SIXTH FRAMEWORK PROGRAMME PRIORITY 1.6. Sustainable Development, Global Change and Ecosystem 1.6.2: Sustainable Surface Transport
506184
Road User Behaviour Model Road User Behaviour Model
Workpackage Title Workpackage No.
WP8
Deliverable No.
D8
Authors (per company, if more than one company provide it together)
Gert Weller, Bernhard Schlag (TUD)
Contributing Authors
Ronald Jorna; Martijn van de Leur (Mobycon) Giovanni Gatti (Poliba)
Status
Final
File Name:
RIPCORD-ISEREST Deliverable D8.doc
Project start date and duration
01 January 2005, 36 Months
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
List of abbreviations ADT
Average daily traffic
AADT
Annual average daily traffic
al.
Alii (others)
CCR
Curvature Change Rate
CCRs
Curvature Change Rate of the single curve
e.g.
Exempli gratia (for example)
est.
estimated
EU
European Union
FFOV
Functional Field of View
km
Kilometer
m
Meter
ms
Millisecond
p.
Page
PDT
Peripheral Detection Task
R2
Regression parameter (proportion of variance explained)
r (or R)
Radius
RECL
Road Environment Construct List
rel.
relative
RHT
Risk Homeostasis Theory
RT
Reaction Time
s
Second
SDLP
Standard deviation of lateral position
SER
Self-Explaining Road
SPF
Safety Performance Function
st.
steady
t
Time
December 2007
2
TUD
Deliverable D8
Dissemination Level (PU)
TLC
Time-to-line / Time-to-lane Crossing
TTC
Time-to-collision / Time-to-contact
TUD
Technische Universität Dresden
UFOV
Useful Field of View
v
Speed
WP
Work-package
%
Percent
°
Degree
τ(t)
Tau
December 2007
3
Contract N. 506184
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Table of contents (main structure) List of abbreviations .................................................................................................... 2 Table of contents (main structure) .............................................................................. 4 Table of contents (detailed structure) ......................................................................... 5 List of Figures ............................................................................................................. 8 List of Tables ............................................................................................................ 10 Executive Summary .................................................................................................. 11 1.
Theoretical Background .................................................................................. 12 1.1.
Introduction .................................................................................................. 12
1.2.
Models of driving behaviour: an overview .................................................... 13
1.3.
Information-processing and perception ....................................................... 15
1.4.
Driving as a self-paced task: Motivational models ....................................... 19
1.5.
Application in rural road design: self-explaining roads................................. 23
2.
Model development and theoretical validation ................................................ 26 2.1.
Overview ..................................................................................................... 26
2.2.
Processes within the model in detail............................................................ 26
3.
Empirical validation: Methodology .................................................................. 32 3.1.
Formulation of Hypotheses .......................................................................... 32
3.2.
Data sources for the testing of the hypotheses ........................................... 33
4.
Empirical validation: Results ........................................................................... 36 4.1.
Hypothesis 1: Affordances and cues (Data Source A)................................. 36
4.2.
Hypotheses 2 and 3: Expectations (Data Source B).................................... 41
4.3.
Hypothesis 4: Workload: Psycho-physiology (Data Source C) .................... 45
4.4.
Hypothesis 4: Workload: Reaction times (Data Source D) .......................... 47
4.5.
Integration of behavioural data in the Safety Performance Function ........... 55
4.6.
Subjective road categorisation .................................................................... 58
5.
Empirical Validation: Conclusions ................................................................... 66
6.
References ..................................................................................................... 68
7.
Annex ............................................................................................................. 73
December 2007
4
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Table of contents (detailed structure) List of abbreviations .................................................................................................... 2 Table of contents (main structure) .............................................................................. 4 Table of contents (detailed structure) ......................................................................... 5 List of Figures ............................................................................................................. 8 List of Tables ............................................................................................................ 10 Executive Summary .................................................................................................. 11 1.
Theoretical Background .................................................................................. 12 1.1.
Introduction .................................................................................................. 12
1.2.
Models of driving behaviour: an overview .................................................... 13
1.3.
Information-processing and perception ....................................................... 15
1.3.1
Attention, mental models and expectations .......................................... 15
1.3.2
Visual perception: the eye and the useful field of view (UFOV) ............ 17
1.4.
Driving as a self-paced task: Motivational models ....................................... 19
1.4.1
Risk Models .......................................................................................... 19
1.4.2
Workload Models .................................................................................. 19
1.4.3
Behavioural adaptation ......................................................................... 22
1.5. 2.
Application in rural road design: self-explaining roads................................. 23 Model development and theoretical validation ................................................ 26
2.1.
Overview ..................................................................................................... 26
2.2.
Processes within the model in detail............................................................ 26
2.2.1
Part I: Affordances and cues ................................................................. 26
2.2.2
Part II: Perceptual invariants ................................................................. 28
2.2.3
Part III: Expected and actual workload and risk .................................... 29
2.2.4
Part IV: Feedback ................................................................................. 30
3.
Empirical validation: Methodology .................................................................. 32 3.1.
Formulation of Hypotheses .......................................................................... 32
3.1.1
Part I: Affordances and cues ................................................................. 32
3.1.2
Part II: Perceptual invariants ................................................................. 32
3.1.3
Part III: Expected and actual workload and risk .................................... 32
3.2.
Data sources for the testing of the hypotheses ........................................... 33
3.2.1
Own simulator experiments (Data Source A) ........................................ 33
3.2.2
Additional collection of data related to available data (Data Source B) . 34
3.2.3
Reanalysis of available data (Data Source C)....................................... 34
3.2.4
Own driving experiments with an equipped vehicle (Data Source D).... 35
December 2007
5
TUD
Deliverable D8
4.
Dissemination Level (PU)
Contract N. 506184
Empirical validation: Results ........................................................................... 36 4.1.
Hypothesis 1: Affordances and cues (Data Source A)................................. 36
4.1.1
Introduction ........................................................................................... 36
4.1.2
Method .................................................................................................. 36
4.1.3
Selected example: warning signs as formal cues ................................. 38
4.1.3.1.
Method and descriptive analysis .................................................... 38
4.1.3.2.
Results (selected example) ............................................................ 39
4.1.3.3.
Discussion (selected example) ....................................................... 40
4.1.4 4.2.
Discussion of Results ............................................................................ 41
Hypotheses 2 and 3: Expectations (Data Source B).................................... 41
4.2.1
Introduction ........................................................................................... 41
4.2.2
Method .................................................................................................. 42
4.2.3
Summary of results ............................................................................... 43
4.2.4
Discussion of results ............................................................................. 44
4.3.
Hypothesis 4: Workload: Psycho-physiology (Data Source C) .................... 45
4.3.1
Introduction ........................................................................................... 45
4.3.2
Method .................................................................................................. 45
4.3.3
Results .................................................................................................. 46
4.3.4
Discussion............................................................................................. 46
4.4.
Hypothesis 4: Workload: Reaction times (Data Source D) .......................... 47
4.4.1
Introduction ........................................................................................... 47
4.4.2
Method .................................................................................................. 47
4.4.3
Selected Analysis .................................................................................. 48
4.4.3.1.
Description of the selected locations .............................................. 48
4.4.3.2.
Results for the selected location ..................................................... 50
4.4.3.3.
Discussion for the selected location ............................................... 53
4.4.4 4.5.
Discussion............................................................................................. 54
Integration of behavioural data in the Safety Performance Function ........... 55
4.5.1
Introduction ........................................................................................... 55
4.5.2
Method .................................................................................................. 55
4.5.3
Results .................................................................................................. 56
4.5.4
Discussion............................................................................................. 57
4.6.
Subjective road categorisation .................................................................... 58
4.6.1
Introduction ........................................................................................... 58
4.6.2
Method .................................................................................................. 59
4.6.3
Results .................................................................................................. 60
December 2007
6
TUD
Deliverable D8
4.6.4
Dissemination Level (PU)
Contract N. 506184
Discussion............................................................................................. 65
5.
Empirical Validation: Conclusions ................................................................... 66
6.
References ..................................................................................................... 68
7.
Annex ............................................................................................................. 73
December 2007
7
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
List of Figures
Figure 1: Proportion of accident causation factors according to Treat et al. (1977). . 12 Figure 2: Overview of different driver behaviour models. ......................................... 13 Figure 3: Combination of performance levels according to Rasmussen (1986) and the hierarchical model according to Michon (1985), modified from Donges (1982, in 1999). ................................................................................................................ 14 Figure 4: The generic error-modelling system (GEMS) as proposed by Reason, (1990). ............................................................................................................... 15 Figure 5: Hypothetical differences in speed and workload in curves with good (left) and inappropriate design (right) (modified from Fuller 2005). ............................ 20 Figure 6: Workload assessment methods and their relationship within general safety assessment. ...................................................................................................... 21 Figure 7: Behavioural adaptation: resulting final outcome in safety. ......................... 22 Figure 8: Process model of behavioural adaptation (Weller & Schlag, 2004). .......... 23 Figure 9: Model of driving behavior on rural roads ................................................... 26 Figure 10: Detailed processes within part III of the driver behaviour model for rural roads: Safe distance keeping by using perceptual invariants proposed by Lee (1976) and Lee & Lishman (1977), tested e.g. by Yilmaz and Warren (1995). Adapted from Bruce et al. (1996). ..................................................................... 28 Figure 11: Detailed processes within part II of the driver behaviour model for rural roads: Expected and actual workload and risk. ................................................. 30 Figure 12: Simulator of the Fraunhofer IVI in Dresden that was used for the simulator experiments (www.ivi.fhg.de). Source: Fraunhofer IVI. ..................................... 33 Figure 13: Experimental vehicle of the Chair of Road Design at TUD. ..................... 35 Figure 14: Bird’s eye view of the simulated road sections in the Fraunhofer IVI simulator. ........................................................................................................... 37 Figure 15: Speed [km/h] averaged across all subjects for a curve with additional signs (bend ahead and guidance signs in the curve; curve K3) and the comparison curve (K19) in the to- and the backwards direction (=R). All curves are left curves; driving direction from left to right side. On the x-axis the distance from curve beginning is shown. The vertical bars mark the beginning (left bar) and end (right bar) of the curve. ............................................................................... 38 Figure 16: Pictures from the low (left) and the high (right) accident rate curve in pair No. 2 as used for the collection of the subjective ratings. .................................. 42 Figure 17: Pictures of the low accident rate road sections 21 (left side) and the high accident rate road section 24 (right side) in the to direction (pictures taken from RoadView TUD). ............................................................................................... 49 Figure 18: Curvature plan of sections 21 (low accident rate curve) and 23 (high accident rate curve) within the whole road stretch. From right to left: driving direction in to-direction; from left to right: driving direction in backwards-direction. .......................................................................................................................... 49 Figure 19: Averaged values for speed, reaction time (measured and interpolated) and fixation duration for the low accident rate curve (section 21). ............................ 51 Figure 20: Averaged values for speed, reaction time (measured and interpolated) and fixation duration for the high accident rate curve (section 24). .......................... 51
December 2007
8
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Figure 21: Relation of the percentage increase in reaction time with the percentage decline in speed (x-axis). Each value on the x-axis represents a single road section. Added is the statistics of a linear regression. ....................................... 56 Figure 22: Results of the hierarchical cluster analysis (dendrogram) (SPSS.14). .... 62 Figure 23: Average factor values for each picture in the different clusters. .............. 63
December 2007
9
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
List of Tables Table 1: Matrix of the data sources used and the hypotheses tested. ...................... 33 Table 2: Simulator experiments: results of the t-Tests for paired samples for different speed parameters for K3 (signs) and respective comparison curve (K19) in both directions. .......................................................................................................... 39 Table 3: Simulator experiments: results of the t-Tests for paired samples for the distances of the maximum speed before the curve (200m to beginning of curve) and the minimum speed after the beginning of the curve. Curves K3 and K19. 40 Table 4: Results of the Wilcoxon-Test for paired samples for the subjective ratings for curves. ............................................................................................................... 40 Table 5: Results of the Wilcoxon-Test for paired samples for the item „The curve is sharp“. ............................................................................................................... 43 Table 6: Results of the Wilcoxon-Test for paired samples for the item „The curve gives good information concerning the following curve path“. ........................... 43 Table 7: Results of the Wilcoxon-Test for paired samples for the item „The road stretch is demanding “. ...................................................................................... 44 Table 8: Results of the Wilcoxon-Test for paired samples for the item „The curve is dangerous“. ....................................................................................................... 44 Table 9: Statistical differences of physiological data and speed between high and low accident rate curves. ......................................................................................... 46 Table 10: Average fixation duration [s]; results of the t-Test for paired samples; low versus high accident rate curve (section 21 versus 24). .................................... 52 Table 11: Average speed [km/h]; results of the t-Test for paired samples; low versus high accident rate curve (section 21 versus 24). ............................................... 52 Table 12: Average reaction times [s] interpolated values; results of the t-Test for paired samples; low versus high accident rate curve (section 21 versus 24). ... 52 Table 13: Results of two linear regression analysis of different reaction time parameters on the percentage decline of speed for different road sections. Once for all twelve sections and once without the extremes on both sides. ............... 56 Table 14: Varimax normalized factor loadings of the RECL items after factor analysis. .......................................................................................................................... 61 Table 15: Road-cluster and factor characteristics combined in a matrix. .................. 63 Table 16: Results of the regression analysis of the three factors on the speed ratings for each road picture.......................................................................................... 64 Table 17: Distinctive objective features between the clusters resulting from the subjective ratings. .............................................................................................. 65
December 2007
10
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Executive Summary The deliverable at hand describes the steps which were undertaken to develop and validate a driver and driving behaviour model for rural roads. First, an overview is given of the theoretical background relevant to these steps. Theories of human perception, information processing, decision making and action in general are included, as well as psychological theories especially developed to explain driver and driving behaviour. The second part of Deliverable D8 introduces the model which was developed based on the theoretical work summarized in the first chapter. This model describes, explains and predicts driver and driving behaviour on rural roads. The third part of Deliverable D8 summarizes the steps which were conducted to validate this driver and driving behaviour model for rural roads and to test the possibility to integrate psychological parameters in a safety performance function (SPF). Depending on the different hypotheses derived from the model, the following data sources were used for this process: - existing driving studies, - additional data collected based on this existing data, - own additional simulator experiments, - own additional driving studies with an equipped vehicle. Additionally, own laboratory experiments were conducted in a study of subjective road categorization. The results found after analysing the data collected in all these studies support the assumptions formulated in the models to a large extent. For example, it was shown that high accident rate curves are systematically underestimated concerning demand and risk. Evidence was found that this underestimation results in inappropriate speed behaviour which could cause accidents. Further on, the influence of cues and affordances in influencing driving behaviour could be shown and is reported in this deliverable exemplarily for road signs. The assumption that workload or risk is higher in high accident rate road sections, compared to low accident rate road sections was only indirectly supported by the empirical data. This finding is in line with the results of the study on subjective road categorisation. While we found in general, that these road categories influence behaviour these categories themselves are built based on affordances and cues, rather than expected workload or risk. Further, we could identify objective criteria which could be used to approximate these subjective categories. These results in turn could provide a valuable input towards harmonizing roads in Europe along selfexplaining road principles. Finally the integration of psychological variables in a safety performance function was tested. Despite of the validation process of the model was successful, it was not possible to assign numeric values to psychological parameters, which would be the prerequisite for an integration in the SPF. However, we could show that characteristic speed parameters could be used to approximate e.g. workload. The application of speed prediction models in the SPF thus allows at least a preliminary solution. By integrating the findings of our studies in future steps, the quality of these speed prediction models could be considerably enhanced. The integration of such psychological factors in speed prediction models for rural roads could ultimately result in a valid safety performance function. The most important future step to do so is to increase the number of cases (i.e. road sections) in the database to be able to derive stable parameter values.
December 2007
11
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
1. Theoretical Background 1.1. Introduction A study originally published by Treat et al. (1977) revealed that human factors are to be blamed for the majority of accidents (see Figure 1). Human (95.4%)
47.8% 6.4%
34.8% 6.4%
1.6% 0.4%
Vehicle (14.8%)
2.6%
Environment (44.2%)
Figure 1: Proportion of accident causation factors according to Treat et al. (1977).
While the statistics suggest that the roads are hardly to be blamed for accidents, analysis on a site basis reveals that human errors occur in specific sites more often than in other sites. This is notably true for rural roads: despite rural roads ranking by far highest concerning the number of people killed the danger associated with them is clearly underestimated by drivers (Ellinghaus & Steinbrecher, 2003). The majority of accidents at these sites is due to a mismatch between environment (the road) and human characteristics. This is depicted by the high proportion of accident causation factors as interaction between environment and human (see Figure 1). The following pages give an overview of how road environments interact with human properties and how both factors have to be taken into account in order to design safer rural roads.
December 2007
12
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
1.2. Models of driving behaviour: an overview This chapter gives a short overview of different driver and driving behaviour models. The aim is to make the reader familiar with the most important terms. While details of some models are explained further in the text, others will only be mentioned here. A more detailed discussion of different models can be found in Michon (1985) or Ranney (1994) and the internal report 8.1 (Weller, Schlag, Gatti, Jorna, & Leur, 2006).
Hierarchical models
Control loop models
(e.g. Michon, 1985)
(e.g. Durth, 1974)
constitute a framework of the driving task in which other theories can be integrated
Taxonomic models
Functional models
emphasis on individual differences, static in nature
take into account complex interactions (road / driver / vehicle) in driving
motivational models
information processing models
cognitive models
theory of direct perception
(e.g. Rumar, 1985)
(Gibson, 1986)
(driving is a self-paced task)
Risk models
(e.g. Wilde, 1994)
Workload models
(e.g. Fuller, 2005)
Figure 2: Overview of different driver behaviour models.
Hierarchical (Michon, 1985) and control loop models (Durth, 1974) serve as a framework for other theories. A widespread hierarchical model developed by Michon (1971, 1979, cited from 1985) and Janssen (1979, cited from Michon, 1985) sees driving as a hierarchical problem solving task that comprises three different levels. These levels can be divided by the specific task requirements on each level, the time frame needed to carry them out, and the cognitive processes involved. The hierarchical task model of Michon finds its equivalent in the distinction between different performance or behaviour levels proposed by Rasmussen (1986). Rasmussen distinguished between knowledge-based, rule-based, and skill-based levels of a task in general. Both models can be combined as proposed by Donges (1982, cited from 1999) (see Figure 3).
December 2007
13
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Knowledge-based Behaviour Identification
Decision
Planning
Route Speed Criteria
Rule-based Behaviour Recognition
Association
Strategic Level
Stored rules
Manoeuvring Level
Feedback Criteria Skill-based Behaviour Feature Formation
Stimulus Reaction Automatisms
Sensory Input
Control Level
Action
Figure 3: Combination of performance levels according to Rasmussen (1986) and the hierarchical model according to Michon (1985), modified from Donges (1982, in 1999).
The left section in Figure 3 represents the different task levels proposed by Rasmussen, while the right section represents the model by Michon. The strategic or navigational level comprises all processes concerning trip decisions, like where to go, when to go, what roads to take, and what modes of transport to use. Decisions on this level are rare and take longest in comparison to the other levels. Due to their nature they are processed in a more or less aware mode, but become habits in case of constant repetition. On the manoeuvring level decisions are made within seconds. Typical manoeuvres are overtaking, turning, or gap acceptance. Behaviour on the manoeuvring level is both influenced by motivational and situational variables. Other terms used to describe the manoeuvring level are tactical or guidance level. Finally, decisions on the control level are made rather automatically within a very short time range as stimulus response reactions. Typical tasks on this level are lane keeping or gear shifting. These are both conducted without conscious information-processing by experienced drivers. The terms operational or stabilisation level are used concurrently. Whether a task is situated on the knowledge-based, rule-based, or skill-based level, depends to a great amount on the familiarity with the task and the environment. Higher order processes situated on the knowledge-based level in general require more cognitive resources than lower level processes. Higher and lower levels of processing are usually referred to as controlled or automatic processing according to Schneider and Shiffrin (1977) and Shiffrin and Schneider (1977). The knowledge whether the behaviour under observation is situated on the automatic or the control level is very important as the strategies to change this behaviour depend on these levels. Only controlled processes can be modified by awareness campaigns, while behaviour on the automatic level needs constant reshaping. The following Figure 4 gives a flow chart of how decision making and problem solving might be arranged in driving. Note that higher order processes are only used when lower order processes do not lead to the desired output. December 2007
14
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Skill-Based Level Routine actions in a familiar environment OK?
Yes
Attentional checks on progress of action
Rule-Based Level
Goal state
OK?
Yes
No No
Problem
Is problem solved?
Consider local state information.
Is the pattern familiar?
Yes Apply stored rule IF (situation) THEN (action).
No
KnowledgeBased Level
Find higher level analogy
None Found Revert to mental model of the problem space. Analyse more abstract relations between structure and function.
Infer diagnosis and formulate corrective actions. Apply actions. Observe results, ... etc.
Subsequent attempts
Figure 4: The generic error-modelling system (GEMS) as proposed by Reason, (1990).
The crucial point for rural road design is that people, in general, rather rely on preprogrammed behavioural sequences found on the skill-based level, than revert to higher-order processes. This is because the latter processes require more resources. Similar, rule-based behaviour will be preferred to knowledge-based behaviour as “… humans, if given a choice, would prefer to act as context-specific pattern recognizers rather than attempting to calculate or optimize” (Rouse, 1981, cited from Reason, 1990, p. 65 ).
1.3. Information-processing and perception 1.3.1 Attention, mental models and expectations Human perception and information processing is influenced by two concurrent systems, a bottom-up and a top-down pathway. In short, top-down processing means that the driver has formed some kind of hypothesis on what to expect in a given situation. Bottom-up processing in contrast means that attention is guided by stimuli in the environment, without higher order cognitive functions. Processes involved in top-down processing are attention, experience, motivation and expectations. Expectations in turn are formed from past experiences. The more December 2007
15
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
similar the new situation is to a past situation, the stronger these expectations will be for the current situation. These expectations in turn help the driver to direct attention to locations where he assumes to find relevant information. The totality of expectations related to a specific situation form a mental model or internal representation of the whole situation. Other terms in relation to mental models are schemata or scripts. All represent implicit or explicit knowledge of situations or actions. Due to its nature, top-down processing requires more time than bottom-up processing. Nevertheless it still increases efficiency and effectiveness in human behaviour due to its simplification in comparison to nature. Second, the use of mental models is automatic rather than conscious and therefore needs less resources in working memory. Top-down processing further guides attention to relevant stimuli and therefore allows an efficient allocation of attentional resources. Finally it allows the driver to actively search and infer missing information. This advantage can easily become a disadvantage when the current situation is misinterpreted, e.g., on the basis of inappropriate expectations and misguided attention. Therefore, internal representations can be the underlying cause behind faulty actions or faulty assumptions themselves (Hacker, 2005; Norman, 1981; Reason, 1990). Further, the stable nature of internal representations makes them hard to be changed by single actions. Concerning top-down processes it should be taken care that the road characteristics are in line with the drivers’ expectations (top-down). In order to change wrong mental models, feedback has to be provided in case of inappropriate behaviour. On the other hand perception is a bottom-up process, meaning, amongst others, that environmental stimuli guide attention as well. Whether attention will be attracted to a stimulus or not, depends on the physical characteristics of this stimulus. As the focus of attention is very narrow due to the characteristics of the eye (see below) the stimuli will first be perceived by peripheral attention. Peripheral attention is captured more easily by moving objects. Stationary objects with low luminance contrasts will be hardly detected by human vision. Therefore, it has to be taken care that non-relevant information does not capture attention (bottom-up) in locations that are supposed to be dangerous while on the other hand relevant information has to be designed to attract attention. The relevance of expectations and mental models for rural road design is in fact already tackled in the engineering concept of “consistency” (e.g., concerning curvature). Consistency, in this context, means that the driver expects the following road section to be similar to the preceding road section, unless indicated by some environmental cue. Besides being used in design guidelines for rural roads (e.g. RAS-L: FGSV, 1995), consistency is an important aspect of safety. Lamm et al. (2006) successfully applied the following three criteria to assess the safety level of rural roads: - design consistency as indicated by the design speed, - operating speed consistency as indicated by differences in V85 between successive elements, - and consistency in driving dynamics, mainly based on side friction. Accidents often occur when the drivers’ expectations do not match the road situation, that is, the road is not consistent.
December 2007
16
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
1.3.2 Visual perception: the eye and the useful field of view (UFOV) Most information needed for driving is taken up predominantly visually. Understanding vision, therefore, helps to understand and explain safe or unsafe behaviour on rural roads. In the retina of the human eye, two different light receptor cells (rods and cones) with different characteristics are to be found. The uneven distribution of these cells in the retina is the reason for an approximately linear degradation of many visual functions with eccentricity from the fovea. Referring to this degradation, often the terms foveal, parafoveal (near but not in the fovea) and peripheral or ambient vision are used. Object identification, which requires deep processing, is only possible in foveal vision and in a very narrow cone around the point of fixation. In contrast to foveal vision, peripheral vision allows a broad area to be scanned without identifying objects. It can be seen as alerting system for saccades (very fast eye movements) to bring the object of interest into foveal vision. Peripheral vision is further very important for the correct perception of speed. These different visual systems are related to two different pathways of information processing in the brain (Milner & Goodale, 1995). Both the areas of foveal and peripheral vision are limited and subject to change. To describe these changes and the areas affected, different terms are in use: - functional field of view (FFOV) - useful field of view / of vision (UFOV) - visual field - tunnel vision UFOV can decrease because of different reasons. One of the reasons is changes in demand or workload (see below). Related to demand, some authors see complexity to be the reason behind diminishing UFOF size (Miura, 1990; Recarte & Nunes, 2000). Decreased UFOV size due to higher speeds is as well reported (e.g., Land & Horwood, 1995). The importance of peripheral vision for speed perception could be shown by Cavallo & Cohen (2001) who found that correct speed estimation is significantly reduced when the size of the visual field, and thus peripheral vision, is diminished. Recarte and Nunes (2000) used the spatial distribution of fixations to describe these changes. However, when discussing effects on peripheral vision it is important to note that the terms introduced above are not used consistently between authors. Within this framework of perceptual processes further characteristics of human perception have to be taken into account when dealing with secondary rural road safety. Some of them are summed up as follows (for further aspects see e.g., Bruce, Green, & Georgeson, 1996): -
-
-
The human eye needs time to adapt to different light conditions. The time for rods and cones to adapt from brightness to darkness takes longer than vice versa and might take up to 30 minutes for rods (von Campenhausen, 1993). This is relevant when entering tunnels or alleyways in daylight. The human eye needs time to accommodate from near to far and vice versa. This accommodation is relevant when drivers direct their attention from inside the car (e.g., speedometer) to outside the car. Accomodation is faster from near to far than vice versa. Human information processing capabilities are limited. When the amount of information is too high, relevant information might not be perceived by the driver.
December 2007
17
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
-
The human eye is only sensitive for light of a very narrow bandwidth and high contrasts. Given contrast sensitivity, it has to be assured that visual information can be perceived in the environment and background where it is presented. - Foveal vision is very restricted but identification of objects is only possible when they are fixated. - Human perception depends on the context and is relative to other stimuli, as shown by psychophysics (Weber, Fechner, Stevens, overview e.g., in Goldstein, 2005) A theory, which stresses the importance of visual perception, was developed by Gibson (1986). This theory of direct perception highlights the importance of characteristics present in the environment and the influence of ecological invariants. Time-to-collision (TTC) or Tau and time-to-line-crossing (TLC) (Godthelp, Milgram, & Blaauw, 1984) are examples of such invariants. Further, Gibson assumes that information is directly picked up from the inherent properties of the objects. These properties are called affordances. Affordances convey a meaning to the observer in the sense of being … – able (e.g., climbable). They thus serve as cue to prompt the respective behaviour at the same time. Contrary to Rumars model (1985), Gibson uses a mere bottom-up approach. Both agree however, that perception is an active process. While Rumar stresses the importance of cognitive factors, Gibson sees movement as the crucial aspect in information acquisition. Movement of the body and the eye help to perceive the property of objects and environments. Therefore, the human body as a whole becomes the organ of perception, and not the eye alone. Through movement, information of depth, distance, or speed is conveyed to the driver. This information is perceived directly from the rate of change in the texture or the so called “optic flow field”. The optic flow field can be imagined as a bunch of vectors created by changes in light due to movement. The focus of the flow field specifies the direction where the observer is heading. Warren et al. (1991) showed that circular heading when negotiating a curve is also derived from the optic flow field. But even without movement, objects convey information through their texture and occlusion of their contour by other objects (examples are given in Bruce et al., 1996). While human perception becomes effective through the use of this information, it can be a source of error itself as is shown by optic illusions. With perception being the basis for action, environmental design to support desirable behaviour is crucial in designing safer roads. The following principles derived from characteristics in visual perception should be known by road designers: - highly textured environments usually diminish speeds - roadside objects should follow road geometry in order to support the drivers expectations - the perceived characteristics of road elements are more important for behaviour than the real characteristics. By applying visual elements the perceived characteristics can be changed. In case mere perceptual measures are not possible due to environmental constraints, road designers can still revert to traditional measures like posting speed limits on signs and enforcing compliance with cameras. In fact, there are several studies that indicate that these measures reduce speed and accidents (for an overview see Elvik & Vaa, 2004).
December 2007
18
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
1.4. Driving as a self-paced task: Motivational models While the models on the role of perception in driving rather highlight common characteristics of the whole driver population, motivational models take into account interactions between general mechanisms and individual differences. The unifying assumption of motivational models is that they stress the self-paced nature of the driving task. Two concepts that could thus be called “motivational” are risk and workload. Closely related is the concept of behavioural adaptation.
1.4.1 Risk Models The central aspect for risk models is the distinction between subjective and objective risk. Klebelsberg (1982) defines objective risk as the measurable probability of having an accident, while subjective risk is the estimated risk by the driver through the perception of the road environment. According to Klebelsberg, situations are unsafe as soon as subjective risk is lower than objective risk. This is because drivers adjust their behaviour according to subjective, not objective risk. The concept of subjective risk as relevant mechanism for driving behaviour was further developed by Wilde (1988; 1994). Originally called theory of risk homeostasis (RHT) it was later termed the theory of target risk. In short, the theory states that accident rates per unit time remain equal, despite objective improvements, as drivers adjust their behaviour so that their subjective risk equals their more or less constant target risk. Elvik & Vaa (2004) sum up the shortcomings of the theory but at the same time agree with other researchers that the theory has identified important mechanisms, which should be taken into account when explaining accident causation mechanisms. A theory applicable on the individual level was developed by Näätänen & Summala (1976).
1.4.2 Workload Models Due to the shortcomings of risk theories, Fuller (2005) developed a theory based on the comparison between task demand and human capability. The resulting outcome of this comparison is the amount of workload a driver experiences. In general, workload is lowest and performance is best at medium levels of demand. Both underand overload caused by a mismatch between demand and capability are detrimental on performance, although compensation due to additional effort invested is possible (see e.g. de Waard, 1996). According to Fuller (2005) driving is save as long as capability exceeds demand. Besides being a function of the objective environmental characteristics, the demand of the driving task at a given time or location, depends on the speed level selected by the driver. The demand of a difficult situation can be substantially decreased by lowering the speed. In order to keep workload at a medium, optimal level, the situation has to convey the necessary information to the driver in advance. The effects of early versus late presentation of appropriate information on workload are depicted in Figure 5.
December 2007
19
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Workload
Workload Begin of curve
Begin of curve
End of curve Speed
Speed [1/ R]
End of curve
[1/ R]
Horizontal alignment
Horizontal alignment Station [m]
Station [m]
Figure 5: Hypothetical differences in speed and workload in curves with good (left) and inappropriate design (right) (modified from Fuller 2005).
In the left image early information leads to early, smooth speed reduction and a subsequent steady level of workload. In the right image, curve characteristics are perceived too late, leading to a high and sudden decrease in speed, which in turn results in a massive increase in workload. Despite Figure 5 suggesting otherwise different forms of demand, capability and workload are distinguished. This distinction is mainly based on Wickens (e.g., 1991) who distinguished resources according to the task characteristics, the senses used to take up and process the information and, finally, the modality with which the resulting action is carried out. Depending on these categories, human resources are regarded as being independent. Therefore, it is often better to present critical information auditory and not visually as the visual system in driving is usually subject to very much other visual information. For the assessment of workload different techniques are in use. Which workload measurement technique is used, depends first of all, on the quality of the measures as described by O`Donnell and Eggemeier (1986, cited from de Waard, 1996; Wickens, 1992) and the requirements and restrictions of the experimental situation. Usually the following five techniques are distinguished: - self-report measures - primary task measures - secondary task measures (dual task paradigm) - physiological measures - visual occlusion. As the most important contributor to the amount of workload in road safety is the amount of demand (not the capacity of the single driver) the road characteristics have to be assessed with equal care. The following characteristics are a selection of the most important elements contributing to demand on rural roads: - vertical and horizontal alignment - deduced parameters like curvature and consistency - road furniture, including lines - surrounding vehicles - environmental conditions at the time of the assessment.
December 2007
20
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Demand and workload assessment together with the five measurement techniques of workload, as explained in the report, are shown in Figure 6 as part of a general safety assessment procedure for rural roads. assessment of objective task demand: (horizontal and vertical alignment; curvature and consistency; environment; etc.) primary task measures (speed acceleration, position, etc.)
assessment of traits & state
secondary task measures assessment of workload (visual, mental, physical)
assessment of longterm consequences of workload (e.g. monotony, fatigue)
self-report measures psycho-physiological measures occlusion (if applicable)
relation to objective consequences (accidents)
Figure 6: Workload assessment methods and their relationship within general safety assessment.
Due to the self-paced nature of the driving task and interactions between parameters, the exact amount of demand is hard to determine. Nevertheless, some approaches provided good results in determining demand. Wagner & Richter (1997) and Wagner (2000) proposed a procedure based on video ratings. They combined several criteria rated in advance as useful by engineers and psychologists. The criteria selected were divided into three groups: - Information-uptake: amount; variability; contrast; spatial and temporal density; visual guidance. - Road quality: surface; orientation possibilities and compatibility with expectations; early perception of danger. - Sensorimotor aspects of car driving: hand; foot; coordination and automatic processing of motor response. The resulting scale (ANSITAX) was presented to different expert groups and resulted in high reliability, both between groups and within groups at different times. Joint assessment by psychologists and engineers proved successful in a study conducted in Switzerland, too (Allenbach, Hubacher, Huber, & Siegrist, 1996).
December 2007
21
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
1.4.3 Behavioural adaptation Behavioural adaptation describes the phenomenon that people adapt their behaviour to changing situational demands. In 1990 the OECD (1990) defined behavioural adaptation as: “… those behaviours, which may occur following the introduction of changes to the road-vehicle-user system and which were not intended by the initiators of the change; Behavioural adaptations occur as road users respond to changes in the road transport system, such that their personal needs are achieved as a result, they create a continuum of effects ranging from a positive increase in safety to a decrease in safety“ (p. 23). Summaries of studies dealing with behavioural adaptation can be found in the OECD report (1990). Whether the net outcome is positive or negative depends on the amount of not intended factors due to behavioural adaptation as shown in Figure 7. Target risk factors (effect as intended by engineering factors)
Net resulting final outcome (accidents, etc.)
Road safety measure Other risk factors (effect of behavioural adaptation not intended) Figure 7: Behavioural adaptation: resulting final outcome in safety.
One could argue similar to RHT that behavioural adaptation implicates that sole engineering measures would not result in a reduction of accidents. In fact there are publications supporting this assumption. When comparing data from a 14 year period (1984-1997) of 50 US states it was found that the downward trend in fatalities is due to demographic factors, an increase in passive safety and improvements in medical technology (Noland, 2003). Improvements in infrastructure did sometimes even have negative effects suggesting behavioural adaptation. Infrastructure included total lane miles, average number of lanes, lane width and percentage of each road class. Curvature, shoulder width, separation of lanes and presence of roadside hazards are not included but it is implicitly assumed that newer roads are built in a safer way. Noland (2003) provoked with the conclusion: “Results strongly refute the hypothesis that infrastructure improvements have been effective at reducing total fatalities and injuries.” (p. 599). Rothengatter (2002) states however, that adaptation in fact occurs but that the effects are not strong enough to eat up positive impacts of safety measures. Somewhat contrary Dulisse (1997) points out that the effects of behavioural adaptation are sometimes even underestimated due to methodological shortcoming (for example, inclusion of drivers who wore seat belts even before wearing was made compulsory). The different findings concerning the amount of behavioural adaptation can be explained by the multiple factors that influence the occurrence of behavioural adaptation. These factors were summarized in a model developed by Weller & December 2007
22
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Schlag (2004) (see Figure 8). Similar aspects are named by Bjørnskau (1994; cited from Elvik & Vaal, 2004). Changes in vehicle or environment Potential changes in:
hTrust
Objective enhancement of safety margins?
Advertising, Information, etc.
No
Yes
Feedback to driver
Subjective enhancement of safety margins?
hSituational
No
Awareness hAttention hWorkload hLocus of
Driver personality: h Sensation Seeking h Age, etc.
Yes No
Control Driving motives
Subjective enhanced utility of adaptation?
Yes
N o A d a p t a t i o n
Adaptation Figure 8: Process model of behavioural adaptation (Weller & Schlag, 2004).
According to this model, the implemented measure has first to provide the objective possibility to change ones behaviour in an unsafe way. Second, the driver has to perceive this possibility. Whether the change is perceived depends on the communication of the measure through media information or advertisements on one hand and on direct feedback to the driver on the other hand. To result in adaptation, the change in behaviour further has to be perceived as being positive for the driver (utility maximization). This function is different between different driver groups (e.g., age groups), as well as within the same driver group (e.g., driver while being in a hurry or not). Independent of this chain of action (objective enhancement, subjective enhancement, utility maximization), there is a second path that leads to adaptation, namely direct change of genuine psychological variables. These changes are a direct outcome of changes in the environment (or the car) and the following changes in the nature of the driving task. When the driving task becomes more easy due to changes in the alignment (straight instead of curved), workload might decrease and speed might be increased as a consequence. In fact, workload is seen as being equally important as risk to explain driving behaviour.
1.5. Application in rural road design: self-explaining roads Research results on information-processing and perception as described in the preceding chapters, were applied in the development of a high successful road design concept, the self-explaining road or SER concept. In short, the term self-explaining already implicates the meaning of SER design: roads designed along SER principles should elicit appropriate behaviour solely due to their perceived design and without further need on the side of the driver to December 2007
23
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
consciously elaborate the required behaviour. Obviously, how this is achieved in detail requires further explanation. Before application in the field of traffic, principles of self-explaining design were developed by Donald A. Norman in his book “The design of everyday things” (1998). According to these principles road design should follow cultural standards or physical analogies as stored in mental models. Only where this mapping principle is not selfexplaining a conceptual model has to be provided with the help of additional cues like signs. These cues should follow the principle of visibility to allow the driver to correctly predict the outcome of his actions. Visibility, first of all, means that information: - has to be physically visible and mentally recognizable (for everyone) - has to be presented at locations and in a way that are in accordance with human expectations - must guide behaviour in a self-explaining way: no explanation needed. However, visibility in Norman’s sense exceeds this meaning as visibility of behavioural outcomes is included, too. It is related to feedback, which communicates the appropriateness of behaviour to the driver. Thus, self-explaining properties have to be added by self-enforcing impact if things are not used appropriately. Affordances in the sense of Gibson have to be provided by the design without additional information. Despite the term self-explaining suggesting otherwise, behaviour associated with a specific design or design element has to be learned in the first place. In case unknown objects or new design elements are encountered, the behaviour elicited is determined by the degree of similarity to the original object. While some roads are highly self-explaining, like motorways, rural roads seem to lack this quality. Therefore, implementing self-explaining road design will lead to substantiate changes in the perceived characteristics of rural roads. Some of these characteristics will not be self-explaining the first time they are encountered. In this case, the appropriate behaviour has to be learned. The ways how this learning is done have to be known in order to be successful. In general, the following four principles are applicable: - Explicit or purposeful learning due to information and education. - Observational learning in the sense of Bandura (for a summary on Bandura, see Gerrig & Zimbardo, 2005; Schlag, 2004). - Contingency management; refers to the way feedback is given, both in time and type (for a review see as well Schlag, 2004). - Stimulus control (antecedent to behaviour). Especially the last three principles will contribute to learning the appropriate behaviour, concurrently. In contrast to the first one, no special effort is required from road authorities, except that the principles are applied consistently throughout the whole road system. These principles were developed amongst others as consequence to human error research in driving. Some conclusions, which were derived from this line of research, were published by Hale et al. already in the nineties (Hale, Stoop, & Hommels, 1990). Theeuwes & Godthelp (1995) and Theeuwes (2000) further elaborated these principles. They were summed up by Theeuwes (2000, p. 21) as follows:
December 2007
24
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
“- Roads should consist of unique road elements (homogeneous within one category and different from all other categories). - Roads should require unique behavior for a specific category (homogeneous within one category and different from all other categories). - Unique behavior displayed on roads should be linked to unique road elements (e.g., woonerfs: obstacles—slow driving, freeway: smooth concrete—fast driving). - The layout of crossings, road sections, and curves should be linked uniquely with the particular road category (e.g., a crossing on a highway should physically and behaviorally be completely different from a crossing on a rural road). - One should choose road categories that are behaviorally relevant. - There should be no fast transitions going from one road category to the next. - When there is a transition in road category, the change should be marked clearly (e.g., with rumble strips). - When teaching the different road categories, one should not only teach the name of, but also the behavior required for, that type of road. - Category-defining properties should be visible at night as well as in the daytime. - The road design should reduce speed differences and differences in direction of movement. - Road elements, marking, and signing should fulfill the standard visibility criteria” (Theeuwes, 2000, p. 21).
December 2007
25
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
2. Model development and theoretical validation 2.1. Overview Based on report 8.1, a model of the regulation of driver and driving behaviour on rural roads has been developed, assuming three main factors and an additional feedback loop, which influence behaviour and thus safety: - Part I: Affordances and cues are used as long as they are present and as long as they are known and perceived by the driver, - Part II: Perceptual invariants are used for the short-term regulation of driving based on visual perception - Part III: Expected and actual level of workload and risk are used in a homeostatic process to regulate behaviour whenever the two other mechanisms are not sufficient and - Part IV: Feedback.
Figure 9: Model of driving behavior on rural roads
Figure 9 gives an overview of these mechanisms and their relationship which will be explained in the following in detail.
2.2.
Processes within the model in detail
Readers who are not familiar with the preceding reports are advised to read the following chapter in order to understand the ideas in the model that will be validated in the course of this report.
2.2.1 Part I: Affordances and cues The driver perceives the road and the road environment ahead and its inherent properties. These properties convey a message to the driver that can be enough to be effective in regulating driver and driving behaviour. In fact, this is the aim of selfexplaining road design (see e.g. Theeuwes, 2000). The question is of course how road and environmental properties regulate behaviour in detail. Environmental properties and suggested behaviour are associated through knowledge in the furthermost sense. This knowledge is learnt and does not have to be explicit. Learning of the association between property and behaviour is achieved through a multitude of ways by (for summaries see e.g. Funke & Frensch, 2006; Koch, 2005; Schlag, 2004): December 2007
26
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
- Classical conditioning, - instrumental learning through operant conditioning, - social learning, - implicit and - explicit (purposeful) learning. Classical conditioning uses innate associations between a certain stimulus and a subsequent behaviour. When the stimulus is shown together with another stimulus (for some time and in a predefined way), this new stimulus will afterwards elicit the behaviour without the original stimulus. Operant conditioning means that a positive or negative consequence follows an act performed by a person before. The close and consistent relationship between a certain antecedent (stimulus) and a certain behaviour is called contiguity and the relationship between a behaviour displayed and a certain consequence is called contingency. Without contiguity the intended behaviour will not become associated with the stimulus. Without contingency the behaviour will not be associated with its consequences (Schlag, 2004). These consequences are rewards or punishments in the furthermost sense of the words. Whereas punishment and reward in traffic safety could be monetary, they are usually constantly given in the form of positive or negative feelings (feeling of safety/danger; feeling of comfort/discomfort; etc.) or to a far lesser extent conflicts and accidents. Social learning means that someone learns from watching someone else doing something and from the consequences of this behaviour. Implicit learning is difficult to define (for a summary see e.g. Frensch, 2006) but usually means that the fact that something is learnt at all is not conscious. Implicit learning is the contrary to explicit learning. Explicit learning is done whenever learning is done on purpose. The second question is how the properties of the environment elicit this knowledge. In psychology there are different theories on how this is done. Two concepts which are useful in our context are the concept of affordances and the concept of cues. The term affordances was created by Gibson (1986) within his theory of direct perception (for a summary of Gibsons theory of affordances see e.g. Jones, 2003). According to Gibson, objects have properties which become affordances in relation to the properties of an individual (here: the driver). An affordance conveys a meaning to the observer in the sense of being … –able, for example being climbable. Similarly, road elements convey a meaning to the driver: the element is drive-able within a certain speed and attention range. This is what we call the “suggested” range of speed and attention. However, the direct approach to perception is not the only possible way how to explain this range of possible behaviours. They can as well be explained by behaviouristic theories that are to a large extent based on conditioned responses. Here, characteristics of the road or environment serve as discriminative stimuli. These discriminative stimuli give a hint to the driver which consequences to expect when showing the respective behaviour. Knowledge and anticipation of these consequences will then result in the respective behaviour being shown in the case of expected positive consequences or not shown in the case of negative consequences (avoidance behaviour). A road sign, for example, can almost be called the “archetype of the discriminative stimulus” (Fuller & Santos, 2002, p. 49). However, this does not mean that a certain behaviour is elicited automatically. There are some predispositions for that: first of all, the sign (or any other discriminative stimulus) has to be perceived, which might be impeded by different filters (see e.g. Rumar, 1985). Second, the wanted behaviour has to be associated with the respective discriminative stimulus which is not necessarily the case. The reason for a lack of association can be a lack of feedback or inconsistent or unreliable information conveyed by the stimulus. The third prerequisite is that the driver perceives his December 2007
27
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
behaviour to be under his own control, which is called self-efficacy. Taken all these prerequisites together, it is to be preferred when the whole situation serves as “integrated” discriminative stimulus (Fuller, 1984). In both cases (single elements or whole situation), the appropriate behaviour can be associated so closely to the stimulus that the stimulus literally prompts or triggers the behaviour. In this case, the stimulus is often termed cue and the terms stimulus - or bottom-up control of behaviour are used. In relation to attention, this kind of control is called automatic, exogenous control of behaviour, in contrast to intentional, endogenous control (Posner, 1980). Automatic, exogenous control of behaviour is faster, less resource consuming (lower workload) (Schneider, Dumais, & Shiffrin, 1984), often more reliable and gives less opportunities for individual differences. Thus, the guidance impact is higher and more reliable and it may be advantageous when fast and appropriate actions are required by almost all drivers at any time. While the mechanisms of exogenous control are valuable for traffic safety, they can lead to errors themselves (besides the ones already named above). This is the case when the wrong affordances or misleading cues are present in a situation which subsequently automatically leads to inappropriate behaviours. Exogenous behavioural guidance can be explained and supported by cues and affordances as well as by the appropriate use of perceptual invariants as described in the next chapter.
2.2.2 Part II: Perceptual invariants Finally, speed and path are regulated by perceptual invariants (Bruce et al., 1996). These perceptual invariants are Tau and Tau dot (Lee, 1976) and the deduced variables TTC (Lee, 1976) and TLC (Godthelp et al., 1984; Van Winsum & Godthelp, 1996). These perceptual invariants are used by the driver to remain within the boundaries of the “lane-tube” (Summala, 1996). A simplified way how drivers use Tau and Tau dot to regulate distance and speed is shown in the following Figure 10.
Figure 10: Detailed processes within part III of the driver behaviour model for rural roads: Safe distance keeping by using perceptual invariants proposed by Lee (1976) and Lee & Lishman (1977), tested e.g. by Yilmaz and Warren (1995). Adapted from Bruce et al. (1996).
December 2007
28
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
In contrast to the two aforementioned mechanisms, perceptual invariants can be calculated without having to know the properties of the individual driver. However useful this might be, perceptual invariants are only used for the short-term regulation of behaviour and therefore are not enough to explain the entity of the complex behaviour regulation on rural roads.
2.2.3
Part III: Expected and actual workload and risk
However, affordances or cues do not necessarily have to be present or they might not be known to the driver. In this case, the driver has to “guess” which behaviour is appropriate. This is done by comparing the expected level of workload and risk with the preferred workload and risk level. Which of these two parameters is actually used is topic of ongoing discussions. In literature evidence for both parameters is found (e.g. Fuller, 2005; Gerald J. S. Wilde, 1994; Gerald J. S. Wilde, 2001). In our studies we found a very strong correlation between rated demand and rated risk which makes it likely that drivers do not really distinguish between those two variables (see report 8.2 for details). Differences will be found however, when drivers are asked to rate the objective risk of an accident (Fuller, 2005). In our research we found that drivers in this case take into account their assumptions of how other drivers will behave, which will not influence their personal behaviour. Regardless of the discussion concerning the relevance of the respective parameters, it is much more important to understand how this process is done. Workload is the effect of situational demand on the driver, depending on the driver’s resources. At this point it is important to notice that situational demand in driving depends on the characteristics of the road and the speed with which this road is driven. This means that workload will differ in the same situation for the same driver when this driver is forced1 to drive through the situation with different speed. Further, workload depends on the capabilities or resources of the driver. These capabilities differ both between drivers but within drivers as well. They depend on the current motivation and the current state of the driver as well as on longer-lasting traits and organic variables like age or driving experience. When different drivers or the same driver at different occasions are forced to drive the same road with the same speed, they will experience different levels of workload. Usually however, driving is a self-paced task, which means that the driver can choose his preferred speed. This is done in order to assure medium levels of workload and risk. In reality workload and risk oscillate around this optimal level which is called homeostasis. This homeostatic regulation is done pro-actively based on expectations concerning the road ahead and re-actively as a result of feedback to the current workload and risk situation. In case of proactive regulation the driver generates expectations concerning workload and risk ahead of the current position. The entity of expectations concerning a situation forms a mental model (for a summary see Weller et al., 2006). To do so, the following variables are combined: - the perceived road ahead with - the information from the road just passed and - the individual knowledge of how situations usually develop.
1
The term “forced” is used here to indicate that the driver usually does not do this. However, it can be achieved in experimental sessions or even when the situation does not have enough degrees of freedom (e.g. due to other cars).
December 2007
29
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
These three input parameters can at the same time result in wrong assumptions. This is the case when: -
the road ahead is perceived as less demanding than it actually is, when the road ahead (static and dynamic situation) differs fundamentally from the road just passed (e.g. in the case of design inconsistencies) or the individual knowledge is inappropriate.
The following Figure 11 sums up the processes described above.
Figure 11: Detailed processes within part II of the driver behaviour model for rural roads: Expected and actual workload and risk.
2.2.4
Part IV: Feedback
Last not least, the selected behaviour itself (be it speed, path or attention) and its consequences, influence future behaviour through feedback. This actual behaviour changes the actual experienced workload and risk and might thus influence the future preferred level of workload and/or risk, as well as the expectations concerning future workload. The actual behaviour influences as well the perceived road ahead through a change in the perceptual invariants. Further, the experiences made with December 2007
30
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
the current behaviour serve as knowledge-base for future situations and might therefore influence directly the perception of the road ahead (e.g. former neutral stimuli become cues through experience, see above). Of course, inappropriate behaviour is enforced as well in case of missing or wrong (here: not negative) feedback. As is well-known from aggression research: if people know that they are doing something wrong but no consequence follows, they perceive this lacking feedback as positive reinforcement, strengthening the wrong behaviour.
December 2007
31
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
3. Empirical validation: Methodology Whether the model developed above is supported by empirical data, was tested in several steps. The results are reported in detail in the internal report 8.2 (Weller & Schlag, 2007) and are summarized in the following sections by using prototypical results.
3.1. Formulation of Hypotheses The empirical validation of a model requires the formulation of consistent hypotheses that can be falsified with empirical data. The model that has been developed above allows the formulation of such hypotheses and the subsequent testing with empirical data. The general research paradigm used is the comparison between two or more road elements that are similar concerning their geometry, but differ in their respective accident rate or other parameters that constitute the independent variable in the hypothesis. The dependant variables can be subjective ratings, as well as driver and driving behaviour data. The formulation of hypotheses follows the three main parts of the model as described above.
3.1.1 Part I: Affordances and cues Hypothesis 1: The presence, respective absence, of cues results in differences concerning both subjective ratings of demand and risk, as well as driving behaviour. The direction of change depends on the message the cue conveys. In the case of warning signs, the respective road element should be rated more dangerous and demanding, while at the same time driving behaviour should be less risky (indicated by e.g. lower speeds).
3.1.2 Part II: Perceptual invariants Perceptual invariants are used for the regulation of driving behaviour on the control level. The fact that drivers use perceptual invariants when driving, has been shown in several publications and is widely accepted (see report 8.1., Weller et al., 2006). Thus, theoretical validation as above may be sufficient. Therefore, no hypotheses concerning perceptual invariants were developed or will be tested in this project.
3.1.3 Part III: Expected and actual workload and risk In this part of the model several assumptions are made. These assumptions concern both the regulation of speed and the general level of safety and are as follows: - Hypothesis 2: If expected geometry or situation do not match actual geometry or situation, the situation is unsafe. - Hypothesis 3: If expected workload or risk is higher than preferred workload or risk, speed will be reduced. - Hypothesis 4: If actual workload or risk is higher than expected workload or risk the situation is unsafe. Using the research paradigm formulated above, workload or risk should be comparable before the curve (expected WL or risk) but different in the curve (actual WL or risk), whereas it should be higher in the high accident rate curve.
December 2007
32
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
3.2. Data sources for the testing of the hypotheses In order to test the above hypothesis, different data sources and different methods were used. These data sources and methods are summarized in the following matrix and are explained in more detail below. As we originally started our work with available data, the order of the data sources in the matrix does not follow chronological order but thematic considerations derived from the model (see last column in Table 1). Table 1: Matrix of the data sources used and the hypotheses tested.
Data source
Method:
Used for testing hypothesis No.:
A.)
Own simulator experiments
Hypothesis 1
B.)
Additional collection of data based on available data
Hypothesis 2 & 3
C.)
Reanalysis of available data
Hypothesis 4
D.)
Own driving experiments with an equipped vehicle
Hypothesis 4
The following sections briefly describe the data and the methods used.
3.2.1 Own simulator experiments (Data Source A) Part I of the model assumes that cues and affordances play a central role in influencing driver and driving behaviour on rural roads. In order to determine the influence of cues and affordances, we conducted experiments in the simulator of the Fraunhofer Institute for Transportation and Infrastructure Systems IVI in Dresden2 (see Figure 12). Details concerning the road and experimental details can be found in the corresponding chapter.
Figure 12: Simulator of the Fraunhofer IVI in Dresden that was used for the simulator experiments (www.ivi.fhg.de). Source: Fraunhofer IVI.
2
Further information concerning the simulator: http://www.ivi.fraunhofer.de/frames/english/projects/eng_fahrsimulator_strasse.htm
December 2007
33
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
The advantage of simulator studies is that they allow the systematic variation of independent variables in a strict experimental laboratory setting. Due to the nature of simulator studies, these independent variables can as well constitute dangerous situations that otherwise cannot be tested. These advantages are of course traded for a lower external validity when compared to driving experiments in the field, which we conducted as well (see Data Source D).
3.2.2 Additional collection of data related to available data (Data Source B) Prior to conducting own driving experiments, existing data was reanalysed with respect to the needs of RiPCORD-iSEREST. This data was collected at TU Dresden during a project funded by DFG3 . The project was a joint project of the Chair of Work- and Organizational Psychology (Prof. P. Richter) and the Chair of Road Planning (Prof. G. Weise). For this past project, 31 subjects drove 12 different stretches of two-lane rural roads in the German federal state of Saxony. The length of the road sections varied between two and seven kilometres, with the majority being around three kilometres in length. Some of the sections were driven again by the same subjects (although only 21 in comparison to the original 31 due to experimental dropout) after one year in order to ensure longitudinal stability. The roads were not altered during this one year period. All drives were recorded on video (front view). In addition to speed, psycho-physiological data was recorded during the drives (see Data Source C). The videos from this study, together with the road geometry and the accident data of these road sections, were used by the Chair of Traffic and Transportation Psychology at TUD to collect subjective ratings in a comparison of high and low accident rate curves.
3.2.3 Reanalysis of available data (Data Source C) The psycho-physiological data from the Richter et al. study described above (Richter, Wagner, Heger, & Weise, 1998; Richter, Weise, Wagner, & Heger, 1996) was reanalysed with respect to the needs of RiPCORD-iSEREST. The following psychophysiological data were recorded during the drives: -
electrocardial measures (ECG): heart rate (beat and beat-to-beat interval) electrooculogram (EOG): blink rate electrodermal measures (EDA): tonic skin conductance level (SCL) and single phasic reactions.
The data of this study was re-analysed to answer questions that were not part of the original data analysis. This reanalysis is described in detail in the respective chapter and report 8.2. (Weller & Schlag, 2007).
3
DFG (Deutsche Forschungsgemeinschaft). Project name: „Fahrverhalten und psychophysiologische Aktivierung von Kraftfahrern als Bewertungskriterien der Gestaltungsgüte von Straßenverkehrsanlagen“ („Driving behaviour and psychophysiological activation of car drivers as assessment criteria for road design“). Deutsche Forschungsgemeinschaft DFG Project Number Ri 671/2-1. Richter et al. (1996); Richter et al. (1998), (Heger & Weise, 1996), (Wagner, 2000).
December 2007
34
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
3.2.4 Own driving experiments with an equipped vehicle (Data Source D) Driving experiments constitute the “via regia” when knowledge concerning real driving behaviour is to be increased. At the same time, driving experiments are a complex, demanding, and costly method. Nevertheless we decided to conduct additional own driving experiments with an equipped vehicle. The following reasons affected this decision: the shortage of appropriate, available existing data (see as well above), the fact that the psycho-physiological data (data source C, see above) turned out to be not as sensitive to our units of analysis (single road elements in contrast to whole road stretches) as expected, the nature of our hypotheses which are related to safety, operationalized as (the absence of) accidents. In order to test Hypothesis 2 which is especially related to safety (see above), additional, new driving experiments were conducted at TUD with an equipped vehicle of the Chair of Road Design at TUD (see Figure 13).
Figure 13: Experimental vehicle of the Chair of Road Design at TUD.
Besides parameters of driving behaviour like speed, driver behaviour was recorded with the help of an integrated, contact-free eye tracker (Smart Eye). The test route consisted of around 2 * 40km of rural roads in the German Federal State of Saxony. Details concerning the test route, the sample and the parameters collected are described in detail below.
December 2007
35
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
4. Empirical validation: Results 4.1. Hypothesis 1: Affordances and cues (Data Source A). 4.1.1 Introduction Part I of the model assumes that cues and affordances play a central role in influencing driver and driving behaviour on rural roads. Determining the validity of this assumption and the extent to which it is valid was the aim of the experiments conducted in the simulator described above. While the entity of the environmental situation sure plays a major role, it is difficult to systematically assess the differentiated effect of single environmental characteristics, let alone their interaction. Although cues are ideally used to convey the right message, they might as well tell a wrong story and thus contribute to the underestimation of demand or subjective risk. The experimental rationale used in this study was to analyze different road sections that were equal concerning their geometry and the geometry of the preceding road section, but differed in their environment. To allow the definite referencing of effects to single independent variables, simulator studies were chosen this time. This additional simulator study constitutes a strict experimental setting and thus allows the control of otherwise unsystematic variance. As is true for the comparison between on-road tests and simulator studies in general, the first approach has higher external and the second one higher internal validity. Thus, combining both might be ideal – as we performed it in RiPCORD-iSEREST.
4.1.2 Method The study was carried out in the driving simulator of the Fraunhofer Institute for Transportation and Infrastructure Systems IVI in Dresden4 (see Figure 12). The sample consisted of 50 participants who were recruited to approximately equal shares from IVI or TUD staff and from an advertisement campaign. All subjects had a valid driving licence. Age varied between 19 and 63 with an average of 37 (SD: 12). Further information concerning the sample characteristics can be found in the internal report 8.2 (Weller & Schlag, 2007) and in Voigt (2007). All curves which are reported here were programmed in the simulator by IVI with a radius of 200 meters and corresponding transition curves before and after the curve. The transition curves had a length of 25 meters each. The total length of the curve was approximately 130 meters. The length of the whole road section was approximately nine kilometres. As the course was driven in both directions, the total length approximated 18 kilometres. Between the two directions was another road section of approximately 15 kilometres which served other purposes and is not reported here. The following Figure shows a bird’s eye view of the simulated road.
4
For further information please visit: www.ivi.fhg.de
December 2007
36
TUD
Deliverable D8
Dissemination Level (PU)
Part 1
Part 2
Part 3
Part 4
Contract N. 506184
Figure 14: Bird’s eye view of the simulated road sections in the Fraunhofer IVI simulator.
The three longer sections are diversified environment, alleyway and monotonous environment. Each of these longer straight sections is around 700 meters long. Between tiles 3 and 7 there is an uphill slope, reducing the sight distance to the following curve and between tiles 11 and 15 there is a downhill slope. Due to technical reasons the two curves at the end and some of the curves in tile 7 and 11 are of different radius than the curves described above. The results concerning these curves will not be reported here. The following environmental variations were tested in this study: warning signs (‘yes’ versus ‘no’) in curves road markings (‘continuous’ versus ‘intermittent line’ versus ‘transverse markings’) in curves environmental influence on straight road sections (‘monotonous’ versus ‘varying’ versus ‘tree-lined alleyway’) ‘uphill’ versus ‘downhill’ slope (‘uphill’ representing high restrictions to foresight while ‘downhill’ represents perfect foresight to the following curve). The effect of the various environmental conditions was assessed by calculating different speed parameters. Speed was analysed for both directions in order to give an estimation of the stability of the data or the possible influence of sequential effects. After the participants finished the simulator drive, they were asked to rate selected road sections from the course driven before. To this aim, video recordings of these stretches from the simulator were shown to the subjects on a computer screen, in an office adjacent to the simulator. The questions were presented on a computerbased questionnaire with five or four point rating scales depending on the question. In the deliverable D8 at hand, we decided to select one of the conditions named above in detail, in order to show a prototypical analysis and interpretation of results. To do so, we selected the example of the warning signs as signs can be seen a good example of a formal cue which is often used in road design, although its application is December 2007
37
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
associated with some disadvantages compared to SER-design. Readers who are interested in the detailed analysis of the other results are again referred to the internal report 8.2. (Weller & Schlag, 2007) whereas all results are discussed at the end of this chapter together with the in-depth analysis of the selected example.
4.1.3 Selected example: warning signs as formal cues As stated above, the following section uses warning signs as prototypical example of the analysis performed with the data of the simulator study. Readers interested in detailed results for the other conditions are referred to internal report 8.2. (Weller & Schlag, 2007). 4.1.3.1.
Method and descriptive analysis
In order to test the influence of “formal” cues on speed in curves, a curve-ahead-sign was placed before one of the curves (K3) together with several red and white chevrons in the curve. The average speed is shown in the following Figure 15, the results of the statistical analysis are shown in Table 2.
Figure 15: Speed [km/h] averaged across all subjects for a curve with additional signs (bend ahead and guidance signs in the curve; curve K3) and the comparison curve (K19) in the toand the backwards direction (=R). All curves are left curves; driving direction from left to right side. On the x-axis the distance from curve beginning is shown. The vertical bars mark the beginning (left bar) and end (right bar) of the curve.
December 2007
38
TUD
Deliverable D8
Dissemination Level (PU)
4.1.3.2.
Contract N. 506184
Results (selected example)
The following Table 2 shows the results of the t-test for paired samples between the experimental curve with warning signs and the comparison curve. Table 2: Simulator experiments: results of the t-Tests for paired samples for different speed parameters for K3 (signs) and respective comparison curve (K19) in both directions.
MAX speed 150m before the curve to the beginning of the curve K19-K3 K19R-K3R MAX speed 75m before the curve to the beginning of the curve K19-K3 K19R-K3R MAX speed 50m before the curve to the beginning of the curve K19-K3 K19R-K3R MAX speed from beginning to end of curve K19-K3 K19R-K3R MAX speed from curve apex to curve end K19-K3 K19R-K3R MIN speed from beginning to end of curve K19-K3 K19R-K3R
AVG diff 14.60 7.94
SD diff 9.12 9.14
T 10.735 5.764
df 44 43
sig. .000 .000
AVG diff 17.99 13.46
SD diff 11.91 11.44
T 10.131 7.801
df 44 43
sig. .000 .000
AVG diff 18.03 13.71
SD diff 11.89 11.36
T 10.171 8.008
df 44 43
sig. .000 .000
AVG diff 13.18 3.82
SD diff 9.47 9.28
T 9.336 2.727
df 44 43
sig. .000 .009
AVG diff 12.00 3.29
SD diff 9.20 9.26
T 8.752 2.357
df 44 43
sig. .000 .023
AVG diff 16.86 7.87
SD diff 10.08 13.80
T 11.219 3.781
df 44 43
sig. .000 .000
(R= backwards direction)
As could be expected from Figure 15 speeds in the curves with signs are lower throughout the whole curve. However, in Figure 15, it can be seen as well that speed is lower already before the 150 meters that were used in the analysis. Probably this is a result of the preceding road element (differences even before speed max was achieved). But while speed was heightened on the comparison curves until approximately 100m before the curve, the maximum speed was reached much sooner on the signed curves. This leads to an increase in the speed differences that this time cannot be attributed to influences of the road just passed. In order to statistically analyse this effect, we calculated the position of maximum speed before the curve (between 200m before the curve and the beginning of the curve) and of minimum speed in the curve (between the beginning of the curve and the end of the curve). The statistical analysis revealed that the subjects did indeed reduce speed much further away from the curve with sign than from the reference curve. The subjects reached their minimum speed much sooner in the curve with signs then in the comparison curve. The results are shown in the following table.
December 2007
39
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Table 3: Simulator experiments: results of the t-Tests for paired samples for the distances of the maximum speed before the curve (200m to beginning of curve) and the minimum speed after the beginning of the curve. Curves K3 and K19.
distance vmax before beginning of the curve curve no AVG SD T K3 153.85 37.12 K19 109.58 51.18 5.017 K3R 134.41 52.49 K19R 97.99 62.27 3.626 distance vmin after beginning of the curve curve no AVG SD T K3 8.39 14.00 K19 37.03 38.33 -4.711 K3R 25.00 41.51 K19R 37.87 35.32 -1.749
df
Sig.
42
.000
42
.001
df
Sig.
42
.000
42
.088
(R= backwards direction)
From a safety point of view, these results have to be interpreted in favour of the signs. The high variation of the values indicates however, that signs do not have the same prompting character for all drivers. Further, the subjective ratings were tested for differences between the curve with warning signs and a comparison curve. As the ratings are not distributed normally, the nonparametric Wilcoxon-Test for paired samples is used. The results are presented in Table 4. Table 4: Results of the Wilcoxon-Test for paired samples for the subjective ratings for curves.
The curve …
comparison curve versus ... warning signs
is requires gives good dangerous reduced information speed concerning further curve path Z -2.979 -3.479 -1.708 Sig. .002 .000 .098 direction of diff. C
mis- seems is deleads narrow mandto ing speeding -3.590 -2.215 -3.617 .000 .031 .000 C>E C
C = comparison curve; E = experimental curve.
4.1.3.3.
Discussion (selected example)
In combination with the speed data the results presented in Table 4 can be interpreted as follows: The curve with warning signs was rated as being more dangerous, more demanding, and requiring more speed reduction than the comparison curve. These subjective ratings are mirrored by the objective data (regardless of direction). The fact that the subjective ratings and speed behaviour are closely related indicates that the message conveyed by the signs has been successfully learnt by the subjects. This knowledge is successfully transferred to behaviour, so that the signs finally act as December 2007
40
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
cues that prompt the behaviour (appropriate behaviour in our case). Of course this at the same time shows the problems intertwined with using signs as cues: in case they are not perceived, no behavioural adaptation results.
4.1.4 Discussion of Results The results, both concerning the warning signs reported above, as well as the additional results reported in the internal report 8.2. (Weller & Schlag, 2007), clearly show that there is a considerable impact of environmental factors, other than road geometry, in influencing driving behaviour. This was demonstrated with the significant speed differences between different curves, as well as between different straight road sections that differed in one parameter but were comparable concerning their geometry. The mechanism describing how these environmental characteristics affect behaviour depends on the nature of these characteristic. Evidence was found for the effectiveness of cues and affordances. For example, driving on alleyways results in a specific behavioural response (significantly higher speed in comparison to road stretch with diversified environment), while at the same time the driver is not aware of these mechanisms, which is shown by the dissociation of behaviour and subjective ratings. Here we find an unconscious but nevertheless strong adaptation in the wrong direction: the environment gives the wrong cues. Other road elements showed significant effects on behaviour but at the same time exhibited corresponding effects in the subjective ratings concerning demand and risk. Here, this road element still acted as cue but the behavioural response was evoked indirectly via the subjective risk associated with this cue. This was the case with road signs. Other elements that were expected to act as affordance or cue, like the variation of sight distance, in our study were not effective concerning behaviour. As these elements are obviously neither related to a certain behavioural response, nor are consciously perceived as requiring a certain behavioural response, they constitute a challenge for traffic safety. Following Gibson (1986, see above), the affordance (behavioural impact or advice) associated with these elements is not clear enough (but presumably could be supported). Concerning the Hypothesis 1, it can be concluded that affordances and cues play an important part in explaining and (in practice) guiding driving behaviour, although their effect might be mediated by conscious processes that make use of stored knowledge.
4.2. Hypotheses 2 and 3: Expectations (Data Source B). 4.2.1 Introduction In order to test the Hypotheses 2 and 3, subjective ratings were collected in an experimental setting in addition to the existing data from the Richter et al. study described above (see Data Source B, chapter 3.2.2 and Data Source C, chapter 3.2.3). The rationale behind our study is explained in the method chapter below. However, the results obtained from these data sources were manifold and presenting them here in their entirety would exceed the scope of this deliverable. Therefore, we decided to briefly summarize the results here, whereas readers interested in the details are asked to consult internal report 8.2 (Weller & Schlag, 2007) where all results are presented. December 2007
41
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
4.2.2 Method The experimental paradigm used was a comparison between two curves that markedly differed in their respective accident rate but were comparable concerning their geometry, both in the curve and within the approaching zone before the curve. First of all, from the roads that were driven in the experiments (Richter et al., 1998; Richter et al., 1996), curves were selected that showed remarkable high accident numbers for a given direction. In a first step, this selection was done by visual impression only, by using maps with symbols representing accidents. The symbols used on these maps were the German symbols used for accident diagrams (see e.g. FGSV, 1998). Accident data in this thesis comprised all accidents from 1993 to 1996. Following this pre-selection of curves, the geometric characteristics of these curves together with the preceding road section were collected. These characteristics included radius, curve length, AADT on the road section, road width, longitudinal gradients and CCR. Further the weighted average curvature of the preceding road section (approximately 400m) were calculated according to (Sossoumihen, 2001). These criteria were used to find corresponding curves within the experimental roads that were characterised by fewer accidents. Before the final decision was made for a pair of curves, the accident rates were calculated. This resulted in eight pairs of curves. In a second step, the video recordings of these curves were retrieved from the videos that were made during the experiment. From these video segments screenshots were taken approximately every 50m, starting from 200m before the curve. The resulting four pictures were presented to the subjects in chronological order for each of the curves under investigation on a computer monitor (19’’). The order of presentation of the curves varied between subjects. The following Figure 16 shows the pictures of pair 2 as an example of the pictures used.
Figure 16: Pictures from the low (left) and the high (right) accident rate curve in pair No. 2 as used for the collection of the subjective ratings.
The subjects were asked to fill out a questionnaire after having seen the four pictures of each curve. The questionnaire consisted of items that were taken from the Road Environment Construct List (RECL) (Steyvers, 1993, 1998) and additional curvespecific ratings. RECL items were translated and some of them adjusted slightly, once subjects reported problems in tests before the actual experiments. All subjective ratings were collected with a four point Likert Scale, ranging from “I totally agree” to December 2007
42
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
“I do not agree at all”. Further details are found in Petermann, Weller, Schlag (2007) and Petermann (2006).
4.2.3 Summary of results In the chapter at hand, prototypical results are summarized. Readers interested in the entirety of results are referred to internal report 8.2. (Weller & Schlag, 2007). In order to test whether expected geometry or situation can account for differences in accident rate (Hypothesis 2), the ratings of the following items were compared between the high and low accident rate curves: “The curve is sharp” (as proxy-variable for expected geometry); “The curve gives good information concerning the following curve path” (as proxy-variable for expected geometry, too). As the ratings were not distributed normally, the nonparametric Wilcoxon-Test for paired samples is used. The results are shown in Table 5 and Table 6. Table 5: Results of the Wilcoxon-Test for paired samples for the item „The curve is sharp“.
Z Sig. direction of difference
Pair 1 -3.760 .000 h
Pair 2 Pair 3 Pair 4 Pair 5 Pair 6 Pair 7 Pair 8 -5.064 -4.105 -0.689 -1.980 -1.930 -4.273 -4.719 .000 .000 .488 .057 .058 .000 .000 h l) (h < l) h < l h
h = high accident rate curve; l = low accident rate curve. Answers: “does not apply” = 1 to “does apply” = 4. Table 6: Results of the Wilcoxon-Test for paired samples for the item „The curve gives good information concerning the following curve path“.
Z Sig. direction of difference
Pair 1 Pair 2 Pair 3 Pair 4 Pair 5 Pair 6 Pair 7 Pair 8 -3.396 -5.432 -3.245 -2.939 -0.985 -1.819 -6.082 -4.446 .001 .000 .001 .003 .356 .077 .000 .000 h>l h>l h>l hl h>l
h = high accident rate curve; l = low accident rate curve. Answers: “does not apply” = 1 to “does apply” = 4.
In short, high accident rate curves are for the majority of cases (six and five respectively out of eight) rated as being less sharp and as giving more information concerning the following curve path, than low accident rate curves. Here it is especially important to remind the reader that radius, CCR and curve length of the curves were comparable. Therefore, it can be stated that expected geometry could in fact be an important determinant in explaining accidents. Hypothesis 3 assumes that speed is regulated by comparing expected and preferred workload or risk. A prerequisite to test this assumption with our experimental paradigm is that preferred workload and risk remain constant for each subject during the time of the experiment (both in the laboratory as well as for the field study in Richter et al.). With this assumption, Hypothesis 3 is supported if: the two curves within each pair differ concerning expected workload and risk, and additionally, if these curves differ concerning speed (lower speeds in case of higher expected demand or risk). To test Hypothesis 3 the two curves do not necessarily have to differ concerning their respective accident rate. However, the fact that they do in our case shows the relevance for road safety, given that Hypothesis 3 is supported by the following results. To test the assumption in Hypothesis 3, demand and subjective risk of the December 2007
43
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
respective curves were rated by the subjects together with ratings concerning the subjective appropriate speed. Further, the actual speed collected by Richter et al. was reanalysed for the curves used in our experiments. The results of the Wilcoxon-Test for the subjective ratings concerning demand and subjective risk are presented in the following Table 7 and Table 8. Table 7: Results of the Wilcoxon-Test for paired samples for the item „The road stretch is demanding “.
Z Sig. direction of difference
Pair 1 Pair 2 Pair 3 Pair 4 Pair 5 Pair 6 Pair 7 Pair 8 -3.853 -5.545 -3.114 -2.372 -3.075 -2.373 -2.858 -4.768 .000 .000 .002 .017 .002 .021 .004 .000 hl h>l h
h = high accident rate curve; l = low accident rate curve. Answers: “does not apply” = 1 to “does apply” = 4.
The road stretches with the high accident rate curves are rated as being less demanding than the road stretches with the low accident rate. This is valid for six out of the eight pairs used in our analysis. While the former item was collected for the whole road stretch (200m before the curve to beginning of curve), the subjects were asked to solely rate the curves themselves for the next item. Table 8: Results of the Wilcoxon-Test for paired samples for the item „The curve is dangerous“.
Z Sig. direction of difference
Pair 1 Pair 2 Pair 3 Pair 4 Pair 5 Pair 6 Pair 7 Pair 8 -3.919 -5.803 -1.658 -1.836 -2.615 -2.256 -3.475 -5.103 .000 .000 .127 .071 .010 .032 .000 .000 hl h>l h
h = high accident rate curve; l = low accident rate curve. Answers: “does not apply” = 1 to “does apply” = 4.
For the majority of cases (five out of eight) the high accident rate curves were rated as less dangerous than the low accident rate curves. Exceptions are the ratings for pairs 3, 4 and 5. This result gives a first and tentative hint in the direction hypothesised by our model: Demand (Table 7) and perceived risk (Table 8) both are underestimated in those high accident rate curves. In order to find out whether these results are relevant for behaviour, further analyses were performed, both of items related to subjective appropriate speed as well as of the objective speed data collected in the Richter et al. study. The results, which are reported in detail in internal report 8.2 (Weller & Schlag, 2007), indeed support Hypothesis 3. In general, speed was higher for the curves which were rated less demanding and risky than the comparison curve. This holds true both for the subjective ratings as well as for the speed data collected in the field.
4.2.4 Discussion of results The results described above (and in detail in internal report 8.2, Weller & Schlag, 2007) clearly support the assumptions both of Hypothesis 2 and of Hypothesis 3. In short, we found considerable evidence that misleading perception as evident in an underestimation of demand and (subjective) risk, leads to maladaptive behaviour, which in turn results in higher accident probabilities. This result is promising as it helps to understand accident occurrence on rural roads. A practical consequence is the recommendation to give appropriate hints in advance (proactive) and to December 2007
44
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
strengthen not only objective safety, but also subjective feelings of unsafety as most relevant for a cautious behaviour. The detailed results together with an in depth discussion of their implication can be found in internal report 8.2 (Weller & Schlag, 2007).
4.3. Hypothesis 4: Workload: Psycho-physiology (Data Source C) 4.3.1 Introduction The model proposed in Figure 9 essentially consists of three factors: expected workload and risk, regulation of driving behaviour by perceptual invariants and regulation of behaviour by cues and affordances. One assumption of the workload and risk part of the driver behaviour model (see Figure 11) is that road elements are unsafe if actual workload or risk is higher than expected workload and risk.
4.3.2 Method To test this assumption, data was used that was collected by Richter et al. (1996, 1998). Our main interest was in the relation of psycho-physiology with actual accidents at single road elements. Therefore, the research paradigm used was again the comparison between low and high accident rate road sections that were comparable concerning their geometry and the geometry of the preceding section. While in the previous chapter subjective ratings were used to assess expected demand, this time psycho-physiological data was used to assess actual workload. Thus, the analysis was tailored to test Hypothesis 4. Accident data from a three year period (1993 to 19965) related to the road geometry for all driven roads were taken from a diploma thesis that was written at TUD (Tscheschlok, 1998). In order to take into account the different stages when negotiating curves (Donges, 1978), data was divided in sections corresponding to: approaching the curve (from 150m before - to curve beginning), negotiating the curve (from beginning to end of the curve), and leaving the curve (from end of the curve to 150m after the curve6 . Using this division, it can be deduced from Hypothesis 4 that workload (or psychophysiological activation) should be similar at the approach section (or even higher for the low accident rate curve), higher in the curve for the high accident rate curve and remaining higher after the curve for the high accident rate curve. As the curves were selected to be equal concerning geometry, they should have the same difficulty, resulting in the same demand when driven with the same speed. For our own analysis, four7 pairs of curves were compared that differed in their accident rate but were comparable concerning their geometry (s. Annex, Table A 1). Details concerning the curves and the methodology can further be found in Wendsche, Uhmann & Meier (2006) and in the internal report 8.2. (Weller & Schlag, 2007).
5
The driving experiments themselves took place in 1995 and 1996.
6
As following curves sometimes interfered, the length for this curve and the corresponding (low or high accident rate) curve had to be shortened. 7
Originaly five pairs were selected, hence the highest pair number is five. Original enumeration was kept in order to ease comparison with the original work.
December 2007
45
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
4.3.3 Results We first referenced psycho-physiological data to the geometrical alignment of the roads by using a software written at TUD (Weiße, 2006). Although the duration of the analyzed sections might differ between subjects due to speed differences, we preferred this solution as it could thus be assured that the same geometric elements were analyzed. As most variables are further analyzed in relation to a predefined time period (e.g. beats per minute) data of different duration can still be compared. Table 9 shows the results of single t-Tests for independent samples that were conducted between high and low accident rate curves for blink rate, heart rate and speed, differentiating between the curve sections before, in and after the curve. In order to conduct this analysis, data was averaged for all curves, resulting in an independent data structure despite the subjects being the same within pairs. In order to conduct this analysis, we decided to tolerate this inconsistency. For the analysis here, the Kolmogorov –Smirnov Test showed further that blink rate differed significantly from a normal distribution. Therefore, for blink rate the parameters of the Mann-Whitney test are additionally shown in brackets. Table 9: Statistical differences of physiological data and speed between high and low accident rate curves.
variable and curve section BR before BR in BR after HR before HR in HR after V before V in V after
accident rate high low high low high low high low high low high low high low high low high low
N 50 50 49 50 50 50 53 53 53 53 53 53 53 53 53 53 53 53
AVG 14.01 15.50 12.94 11.01 18.12 14.97 81.82 82.43 82.81 82.96 81.80 82.38 78.12 73.23 65.22 60.27 70.37 64.61
SD 12.29 14.58 18.27 16.73 22.90 21.30 14.75 14.13 15.05 14.58 13.37 13.65 8.19 8.27 7.22 7.10 8.57 10.37
T (for BR: Mann Whitney -U)
df
Sig.
-0.553 (1248)
98
.581 (.993)
0.548 (1174)
97
.585 (.697)
0.713 (1135)
98
.478 (.401)
-0.218 104
.828
-0.053 104
.958
-0.222 104
.825
3.057 104
.003**
3.558 104
.001**
3.117 104
.002**
*: p < .05; **: p < .01; BR: blink rate; HR: heart rate; V: speed.
Table 9 shows that high and low accident rate curves do not differ significantly concerning the physiological data blink rate and heart rate. They do however differ significantly concerning speed: curves with higher accident rates are characterised by higher speeds throughout the whole curve (before, in and after the curve).
4.3.4 Discussion According to Hypothesis 4, we expected higher workload in and after the curve, in the high accident rate curves (see above). Our results did not support these assumptions as there were no significant differences at all in the psycho-physiological data December 2007
46
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
between both curve types. Contrary to workload, higher speeds were indeed found for the high accident rate curves. The integration of the speed and workload results allows for two explanations: either the assumption concerning workload formulated in Hypothesis 4 is wrong, which means that low and high accident rate curves cannot be distinguished by workload, or the prerequisites of the comparison between high and low accident rate curves were violated. Although at present it cannot be ruled out that the first explanation is valid, there is in fact some evidence for the validity of the second explanation: the (minor) differences in geometry between the high and low accident rate curves (see Annex Table A 1), might constitute systematic differences in demand. This would have violated the prerequisite according to which both curves should be equal in demand when driven with the same speed. As this does not constitute a satisfying answer, further effort was invested to test the assumptions formulated in Hypothesis 4. These additional driving experiments will be reported in the following.
4.4. Hypothesis 4: Workload: Reaction times (Data Source D) 4.4.1 Introduction In the preceding section, a first test of Hypothesis 4, concerning the ratio between expected and actual workload and risk and its relation to accidents, was carried out with the help of existing psycho-physiological data that was reanalysed. While the results did not support the assumptions formulated in Hypothesis 4, restrictions concerning the applicability of the existing data to the new assumptions could have accounted for this weakness as was described above. In order to overcome this weakness, we collected new data during own, additional driving experiments with an equipped vehicle. However, this time reaction times collected with the help of a peripheral detection task (PDT, see below) served as parameter for workload (for a description of the rational behind this method, see report 8.1, Weller et al., 2006). Additionally, gaze data was recorded that allows further analysis.
4.4.2 Method The equipped vehicle of the Chair of Road Designs at TUD (see Figure 13 for pictures) together with the program “RoadView” (Dietze, 2007) allowed the collection and analysis of the following data that was used for our analysis: Road geometry (radius, length, coordinates, uphill grade, etc., for a more detailed description, see report 10.1. (Dietze et al., 2005)); Accident data (see below); Driving behaviour data: speed and acceleration; Driver behaviour data: o Smart Eye gaze data (see below), o Reaction time from the peripheral detection task (PDT) (see below). All data was recorded with time stamps which were synchronized and referenced to the location that was available through the high precision GPS based system APPLANIX. The peripheral detection task (PDT) required a manual reaction to signals shown on a touch screen monitor (19’’). The screen was installed near the dash board to the right of the driver and was adjusted for each subject to allow easy reach (see Figure December 2007
47
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
13 for a picture). The time between onset of the signal and reaction was measured as reaction time and served as proxy variable for workload (see report 8.1. Weller et al., 2006). The PDT had to be performed by the subjects at selected locations. Start and stop instructions were given to the subjects by the instructor, who was seated on the back seat of the car. Gaze coordinates were recorded with the Smart Eye System8 which was already integrated in the measuring vehicle at the time of our test drives. For our purpose, fixation durations are the gaze parameter of choice as this parameter is usually used as proxy variable for workload (see report 8.1. Weller et al., 2006). As the PDT necessarily changes natural gaze behaviour, the fixation durations analyzed during PDT use do not represent natural values. However, differences between two or more locations still allow comparative analysis which serves our purposes. Fixations were detected following an algorithm developed by Jacob (1995) which was adjusted to Smart Eye and the requirements of the test situation (see, Schulz & Lippold, 2006). Concerning the selection of the test route, the safety related hypothesis required the presence of several locations with a “reasonable” number of driving accidents and, at the same time, road sections with similar geometry which could be used as comparative sections. This resulted in a test route with a length of approximately 40 km situated 20km north of Dresden. This test route was driven in both directions, thus resulting in a total of 80km. Further details concerning the selection process and the test route can be found in the internal report 8.2 (Weller & Schlag, 2007). Within this test route, two high accident rate locations were present which could be used together with their respective comparison curves for statistical data analysis. Similar to the simulator study, we decided to report only one prototypical analysis which is shown below. Readers who are interested in the in-depth analysis of the second location are referred to the internal report 8.2. (Weller & Schlag, 2007). The driver-sample consisted of 15 subjects aged between 25 and 47 years (avg. 30) who were all employees of TUD. The participants were recruited via mailing-lists of the institutes. All subjects had a valid driver licence and wore no glasses. The average annual mileage driven across the last three years was 15000 km/year (Min.: 5000 to max.: 40000; SD 10000). In order to allow detailed statements concerning the development of a situation we decided not to use the average for whole sections, but for 25 meter subsections within each section. The statistical analysis was then conducted for these subsections.
4.4.3 Selected Analysis As already stated, the following section shows the prototypical analysis of one of the two high accident rate locations, which were analysed in detail in the internal report 8.2. (Weller & Schlag, 2007). 4.4.3.1.
Description of the selected locations
To test the assumption formulated in Hypothesis 4 (see above) the two curves depicted in the following Figure 17 were selected: one with a high amount of accidents (three driving accidents with major personal damage; one fatal driving accident and two driving accidents with property damage only) and the other with no (registered) accidents at all. As both curves are on the same road section, annual average daily traffic is the same on both curves. The road section in its entity is 8
Company and system information at: http://www.smarteye.se
December 2007
48
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
limited by a cross road with traffic light in a village on one side and a village on the other side. Both curves were driven in both directions. Subjects performed the PDT on both to directions (section 21 and 24) and the backwards direction for section 24. The Radius for the low accident curve is 135m (length = 49m), the radius for the high accident curve is 141m (length = 53m).
Figure 17: Pictures of the low accident rate road sections 21 (left side) and the high accident rate road section 24 (right side) in the to-direction (pictures taken from RoadView TUD).
None of the curves has transition curves. There was a speed camera near the high accident rate curve in the backwards direction (discussed below). The whole situation with both curves is depicted in the following Figure 18.
Figure 18: Curvature plan of sections 21 (low accident rate curve) and 23 (high accident rate curve) within the whole road stretch. From right to left: driving direction in to-direction; from left to right: driving direction in backwards-direction.
Given that the assumptions made by the model are valid, we would assume that the high accident rate curve is characterized by
December 2007
49
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
either higher reaction times (i.e. higher workload), or higher speed, or both in comparison to the low accident rate curve. The explanation according to the model would be as follows: Higher reaction times at the high accident rate curve, with no speed differences between high and low accident rate curve, occur when the high accident rate curve requires even more speed reduction than the low accident rate curve, independent of geometry. In this case, expectations concerning geometry and situation for the high accident rate curve were wrong. Higher speeds at the high accident rate curve, with no reaction time differences between high and low accident rate curve, could be the result of missing feedback in the high accident rate curve: the driver does not reduce speed appropriately but nevertheless does not get enough feedback that would lead to higher workload. Another explanation concerns speed in the low accident rate curve: this speed might be reduced more than would have actually been necessary from a safety point of view, either due to expectations or due to cues. In this case speed would be lower while at the same time workload would not be affected as in both curves sufficient resources would be available. Finally, workload and speed might be higher in the high accident rate curve. This is the case when speed reduction did not occur due to missing cues or wrong expectations. This inappropriate speed will then lead to less resources being available as these resources are needed to keep the perceptual invariants on a constant level, and subsequent higher reaction times. The first comparison is between the low (section 21) and the high accident rate curve (section 24). As no PDT was performed in the backwards direction, the to-direction is used for the statistical analysis. As already stated above, each section was divided into subsections of a length of 25m each. -
4.4.3.2.
Results for the selected location
The following figures (Figure 19 and Figure 20, see next page) show the values for the respective sections, including speed, fixation durations and reaction time (measured and interpolated). At first sight, both road sections show similar patterns in the analysed variables. In case this similarity is supported by statistical analysis, the assumption that e.g. high accident rate curves are characterized by a lack of anticipating behaviour, would have to be rebutted. Before applying the t-tests, all values within the subsections were tested for normal distribution, using the Kolmogorov-Smirnov-Test. As this test showed no significant deviations from the assumption of normally distributed values, the t-test for paired samples could be used.
December 2007
50
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Figure 19: Averaged values for speed, reaction time (measured and interpolated) and fixation duration for the low accident rate curve (section 21).
Figure 20: Averaged values for speed, reaction time (measured and interpolated) and fixation duration for the high accident rate curve (section 24).
The following tables show the results for speed, reaction time and fixation duration within corresponding subsections. The reference point for matching the 25m subsections within pairs was the respective beginning of the curve.
December 2007
51
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Table 10: Average fixation duration [s]; results of the t-Test for paired samples; low versus high accident rate curve (section 21 versus 24).
Distance to curve begin [m] 75 50 25 0 -25 -50 -75
AVG (low) (SD)
AVG (high) (SD)
Diff (SD)
T
df
Sig.
0.21 (0.12) 0.26 (0.12) 0.28 (0.09) 0.32 (0.17) 0.26 (0.15) 0.35 (0.28) 0.33 (0.03)
0.16 (0.07) 0.24 (0.10) 0.24 (0.12) 0.21 (0.05) 0.27 (0.18) 0.28 (0.15) 0.21 (0.11)
0.05 (0.17) 0.02 (0.17) 0.04 (0.12) 0.10 (0.14) -0.01 (0.20) 0.07 (0.36) 0.12 (0.08)
0.67 0.27 0.93 1.95 -0.10 0.40 2.65
4 4 5 6 6 3 2
.538 .803 .395 .099 .920 .719 .118
25m subsections; *: p < .05; **: p < .01. Table 11: Average speed [km/h]; results of the t-Test for paired samples; low versus high accident rate curve (section 21 versus 24).
Distance to curve begin [m] 125 100 75 50 25 0 -25 -50 -75
AVG (low) (SD)
AVG (high) (SD)
Diff (SD)
T
df
Sig.
81.38 (8.67) 80.49 (8.96) 79.24 (8.96) 76.99 (8.68) 73.69 (8.74) 70.75 (8.82) 70.59 (8.67) 72.21 (8.37) 73.94 (8.34)
77.74 (7.64) 76.46 (7.68) 74.75 (7.43) 72.58 (6.92) 69.59 (6.94) 67.57 (6.72) 67.24 (6.12) 68.05 (5.49) 68.81 (5.18)
3.64 (5.40) 4.04 (5.89) 4.49 (6.00) 4.41 (5.95) 4.10 (6.46) 3.17 (6.51) 3.35 (5.95) 4.16 (5.38) 5.13 (5.26)
2.13 2.17 2.37 2.35 2.01 1.54 1.78 2.45 3.08
9 9 9 9 9 9 9 9 9
.062 .058 .042* .044* .076 .158 .109 .037 .013*
25m subsections; *: p < .05; **: p < .01. Table 12: Average reaction times [s] interpolated values; results of the t-Test for paired samples; low versus high accident rate curve (section 21 versus 24).
Distance to curve begin [m] 125 100 75 50 25 0 -25 -50 -75
AVG (low) (SD)
AVG (high) (SD)
Diff (SD)
T
df
Sig.
0.78 (0.24) 0.86 (0.25) 0.89 (0.30) 0.93 (0.35) 1.09 (0.52) 1.47 (0.67) 1.86 (0.92) 2.05 (1.30) 1.43 (0.87)
0.89 (0.27) 0.87 (0.25) 0.91 (0.40) 0.93 (0.32) 0.98 (0.42) 1.28 (0.58) 1.65 (0.79) 1.56 (0.90) 1.10 (0.66)
-0.11 (0.20) -0.01 (0.27) -0.02 (0.30) 0.01 (0.25) 0.10 (0.42) 0.18 (0.59) 0.21 (0.92) 0.48 (1.48) 0.33 (0.94)
-1.73 -0.13 -0.21 0.10 0.82 1.02 0.76 1.08 1.18
10 10 10 10 10 10 10 10 10
.115 .897 .835 .926 .430 .330 .464 .305 .264
25m subsections; *: p < .05; **: p < .01.
Except for the significant differences in speed (lower for the high accident rate curve, see Table 11) there are no significant differences within corresponding subsections between low and high accident rate curve.
December 2007
52
TUD
Deliverable D8
Dissemination Level (PU)
4.4.3.3.
Contract N. 506184
Discussion for the selected location
The analysis above showed higher speeds for the low accident rate curve (at least in some subsections) and no significant differences in fixation duration and reaction times, which both were used as proxy variables for workload. These results contradict the assumptions formulated in Hypothesis 4. Before entirely refuting this hypothesis, restrictions concerning the scope of the analysis conducted above, shall be discussed. First of all, the analysis for the backwards direction was not conducted as no reaction time values were collected for this direction. The decision not to use the PDT for these sections in the backwards direction was made because of the specific situation for section 24 (the high accident rate curve): for the backwards direction this curve is situated right after the village and approximately 400m after the speed camera (see Figure 18). Thus, it was assumed that speeds are considerably reduced and that, therefore, this direction cannot be responsible for the accidents. Of course, the speed camera could have been positioned after the accidents occurred and would thus not have influenced speed when the accidents occurred. However, according to the responsible authority, the speed camera was present (although not working all the time!) even before the three year period during which the accidents were collected for our analysis. It was further positioned at this location to protect pedestrians as there is no pavement along this road through the village and not because of the accidents in the following curve. The detailed accident analysis revealed further that all except one9 accident were in fact caused by drivers driving in the direction of our analysis. Therefore, the explanation for the non-existing significant differences that the accidents were caused by drivers driving the opposite direction can be ruled out. The second explanation is of rather general nature concerning the rationale behind field experiments when conducted for the purpose of explaining accidents: on one hand these experiments necessarily have to rely on a relatively small sample of drivers, while on the other hand accidents are rare events from a statistical point of view (see report 8.1., Weller et al., 2006). The field experiments will only show significant differences between low and high accident rate road section, if the accidents at this location are caused by a change in the monitored parameter (workload in our case) that is valid for all drivers (including of course the sample). It will not contribute to explaining accidents at this location if the accidents are caused by a minority of drivers who behave very different from the average driver. This could be for example very high speeds or inattention in an unforgiving environment. In this case it is unlikely that such drivers are in the sample and thus the data collected in these experiments cannot explain the accidents. This is the case, even if the cause of the accidents is to be found in the monitored parameter. At present this explanation cannot be ruled out. Finally the accidents at the location under inspection could be caused by situation specific factors that are not related to the monitored parameter. This can only be revealed by detailed analysis of the accidents that occurred at this location. In fact the accidents that occurred at section 24 show such peculiarities: all happened when the road surface was wet. Thus, the problem at this location could be attributed to low friction coefficients in wet conditions. This was not tested during our experiments and 9
This one accident was caused by a motorcycle rider for whom the speed camera obviously played no role (no number plates on front side of motorcycles; accident cause: exceeding the speed limit).
December 2007
53
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
as most drivers were conducted in dry conditions the experiments would not show these peculiarities. We have to conclude that this is indeed a possible explanation. Summing up the discussion, it cannot be ruled out that the hypothesis is true despite the results found do not support this interpretation.
4.4.4 Discussion The aim of the driving experiments reported here was to further validate the model, especially Hypothesis 4. According to this hypothesis, it was assumed that higher workload is found at high accident rate road sections when compared to low accident rate road sections. To test this assumption two different road sections with a high amount of driving accidents were compared with respective road sections which were matched concerning geometry but showed no accidents. One of these locations was reported above; readers interested in the second one are referred to internal report 8.2. (Weller & Schlag, 2007). The results did not support the assumptions concerning workload formulated in Hypothesis 4. However, they did not rebut these assumptions either, as the results could be explained in line with the general assumptions of the model. The insignificant differences in workload between high and low accident rate curve had to be attributed to an appropriate speed reduction at the high accident rate road sections. It is assumed that drivers not showing this reduction in speed, either because they did not perceive the cues indicating this reduction or because potential cues are deliberately disregarded, are responsible for the accidents that happened at the high accident rate roar sections. As was discussed above, testing this assumption cannot be done within driving experiments but might require simulator studies. Further the result shows the importance of driver training and education as well as of enforcement and driver selection.
December 2007
54
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
4.5. Integration of behavioural data in the Safety Performance Function 4.5.1 Introduction After the comparisons between sections with high versus low accident rates as conducted above, the following chapter is used to combine different measures in order to find possible parameters for the integration in the safety performance function (SPF). Prior to our analysis, the general ideas behind this step should be discussed. First of all, the psychological parameters should not replace the engineering parameters but heighten the quality of the SPF by adding psychological aspects. Statistically this implies that engineering and psychological parameters are (more or less) independent. In theory a (highly unlikely) correlation of 1 would mean that one of the parameters can be replaced by the other parameter without any loss of information. In this case one would choose the parameter that requires less effort concerning data collection -while at the same time sustaining data quality. One important approach in Workpackage 10 was the segmentation of road stretches in several sub-sections according to characteristic speed differences (see the corresponding deliverables in Workpackage 10. With the outstanding importance of this parameter, we wanted to know how it varies with reaction time in the peripheral detection task. Reaction time in this case is a proxy variable for workload. Given that workload could add additional information to the SPF, reaction time should be independent on one hand (i.e. no significant correlation), but on the other hand should show a characteristic pattern (e.g. a U-shaped function). In case this characteristic pattern could be described by a mathematical function, speed differences could be weighted following this function.
4.5.2 Method To conduct this analysis, sections were selected from the course that were characterised by a curve in which the PDT was conducted. Further, reaction time values had to be available prior to the curve in order to calculate differences. For the resulting 12 sections, the difference between the minimum and the maximum value in each section were assessed, together with the resulting percentage difference. The section started at approximately 150m before the curve and ended approximately 50m after the curves.
December 2007
55
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
4.5.3 Results The results are depicted in the following Figure 21.
Figure 21: Relation of the percentage increase in reaction time with the percentage decline in speed (x-axis). Each value on the x-axis represents a single road section. Added is the statistics of a linear regression.
Different reaction time parameters (minimum, maximum, mean, total change and percentage change) for the sections shown in Figure 21 were used as input variables for a stepwise linear regression analysis to find which parameters best describe the percentage decline in speed. In order to ensure that no non-linear relation exists in our data, the graphs of each of these parameters were inspected manually. Further a second regression analysis was calculated leaving out the two values situated on the extremes of the x-axis (7,9 and 19,64). This resulted in the following regression parameters. Table 13: Results of two linear regression analysis of different reaction time parameters on the percentage decline of speed for different road sections. Once for all twelve sections and once without the extremes on both sides.
Percent increase in RT Percent increase in RT (without extremes)
B 0,483**
Beta 0,686
R square 0,471
F 8,902
p ,014
0,500**
0,927
0,859
48,693
,000
The only parameter that met the criteria of the regression analysis was the percentage change in reaction time. As can be seen, a linear relation matches the data very well. When the two extremes (out of the twelve values) are left out, the values get really satisfying. In accordance with the introduction to this chapter, this December 2007
56
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
means that the inclusion of the reaction time does not add value to a SPF that uses the percentage decline in speed - at least not at this stage. However, the fact that the exclusion of the two extreme values at the end of the xaxis results in such large improvement of the quality of the regression function (see Table 13) require that these sections are inspected closer. In fact, the analysis revealed that the section with the smallest percentage decrease in speed is situated within a very curved section. Therefore, speed is already low at the very beginning, resulting in the small differences between maximum and minimum speed. At the same time the radii within this section are very small (Rmin= 83m). This could be interpreted as an increase in reaction time that is independent of speed reduction, given that the demand exceeds a certain level as indicated by the small radii in this section. On the other extreme of the x-axis, the curve with the high percentage decline in speed and the relatively low percentage change in reaction time, will be analysed closer. This section corresponds to section 50 which is characterised by a speed limit of 70km/h that is valid only during curve driving (see Weller & Schlag, 2007). When explaining the large percentage decline in speed, there are two different mechanisms to be considered: first of all, our subjects complied with the speed limit. Second, the speed limit at this site is obviously not selfexplaining which explains the high maximum speeds before the curve where the speed limit is not yet adhered to. These two mechanisms result in the high percentage decline in speed at this location. The comparatively low change in reaction time in turn is a direct result of this externally induced reduction of speed. This reduction allows the subjects to keep their relative performance in the PDT. The fact that this value deviates from the regression line in Figure 21 indicates the external reason for these values. The fact that the same phenomenon does not apply for section 09 (value 16.05 on the x-axis in Figure 21) which is the corresponding section for the to-direction, could be explained in two ways: first, it could be that subjects were not yet used enough to the PDT at the beginning of the course. This however, cannot be the only explanation, as subjects had lower reaction times in section 02 (corresponding to values 7.94 on the x-axis in Figure 21) which is situated before section 09. It is more likely that the curve 09 / 50 indeed is more demanding than other curves but that this higher demand is partly compensated for by the high experience with the PDT the subjects had collected near the end of the course.
4.5.4 Discussion The discussion of the results above show that the additional analysis of reaction time in relation to speed, reveals important aspects, even if it cannot yet be used for the SPF algorithm. This step requires a larger sample of data. This enlargement of data this time refers to the driven sections rather than the number of subjects. An integration of the percentage increase of reaction time, together with the percentage decrease in speed, could result in a valuable function to predict accidents.
December 2007
57
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
4.6. Subjective road categorisation 4.6.1 Introduction One important aspect of driver and driving behaviour on (rural) roads is that it is to a large part based on expectations. These expectations are in turn built based on the perception of the current road situation on one hand and past experiences and knowledge on behalf of the driver on the other hand (see as well the model in Figure 11). The closer the perceived road ahead matches a prototypical road schema stored in knowledge, the more likely that a certain behaviour associated with this knowledge will be displayed in the situation. The knowledge of this relationship is as well used in the concept of self-explaining road design (SER-design). Therefore, one important step towards self explaining road design in Europe is increasing the knowledge on how drivers categorise roads. By using this knowledge roads can be designed which would be homogenous within one category but heterogenous between categories (see e.g. Matena et al., 2006, Figure 15). While the difference between higher-order categories (motorways like the German Autobahn versus rural or city road) is often quite clear, data collected within some of these higher-order categories, like rural roads, reveal that there is a considerable variation in behaviour. The fact that behaviour varies more within rural roads in general, is a direct consequence of: the partly historic roots of these roads, the different functions these roads have to fulfil, the higher legal speed limits that allow larger variance, the dependency on landscape features like slopes which is taken for granted due to a lack of resources that could be spent to avoid this dependency, and o the large network length that makes redesigning time- and cost-consuming.
o o o o
The resulting large variation of alignment and appearance produces the large variation in behaviour. However, the existence of official subcategories within the “rural road” category, already suggests that some of these roads do have some common features. These official subcategories are usually as well associated with certain design standards or design elements and sometimes even with different legal speed limits. One possible explanation for the high amount of (severe) accidents on rural roads might be that there is a discrepancy between behaviour “officially” expected and behaviour displayed in reality. While other research conducted so far in workpackage 8 was directed towards single locations differing in accident rate, the research conducted here aims at larger road entities. The length of these entities depends on how they are perceived: as long as they are perceived as entity, they are treated as entity. However, knowledge on which features are relevant for rural roads in order to be perceived as entity is lacking. Therefore, we conducted a study that aimed at diminishing this lack of knowledge. As this part of WP 8 was not yet reported, the results are presented here in comparative detail.
December 2007
58
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
4.6.2 Method This study consisted of the following steps which are later described in more detail: o first, relevant behavioural dimensions were identified that could distinguish different rural roads, o second, pictures of rural roads were subsequently rated along those dimensions by subjects in a laboratory setting, o third, the ratings of these road pictures were later used to categorise the roads by using statistical analysis, and o finally, an attempt was made to re-translate this behavioural clustering in objective features. This last step is especially relevant, as it might help to identify possible cues or affordances that convey a behavioural message to the driver and which could later be used intentionally to influence behaviour (see as well, Figure 9). In order to be of use in a study of road categorisation, the dimensions and parameters along which the roads are rated, have to be able to distinguish between different roads and they have to be behavioural relevant. From former studies conducted by Riemersma (1988) and Steyvers (Steyvers, 1993, 1998; Steyvers, Dekker, Brookhuis, & Jackson, 1994), some relevant dimensions are already known (for a description of these studies see as well report 8.1. Weller et al., 2006). Different from Riemersma (1988) who included motorways as well, we used only rural roads, which thus required additional steps to identify relevant dimensions. Further, all studies so far are based on data collected by using Dutch roads. These roads might differ both in appearance as well as concerning the appropriate behaviour in comparison to other European countries. Concerning the research methodology, our study was conducted with close proximity to the aforementioned other studies. First of all, pictures were taken of a variety of rural roads in the German Federal State of Saxony and Brandenburg. It was taken care that these pictures comprised a large variety of rural roads. Care was spent to select a broad variety of different characteristics. All pictures were taken in summer in comparable dry whether conditions from the perspective of a car driver, both in height and in position on the road. No other road users were visible on the pictures. Some pictures were edited on the computer so that the edited picture differed from the original picture in just one element. All in all, 25 pictures were thus gathered. These 25 pictures were combined to 20 groups, each consisting of three printed-out pictures. These triads were then presented to a sample of 10 subjects which were recruited from university and through personal contacts. All subjects hold a valid driving licence between 5 and 35 years (AVG: 11 years) and drove an average of 17300km/a. The subjects were aged between 23 and 56 years with an average of 29.8 years. Each triad was presented to the subjects in succession. Subjects were asked to name what two of the pictures had in common that differed from the third picture. This technique is called repertory grid technique and was developed by Kelly within his theory of personal constructs (1955, cited from Riemersma, 1988). Its main advantage is that subjects are free to name their own parameters and thus allow previous unknown parameters to be found. In order to avoid that obvious parameters December 2007
59
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
were named, some triads were grouped so that it only contained similar elements, e.g. only sections with a forthcoming curve. On the other hand, in order to ensure highest variability in the parameters, other triads were grouped so as to contain pictures with the highest possible variability in road characteristics. The results of this first step showed that the subjects exclusively named objective criteria both to describe differences between, as well as communalities within the triads (see Annex Table A 2). However, during this first step of the study we had the impression that the subjects understood the task as some kind of “search task”, which can be regarded as being far from real behaviour in the field, where top-down guided search is only used when bottom-up control does not provide the necessary cues. Therefore, we decided to use own items which we regarded relevant together with the items of the Road Environment Construct List RECL used in former studies (Steyvers, see above). All items were rated on a six point Likert scale with verbal anchors. Low values indicate dissent, high values consent to the items. Besides the items reported in the following, the entire questionnaire consisted of more items which are in detail described in Friedel (2005). Additional analysis concerning the structure of the RECL items can be found in Rammin (2006). In the second step of the study, the items introduced above were used to rate 21 pictures of the rural roads which were selected from the 25 pictures used in the first step of the study. The subject sample this time consisted of 46 people (45,7 % female and 54,3 % male) with an age between 20 and 65 years (AVG: 40; SD 15 years). All subjects had a valid driving licence for between 2 and 47 years (AVG: 20; SD: 14 years) at the time of the study. The average annual distance driven (last two years) varied between 500 and 80,000 km (AVG: 21,152 km, SD: 17,447 km). The pictures were presented on a computer screen to the subjects who were asked to fill in the questionnaire (paper & pencil) without time constraints. The order of presentation varied, so that the first picture of one experimental run was the last picture for the next subject. The ratings were coded in the statistic software SPSS10 and used for the subsequent analysis.
4.6.3 Results In order to reduce the data to a reasonable number of variables, a factor analysis was performed with the RECL items. Prior to this step, the data was structured following Steyvers (1993) which resulted in a 16 variables times 966 ratings matrix (21 pictures times 46 subjects). Instead of following the original coding schema of Steyvers (see Annex, Table A 3) we conducted own factor analysis. The reason was that our study solely focused on secondary rural roads for which prior unknown factors might be more appropriate. Conducting our own analysis allowed these factors to emerge, while sustaining the possibility of strengthening the original factor structure. The visual analysis of the screeplot (see Annex Figure A 1) suggested to use two, maybe three factors (Eigenvalues exceeding 1). After comparing the two and three factor solution, we decided to preset the number of factors to be extracted to three. The results are presented in the following Table 14.
10
www.spss.com
December 2007
60
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Table 14: Varimax normalized factor loadings of the RECL items after factor analysis.
Variable Monotonous Lowers concentration Boring Lowers alertness Changeable Increases wakefulness Increases attention Relaxing Enjoyable Gives a good view Peaceful Spacious Irritating Threatening Demanding Dangerous Explained Variance [%]:
Factor I
Factor II
Factor III
0.854 0.832 0.802 0.782 -0.780 -0.710 -0.701 0.188 0.062 0.272 0.041 0.278 -0.089 -0.070 -0.160 -0.103 28.26
0.032 0.294 -0.050 0.302 0.114 -0.413 -0.480 0.805 0.803 0.651 0.630 0.591 -0.179 -0.411 -0.407 -0.507 23.01
-0.163 0.051 -0.138 0.006 0.240 0.083 0.072 -0.206 -0.388 -0.433 -0.302 -0.395 0.837 0.717 0.698 0.626 17.75
The total explained variance of all three factors amounts to 69.02 percent. Our factor structure showed some differences to the Steyvers solution (see Annex Table A 3): the items loading negatively on Factor I (Hedonic Value) in the Steyvers studies form a factor of their own (Factor III) in our analysis. Although the items could be summarized as negative hedonic factor, we think that the structure of items together with the theoretical basis summed up before, suggests the name “Danger by (information) overload” which we in short termed “Demand”. Our Factor II aggregates the remaining items of the Hedonic Value factor in Steyvers which all have a positive connotation. Based on the items subsumed in this factor, the picture of cruising through an open landscape literally comes to mind. We thus decided to name this factor “Comfort”. The last factor in our analysis (Factor I in Table 14) unites the items of the factors “Activational Value” and “Perceptual Variation” in the Steyvers studies. These items have a negative connotation and describe a state of monotony induced by underload. This factor thus was termed “Monotony”. The factor values of each of our three factors for each rating were calculated with SPSS and saved as additional variables. These factor values were further averaged for each picture and a cluster analysis was calculated based on these values. The results (dendrogram) are shown in the following Figure 22.
December 2007
61
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Figure 22: Results of the hierarchical cluster analysis (dendrogram) (SPSS.14).
The overview of the single clustering steps performed in the analysis suggests using the results after step 18 (see Annex, Table A 4). After this step further aggregation of pictures to clusters would result in a marked increase in the distance between clusters within the subsequent cluster. This resulted in three clusters (to the left of the blue line in Figure 22) representing the following pictures (sorted in dendrogram order): -
Cluster I: 6; 16; 9; 17; 2; 19 Cluster II: 11; 14; 12; 18; 4; 7; 20; 21; 3 Cluster III: 1; 13; 8; 5; 10; 15.
The average values for each factor and each picture are shown in the following Figure 23. The pictures are sorted by cluster and by rated appropriate speed within each cluster.
December 2007
62
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Figure 23: Average factor values for each picture in the different clusters.
To determine the stability of this clustering and as input for the further steps, an ANOVA was calculated for the three factors between the three clusters. The ANOVA showed significant differences both for the overall differences (Pillai: 1.725; F: 35.604; df: 6 / 34; Sig. .000) as well as for each factor (see Annex, Table A 5). Figure 23 together with the results displayed in Table A 6 (Annex) can be summarized in the following matrix (see Table 15). Table 15: Road-cluster and factor characteristics combined in a matrix.
Cluster I Cluster II Cluster III
Factor I
Factor II
Factor III
Low Low High
Low (High) High
High (Low) Low
According to this matrix, Cluster I roads are perceived as being highly demanding and not at all monotonous or comfortable. Cluster III roads represent the opposite pole, they are perceived as being highly monotonous, highly comfortable and not at all demanding. While perceived demand varies in the Cluster II roads, they differ from Cluster I roads by higher comfort and from Cluster III roads by their lower monotony. Thus, the cluster solution allows interpretation and suggests a clear distinction between the clusters with a certain instability for the Cluster II roads. Following the steps above, there are two research questions that have to be answered in order to be able to categorise roads along objective criteria which are behaviourally relevant: -
Is the subjective categorisation found so far mirrored by behaviour? If so, can objective criteria be identified which can be used as proxy variables for the subjective categorisation?
December 2007
63
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
To answer the first question a discriminant analysis and a regression analysis were calculated. Both used the subjective ratings concerning the appropriate speed [km/h] on the roads under investigation as proxy variable for measured speed. Measuring speeds on the road on all investigated roads would have exceeded the project resources. The discriminant analysis was performed to find out whether speed could be used to approximate the clustering found in the cluster analysis. The analysis revealed that only two out of the 21 roads were classified wrong with the following function: -7.825 * 0.101* speed [km/h]. This is an exceptionally good result, given that only one variable was used to predict the clusters. The two misclassified roads were the roads number 9 and 19, which were both classified as belonging to Cluster II instead of Cluster I. The reason for this misclassification can be seen in the peculiarities of these roads: road number 19 is an alleyway and road number 9 shows a building on the left close to the road, which might have influenced the ratings. Road number 19, the alleyway, constitutes an exception again below (see Table 17). As other roads with alleyway characteristics were present in the study as well, the mere alleyway characteristic does not suffice to draw inferences concerning speed or other subjective ratings. Still the results show that alleyways constitute a special case and need future in depth investigations. The regression analysis was calculated on the basis of the factor values of the RECL items (see above) on subjective speed. Success in this step would indicate that speed is indeed influenced by the overall impression of the road (which would include affordances as well) and thus constitute a further step towards model validation. A linear regression analysis with the three factor values for each of the 21 road pictures was performed, using the inclusion method, which means that all three variables remain in the analysis, even if their share in explaining variance is not significant. To ensure the quality of the regression, tolerance (VIF-) statistics and the Durbin-Watson coefficient were as well calculated and showed satisfying results (tolerance more than 0.1; DB-coefficient between 1.5 and 2.5) (see Brosius, 2002). Table 16: Results of the regression analysis of the three factors on the speed ratings for each road picture.
Statistics for the variables B in the regression Constant 77.168 Factor I (Monotony) 15.294 Factor II (Comfort) 27.323 Factor III (Demand) 8.547 Statistics for the Sum of Squares regression Regression 6392.575 Residuals 638.533 Corrected R Square 0.893
Standard Error 1.339 2.465 5.136 7.491 df 3 17
Beta
0.533 0.739 0.166 AVG sum of squares 2130.858 37.561
T
Sig.
57.612 6.203 5.320 1.141 F
.000 .000 .000 .270 Sig.
56,731
.000
The results in Table 16 show that the regression with all three factors explains almost 90 % of the variance in the speed data. Thus, speed cannot only be used to approximate the cluster solution (see the results for the discriminant analysis) but as well vice versa, the factor values of the three variables allow the prediction of speed. December 2007
64
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Factor three, however, has by far the lowest Beta weight and is not significant. This means that subjective feeling of risk and demand might be less important in predicting speed than assumed in the model. However, the other two factors might be interpreted as representing affordances, and thus strengthen the respective part of the model. As already pointed out above, the second important step after having shown the relevance of the cluster solution for behaviour, is to identify possible objective characteristics which might serve as proxy variables for the subjective ratings. This step was conducted by visual inspection of the pictures used in the study which were grouped according to the clusters. This visual analysis of the characteristics of the roads within each cluster revealed the following matrix of common features within and distinctive features between each cluster. Table 17: Distinctive objective features between the clusters resulting from the subjective ratings.
Surface Road Width Road Markings (centre line) Sight Distance Horizontal alignment
Cluster I
Cluster II
Cluster III
Poor Very narrow No11
Between I & III Between I & III
Good Wide Yes Very high Low CCR
High CCR
According to Table 17, roads can be classified as belonging to Cluster I as soon as they have a poor road surface, are narrow and do have no centre line. The other extreme is constituted by the roads belonging to Cluster III which have a good road surface, are wide and have centre markings. However, the distinctive feature to the roads classified as belonging to Cluster II, is that Cluster III roads have a very low curvature change rate (CCR) which additionally results in the very high sight distances found on these roads. The free cells in Table 17 indicate that the respective parameter can take any value within the extremes denoted by the other Clusters.
4.6.4 Discussion The study on the subjective categorisation of rural roads revealed that drivers distinguish between three different road categories which can be described with comparatively few objective criteria. These categories are as well relevant for behaviour, whereas in the present study solely rated appropriate speed could be used as proxy for measured speed on the respective roads. Further, it could be shown that (rated) speed can be inferred from the subjective ratings of the respective road. The results emphasize the importance of affordances and cues in the model. In contrast, the demand- (and risk) factor had no significant influence in predicting speed. This could mean that expected workload or risk are indeed only used in regulating driving behaviour in case the roads are not selfexplaining, i.e. relevant cues are missing or even inappropriate or the impression of the road does not form a behaviourally relevant affordance.
11
With the exception of road no. 19, an alleyway which had an intermittent centre line.
December 2007
65
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
5. Empirical Validation: Conclusions The overall results of the different empirical research strategies revealed that most of the assumptions formulated in the hypotheses are supported by the empirical data, whilst other hypotheses have to be tested more extensively. Firstly, evidence was found for a systematic underestimation of workload and risk of high accident rate curves (see chapter 4.2). This underestimation results in inappropriate speed behaviour which could cause accidents at the respective locations. A practical consequence concerning the application of this result for traffic safety would be to support the driver prior to the curve with appropriate information on one hand and feedback on the other hand. The increase in objective safety is in this case achieved by increasing the subjective feeling of un-safety. The necessary information could be conveyed to the driver in a proactive way supporting early adaptation with the help of affordances and cues. Secondly, the influence of affordances and cues in influencing driving behaviour could be shown (see chapter 4.1). This was done with the help of additional simulator studies. For example, we could show that simple curve warning signs resulted in a reduction in speed which was detectable at a distance significantly further away from the curve, compared to curves without these signs. As already discussed when introducing the model, the use of such explicit signs is unfortunately associated with some disadvantages. Therefore, they should only be used when no implicit cues or affordances could be used. Such implicit cues could be for example an optical (not physical) narrowing of the road, which would be effective by decreasing the subjective level of safety, thus evoking more precautious behaviour. Thirdly, the assumption that workload is higher in high accident rate road sections compared to low accident rate road sections, was only indirectly supported by the empirical data. Neither reanalysing existing psycho-physiological data (see chapter 4.3) nor analysing the data of own additional driving experiments conducted in the field (see chapter 4.4) showed this expected pattern. It was found that within our sample of subjects, a marked reduction of speed occurred at the high accident rate road sections. This could explain the non-significant differences in workload between high and the low accident rate road sections, as lowering driving speeds modulates workload. At the same time this means (given the assumptions formulated in the hypotheses are right), that the accidents are caused by drivers behaving differently from our sample of drivers. In fact the known characteristics of accidents (see report 8.1. Weller et al., 2006) do support this assumption. These few drivers would drive faster because they intentionally or unintentionally disregard the cues at the high accident rate road sections that caused the drivers in our sample to reduce speed. On one hand, this reveals the importance of driver education and training. On the other hand, it reveals the importance of building roads that are self-explaining and self-enforcing in the sense that they prevent all (!) drivers from showing such inappropriate behaviour. This could again be achieved by affordances and implicit cues which prompt the appropriate behaviour. At the same time, self enforcing roads would provide the driver with enough feedback (both positive and negative) and thus encourage appropriate and discourage, or even prevent, inappropriate behaviour. In the strictest application of such road design, there would be very limited margins for the impact of differential aspects distinguishing drivers or groups of drivers. Thus, a self-organizing loop between driver and environment is installed, leading the driver in December 2007
66
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
a fast and rather instinctive way to the behaviour required, which in turn is reinforced by positive consequences respectively by avoidance of negative consequences. Another component in the model are expectations and mental models which are important both in the perception as well as in the interpretation of e.g. cues. The perceived road category is supposed to form a vital part in mental models which influence behaviour to a large extent. To determine both this influence as well as gain further insight how these categories are built, we conducted an additional study on the subjective categorisation of rural roads. This study revealed that drivers distinguish between three different rural road categories which can be described with comparatively few objective criteria. These categories are also relevant for behaviour, at least concerning rated appropriate speed. Further, it could be shown that (rated) speed can be inferred from the rated impression of the respective road. Thus, visual impression and subjective categorization give important hints for speed choice. Here factors close to affordances proved to be more important than expected demand and risk. The results support the assumption that expected workload or risk are indeed only used in regulating driving behaviour in case the roads are not self-explaining, i.e. relevant cues are missing or even inappropriate or the impression of the road as such does not form a behaviourally relevant affordance. Further, workpackage 8 aimed at the integration of psychological parameters in a safety performance function (SPF). The fact that speed differences could be used as proxy variable for workload (see chapter 4.5) is regarded as important finding towards this integration. As speed can further be approximated by different models using geometrical data as input variables (see Deliverable D10: Dietze et al., 2007), the integration of psychology in the SPF at present and with the data available has to be done by using such models. Further assignment of numeric values to psychological parameters was not feasible within the project, at least not with sufficient generalisability for other situations than the ones tested in the experiments. However, our findings throughout the validation process of the model (see the report at hand) showed that driver and driving behaviour, as well as the (resulting) accidents, are influenced by a variety of psychological factors. Although these factors vary as well with geometry, we found that a considerable variation in these parameters is independent of geometry. In the project we were able to identify a number of factors which explain this variation. The integration of such factors in speed prediction models for rural roads could ultimately result in a valid safety performance function which integrates psychological factors in a cost-effective way. The most important future step to do so is to increase the number of cases (i.e. road sections) in the database to be able to derive stable parameter values.
December 2007
67
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
6. References Allenbach, R., Hubacher, M., Huber, C. A., & Siegrist, S. (1996). Verkehrstechnische und -psychologische Sicherheitsanalyse von Strassenabschnitten. (Vol. 31). Bern: bfu. Brosius, F. (2002). SPSS 11. Bonn: mitp Verlag. Bruce, V., Green, P. R., & Georgeson, M. A. (1996). Visual perception. Physiology, Psychology and Ecology (3 ed.). Hove: Psychology Press. Cavallo, V., & Cohen, A. S. (2001). Perception. In P.-E. Barjonet (Ed.), Traffic psychology today (pp. 63-89). Boston, Dordrecht, London: Kluwer Academic Publisher. de Waard, D. (1996). The Measurement of Drivers' Mental Workload. The Traffic Research Centre VSC, University of Groningen, The Netherlands. Retrieved 07.03.2007, from http://www.home.zonnet.nl/waard2/mwl.htm Dietze, M. (2007). RoadView. Visualisation and analysis software for roads. Dresden: Technische Universität Dresden. Dietze, M., Ebersbach, D., Lippold, C., Mallschützke, K., & Gatti, G. (2005). Road Geometry, Driving Behaviour and Road Safety. Safety Performance Functions - Internal Report I. RIPCORD-ISEREST. EU FP 6, Contract No. 50 61 84. Retrieved 31.07.2007, from http://ripcord.bast.de/pdf/ri-tud-wp10-r1v1_final.pdf Dietze, M., Ebersbach, D., Lippold, C., Mallschützke, K., Gatti, G., & Wieczynski, A. (2007). RiPCORD-iSEREST Deliverable D10: Safety Performance Function, from http://www.ripcord-iserest.com/ Donges, E. (1978). A Two-Level Model of Driver Steering Behavior. Human Factors, 20(6), 691-707. Donges, E. (1999). A Conceptual Framework for Active Safety in Road Traffic. Vehicle System Dynamics, 32(2/3), 113-128. Dulisse, B. (1997). Methodological Issues in Testing the Hypothesis of Risk Compensation. Accident Analysis and Prevention, 29(3), 285 - 292. Durth, W. (1974). Ein Beitrag zur Erweiterung des Modells für Fahrer, Fahrzeug und Straße in der Straßenplanung. In Straßenbau und Straßenverkehrstechnik, H.163. Bonn- Bad Godesberg. Ellinghaus, D., & Steinbrecher, J. (2003). Fahren auf Landstraßen. Traum oder Albtraum? Uniroyal Verkehrsuntersuchung 28. Hannover: Continental AG. Elvik, R., & Vaa, T. (2004). The handbook of road safety measures. Amsterdam: Elsevier. Forschungsgesellschaft für Straßen- und Verkehrswesen (FGSV). (1998). Merkblatt für die Auswertung von Straßenverkehrsunfällen - Teil 1; Führen und Auswerten von Unfalltypen-Steckkarten; Empfehlungen Nr. 12. Köln: FGSV. Forschungsgesellschaft für Straßen- und Verkehrswesen (FGSV) (Ed.). (1995). RASL: Richtlinien für die Anlage von Straßen, Teil: Linienführung. Köln: FGSV. Frensch, P. A. (2006). Implizites Lernen. In J. Funke & P. A. Frensch (Eds.), Handbuch der Allgemeinen Psychologie - Kognition (pp. 229-238). Göttingen: Hogrefe. Friedel, T. (2005). Subjektive Kategorisierung von Landstraßen (Subjective Categorisation of Rural roads). Unpublished Diploma Thesis, TU Dresden; Chair of Traffic and Transportation Psychology, Dresden. Fuller, R. (1984). A conceptualization of driving behaviour as threat avoidance. Ergonomics, 27(11), 1139-1155. December 2007
68
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Fuller, R. (2005). Towards a general theory of driver behaviour. Accident Analysis and Prevention, 37(3), 461–472. Fuller, R., & Santos, J. A. (2002). Human Factors for Highway Engineers. Amsterdam: Pergamon. Funke, J., & Frensch, P. A. (Eds.). (2006). Handbuch der Allgemeinen Psychologie Kognition. Göttingen: Hogrefe. Gerrig, R. J., & Zimbardo, P. G. (2005). Psychology and Life (17. ed.). Boston: Pearson. Gibson, J. J. (1986). The ecological approach to visual perception. Hillsdale (New Jersey): Lawrence Erlbaum. Godthelp, H., Milgram, P., & Blaauw, G. J. (1984). The development of a time-related measure to describe driving strategy. Human Factors, 26(3), 257-268. Goldstein, E. B. (Ed.). (2005). Blackwell handbook of sensation and perception. Oxford: Blackwell. Hacker, W. (2005). Allgemeine Arbeitspsychologie. Psychische Regulation von Wissens-, Denk- und körperlicher Arbeit (2 ed.). Bern: Hans Huber. Hale, A. R., Stoop, J., & Hommels, J. (1990). Human error models as predictors of accident scenarios for designers in road transport systems. Ergonomics, Vol. 33, Numbers 10 and 11(Special Issue: Errors in the Operation of Transport Systems), 1377-1387. Heger, R., & Weise, G. (1996). Fahrverhalten und Aktivierung von Kraftfahrern als Kriterien zur Bewertung der Linienführung von Außerortsstraßen. In B. Schlag (Ed.), Fortschritte der Verkehrspsychologie 1996, Kongressbericht. Dresden: Deutscher Psychologen Verlag GmbH. Jacob, R. J. K. (1995). Eye Tracking in Advanced Interface Design. In W. Barfield & T. A. Furness III (Eds.), Virtual Environments and Advanced Interface Design (pp. 258-288). Oxford: Oxford University Press. Jones, K. S. (2003). What Is an Affordance? Ecological Psychology, 15(2), 107-114. Koch, I. (2005). Konditionieren und implizites Lernen. In J. Müsseler & W. Prinz (Eds.), Allgemeine Psychologie (pp. 386-431). Heidelberg: Spektrum Akademischer Verlatg. Lamm, R., Beck, A., Ruscher, T., Mailaender, T., Cafiso, S., & La Cava, G. (2006). How to Make Two-Lane Rural Roads Safer. Scientific Background and Guide for Practical Application: Witpress (in print). Land, M. F., & Horwood, J. M. (1995). Which Parts of the Road Guide Steering. Nature, 377, 339-340. Lee, D. N. (1976). A theory of visual control of braking based on information about time-to-collision. Perception, 5, 437-459. Lee, D. N., & Lishman, R. (1977). Visual control of locomotion. Scandinavian Journal of Psychology, 18(3), 224-230. Matena, S., Weber, R., Louwerse, R., Drolenga, H., Vaneerdewegh, P., Pokorny, P., et al. (2006). Road categorisation and design of self explaining roads. Internal Report RI-BASt-WP3-R1-V03_Road_Categorisation_and_SER. RIPCORDISEREST. EU FP 6, Contract No. 50 61 84. Retrieved 2007.08.14, from http://ripcord.bast.de/pdf/ri-bast-wp3-r1-V3road_categorisation_and_ser_final.pdf Michon, J. A. (1985). A critical view of driver behavior models: what do we know, what should we do? In L. Evans & R. C. Schwing (Eds.), Human behavior and traffic safety (pp. 485 - 524). New York - London: Plenum Press. Milner, D., & Goodale, M. A. (1995). The visual brain in action. Oxford: Oxford University Press. December 2007
69
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Miura, T. (1990). Active function of eye movement and useful field of view in a realistic setting. In R. Groner, d’Ydewalle & R. Parham (Eds.), From Eye to Mind: Information Acquisition in Perception, Search, and Reading (pp. 119127). Amsterdam: North-Holland. Näätänen, R., & Summala, H. (1976). Road-user behavior and traffic accidents. Amsterdam, Oxford: North-Holland Publishing Company. Noland, R. B. (2003). Traffic fatalities and injuries: the effect of changes in infrastructure and other trends. Accident Analysis and Prevention, 35(4), 599– 611. Norman, D. A. (1981). Categorization of action slips. Psychological Review, 88, 1-15. Norman, D. A. (1998). The Design of everyday things. London: The MIT Press. OECD. (1990). Behavioural adaptations to changes in the road transport system. Paris: OECD expert group. Petermann, I. (2006). Visuelle Wahrnehmung und subjektive Einschätzung von Kurven und von deren Annäherungsbereichen. der Beitrag psychologischer Einflussgrößen zur Erklärung des Unfallgeschehens auf Außerortsstraßen. Unpublished Diploma thesis, TU Dresden; Chair of Traffic and Transportation Psychology, Dresden. Petermann, I., Weller, G., & Schlag, B. (2007). Beitrag des visuellen Eindrucks zur Erklärung des Unfallgeschehens in Landstraßen (The Contribution of visual Impression in explaining Curve Accidents on Rural Roads). Straßenverkehrstechnik, 51(8), 401-408. Posner, M. I. (1980). Orienting of attention. Quarterly Journal of Experimental Psychology, 32(1), 3-25. Rammin, C. (2006). Die Anwendung der Road Environment Construct List (RECL) zur subjektiven Bewertung von Landstraßen. (Application of the Road Environment Construct List (RECL) for the subjective appraisal of rural roads). Unpublished Student Paper, TU Dresden; Chair of Traffic and Transportation Psychology, Dresden. Ranney, T. A. (1994). Models of driving behavior: a review of their evolution. Accident Analysis & Prevention, 26(6), 733-750. Rasmussen, J. (1986). Information processing and human-machine interaction. An approach to cognitive engineering. New York: North-Holland. Reason, J. T. (1990). Human Error. Cambridge: Camebridge University Press. Recarte, M. A., & Nunes, L. M. (2000). Effects of verbal and spatial-imagery tasks on eye fixations while driving. Journal of Experimental Psychology: Applied, 6(1), 31-43. Richter, P., Wagner, T., Heger, R., & Weise, G. (1998). Psychophysiological analysis of mental load during driving on rural roads – a quasi-experimental field study. Ergonomics, 41(5), 593-609. Richter, P., Weise, G., Wagner, T., & Heger, R. (1996). Fahrverhalten und psychophysiologische Aktivierung von Kraftfahrern als Bewertungskriterien der Gestaltungsgüte von Straßenverkehrsanlagen. Schlussbericht zum Forschungsprojekt Ri 671/2-1 im Auftrag der Deutschen Forschungsgemeinschaft DFG. Dresden: Technische Universität Dresden. Riemersma, J. B. J. (1988). An empirical study of subjective road categorization. Ergonomics, 31(4), 621 - 630. Rothengatter, T. (2002). Drivers' illusions--no more risk. Transportation Research Part F: Traffic Psychology and Behaviour, 5(4), 249-258.
December 2007
70
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Rumar, K. (1985). The role of perceptual and cognitive filters in observed behaviour. In L. Evans & R. C. Sching (Eds.), Human Behaviour and Traffic Safety (pp. 151 - 170). New York: Plenum Press. Schlag, B. (2004). Lern- und Leistungsmotivation. (2 ed.). Wiesbaden: VS Verlag. Schneider, W., Dumais, S. T., & Shiffrin, R. M. (1984). Automatic and control processing and attention. In R. Parasuraman & D. R. Davies (Eds.), Varieties of attention (pp. 1-27). London: Academic Press. Schneider, W., & Shiffrin, R. M. (1977). Controlled and automatic human information processing: I. Detection, search, and attention. Psychological Review, 84(1), 1-66. Schulz, R., & Lippold, C. (2006). Orientierungssichtweite - Definition und Beurteilung. Schlussbericht (No. Bast FE 02.0231/2003/AGB). Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic human information processing: II. Perceptual learning, automatic attending and a general theory. Psychological Review, 84(2), 127-190. Sossoumihen, A. J. (2001). Entwicklung eines Rahmenkonzeptes zur Bewertung der Linienführung von Außerortsstraßen nach der Zielfunktion Fahrsicherheit. Unpublished Dissertation, Fakultät Verkehrswissenschaften “Friedrich List” TU Dresden. Steyvers, F. J. J. M. (1993). The measurement of road environment appreciation with a multi-scale construct list. In A.G. Gale, I. D. Brown, C. M. Haslegrave, H. W. Kruysse & S. P. Taylor (Eds.), Vision in Vehicles-IV (pp. 203-212). Amsterdam: North-Holland. Steyvers, F. J. J. M. (1998). Categorisation and appraisal of rural two-lane undivided 80-km/h roads. In A.G.Gale, I.D. Brown, C.M. Haslegrave & S. P. Taylor (Eds.), Vision in vehicles - VI (pp. 271-278). Amsterdam: Elsevier. Steyvers, F. J. J. M., Dekker, K., Brookhuis, K. A., & Jackson, A. E. (1994). The experience of road environments under two lighting and traffic conditions: application of a Road Environment Construct List. Applied Cognitive Psychology, 8(5), 497-511. Summala, H. (1996). Accident risk and driver behaviour. Safety Science, 22(1-3), 103-117. Theeuwes, J. (2000). Commentary on Räsänen and Summala, “Car Drivers’ Adjustments to Cyclists at Roundabouts”. Transportation Human Factors, 2(1), 19-22. Theeuwes, J., & Godthelp, H. (1995). Self-explaining roads. Safety Science, 19, 217225. Treat, J. R., Tumbas, N. S., McDonald, S. T., Shinar, D., Hume, R. D., Mayer, R. E., et al. (1977). Tri-level study of the causes of traffic accidents. Volume I: Casual factor tabulations and assessment. Final report (No. DOT-HS-034-3534). Washington: National Highway Traffic Safety Administration. Tscheschlok, N. (1998). Analyse von Straßenverkehrsunfällen unter Berücksichtigung der mentalen Beanspruchung der Kraftfahrer, des Fahrverhaltens und der Streckencharakteristik auf Außerortsstraßen. Unpublished Diplomarbeit, Technische Universität Dresden, Dresden. Van Winsum, W., & Godthelp, H. (1996). Speed Choice and Steering Behavior in Curve Driving. Human Factors, 38(3), 434-441. Voigt, J. (2007). Der Einfluss der Straßengestaltung auf das Fahrverhalten unter Berücksichtigung der subjektiven Einschätzung der einzelnen Gestaltungselemente. Unpublished Diploma thesis, TU Dresden; Chair of Traffic and Transportation Psychology, Dresden. December 2007
71
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
von Campenhausen, C. (1993). Die Sinne des Menschen. Einführung in die Psychophysik der Wahrnehmung (2 ed.). Stuttgart: Georg Thieme Verlag. Wagner, T. (2000). Die Integration von anforderungsanalytischen und biopsychologischen Methoden zur Bewertung von Arbeitssystemen - eine Felduntersuchung zu Fahraufgaben auf Außerortsstraßen. Dissertationsschrift. Dresden: Technische Universität Dresden. Wagner, T., & Richter, P. (1997). Anforderungsanalytische Streckenbewertung und deren Anwendbarkeit für die Anlage von Außerortsstraßen eine psychologisch-ingenieurwissenschaftliche Perspektive. In B. Schlag (Ed.), Fortschritte der Verkehrspsychologie 1996, Kongressbericht. Dresden: Deutscher Psychologen Verlag GmbH. Warren, W. H., Mestre, D. R., Blackwell, A. W., & Morris, M. W. (1991). Perception of Circular Heading From Optical Flow. Journal of Experimental Psychology: Human Perception and Performance., 17(1), 28-43. Weiße, B. (2006). Channel merger. Program developed for data synchronisation. Unpublished. Dresden: Technische Universität Dresden. Weller, G., & Schlag, B. (2004). Verhaltensadaptation nach Einführung von Fahrerassistenzsystemen. In B. Schlag (Ed.), Verkehrspsychologie. Mobilität – Verkehrssicherheit – Fahrerassistenz. (pp. 351-370). Lengerich: Pabst Science Publ. Weller, G., & Schlag, B. (2007). RiPCORD-iSEREST. Internal Report 8.2.: Theoretical and empirical validation of a driver behaviour model for rural roads, from http://www.ripcord-iserest.com/ Weller, G., Schlag, B., Gatti, G., Jorna, R., & Leur, M. v. d. (2006). Human Factors in Road Design. State of the art and empirical evidence. Internal Report 8.1 RiPCORD-iSEREST. Retrieved 07.03.2007, from http://ripcord.bast.de/pdf/ri_tud_wp8_r1_v5_human_factors_final.pdf Wendsche, J., Uhmann, S., & Meier, J. (2006). Beanspruchungsmessung in Kurven: Untersuchung physiologischer und fahrverhaltensbezogener Parameter in Kurven mit unterschiedlicher Unfallhäufigkeit. Bericht zur Forschungsorientierten Vertiefung. Unpublished. Dresden: Technische Universität. Wickens, C. D. (1991). Processing resources and attention. In D. L. Damos (Ed.), Multiple-task performance. London: Taylor & Francis. Wickens, C. D. (1992). Engineering psychology and human performance (2 ed.). New York: Harper-Collins. Wilde, G. J. S. (1988). Risk homeostasis theory and traffic accidents: propositions, deductions and discussions of dissension in recent reactions. Ergonomics, 31(4), 441 - 468. Wilde, G. J. S. (1994). Target risk. Toronto: PDE. Wilde, G. J. S. (2001). Target risk 2. A new psychology of safety and health ; what works? What doesn't? And why ... Toronto: PDE Publications. Yilmaz, E. H., & Warren, W. H. (1995). Visual control of braking: A test of the [taudot] hypothesis. Journal of Experimental Psychology: Human Perception and Performance, 21(5), 996-1014.
December 2007
72
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
7. Annex Table A 1: Curve characteristics of the curves used for the re-analysis of the psychophysiological data.
Pair No.
2 3 4 5
kind of curve (high versus low accident numbers) high low high low
curve direction
radius [m]12
total length [m]
CCR [gon/km]
accident number (all accidents; 19931996)
ADT [cars/24h]
accident rate [A/(10^6 cars*km)]
left left left left
64 58 83 149
979 795 528 428
10 3 20 4
5370 5370 10000 3170
26.6 8.8 22.0 7.7
high low high low
left left right right
65 80 85 (+) 130 / 98 (+) 105 100 (+) 80 75
64 100 58 55
606 392 795 848
12 0 6 1
5370 4100 5370 3170
31.9 0 17.6 5.2
Table A 2: Criteria named in a first step by subjects to distinguish rural roads.
12
criteria named by the subjects
poles that described these criteria
road markings road surface sight distance ahead sight distance to the side greenery / planting alleyway buildings curvature traffic-lanes road width hard shoulder / space to the side foot-path ditch (at the side of the road) reflexion post kerb kerb overgrown roadside crash barrier tree mirrors (marking of the trees) clarity vertical alignment direction signs / traffic lights
yes vs. no / if yes: clear vs. unclear good vs. bad good vs. bad good vs. bad a lot of vs. little yes vs. no yes vs. no heavy vs. little 2 vs. 1 (per direction ) or 3 vs. 2 (in all) wide vs. narrow yes vs. no yes vs. no yes vs. no yes vs. no yes vs. no yes vs. no loose / unfortified vs. fastened yes vs. no yes vs. no good vs. bad hilly vs. even yes vs. no
(+): plus transitions curves.
December 2007
73
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Table A 3: Original calculations of RECL items ( © 2005 University of Groningen).
Calculation key for the three RECL-factors Hedonic Value (factor 1), Activational Value (factor 2) and Perceptual Variation (factor 3). Positive constructs can be summated directly, negative constructs have to be mirror-reversed (= 7 – construct score). Scoring: raw scores from left to right in the RECL-form with 1, 2, 3, 4, 5 or 6. Construct-number
Construct
01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16
Lowers alertness Changeable Threatening Lowers concentration monotonous Dangerous Demanding Irritating Relaxing Increases attention Gives a good view Enjoyable Spacious Peaceful Boring Increases wakefulness
sign and loading on factor nr. -2 +3 -1 -2 -3 -1 -1 -1 +1 +2 +1 +1 +1 +1 -3 +2
© 2005 University of Groningen
Figure A 1: Screeplot of the RECL items as found for our data.
December 2007
74
TUD
Deliverable D8
Dissemination Level (PU)
Contract N. 506184
Table A 4: Overview of steps performed in the cluster analysis. Zusammengeführte Cluster Schritt 1
Erstes Vorkommen des Clusters Nächster Schritt
Cluster 1 6
Cluster 2 16
Koeffizienten ,006
Cluster 1 0
Cluster 2 0
2
7
20
,012
0
0
6
3
6
9
,024
1
0
4
4
6
17
,033
3
0
15
5
12
18
,065
0
0
9
6
7
21
,096
2
0
11
7
1
13
,102
0
0
10
8
5
10
,136
0
0
13
9
4
12
,159
0
5
14
10
1
8
,170
7
0
16
11
3
7
,176
0
6
14
12
11
14
,223
0
0
17
13
5
15
,255
8
0
16
14
3
4
,291
11
9
17
15
2
6
,357
0
4
18
16
1
5
,638
10
13
20
17
3
11
,755
14
12
19
18
2
19
,793
15
0
19
19
2
3
1,423
18
17
20
20
1
2
2,648
16
19
0
3
Table A 5: Between subject effects of the ANOVA for the factors and clusters.
Variable Factor I (Monotony) Factor II (Comfort) Factor III (Demand)
Sum of squares 6.633 4.460 1.462
df 2 2 2
AVG sum of squares 3.317 2.230 0.731
F
Sig.
31.489 59.133 10.988
.000 .000 .001
Part. Eta Square 0.778 0.868 0.550
Table A 6: Results of the Scheffeé-Test following the ANOVA
Variable
Cluster No.
Cluster No.
Diff.
Factor I (Monotony)
1 1 2 1 1 2 1 1 2
2 3 3 2 3 3 2 3 3
0.400 -0.949 -1.349 -0.863 -1.162 -0.299 0.417 0.692 0.275
Factor II (Comfort) Factor III (Demand)
December 2007
75
Standard error 0.171 0.187 0.171 0.102 0.112 0.102 0.136 0.149 0.136
Sig. .092 .000 .000 .000 .000 .030 .023 .001 .158
TUD
