight had an average valence rating of 2.97 (negative valence), eight had an average of 4.9 (neutral valence) and the remaining eight 7.12 (positive valence), with arousal controlled for. Each image was controlled for spatial frequency (Delplanque et al., 2007) by ensuring that image file size in bytes, luminance levels and colour saturation for red, green and blue channels were relatively similar. IAPS picture codes and details of valence, arousal and spatial frequency ratings are presented in Appendix B. The means and standard deviations for

Fig. 1. A timeline of the experimental procedure as shown on screen using hazard clip one (see Appendix A) as an example. 2 s before the oncoming hazard begins, an image is overlaid onto the video clip at 50% transparency. This lasts for 50 frames and ends when the hazard (and consequently the ‘dangerous’ measurement window) begins according to the Driving Standards Agency. The measurement window ends when the hazard ends. A ‘safe’ measurement window was also presented in each clip and followed the same procedure.

M.P. Jones et al. / Accident Analysis and Prevention 73 (2014) 296–304

each of these characteristics were analysed in a One-way ANOVA using SPSS. With the exception of the standard deviation of colour saturation in the red channel (F(2, 21) = 4.179, MSE = 412.559, p < 0.05), no significant differences were found for image characteristics (all p’s < 0.05). For the purposes of ethics, sexual or violent pictures were removed in the selection process. Images were edited onto each of the hazard perception clips on a Mac computer, using Quicktime and Adobe Premiere Pro software. These edited clips were then recorded onto Sony MiniDV video cassettes using a Sony video recorder. The images were then played back onto a 18-in. Sony monitor, which was connected to a computer for eye movement calibration, and also to a MiniDV player for stimulus presentation. All footage was presented with an aspect ratio of 4:3 and a viewing distance of approximately 60 cm once participants’ heads were positioned in the chin and head rest. WinCAL software was used to help enable calibration. Eye movements were calibrated using iViewX software on a computer, and eye movements were tracked and recorded with a SensoMotoric Instruments IView X RED eye tracker. The sampling rate of the eye tracker was 50 Hz, and spatial resolution was better than 0.5 . Perceptions of hazard were measured with a Biopac variable response transducer, and physiological data was measured using a Biopac MP150 amplifier. Galvanic skin responses and heart rate were recorded on the non-dominant hand using finger straps, with isotonic recording gel used to record skin response (GEL101, Biopac Systems Inc., Goleta CA., USA) This was connected to a computer, which recorded the physiological and hazard perception data using Acqknowledge. 2.4. Procedure Prior to the experiment participants were attached to the heart rate and galvanic skin response equipment, and eye movements were calibrated in order for successful eye movement recording to occur. They were then informed that they would be watching a series of hazard perception clips. Their main task was to concentrate on the road as much as possible, as if they were the driver of the car. If at any point the participant sensed a hazard was about to occur they were to increase their hazard response on the response transducer. If on the other hand they felt that they were in a safe driving situation they were to decrease their response on the transducer. Responses could be anywhere on the scale, ranging from a score of 0–9, with 0 indicating feelings of safety and 9 indicating that the participant would need to immediately brake or take evasive action in order to avoid a crash. They were to respond in this way continuously throughout the clips. They then performed the task, which lasted for 10 min. Afterwards all

299

physiological equipment was removed; they were thanked for their time and debriefed on the purposes of the research. 3. Results All data collected was analysed using a 2  3 repeated-measures ANOVA in SPSS, and results were deemed significant at a threshold of p < 0.05; any significant results were also subject to planned comparisons in SPSS. Risk perception was measured by allowing participants to continuously respond on a 9-point scale using a Biopac variable response transducer, according to how hazardous or safe they felt whilst watching the clips. Measures selected for analysis were response given per minute, and a calculated hazard perception score. Response rate per minute was calculated by smoothing the data recorded from Acqknowledge, noting the amount of upward peaks before a significant change in response for each condition. Peaks were excluded if their response value was lower than 10% of the maximum hazard response from the participant. Hazard scores were calculated by recording the time it took for the first significant upward peak to occur within a measurement window, and assigning a score between 0 and 5 according to how early or late this response was made, with 5 indicating an early response and 0 indicating an extremely late or non-existent response. Mean response rate per minute and mean calculated hazard perception scores are presented in Fig. 2a and b. Mauchley’s assumption of sphericity was violated for the interaction between measurement window and emotion (p = 0.04), so a Greenhouse-Geisser correction was made. No main effect of emotional image was found for response rate per minute (F(2, 70) = 0.706, MSE = 4.856, p > 0.05). However, a significant main effect was found for measurement window (F(1,35) = 184.132, MSE = 4.274, p < 0.05). Response rate in the dangerous window (m = 5.02) was significantly higher than in the safe window (m = 1.20). A significant interaction was also found between variables (F(2,70) = 5.053), MSE = 3.69, p < 0.05). Helmert contrasts on interactions showed that a higher difference was found between windows for neutral valence images (m difference = 4.69) than for emotional images regardless of positive (m difference = 3.93) or negative (m difference = 2.83) valence (F(1,35) = 5.612, MSE = 5.558, p < 0.05). There were also significantly greater differences between measurement windows for responses per minute for positively valenced images than for negatively valenced images (F(1,35) = 4.254, p < 0.05). No main effect of emotional image was found for hazard perception scores (F(2,70) = 2.617, MSE = 0.380, p > 0.05). However, a significant main effect of measurement window was found (F(1.35) = 114.754, MSE = 0.403,

Fig. 2. a: The mean response per minute given by participants according to image valence and window of measurement; b: The mean calculated hazard perception score according to image valence and measurement window. All error bars represent standard error of the mean.

300

M.P. Jones et al. / Accident Analysis and Prevention 73 (2014) 296–304

p < 0.05). Hazard perception scores were significantly higher in the dangerous window (m = 1.45) than in the safe window (m = 0.52). A significant interaction was also found between variables (F (2,70) = 4.395, MSE = 0.443, p < 0.05). Helmert contrasts on interactions found that differences between measurement windows for hazard scores were significantly higher for neutral valence images (m difference = 1.284) than for positive (m difference = 0.854) and negative (m difference = 0.639) valence images (F(1,35) = 6.112, MSE = 0.853, p < 0.05). However no significant comparisons were found between positive and negatively valenced images (F (1,35) = 1.315, MSE = 0.634, p > 0.05). In summary, neutrally valenced images overlaid onto hazard perception clips resulted in a higher difference between measurement windows, in terms of both response rate per minute and hazard perception scores, than positive and negative valence images. Positive images also produced a higher difference between windows for response rate per minute than negative images. 3.1. Measures of physiology Physiological measures selected for analysis included mean skin conductance, amount of electro dermal responses (EDRs) according to measurement window, and heart rate. The amount of EDRs were recorded by smoothing the GSR data according to recommendations by Kim et al. (2004); any EDRs with a value of less than 10% of the maximum galvanic skin response were excluded from analysis. Heart rate was recorded by counting the amount of heart beats within each measurement window and dividing this value by 60. Mean electrodermal response and heart rate are presented in Fig. 3a and b. For the amount of EDRs, no significant main effects of window (F(1,35) = 0.175,MSE = 4.922, p > 0.05) or emotion (F(2,70) = 1.365, MSE = 4.691, p > 0.05) was found. However, a significant interaction was found between variables (F(2,70) = 9.489, MSE = 5.336, p < 0.05). Helmert contrasts on interactions found that whilst there were no significant differences between measurement windows for positively and negatively valenced images (F(1,35) = 2.172, MSE = 8.314, p > 0.05), neutrally valenced images produced significantly higher differences between measurement windows for EDRs (m difference = 2.581 EDRs per minute) than positive (m difference = 1.299 EDRs per minute) and negative (m difference = 0.297 EDRs per minute) valenced images (F(1.35) = 14.158, MSE = 9.772, p = 0.001). For heart rate data, Mauchely’s assumption of sphericity was violated for both the main effect of emotion and the general interaction (both p’s < 0.05), thus a Greenhouse-Geisser correction was used. No main effects of window (F(1,35) = 0.000, MSE = 16.68, p > 0.05) or emotion (F(2,70) = 0.101, MSE = 3.73,

p > 0.05) were found, and no significant interaction was found between variables (F(2,70) = 1.924, MSE = 38.14, p > 0.05). Therefore, neutrally valenced images produced a larger difference between measurement windows in terms of EDRs per minute than positive and negatively valenced images. 3.2. Measures of visual search The measures of visual search selected for analysis were mean fixation durations and the spread of horizontal and vertical search, measured as standard deviations of the fixation locations of X and Y (horizontal and vertical respectively). These measures were recorded during the 24 possible measurement windows within the procedure. Due to problems with data recording, 10 participants were excluded, leaving 26 participants’ data available for analysis. All eye movement measurements according to condition are presented in Table 1. For mean fixation duration data, Mauchley’s assumption of sphericity was violated in both the main effect of emotion and the interaction, so a Greenhouse-Geisser correction was applied. No main effects of window (F(1, 25) = 2.037, MSE = 131444, p > 0.05) or emotion (F(2, 50) = 0.911, MSE = 74893, p > 0.05) were found. A trend towards a significant interaction was found (F(2,70) = 3.523, MSE = 129516, p = 0.056). Helmert contrasts on this interaction show that a trend towards significantly smaller differences between measurement windows in terms of fixation durations with neutrally valenced images (m difference = 9.104 ms) than for positive (m difference = 98.08 ms) and negative (m difference = 186.3 ms) (F(1,35) = 3.950, MSE = 211163, p = 0.058), but no trend towards a difference between positive and negatively valenced images (F(1,35) = 2.022, MSE = 80020, p > 0.05). For the spread of horizontal search, Mauchley’s assumption of sphericity was violated for both emotion and interaction, so a GreenhouseGeisser correction was applied. A significant main effect of window was found (F(1,25) = 26.013,MSE = 533.7, p < 0.001). Spread of search was significantly greater in dangerous windows (m = 294.8 pixels) than in safe windows (m = 227.5 pixels). However, no significant main effect of emotion was found (F(2,32.7) = 0.04, MSE = 1211, p > 0.05) and no interaction was found (F (2,35.684) = 0.356,MSE = 2378, p > 0.05). For the spread of vertical search, no significant main effect was found for either measurement window (F(1.35) = 0.282, MSE = 224.4, p > 0.05) or emotion (F (2,70) = 0.34, MSE = 9265, p > 0.05), and no significant interaction was found (F(2,70) = 2.389, MSE = 134.5, p > 0.05). To summarise, a trend was found towards neutrally valenced images producing significantly smaller differences between measurement windows, in terms of fixation durations, was found compared to positive and negatively valenced images.

Fig. 3. a: Mean electro dermal response in MicroSiemens according to image valence and measurement window; b: heart rate in beats per minute (BPM) according to image valence and measurement window. All error bars represent standard error of the mean.

M.P. Jones et al. / Accident Analysis and Prevention 73 (2014) 296–304

301

Table 1 Eye movement data according to image valence and measurement window. Measurement window

Image valence

Mean fixation duration (ms)

Standard deviation of horizontal position (pixels)

Safe

Neutral Positive Negative

660.732332 647.9390063 756.6140481

106.0390329 383.1705612 193.2794568

44.99322089 97.91925472 54.55903241

Dangerous

Neutral Positive Negative

669.8360482 549.85029 570.3624848

462.9677297 185.1636274 236.3157697

104.2284213 52.33590912 65.6488838

4. Discussion The aim of this study was to investigate the relationship between image valence and a driver’s perception of hazard in the context of a hazard perception test and situational awareness. This was to be done without having the methodological problem of looking away from the road to identify an image, which was achieved by overlaying emotional images onto the hazard perception clips. Overall, our results demonstrated that images of both positive and negative valence, in comparison to neutral valence, resulted in significantly lower differences between measurement windows in terms of subjective perceptions of risk, lower electrodermal responses, and a trend of fixating for shorter periods of time. Thus the emotion portrayed by the IAPS stimuli resulted in a reduction in ability to detect hazards. The conclusion based on inhibition of sensitivity to hazards can be based on the findings from risk perception and physiological data. For the risk perception data, the fact that dangerous windows produced significantly responses per minute and hazard perception scores than safe windows indicates that people were really taking part in the task. Significant interactions for response per minute and hazard perception scores both demonstrated that in a hazardous situation, positive and negative images produced significantly lower results than the baseline neutral images. This suggests that during the dangerous measurement windows, responses were inhibited and participants were less sensitive to the hazard that was about to appear. This pattern is also evident in the physiological data, None of the heart rate data, which has previously been indicative of cognitive mental workload, came out as significant. Nevertheless, the pattern of descriptive results (as seen in Fig. 2b) indicates that a similar pattern would be occurring; it would be interesting for future research to explore this. The idea of emotion resulting in inhibition may be relevant to research into the emotional sensitisation of advertising. For example, longitudinal research into bloody and shocking advertising in Spain as a result of increasing punishment strategies to limit road fatalities found a saturation effect (Castillo-Manzano et al., 2012). Adverts that were too shocking resulted in an avoidance pattern similar to that reported by the inverted ‘U’ hypothesis (Yerkes and Dodson, 1908). In this context of this research, whilst the images do not necessarily reflect the same degree of arousal as the Spanish study, they are inhibiting the driver's ability to make a sufficient hazard response, which in this case is not necessarily advantageous. What is more difficult to explain, however, are the results from the eye movement data. On the one hand, a trend towards significantly longer mean fixation durations was found for neutral images in comparison to both positive and negative images, suggesting a potential for emotional images to inhibit perception of risk in this context. On the other hand, contrary to our hypotheses, spread of search data did not suggest any cognitive tunnelling effect as a result of viewing negatively valenced images (Briggs et al., 2011), nor was a broaden-and-build effect found in

Standard deviation of vertical position (pixels)

response to positively valenced images (Frederickson, 2001). One possible explanation could be that by overlaying images over the hazard perception clips at full screen, participants lost potential reference points within the road, therefore leading to nonsignificant findings related to spread of search. The decision to overlay images onto the clips was made based on the belief that other possible ways, such as briefly cutting away from the video to present the image, would have resulted in participants having to effectively briefly look away from the road to see the image. This would have resulted in this study becoming subject to the same methodological limitation we propose for Trick et al.’s (2012) study. Nonetheless alternative ways to present IAPS images without having to look away from the road should be considered in future research. Interestingly, results from the standard deviation of horizontal gaze opposed previous research into eye gaze during hazardous situations. Whilst studies such as Chapman and Underwood’s (1998) have shown a reduction in visual search in response to hazardous situations, this study found an increase in visual search during hazard windows as opposed to safe windows. One methodological explanation for these results could be the selection of measurement windows as a control variable. For example, in Chapman and Underwood’s (1998) study, a safe window was defined as the remaining part of a hazard perception clip that did not contain a hazard. However our study the safe measurement window was strictly controlled; it lasted for the same amount of time as the dangerous window and featured in a symmetrical point in comparison to the hazardous window. Whilst this acted as a good and strong control it could also be the reason why the spread of horizontal search provided findings opposing previous research. This should be considered in future studies looking at general hazard perception. Another thing to consider is that whilst there were controls on arousal and spatial frequency for emotional images, it is possible that some aspect of the positive and negative valenced images may have captured attention in a way that neutral images did not, hence a reduction in perception of risk was found. Specifically, it may have been possible that arousal of an image could have an effect on hazard perception rather than emotional valence; hence the appropriate emotional effects were not found. In research into advertising, it has been found that arousal has a better effect on advert evaluation than valence (Gorn et al., 2001). In Trick et al.’s (2012) study, braking times were affected by image arousal, which perhaps is more related to perceiving a hazard than previously thought. In future research it may be beneficial to research the effects of arousal, if not in isolation then in conjunction with valence, to assess their contribution to hazard perception. Alternatively, these findings may be a reflection of the potential limits of generalizability. Firstly, whilst significant results were obtained for calculated hazard perception scores, The average calculated score out of five was low for each condition. The current Hazard Perception test contains fourteen clips containing a total of fifteen developing hazards; scores for

302

M.P. Jones et al. / Accident Analysis and Prevention 73 (2014) 296–304

each developing hazard are given out of five, depending on how early the hazard is detected, and 44 points out of 75 are required to pass. If the participants in this study had taken a genuine Hazard Perception test, and scored the highest possible average for each clip according to condition (this was 1.64, in the dangerous window condition after a neutral image had been presented), none of them would pass the test. This suggests that whilst hazard score analyses are reflective of our conclusions on a reduction in sensitivity to hazards as a result of emotional image presentation, the magnitude of such scores may not necessarily reflect real world ones. Secondly, the fidelity of the task may reflect limits of generalizability. Whilst recent studies into the general relationship between emotion and driving behaviour have used a similar level of fidelity and found significant results (Schmidt-Daffy et al., 2012; Schmidt-Daffy, 2013; Hu et al., 2013), it would be interesting to see how our findings would be reflected in a simulator with more fidelity. For example, Trick et al. (2012) used a driving simulator within their study, and some studies have even looked at emotion on real roads (Fairclough et al., 2006); it may be of interest to see, if participants do show a reduction in sensitivity to oncoming hazards when presented with an emotional image, how this is affected by the introduction of basic driving skills such as steering and braking. It may also be of interest to look at participants of a wider variety of ages in future research. In this study participants had an average age of 20.2. Research has suggested that the dorsolateral prefrontal cortex, an area of the brain associated with decision making and assessment of risk, is still developing until the age of 25 (Dahl and Spear, 2004), hence this could explain why younger drivers are more likely to take risks without necessarily thinking about the consequences (Rhodes and Pivik, 2011). Additionally, research into older driver behaviour is mixed. Whilst there is a negative stereotype that older drivers are more dangerous (Lambert et al., 2013), and some research may support this theory (Thompson et al., 2012), people within this age category consciously self-regulate their driving behaviour without realising that their own age-related changes are a potential for risk (Siren and Kjaer, 2011). This is reflected in eye movement data into risk perception, which has shown that older drivers have both better scanning patterns and more of an engagement in risk recognition behaviour (Pradhan et al., 2005). Thus it would be interesting to see whether our findings are similar in older drivers. In conclusion, the aim of this study was to investigate the impact of image valence on the ability to perceive a hazard in terms of risk perception, physiology and eye movements. Rather than finding that emotional valence plays the role of a distracter in the same way that Trick et al. (2012) did, an reduction in sensitivity to hazard perception was found. By taking away the task of looking away from the road, the notion of emotion as distraction is questioned and the notion of emotion resulting in reduction to hazard sensitivity is proposed. The emotional valence of an image, based on these findings, reduces the ability to perceive danger, whether that be through subjective perception of risk, or physiological response. By using stimuli from a widely known and important test of hazard perception, this study has strong implications for the fields of emotion and driving in the context of inhibition; future research taking this perspective into a context such as a driving simulator can help. Current research has investigated distraction using lexical emotional manipulation and found significant effects (Chan and Singhal, 2013), so to extend our findings using IAPS images

would further validate this role of valence as a hazard inhibitor in real-world settings Appendix A. Table A Descriptions of driving standard’s agency hazard perception clips. Clip Environment Description number

Hazard Window appearance One

Window Two

1

Urbanstreet

Pedestrian Late cross road, gets into car and pulls away

Dangerous Safe

2

Urbanstreet

Cyclist pulls out Late onto the main road from a junction

Dangerous Safe

3

Motorway

Car pulls out from the hard shoulder onto the main road

Late

Dangerous Safe

4

Rural

Pedestrian on a Late motorised scooter pulls out into main road from pavement

Dangerous Safe

5

Urbanstreet

Pedestrian crosses road from behind a parked van

Late

Dangerous Safe

6

Urbanstreet

Late Motorbike pulls out onto main road from side road

Dangerous Safe

7

Urbanstreet

Pedestrian runs Early into the main road

Safe

Dangerous

8

Urbanstreet

Schoolchildren walk across a zebra crossing

Early

Safe

Dangerous

9

Urbanstreet

Car pull out into the main road from a junction

Early

Safe

Dangerous

10

Rural

Early Van pulls out onto main road from side road

Safe

Dangerous

11

Rural

Motorbike is passing from the other side of the road

Early

Safe

Dangerous

12

Urban

Lorry is turning Early to go into a side road

Safe

Dangerous

Appendix B. Table B Information on pictures from the International Affective Picture System (IAPS).

935.84 993.96 417.51 611.32 544.99 891.37 291.44 243.25 993.07 442.61 261.89 262.03 306.16 354.35 404.34 1014.42 622.15 1213.58 558.33 225.17 362.43 184.4 307.4 611.22 51.32 68.35 60.01 92.61 68.23 69.62 45.85 85.58 39.89 40.52 49.86 84.61 79.12 44.65 58.23 47.19 64.74 64.81 68.57 84.3 71.33 69.9 82.68 49.65

blue (sd) blue (m)

40.82 91.51 61.86 109.52 114.19 70.42 52.2 128.18 33.52 111.53 30.5 176.1 71.1 26.15 52.66 92.98 132.57 89.74 102.54 156.83 71.45 73.22 60.93 67.98 66.38 61.45 72.31 88.33 58 59.1 57.2 78.77 60 62.16 68.83 84.04 88.66 54.49 64.89 65.12 70.53 58.79 83.54 75.91 64.29 70.03 80.55 51.45

green (sd) green (m)

77.42 94.44 82.17 101.47 107.74 59.29 85.29 125.6 89.81 131.9 83.48 179.84 82.92 47.75 63.86 125.63 161.07 79.69 124.26 151.17 61.54 65.79 72.85 89.22 94.76 64.08 85.72 87.31 58.94 72.59 68.62 73.18 79.28 86.55 78.31 88.45 96.21 79.19 95.21 65.06 70.97 59.61 73.78 70.83 65.66 72.85 77.37 62.48

red (sd) red (m)

103.48 118.77 119.17 115.3 111.45 72.18 123.35 137.2 102.01 143.11 142.31 176.62 116.93 88.78 127.94 159.06 146.58 74.91 165.38 153.54 71.28 64.77 101.55 109.62 70.73 59.69 73.04 88.18 58.12 62.01 58.13 76.84 62.79 64.29 66.28 82.17 88.36 57.42 68.07 60.61 69.27 58.76 76.66 74.46 63.18 70.18 78.36 53.48

lum (sd) lum (m)

81.21 101.41 91.04 106.5 109.56 64.38 93.07 129.37 87.28 133.02 95.3 178.46 91.82 57.68 81.85 132.07 153.59 79.36 134.2 152.51 65.55 66.3 80.15 93 2.03 2.04 2.2 1.88 2.22 1.88 1.82 1.96 2.49 2.62 2.26 2.37 2.4 2.27 2.38 2.47 2.15 2.1 2.48 2.26 2.03 2.36 2.15 2.35

Aro (sd) Aro (m)

4.42 3.32 3.79 3.71 3.77 2.93 2.43 3.26 5.22 5.07 4.64 4.54 4.77 5.63 4.46 5.42 5.88 4.5 5.09 4.64 5.85 5.43 5.09 5.16 1.68 1.08 1.69 1.24 1.46 1.06 1.52 1.08 2.02 1.59 1.6 1.73 1.47 1.7 1.8 1.72 1.9 1.89 1.73 1.58 1.73 1.59 1.42 1.48

Val (sd) IAPS code

1945 7036 5532 2595 7595 2393 2840 2745.1 1722 5660 7470 2030 1999 7501 7280 2345 1090 9341 7361 9041 2691 6242 2900 9340

Description

Turtle Shipyard Mushrooms Women Traffic Factory worker Chess Shopping Jaguars Mountains Pancakes Woman Mickey City Wines Children Snake Pollution Meat slicer Scared child Riot Gang Crying boy Garbage

Val (m)

303

References

4.59 4.88 5.19 4.88 4.55 4.87 4.91 5.31 7.04 7.27 7.08 6.71 7.43 6.85 7.2 7.41 3.7 3.38 3.1 2.98 3.04 2.69 2.45 2.41

Size (bytes)

M.P. Jones et al. / Accident Analysis and Prevention 73 (2014) 296–304

Arnau-Sabatés, L., Sala-Roca, J., Jariot-Garcia, M., 2012. Emotional abilities as predictors of risky driving behaviour among a cohort of middle aged drivers. Acc. Anal. Prev. 45, 818–825. Ben-Ari, O., 2012. The effects of positive emotion priming on self-reported reckless driving. Acc. Anal. Prev. 45, 718–725. Briggs, G.F., Hole, G.J., Land, M.F., 2011. Emotionally involving telephone conversations lead to driver error and visual tunnelling. Transport. Res. Part F 14, 313–323. Brodsky, W., 2002. The effects of music tempo on simulated driving performance and vehicular control. Transport. Res. Part F 4, 219–241. ~ no, M., Pedregal, D.J., 2012. How many lives can Castillo-Manzano, J.I., Castro-Nu bloody and shocking road advertising save? The case of Spain. Transport. Res. Part F 15, 174–187. Chan, M., Singhal, A., 2013. The emotional side of cognitive distraction: implications for road safety. Acc. Anal. Prev. 50, 147–154. Chapman, P.R., Underwood, G., 1998. Visual search of driving situations: danger and experience. Perception 27, 951–964. Chisholm, S.L., Caird, J.K., Lockhart, J., 2008. The effects of practise with MP3 players on driving performance. Acc. Anal. Prev. 40, 704–713. Crundall, D., van Loon, E., Underwood, G., 2006. Attraction and distraction of attention with outdoor media. Acc. Anal. Prev. 38, 671–677. Dahl, R.E., Spear, L.P., 2004. Adolescent brain development: vulnerabilities and opportunities. Ann. N. Y. Acad. Sci. 1–469. Delplanque, S., N'diaye, K., Scherer, K., Grandjean, D., 2007. Spatial frequencies or emotional effects? A systematic measure of spatial frequencies for IAPS pictures by a discrete wavelet analysis. J. Neurosci. Methods 165, 144–150. Driver, J., Baylis, G.C., 1989. Movement and visual attention: the spotlight metaphor breaks down. J. Exp. Psychol. 15 (3), 448–456. Easterbrook, J.A., 1959. The effect of emotion on cue utilisation and the organisation of behaviour. Psychol. Rev. 66 (3), 183–201. Fairclough, S.H., Tattersall, A.J., Houston, K., 2006. Anxiety and performance in the British driving test. Transport. Res. Part F: Traffic Psychol. Behav. 9 (1), 43–52. Frederickson, B.L., 2001. The role of positive emotions in positive psychology: the broaden-and-build theory of positive emotions. Am. Psychol. 56 (3), 218–226. Gorn, G., Pham, M.T., Sin, L.Y., 2001. When arousal influences ad evaluation and valence does not (and vice versa). J. Consum. Psychol. 11 (1), 43–55. Holahan, C., 1977. Relationship Between Roadside Signs and Traffic Accidents: A Field Investigation. Council for Advanced Transportation Studies. University of Texas Report No. 54. Horswill, M.S., McKenna, F.P., 2004. Drivers’ Hazard Perception Ability: Situation Awareness on the Road. In: Banbury, S., Tremblay, S. (Eds.), A Cognitive Approach to Situation Awareness: Theory and Application. Ashgate Publishing Limited, Hampshire, England. Hu, T., Xie, X., Li, J., 2013. Negative or positive? The effect of emotion and mood on risky driving. Transport. Res. Part F: Traffic Psychol. Behav. 16, 29–40. Hughes, P.K., Cole, B.L., 1986. What attracts attention when driving? Ergonomics 29, 377–391. Johnson, M.J., Chahal, T., Stinchcombe, A., Mullen, N., Weaver, B., Bédard, M., 2011. Physiological responses to simulated and on-road driving. Int. Psychophysiol. 81 (3), 203–208. Johnston, A.W., Cole, B.L., 1976. Investigations of distractions by irrelevant information. Aus. Road Res. 6 (3), 3–23. Kim, K.H., Bang, S.W., Kim, S.R., 2004. Emotion recognition system using short-term monitoring of physiological signals. Med. Biol. Eng. Comput. 42, 419–427. Kinnear, N., Kelly, S.W., Stradling, S., Thomson, J., 2013. Understanding how drivers learn to anticipate risk on the road: a laboratory experiment of affective anticipation of road hazards. Acc. Anal. Prev. 50, 1025–1033. Lambert, A.E., Seegmiller, J.K., Stefanucci, J.K., Watson, J.M., 2013. On working memory capacity and implicit associations between advanced age and dangerous driving sterotypes. Appl. Cognitive Psychol. 27 (3), 306–313. Lang, P.J., Bradley, M.M., Cuthbert, B.N., 2008. International Affective Picture System (IAPS): affective ratings of pictures and instruction manual. Technical Report A8. University of Florida, Gainesville, FL. Lansdown, T.C., 1997. Visual allocation and the availability of driver information. In: Rothengatter, T., Carbonell, E. (Eds.), Traffic and Transport Psychology: Theory and Application. Pergamon Press, Oxford. Lansdown, T.C., 2012. Individual differences and propensity to engage with invehicle distractions – a self-report study. Transport. Res. Part F: Traffic Psychol. Behav. 15, 1–8. McMonagle, J.C., 1952. The effect of roadside features on traffic accidents. Traffic Quarterly 6 (2), 228–243. Megías, A., Maldonado, A., Cándido, A., Catena, A., 2011. Emotional modulation of urgent and evaluative behaviours in risky driving scenarios. Acc. Anal. Prev. 43 (3), 813–817. Mesken, J., Hagenzieker, M.P., Rothengatter, T., de Waard, D., 2007. Frequency, determinants and consequences of different drivers' emotions: an on-the-road study using self-reports, (observed) behaviour, and physiology. Transport. Res. Part F: Traffic Psychol. Behav. 10 (6), 458–475. Miyazawa, S., Iwasaki, S., 2009. Effect of negative emotion on visual attention: automatic capture by fear-related stimuli. Jpn. Psychol. Res. 51 (1), 13–23. Neale, V., Dingus, T., Klauer, G., Sudweeks, J., Goodman, M., 2005. An Overview of the 100-Car Naturalistic Study and Findings. DOT HS Publication, pp. 05–400.

304

M.P. Jones et al. / Accident Analysis and Prevention 73 (2014) 296–304

Pêcher, C., Lemercier, C., Cellier, J., 2009. Emotions drive attention: effects on driver’s behaviour. Saf. Sci. 47, 1254–1259. Pradhan, A.K., Hammel, K.R., DeRamus, R., Pollatsek, A., Noyce, D.A., Fisher, D.L., 2005. Using eye movements to evaluate effects of driver age on risk perception in a driving simulator. Hum. Factors 47 (4), 840–852. Previc, F.H., 1998. The neuropsychology of 3-D space. Psychol. Bull. 124, 123–164. Reed, M.P., Green, P.A., 1999. Comparison of driving performance on-road and in a low-cost simulator using a concurrent telephone dialling task. Ergonomics 42 (8), 1015–1037. Reed-Jones, J., Trick, L.M., Matthews, M., 2008. Testing assumptions implicit in the use of the 15-second rule as an early predictor of whether an in-vehicle device produces unacceptable levels of distraction. Acc. Anal. Prev. 40, 628–634. Rhodes, N., Pivik, K., 2011. Age and gender differences in risky driving: the roles of positive affect and risk perception. Acc. Anal. Prev. 43 (3), 923–931. Rowe, G., Hirsh, J.B., Anderson, A.K., Smith, E.E., 2007. Positive affect increases the breadth of attentional selection. PNAS Proceedings of the National Academy of Sciences of the United States of America 104 (1), 383–388. Schmidt-Daffy, M., 2013. Fear and anxiety while driving: differential impact of task demands: speed and motivation. Transport. Res. Part F: Traffic Psychol. Behav. 16, 14–28.

Schmidt-Daffy, M., Brandenburg, S., Beliavski, A., 2012. Velocity, safety, or both? How do balance and strength of goal conflicts affect drivers’ behaviour, feelings and physiological responses. Acc. Anal. Prev. 55, 90–100. Siren, A., Kjaer, M.R., 2011. How is the older road users' perception of risk constructed? Transport. Res. Part F: Traffic Psychol. Behav. 14 (3), 222–228. Summala, H., Nieminen, T., Punto, M., 1996. Maintaining lane position with peripheral vision during in-vehicle tasks. Hum. Factors 38, 442–451. Thompson, K.R., Johnson, A.M., Emerson, J.L., Dawson, J.D., Boer, E.R., Rizzo, M., 2012. Distracted driving in elderly and middle-aged drivers. Acc. Anal. Prev. 45, 711–717. Trick, L.M., Brandigampola, S., Enns, J.T., 2012. How fleeting emotions affect hazard perception and steering while driving: the impact of image arousal and valence. Acc. Anal. Prev. 45, 222–229. Yerkes, R.M., Dodson, J.D., 1908. The relation of strength of stimulus to rapidity of habit-formation. J. Comp. Neurol. Psychol. 18, 459–482. Young, M.S., Mahfoud, J.M., Stanton, N.A., Walker, G.H., Salmon, P.M., Jenkins, D.P., 2007. Eyes front! Are roadside billboards bad for driver attention? Traffic Eng. Control 48 (8), 365–367. Zwahlen, H.T., Adams, C.C., de Bald, D.P., et al., 1988. Safety aspects of CRT touch panel controls in automobiles. In: Gale, A.G. (Ed.), Vision in Vehicles II. The North Holland Press, Netherlands, pp. 335–344.