- Email: [email protected]
Typical patterns in road-traffic accidents during driver training
Accident Analysis and Prevention 36 (2004) 603–608
Typical patterns in road-traffic accidents during driver training An explorative Swedish national study Hans-Yngve Berg a,b,∗ , Nils P. Gregersen a,c , Lucie Laflamme d a
d
Division of Social Medicine & Public Health Science, Department of Health and Society, Faculty of Health Sciences, SE-581 85 Linköping, Sweden b Swedish National Road Administration, SE-781 87 Borlänge, Sweden c Swedish National Road and Transport Research Institute (VTI), SE-581 95 Linköping, Sweden Division of Social Medicine, Department of Public Health Sciences, Karolinska Institutet, SE 171 76 Stockholm, Sweden Received 29 March 2001; received in revised form 7 April 2003; accepted 10 April 2003
Abstract A new law came into force in Sweden on 1 September 1993, which makes instructor-assisted driving practice possible at the age of 16 years instead of the previous 17 years and 6 months. When the age limit was reduced, the possibility that this would lead to more road-traffic accidents (RTA) during driving practice was discussed. The aim of this study was to highlight typical road-traffic accident patterns and to discuss their potential for improved training and targeted prevention. A total of 11 variables (41 categories) descriptive of the 1081 RTA involving novice drivers and police registered during the period 1994–1999 were analysed simultaneously using in turn, two multivariate analysis techniques: the Factorial Analysis of Correspondence (FAC) and the Hierarchical Ascendant Classification (HAC). Four accident classes were identified and quantified, of which the first two were more typical of rural areas: (1) accidents in rural areas on straight stretches and related to speed limit 70 km/h (n = 306); (2) accidents in rural areas on straight stretches and high-speed related (n = 97); (3) accidents in built-up areas, low-speed related and of the type rear-end (n = 298); and (4) accidents in built-up areas, at road junctions and low-speed related (n = 380). Together, these classes point to a variety of opportunities to develop ways of working with targeted prevention. Instead of adopting a general attempt to counteract the relationship between individual variables and accidents, it is possible instead to focus on a whole context and its relationship with its typical accidents and any resulting injuries. This, in its turn, allows greater specificity in the build up of the Swedish licence and training regulations and its corresponding course curriculum. © 2003 Elsevier Ltd. All rights reserved. Keywords: Traffic accidents; Traffic safety; Learner drivers; Driver training; Factorial Analysis of Correspondence; Hierarchical Ascendant Classification
1. Introduction A system under which an individual is permitted to choose for him/herself whether to learn to drive in a driving school and/or through private instruction has been in place in Sweden since the start of the 1920s. A new law came into force on 1 September 1993, which makes instructor-assisted driving practice possible at the age of 16 years instead of the previous 17 years and 6 months. For a person to be able to practice driving, it is necessary for him/her to obtain a learner’s permit regardless of the learner driver’s age. The learner driver can then choose to practice driving either with a lay instructor or with a driving school; a combination of these two possibilities is most common. ∗ Corresponding author. Tel.: +46-243-750-63; fax: +46-243-751-93. E-mail address: [email protected] (H.-Y. Berg).
0001-4575/$ – see front matter © 2003 Elsevier Ltd. All rights reserved. doi:10.1016/S0001-4575(03)00068-X
An instructor’s permit is required in order to act as a lay instructor, who more often than not is one or other of the parents. In order to obtain a permit, an instructor must be aged more than 24 years, have held a full driving licence for a car for five years and not have too many demerit points in the national Swedish driver register. Once the learner driver and the lay instructor have obtained their permits, no further restrictions apply. They can practice driving whenever and wherever they wish. This also applies to the driving school. When the age limit was reduced, it was put forward that this could lead to an increase in the number of accidents during driving practice. Consequently, the registration form completed by police officers when called on an accident site was modified so as to incorporate two additional fields, i.e. with the text “driving school” or “lay instruction”. This, in turn, enabled monitoring of reported police accidents occurring during practice driving.
604
H.-Y. Berg et al. / Accident Analysis and Prevention 36 (2004) 603–608
Gregersen et al. (2003) have calculated the size of the accident problem while practising for a licence for a private car (class B). They also estimated the size of the public health problem and also the road safety problem from a health risk perspective. In total, they analysed 444 driving practice accidents with personal injury during the period 1 January 1994–31 December 2000 among learner drivers of all ages. The results show that in the 444 injury accidents, 827 persons were injured and 22 persons killed, which is equivalent to approximately three persons per year. The younger the driver, the greater the number of injured or killed per accident. From the national licence register, it was found that 229,000 learner drivers had started to practice from the age of 16 during 1994–1996. They were involved in 63 injury accidents. The population of newly licensed drivers aged 18–19 during the same period was 269,000 and as drivers they were involved in 2461 injury accidents. This gives a health risk of 0.28 for learner drivers and 9.15 for newly licensed drivers. The accident risk for learner drivers was 0.47 and 4.77 for newly licensed drivers. Gregersen et al. (2003) concluded in their study that there is a problem with accidents during driving practice which needs to be addressed, especially with regard to the Swedish “Vision Zero”, approved by the Swedish Government, which aims at zero killed or seriously injured in road traffic. They also concluded that the problem is minor compared to the situation after licensing and that the focus in developing safety measures should be on the accident situation of novice drivers rather than learner drivers. However, their analysis also showed, that the driving practice accidents were slightly more severe, both in terms of costs and number of killed and injured per accident. This is not acceptable in the perspective of the “Vision Zero”. In that respect, there is a need for increased knowledge about where, when and how accidents take place during driving practice as well as who are concerned. The data gathered by the registration form completed by police officers makes it possible to identify particular circumstances of road traffic accidents among learner drivers
and with help of this knowledge design suitable countermeasures.
2. Aim The aim of this nationwide study was to increase knowledge about the circumstances of occurrence of accidents during driving practice. Such results are expected to serve as a basis to lay the groundwork for the design of preventive measures—and course curriculum—adapted as much as possible to target situations and groups. 3. Materials and methods 3.1. Data source and selected variables The additions made to the police report form in 1993, combined with the information normally collected made it possible to document where driving practice accidents occur, when, and who is involved in them. Indeed, apart from entering details about the type of driving practice, the form also informs about, e.g. the date, time and place where the accident occurred, the age and sex of the driver, type of accident, speed limit, severity of the accident, and the weather and road conditions at the time. After completion by the police on the site, the data on each form is then entered in a national database under the responsibility of the Swedish National Road Administration (SNRA). For the current study, the 11 variables listed in Table 1 were extracted from the national register and coded into the categories specified in the second column of the table. The seven operating regions into which the SNRA divides Sweden are geographically identified in Fig. 1. Some of them are almost exclusively rural while others are mainly urban. The northern and central regions are sparsely populated with many sections of road subject to 90 and 110 km/h
Table 1 Variables, categories and number of accidents in the analysis Variable
Categories and number in parentheses in each category
Answer
Type of accident
Single (191), meeting (78), overtaking (55), rear-end (150), turning (156), crossing (178), vulnerable (55), hoofed wild animals (36), other (182) Fatal/severe (105), minor injury (335), without injury (641) North (34), central (93), Stockholm (178), west (260), Mälardalen (135), southeastern (169), Skåne (212) Straight stretch (471), road junction (503), other (e.g. roundabout, area for market place, complicated road crossing, etc. 107) 90/110 (54), 70 (205), 20/30/50 (822) Built-up (614), rural (467) Driving school (195), lay-instruction (886) 06:00–18:00 h (788), 18:00–06:00 h (293) Winter (212), spring (251), summer (339), autumn (279) 16–17 (402), 18–24 (193), 25–40 (249), >40 (237) Men (675), women (406)
What?
Seriousness of the accidents Region Places of occurrence Speed limit (km/h) Type of area Type of driver training Time Season Driver’s age (years) Gender
What? Where? Where? Where? Where? Where? When? When? Who? Who?
H.-Y. Berg et al. / Accident Analysis and Prevention 36 (2004) 603–608
605
On the other hand, fog can form during the summer and the spring. The analysis covers all types of vehicles. No selection is made in respect of the question of blame for the accidents. 3.2. Data treatment
Fig. 1. Sweden divided into seven regions.
speed limits. The traffic density is not as high as in the other regions, but it can sometimes increase during the rush hour in built-up areas. The two regions have very often hard winters with constant large quantities of snow, icy roads and little daylight. The winters are long, and the spring, summer and autumn are rather shorter than in the rest of Sweden. The hours of daylight are noticeably longer during the summer in northern Sweden, since the sun never sets above a certain latitude for several months. The temperature in the region fluctuates dramatically between winter and summer, from approximately −30 to +25 ◦ C. The Mälardalen and Stockholm regions are densely populated with several large cities and many sections of road that are subject to speed limits of 50, 70, 90 and 110 km/h. Winter in the two regions is not anything like as hard as in northern Sweden, although they have distinctly wintry weather. Road conditions vary, but the ground is often bare of snow for the entire winter. The roads can sometimes freeze over, resulting in a severely slippery surface. The hours of daylight during the winter are rather longer than in northern Sweden, but shorter during the summer months. The summer is similar to northern Sweden, but arrives rather earlier in the year. The southern regions, western, southeastern and Skåne, are densely populated with larger cities and many roads with speed limits of 50 and 70 km/h. The three regions have milder winters than in the rest of Sweden, although it can sometimes be really cold with heavy snowfalls. The summer is more or less as in the rest of Sweden, but the spring comes earlier than in northern and central Sweden.
To highlight the most typical circumstances of occurrence of the driving practice accidents the coded values of these variables were analysed simultaneously, employing a classification method based on the use of two multivariate analysis techniques applied in sequence: the Factorial Analysis of Correspondence (FAC) and the Hierarchical Ascendant Classification (HAC) (Fénelon, 1981; Greenacre, 1984; Benzécri, 1985). These are data-reduction techniques that focus on the attributes (categories) of variables rather than on any set of variables taken as a whole. The FAC and HAC have been extensively applied in the arena of occupational injuries, and have been described in greater detail in previous studies by Laflamme et al. (1991, 1993) and Laflamme and Blank (1996). The HAC is a classification technique that divides events under investigation into a number of (unempty) classes in such a way that each individual belongs to one and just one class. It maximizes the variance between classes and minimizes that within classes. The inter-class inertia is a measure of the separateness or distinctness of classes (Fénelon, 1981; Greenacre, 1984; Benzécri, 1985). The higher the inter-class variance, the greater is the difference between classes. The intra-class variance is the measure of the internal consistency of a class (i.e. its compactness, or the degree of similarity between the people forming the class). The lower the intra-class variance, the greater is internal class consistency. The class system built using the HAC is based on the coordinates of each event studied on the most significant factors obtained by a FAC into which are included the coded values of the variables of interest. This procedure has the methodological advantage of ensuring that any classification will be performed on a refined data set, and structured on the basis of the most significant interrelations between the categories of the variables of interest. In the current study, the HAC was performed on the first two factors of the FAC, i.e. using the coordinates of the accidents analysed on the first two factorial axes, representing cumulatively 36.1% of the total variance of the core of data.
4. Results During the period 1994–1999, which is the period covered by the study, a total of 1081 accidents during practice driving were reported by the police (Table 1). During this period, driving licences were issued to approximately 497,000 persons. Of the 1081 accidents registered, single accidents were most common, followed by crossing and turning accidents; accidents involving rear impact occupied
606
H.-Y. Berg et al. / Accident Analysis and Prevention 36 (2004) 603–608
the middle ground. Meeting and overtaking accidents, collisions with cycles, mopeds or pedestrians and accidents with hoofed wild animals were less common. A significant number of accidents were also difficult to allocate to the other categories, and these have accordingly been classified as “other accidents”. They accounted for a large proportion of all accidents. This is almost as many as for single accidents. The results of the HAC put into light four typical accident patterns. They are described in the text below, based on the categories of the variables that significantly contributed to the formation of each of them (P > 0.05). These can also be
found in italics in Table 2 where all variables and categories are listed, and where the percentage distribution of the latter within each variable by class and for all accidents aggregated is given. 4.1. Class 1 Rural accidents on straight stretches and high-speed related (n = 306; 28.1% variance): Class 1 is a group of accidents that occurred in rural areas, where the speed limit is most typically of 70 km/h. Accidents that took place on a straight stretch and when meeting and single accidents are
Table 2 Description of the four classes resulting from the HAC Variable
Category
Class 1 (n = 306)
Class 2 (n = 97)
Class 3 (n = 298)
Class 4 (n = 380)
Total (n = 1081)
Type of area
Built-up Rural
10.8 89.2
6.2 93.8
81.5 18.5
87.4 12.6
56.8 43.2
Speed limit (km/h)
20/30/50 70 90/110
46.7 52.9 0.3
15.5 29.9 54.6
99.0 1.0 0.0
97.1 2.9 0.0
76.0 19.0 5.0
Type of accident
Single Meeting Overtaking Rear-end Turning Crossing Vulnerable Hoofed wild animals Other
27.1 24.5 3.6 5.2 7.8 1.6 2.3 0.0 27.8
30.9 1.0 14.4 5.2 1.0 1.0 1.0 37.1 8.2
12.1 0.3 9.7 40.6 6.7 3.7 2.3 0.0 24.5
11.1 0.3 0.3 2.1 29.2 42.4 10.5 0.0 4.2
17.7 7.2 5.1 13.9 14.4 16.5 5.1 3.3 16.8
Place of occurrence
Straight stretch Road junction Other
82.0 16.0 2.0
92.8 6.2 1.0
42.3 28.5 29.2
1.1 95.5 3.4
43.6 46.5 9.9
Driver’s age (years)
16–17 18–24 25–40 >40
47.4 15.0 20.9 16.7
61.9 9.3 14.4 14.4
19.8 23.8 28.9 27.5
36.3 17.6 22.4 23.7
37.2 17.9 23.0 21.9
Type of driver training
Driving school Lay-instruction
9.8 90.2
6.2 93.8
29.5 70.5
18.7 81.3
18.0 82.0
Time
06:00–18:00 18:00–06:00
67.6 32.4
66.0 34.0
79.9 20.1
73.4 26.6
72.9 27.1
Region
North Central Stockholm West Mälardalen southeastern Skåne
0.3 14.4 7.5 24.8 18.0 21.6 13.4
21.6 10.3 5.2 8.2 17.5 29.9 7.2
0.3 3.0 36.2 28.2 7.7 8.4 16.1
2.9 7.9 11.1 24.2 10.5 12.9 30.5
3.1 8.6 16.5 24.1 12.5 15.6 19.6
Season
Winter Spring Summer Autumn
23.9 18.0 32.7 25.5
17.5 22.7 29.9 29.9
10.4 26.5 37.6 25.5
23.9 25.0 25.8 25.3
19.6 23.2 31.4 25.8
Severity
Fatal/severe Minor injury Without injury
17.3 29.4 53.3
11.3 30.9 57.7
6.4 29.2 64.4
5.8 33.7 60.5
9.7 31.0 59.3
Gender
Men Women
63.7 36.3
67.0 33.0
60.7 39.3
61.6 38.4
62.4 37.6
The categories of the variables that significantly contributed to the formation of each class are marked in italics (percentage within each variable).
H.-Y. Berg et al. / Accident Analysis and Prevention 36 (2004) 603–608
over-represented. Other types of accidents in excess proportions are “other” (e.g. car–train, car–other animal than hoofed, car–parked car, etc.). A larger proportion of accidents than expected by chance were severe or fatal. There was also an over-representation of motor vehicle drivers belonging to the younger age category (16 years), of drivers receiving lay-instruction and the time 18:00–06:00 h. Regions over-represented are central, Mälardalen and southeastern. 4.2. Class 2 Rural accidents on straight stretches and high-speed related (n = 97; 12.8% variance): Class 2 is a second group of accidents occurring to a larger extent than expected in rural areas and on straight stretches. The speed limit here is more often high, i.e. 70 or 90/110 km/h, though there is even a slight over-representation of 90/110 km/h. Accidents belonging to the category type hoofed wild animals and single are more frequent than expected. Regions over-represented are the north and southeast parts of Sweden. As in Class 1, motor vehicle drivers belonging to the younger age category (16 years) and drivers receiving lay-instruction are over-represented. Another typical type of accident in this class is overtaking. 4.3. Class 3 Accidents in built-up areas, low-speed related and of the type rear-end (n = 298; 24.4% variance): accidents in Class 3 are most typically of the type “rear-end”, occur where the speed limit is of 20/30/50 km/h, and at “other places” (e.g. roundabout, area for market place, complicated road crossing, etc.). They are most frequently found than expected by chance in built-up areas and in the Stockholm region. Drivers trained in driving schools are over-represented. Other types of accidents over-represented are “other” and overtaking. The three older categories of driver’s age (18–24, 25–40 and >40 years) are more common than expected and so are accidents occurring in the time period 06:00–18:00 h and in the summer season. 4.4. Class 4 Accidents in built-up areas, at road junctions and low-speed related (n = 380; 29.0% variance): accidents in Class 4 took place almost exclusively at road junctions; accordingly, accidents at crossings are over-represented (as well as the types turning and vulnerable but to a lesser extent). Those accidents occurred in greater proportions than expected by chance in built-up areas as well as where the speed limit was 20/30/50 km/h. The region of Skåne is over-represented. The winter season is slightly over-represented.
607
5. Discussion 5.1. Main results Besides the already recognized problem of accident occurrence during driver training (Gregersen et al., 2003), the results of the current study highlight the variety of circumstances under which accidents may occur. In other words, there are a variety of situations that are more at risk for learner drivers—and possibly other drivers—and need to be addressed during the training period. This, in turn, suggests that accident and injury prevention in this group may benefit from a number of targeted strategies, as a complement to already existing general ones. The description of the four classes gives valuable information where, to whom and how to start a preventive work. One general factor that ought to receive consideration in the conception of preventive measures is undeniably the shown relationship between severe or fatal outcome, rural areas, the youngest learners, lay-instruction and single and meeting accidents on roads with high speed limit that comes out clearly in Classes 1 and 2. A similar results have been observed by Gregersen and Nyberg (2002) who compared accidents during driver training among drivers of age 16–17 years and among new drivers of age 18–19 years. Most of the accidents in both groups occurred on roads with a speed limit of 50 km/h. Also, the learner drivers were under-represented in accidents occurring at speed limit 50 km/h, proportionally represented in those occurring at speed limit 70 km/h and over-represented at speed limit 90 or 110 km/h. This is an important source of worry given the well-known relationship between high speed and high accident risk and accident severity (Englund et al., 1998). Preventive efforts should therefore be directed primarily to the reduction of the youngest learners accidents in rural areas on high-speed roads. Targeted information to the youngest learner group and/or special made education of lay-instructors is also a possible way of working with this problem. The results also suggest that a geographical focus and selection of measures is preferable and could lead to higher success in the safety work. In the light of the other classes obtained some other typical scenarios may be described and examples can be provided that have a more specific road type and regional connotation. The results show, in the case of Class 3 for example, that a preventive work with the aim to reduce the number of low speed “rear-end” accidents in built-up areas could improve safety during for driving school instructors and their learner drivers. This preventive work could be carried out by education and by special training of driving school instructors. The outcome of such training will be a safer training on low speed roads in built-up areas. This work could preferably be carried out in cooperation between authorities responsible for road safety and driving school associations and the risk pattern shown in Class 3 could serve as a base for an even more in-depth investigation of the problem.
608
H.-Y. Berg et al. / Accident Analysis and Prevention 36 (2004) 603–608
In that vein, the particular circumstances of rural and urban accidents are well exemplified by the classes. Although it may be difficult to target learners on the basis of where they live because they may practice driving all over the country, it can be important to pay attention to this dimension in the driver training and driver licence regulations. The Swedish National Road Administration constantly works to improve these regulations. An important task in this work is to give the possibility for the learner to practice as much as possible to experience a well-known positive effect on traffic safety after obtaining a driving licence (Gregersen et al., 2000). A very important ethical aspect in this work is to try to minimize the accident risk during practice. This could be done by changes in the regulations regarding when, with whom and where driver training is allowed. Some specific rules for this during different phases of the driver education could be needed. The four accident patterns shown in this study also give many ideas and examples for a preventive work which do not need any changes in the regulations. One idea is to start a development work of an initial education and a “practice manual” for lay-instructors. This measures could serve as guidance for the lay instructor, include examples of practices and information of the accident risk during practice and ways of minimising it. Examples of risky situations could be developed after the four accident patterns shown in this study. 5.2. Limitations It is important to consider that the findings are based on a large number of different approximations. If univariate or bivariate statistics had been used in the study, it would have been necessary to make many more analyses. The precision would perhaps have been greater in every individual analysis, but the general view of the result, would have been, because of too many cross tables, drastically reduced. Correspondence analysis techniques instead indicate the structure of different dependent conditions. the main purpose of correspondence analysis is to reveal the structure of a complex data matrix by replacing the raw data with a more simple data matrix without losing essential information (Clausen, 1988, p. 1). Further, it ought to be reemphasized that the analysis of accident severity is based on information about the most severe injury in each accident irrespective of who was injured. It was thus not possible to determine from the data at hand whether the person who was behind the wheel practicing
driving was injured or whether the same person was legally to blame for any of the accidents. This means that a person practicing driving may have suffered a rear impact accident, or equally that he/she may have driven into the rear of someone else. The same applies to meeting, turning and crossing accidents. A person may have been involved in one of these accidents without having been to blame. As far as single accidents are concerned, the question of blame is clearer. It is probable in this case that the person who was practicing driving was to blame.
Acknowledgements Many thanks to Arne Land, statistician at VTI, and to The Swedish National Road Administration for sharing their accident database which is the base for this study and to Professor Per Bjurulf for his helpful comments on the writing of this article. References Benzécri, J.-P., 1985. Introduction à la classification ascendante hiéarchique d’apres un example de données économiques. Cahiers de l’Analyse des Données. X-3, 279–302. Clausen, S.-E., 1988. Applied correspondence analysis: an introduction. Series on Quantitative Applications in the Social Sciences. Sage, Thousand Oaks, CA, pp. 7–121. Englund, A., Gregersen, N.P., Hydén, C., Lövsund, P., Åberg, L., 1998. Trafiksäkerhet. En kunskapsöversikt, Studentlitteratur, Lund, Sweden. Fénelon, J.P., 1981. Qu’est-ce que l’analyse des données? Lefonen, Paris. Greenacre, M.J., 1984. Theory and Applications of Correspondence Analysis. Academic Press, London. Gregersen, N.P., Nyberg, A., 2002. Privat övningskörning. En undersökning om hur den nyttjas och om dess för—och nackdelar för trafiksäkerheten. VTI rapport 484, Linköping, Sweden. Gregersen, N.P., Berg, H.-Y., Engström, I., Nolén, S., Nyberg, A., 2000. Sixteen years age limit for learner drivers in Sweden—an evaluation of safety effects. Accid. Anal. Prev. 32, 25–35. Gregersen, N.P., Nyberg, A., Berg, H.-Y., 2003. Accident involvement among learner drivers—an analysis of the consequences of supervised practice. Accid. Anal. Prev. 35, 725–730. Laflamme, L., Backström, T., Döös, M., 1993. Typical accidents encountered by assembly workers: six scenarios for safety planning identified using multivariate methods. Accid. Anal. Prev. 25 (4), 399– 410. Laflamme, L., Blank, V.L.G., 1996. Age-related accident risks: longitudinal study of Swedish iron ore miners. Am. J. Ind. Med. 30, 479–487. Laflamme, L., Döös, M., Backström, T., 1991. Identifying patterns using the FAC and HAC; their application to accidents at the engine workshops of an automobile and truck factory. Safety Sci. 14, 13–33.
Typical patterns in road-traffic accidents during driver training An explorative Swedish national study Hans-Yngve Berg a,b,∗ , Nils P. Gregersen a,c , Lucie Laflamme d a
d
Division of Social Medicine & Public Health Science, Department of Health and Society, Faculty of Health Sciences, SE-581 85 Linköping, Sweden b Swedish National Road Administration, SE-781 87 Borlänge, Sweden c Swedish National Road and Transport Research Institute (VTI), SE-581 95 Linköping, Sweden Division of Social Medicine, Department of Public Health Sciences, Karolinska Institutet, SE 171 76 Stockholm, Sweden Received 29 March 2001; received in revised form 7 April 2003; accepted 10 April 2003
Abstract A new law came into force in Sweden on 1 September 1993, which makes instructor-assisted driving practice possible at the age of 16 years instead of the previous 17 years and 6 months. When the age limit was reduced, the possibility that this would lead to more road-traffic accidents (RTA) during driving practice was discussed. The aim of this study was to highlight typical road-traffic accident patterns and to discuss their potential for improved training and targeted prevention. A total of 11 variables (41 categories) descriptive of the 1081 RTA involving novice drivers and police registered during the period 1994–1999 were analysed simultaneously using in turn, two multivariate analysis techniques: the Factorial Analysis of Correspondence (FAC) and the Hierarchical Ascendant Classification (HAC). Four accident classes were identified and quantified, of which the first two were more typical of rural areas: (1) accidents in rural areas on straight stretches and related to speed limit 70 km/h (n = 306); (2) accidents in rural areas on straight stretches and high-speed related (n = 97); (3) accidents in built-up areas, low-speed related and of the type rear-end (n = 298); and (4) accidents in built-up areas, at road junctions and low-speed related (n = 380). Together, these classes point to a variety of opportunities to develop ways of working with targeted prevention. Instead of adopting a general attempt to counteract the relationship between individual variables and accidents, it is possible instead to focus on a whole context and its relationship with its typical accidents and any resulting injuries. This, in its turn, allows greater specificity in the build up of the Swedish licence and training regulations and its corresponding course curriculum. © 2003 Elsevier Ltd. All rights reserved. Keywords: Traffic accidents; Traffic safety; Learner drivers; Driver training; Factorial Analysis of Correspondence; Hierarchical Ascendant Classification
1. Introduction A system under which an individual is permitted to choose for him/herself whether to learn to drive in a driving school and/or through private instruction has been in place in Sweden since the start of the 1920s. A new law came into force on 1 September 1993, which makes instructor-assisted driving practice possible at the age of 16 years instead of the previous 17 years and 6 months. For a person to be able to practice driving, it is necessary for him/her to obtain a learner’s permit regardless of the learner driver’s age. The learner driver can then choose to practice driving either with a lay instructor or with a driving school; a combination of these two possibilities is most common. ∗ Corresponding author. Tel.: +46-243-750-63; fax: +46-243-751-93. E-mail address: [email protected] (H.-Y. Berg).
0001-4575/$ – see front matter © 2003 Elsevier Ltd. All rights reserved. doi:10.1016/S0001-4575(03)00068-X
An instructor’s permit is required in order to act as a lay instructor, who more often than not is one or other of the parents. In order to obtain a permit, an instructor must be aged more than 24 years, have held a full driving licence for a car for five years and not have too many demerit points in the national Swedish driver register. Once the learner driver and the lay instructor have obtained their permits, no further restrictions apply. They can practice driving whenever and wherever they wish. This also applies to the driving school. When the age limit was reduced, it was put forward that this could lead to an increase in the number of accidents during driving practice. Consequently, the registration form completed by police officers when called on an accident site was modified so as to incorporate two additional fields, i.e. with the text “driving school” or “lay instruction”. This, in turn, enabled monitoring of reported police accidents occurring during practice driving.
604
H.-Y. Berg et al. / Accident Analysis and Prevention 36 (2004) 603–608
Gregersen et al. (2003) have calculated the size of the accident problem while practising for a licence for a private car (class B). They also estimated the size of the public health problem and also the road safety problem from a health risk perspective. In total, they analysed 444 driving practice accidents with personal injury during the period 1 January 1994–31 December 2000 among learner drivers of all ages. The results show that in the 444 injury accidents, 827 persons were injured and 22 persons killed, which is equivalent to approximately three persons per year. The younger the driver, the greater the number of injured or killed per accident. From the national licence register, it was found that 229,000 learner drivers had started to practice from the age of 16 during 1994–1996. They were involved in 63 injury accidents. The population of newly licensed drivers aged 18–19 during the same period was 269,000 and as drivers they were involved in 2461 injury accidents. This gives a health risk of 0.28 for learner drivers and 9.15 for newly licensed drivers. The accident risk for learner drivers was 0.47 and 4.77 for newly licensed drivers. Gregersen et al. (2003) concluded in their study that there is a problem with accidents during driving practice which needs to be addressed, especially with regard to the Swedish “Vision Zero”, approved by the Swedish Government, which aims at zero killed or seriously injured in road traffic. They also concluded that the problem is minor compared to the situation after licensing and that the focus in developing safety measures should be on the accident situation of novice drivers rather than learner drivers. However, their analysis also showed, that the driving practice accidents were slightly more severe, both in terms of costs and number of killed and injured per accident. This is not acceptable in the perspective of the “Vision Zero”. In that respect, there is a need for increased knowledge about where, when and how accidents take place during driving practice as well as who are concerned. The data gathered by the registration form completed by police officers makes it possible to identify particular circumstances of road traffic accidents among learner drivers
and with help of this knowledge design suitable countermeasures.
2. Aim The aim of this nationwide study was to increase knowledge about the circumstances of occurrence of accidents during driving practice. Such results are expected to serve as a basis to lay the groundwork for the design of preventive measures—and course curriculum—adapted as much as possible to target situations and groups. 3. Materials and methods 3.1. Data source and selected variables The additions made to the police report form in 1993, combined with the information normally collected made it possible to document where driving practice accidents occur, when, and who is involved in them. Indeed, apart from entering details about the type of driving practice, the form also informs about, e.g. the date, time and place where the accident occurred, the age and sex of the driver, type of accident, speed limit, severity of the accident, and the weather and road conditions at the time. After completion by the police on the site, the data on each form is then entered in a national database under the responsibility of the Swedish National Road Administration (SNRA). For the current study, the 11 variables listed in Table 1 were extracted from the national register and coded into the categories specified in the second column of the table. The seven operating regions into which the SNRA divides Sweden are geographically identified in Fig. 1. Some of them are almost exclusively rural while others are mainly urban. The northern and central regions are sparsely populated with many sections of road subject to 90 and 110 km/h
Table 1 Variables, categories and number of accidents in the analysis Variable
Categories and number in parentheses in each category
Answer
Type of accident
Single (191), meeting (78), overtaking (55), rear-end (150), turning (156), crossing (178), vulnerable (55), hoofed wild animals (36), other (182) Fatal/severe (105), minor injury (335), without injury (641) North (34), central (93), Stockholm (178), west (260), Mälardalen (135), southeastern (169), Skåne (212) Straight stretch (471), road junction (503), other (e.g. roundabout, area for market place, complicated road crossing, etc. 107) 90/110 (54), 70 (205), 20/30/50 (822) Built-up (614), rural (467) Driving school (195), lay-instruction (886) 06:00–18:00 h (788), 18:00–06:00 h (293) Winter (212), spring (251), summer (339), autumn (279) 16–17 (402), 18–24 (193), 25–40 (249), >40 (237) Men (675), women (406)
What?
Seriousness of the accidents Region Places of occurrence Speed limit (km/h) Type of area Type of driver training Time Season Driver’s age (years) Gender
What? Where? Where? Where? Where? Where? When? When? Who? Who?
H.-Y. Berg et al. / Accident Analysis and Prevention 36 (2004) 603–608
605
On the other hand, fog can form during the summer and the spring. The analysis covers all types of vehicles. No selection is made in respect of the question of blame for the accidents. 3.2. Data treatment
Fig. 1. Sweden divided into seven regions.
speed limits. The traffic density is not as high as in the other regions, but it can sometimes increase during the rush hour in built-up areas. The two regions have very often hard winters with constant large quantities of snow, icy roads and little daylight. The winters are long, and the spring, summer and autumn are rather shorter than in the rest of Sweden. The hours of daylight are noticeably longer during the summer in northern Sweden, since the sun never sets above a certain latitude for several months. The temperature in the region fluctuates dramatically between winter and summer, from approximately −30 to +25 ◦ C. The Mälardalen and Stockholm regions are densely populated with several large cities and many sections of road that are subject to speed limits of 50, 70, 90 and 110 km/h. Winter in the two regions is not anything like as hard as in northern Sweden, although they have distinctly wintry weather. Road conditions vary, but the ground is often bare of snow for the entire winter. The roads can sometimes freeze over, resulting in a severely slippery surface. The hours of daylight during the winter are rather longer than in northern Sweden, but shorter during the summer months. The summer is similar to northern Sweden, but arrives rather earlier in the year. The southern regions, western, southeastern and Skåne, are densely populated with larger cities and many roads with speed limits of 50 and 70 km/h. The three regions have milder winters than in the rest of Sweden, although it can sometimes be really cold with heavy snowfalls. The summer is more or less as in the rest of Sweden, but the spring comes earlier than in northern and central Sweden.
To highlight the most typical circumstances of occurrence of the driving practice accidents the coded values of these variables were analysed simultaneously, employing a classification method based on the use of two multivariate analysis techniques applied in sequence: the Factorial Analysis of Correspondence (FAC) and the Hierarchical Ascendant Classification (HAC) (Fénelon, 1981; Greenacre, 1984; Benzécri, 1985). These are data-reduction techniques that focus on the attributes (categories) of variables rather than on any set of variables taken as a whole. The FAC and HAC have been extensively applied in the arena of occupational injuries, and have been described in greater detail in previous studies by Laflamme et al. (1991, 1993) and Laflamme and Blank (1996). The HAC is a classification technique that divides events under investigation into a number of (unempty) classes in such a way that each individual belongs to one and just one class. It maximizes the variance between classes and minimizes that within classes. The inter-class inertia is a measure of the separateness or distinctness of classes (Fénelon, 1981; Greenacre, 1984; Benzécri, 1985). The higher the inter-class variance, the greater is the difference between classes. The intra-class variance is the measure of the internal consistency of a class (i.e. its compactness, or the degree of similarity between the people forming the class). The lower the intra-class variance, the greater is internal class consistency. The class system built using the HAC is based on the coordinates of each event studied on the most significant factors obtained by a FAC into which are included the coded values of the variables of interest. This procedure has the methodological advantage of ensuring that any classification will be performed on a refined data set, and structured on the basis of the most significant interrelations between the categories of the variables of interest. In the current study, the HAC was performed on the first two factors of the FAC, i.e. using the coordinates of the accidents analysed on the first two factorial axes, representing cumulatively 36.1% of the total variance of the core of data.
4. Results During the period 1994–1999, which is the period covered by the study, a total of 1081 accidents during practice driving were reported by the police (Table 1). During this period, driving licences were issued to approximately 497,000 persons. Of the 1081 accidents registered, single accidents were most common, followed by crossing and turning accidents; accidents involving rear impact occupied
606
H.-Y. Berg et al. / Accident Analysis and Prevention 36 (2004) 603–608
the middle ground. Meeting and overtaking accidents, collisions with cycles, mopeds or pedestrians and accidents with hoofed wild animals were less common. A significant number of accidents were also difficult to allocate to the other categories, and these have accordingly been classified as “other accidents”. They accounted for a large proportion of all accidents. This is almost as many as for single accidents. The results of the HAC put into light four typical accident patterns. They are described in the text below, based on the categories of the variables that significantly contributed to the formation of each of them (P > 0.05). These can also be
found in italics in Table 2 where all variables and categories are listed, and where the percentage distribution of the latter within each variable by class and for all accidents aggregated is given. 4.1. Class 1 Rural accidents on straight stretches and high-speed related (n = 306; 28.1% variance): Class 1 is a group of accidents that occurred in rural areas, where the speed limit is most typically of 70 km/h. Accidents that took place on a straight stretch and when meeting and single accidents are
Table 2 Description of the four classes resulting from the HAC Variable
Category
Class 1 (n = 306)
Class 2 (n = 97)
Class 3 (n = 298)
Class 4 (n = 380)
Total (n = 1081)
Type of area
Built-up Rural
10.8 89.2
6.2 93.8
81.5 18.5
87.4 12.6
56.8 43.2
Speed limit (km/h)
20/30/50 70 90/110
46.7 52.9 0.3
15.5 29.9 54.6
99.0 1.0 0.0
97.1 2.9 0.0
76.0 19.0 5.0
Type of accident
Single Meeting Overtaking Rear-end Turning Crossing Vulnerable Hoofed wild animals Other
27.1 24.5 3.6 5.2 7.8 1.6 2.3 0.0 27.8
30.9 1.0 14.4 5.2 1.0 1.0 1.0 37.1 8.2
12.1 0.3 9.7 40.6 6.7 3.7 2.3 0.0 24.5
11.1 0.3 0.3 2.1 29.2 42.4 10.5 0.0 4.2
17.7 7.2 5.1 13.9 14.4 16.5 5.1 3.3 16.8
Place of occurrence
Straight stretch Road junction Other
82.0 16.0 2.0
92.8 6.2 1.0
42.3 28.5 29.2
1.1 95.5 3.4
43.6 46.5 9.9
Driver’s age (years)
16–17 18–24 25–40 >40
47.4 15.0 20.9 16.7
61.9 9.3 14.4 14.4
19.8 23.8 28.9 27.5
36.3 17.6 22.4 23.7
37.2 17.9 23.0 21.9
Type of driver training
Driving school Lay-instruction
9.8 90.2
6.2 93.8
29.5 70.5
18.7 81.3
18.0 82.0
Time
06:00–18:00 18:00–06:00
67.6 32.4
66.0 34.0
79.9 20.1
73.4 26.6
72.9 27.1
Region
North Central Stockholm West Mälardalen southeastern Skåne
0.3 14.4 7.5 24.8 18.0 21.6 13.4
21.6 10.3 5.2 8.2 17.5 29.9 7.2
0.3 3.0 36.2 28.2 7.7 8.4 16.1
2.9 7.9 11.1 24.2 10.5 12.9 30.5
3.1 8.6 16.5 24.1 12.5 15.6 19.6
Season
Winter Spring Summer Autumn
23.9 18.0 32.7 25.5
17.5 22.7 29.9 29.9
10.4 26.5 37.6 25.5
23.9 25.0 25.8 25.3
19.6 23.2 31.4 25.8
Severity
Fatal/severe Minor injury Without injury
17.3 29.4 53.3
11.3 30.9 57.7
6.4 29.2 64.4
5.8 33.7 60.5
9.7 31.0 59.3
Gender
Men Women
63.7 36.3
67.0 33.0
60.7 39.3
61.6 38.4
62.4 37.6
The categories of the variables that significantly contributed to the formation of each class are marked in italics (percentage within each variable).
H.-Y. Berg et al. / Accident Analysis and Prevention 36 (2004) 603–608
over-represented. Other types of accidents in excess proportions are “other” (e.g. car–train, car–other animal than hoofed, car–parked car, etc.). A larger proportion of accidents than expected by chance were severe or fatal. There was also an over-representation of motor vehicle drivers belonging to the younger age category (16 years), of drivers receiving lay-instruction and the time 18:00–06:00 h. Regions over-represented are central, Mälardalen and southeastern. 4.2. Class 2 Rural accidents on straight stretches and high-speed related (n = 97; 12.8% variance): Class 2 is a second group of accidents occurring to a larger extent than expected in rural areas and on straight stretches. The speed limit here is more often high, i.e. 70 or 90/110 km/h, though there is even a slight over-representation of 90/110 km/h. Accidents belonging to the category type hoofed wild animals and single are more frequent than expected. Regions over-represented are the north and southeast parts of Sweden. As in Class 1, motor vehicle drivers belonging to the younger age category (16 years) and drivers receiving lay-instruction are over-represented. Another typical type of accident in this class is overtaking. 4.3. Class 3 Accidents in built-up areas, low-speed related and of the type rear-end (n = 298; 24.4% variance): accidents in Class 3 are most typically of the type “rear-end”, occur where the speed limit is of 20/30/50 km/h, and at “other places” (e.g. roundabout, area for market place, complicated road crossing, etc.). They are most frequently found than expected by chance in built-up areas and in the Stockholm region. Drivers trained in driving schools are over-represented. Other types of accidents over-represented are “other” and overtaking. The three older categories of driver’s age (18–24, 25–40 and >40 years) are more common than expected and so are accidents occurring in the time period 06:00–18:00 h and in the summer season. 4.4. Class 4 Accidents in built-up areas, at road junctions and low-speed related (n = 380; 29.0% variance): accidents in Class 4 took place almost exclusively at road junctions; accordingly, accidents at crossings are over-represented (as well as the types turning and vulnerable but to a lesser extent). Those accidents occurred in greater proportions than expected by chance in built-up areas as well as where the speed limit was 20/30/50 km/h. The region of Skåne is over-represented. The winter season is slightly over-represented.
607
5. Discussion 5.1. Main results Besides the already recognized problem of accident occurrence during driver training (Gregersen et al., 2003), the results of the current study highlight the variety of circumstances under which accidents may occur. In other words, there are a variety of situations that are more at risk for learner drivers—and possibly other drivers—and need to be addressed during the training period. This, in turn, suggests that accident and injury prevention in this group may benefit from a number of targeted strategies, as a complement to already existing general ones. The description of the four classes gives valuable information where, to whom and how to start a preventive work. One general factor that ought to receive consideration in the conception of preventive measures is undeniably the shown relationship between severe or fatal outcome, rural areas, the youngest learners, lay-instruction and single and meeting accidents on roads with high speed limit that comes out clearly in Classes 1 and 2. A similar results have been observed by Gregersen and Nyberg (2002) who compared accidents during driver training among drivers of age 16–17 years and among new drivers of age 18–19 years. Most of the accidents in both groups occurred on roads with a speed limit of 50 km/h. Also, the learner drivers were under-represented in accidents occurring at speed limit 50 km/h, proportionally represented in those occurring at speed limit 70 km/h and over-represented at speed limit 90 or 110 km/h. This is an important source of worry given the well-known relationship between high speed and high accident risk and accident severity (Englund et al., 1998). Preventive efforts should therefore be directed primarily to the reduction of the youngest learners accidents in rural areas on high-speed roads. Targeted information to the youngest learner group and/or special made education of lay-instructors is also a possible way of working with this problem. The results also suggest that a geographical focus and selection of measures is preferable and could lead to higher success in the safety work. In the light of the other classes obtained some other typical scenarios may be described and examples can be provided that have a more specific road type and regional connotation. The results show, in the case of Class 3 for example, that a preventive work with the aim to reduce the number of low speed “rear-end” accidents in built-up areas could improve safety during for driving school instructors and their learner drivers. This preventive work could be carried out by education and by special training of driving school instructors. The outcome of such training will be a safer training on low speed roads in built-up areas. This work could preferably be carried out in cooperation between authorities responsible for road safety and driving school associations and the risk pattern shown in Class 3 could serve as a base for an even more in-depth investigation of the problem.
608
H.-Y. Berg et al. / Accident Analysis and Prevention 36 (2004) 603–608
In that vein, the particular circumstances of rural and urban accidents are well exemplified by the classes. Although it may be difficult to target learners on the basis of where they live because they may practice driving all over the country, it can be important to pay attention to this dimension in the driver training and driver licence regulations. The Swedish National Road Administration constantly works to improve these regulations. An important task in this work is to give the possibility for the learner to practice as much as possible to experience a well-known positive effect on traffic safety after obtaining a driving licence (Gregersen et al., 2000). A very important ethical aspect in this work is to try to minimize the accident risk during practice. This could be done by changes in the regulations regarding when, with whom and where driver training is allowed. Some specific rules for this during different phases of the driver education could be needed. The four accident patterns shown in this study also give many ideas and examples for a preventive work which do not need any changes in the regulations. One idea is to start a development work of an initial education and a “practice manual” for lay-instructors. This measures could serve as guidance for the lay instructor, include examples of practices and information of the accident risk during practice and ways of minimising it. Examples of risky situations could be developed after the four accident patterns shown in this study. 5.2. Limitations It is important to consider that the findings are based on a large number of different approximations. If univariate or bivariate statistics had been used in the study, it would have been necessary to make many more analyses. The precision would perhaps have been greater in every individual analysis, but the general view of the result, would have been, because of too many cross tables, drastically reduced. Correspondence analysis techniques instead indicate the structure of different dependent conditions. the main purpose of correspondence analysis is to reveal the structure of a complex data matrix by replacing the raw data with a more simple data matrix without losing essential information (Clausen, 1988, p. 1). Further, it ought to be reemphasized that the analysis of accident severity is based on information about the most severe injury in each accident irrespective of who was injured. It was thus not possible to determine from the data at hand whether the person who was behind the wheel practicing
driving was injured or whether the same person was legally to blame for any of the accidents. This means that a person practicing driving may have suffered a rear impact accident, or equally that he/she may have driven into the rear of someone else. The same applies to meeting, turning and crossing accidents. A person may have been involved in one of these accidents without having been to blame. As far as single accidents are concerned, the question of blame is clearer. It is probable in this case that the person who was practicing driving was to blame.
Acknowledgements Many thanks to Arne Land, statistician at VTI, and to The Swedish National Road Administration for sharing their accident database which is the base for this study and to Professor Per Bjurulf for his helpful comments on the writing of this article. References Benzécri, J.-P., 1985. Introduction à la classification ascendante hiéarchique d’apres un example de données économiques. Cahiers de l’Analyse des Données. X-3, 279–302. Clausen, S.-E., 1988. Applied correspondence analysis: an introduction. Series on Quantitative Applications in the Social Sciences. Sage, Thousand Oaks, CA, pp. 7–121. Englund, A., Gregersen, N.P., Hydén, C., Lövsund, P., Åberg, L., 1998. Trafiksäkerhet. En kunskapsöversikt, Studentlitteratur, Lund, Sweden. Fénelon, J.P., 1981. Qu’est-ce que l’analyse des données? Lefonen, Paris. Greenacre, M.J., 1984. Theory and Applications of Correspondence Analysis. Academic Press, London. Gregersen, N.P., Nyberg, A., 2002. Privat övningskörning. En undersökning om hur den nyttjas och om dess för—och nackdelar för trafiksäkerheten. VTI rapport 484, Linköping, Sweden. Gregersen, N.P., Berg, H.-Y., Engström, I., Nolén, S., Nyberg, A., 2000. Sixteen years age limit for learner drivers in Sweden—an evaluation of safety effects. Accid. Anal. Prev. 32, 25–35. Gregersen, N.P., Nyberg, A., Berg, H.-Y., 2003. Accident involvement among learner drivers—an analysis of the consequences of supervised practice. Accid. Anal. Prev. 35, 725–730. Laflamme, L., Backström, T., Döös, M., 1993. Typical accidents encountered by assembly workers: six scenarios for safety planning identified using multivariate methods. Accid. Anal. Prev. 25 (4), 399– 410. Laflamme, L., Blank, V.L.G., 1996. Age-related accident risks: longitudinal study of Swedish iron ore miners. Am. J. Ind. Med. 30, 479–487. Laflamme, L., Döös, M., Backström, T., 1991. Identifying patterns using the FAC and HAC; their application to accidents at the engine workshops of an automobile and truck factory. Safety Sci. 14, 13–33.