Accident Analysis and Prevention 81 (2015) 96–106

Contents lists available at ScienceDirect

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

Temporal and spatial patterns of suicides in Stockholm’s subway stations Adriaan Uittenbogaard * , Vania Ceccato Housing and Safety Research Group at CEFIN, School of Architecture and the Built Environment, Royal Institute of Technology (KTH), Drottning Kristinas väg 30, 100 44 Stockholm, Sweden

A R T I C L E I N F O

A B S T R A C T

Article history: Received 4 November 2014 Received in revised form 12 February 2015 Accepted 20 March 2015 Available online xxx

This paper investigates the potential temporal and spatial variations of suicides in subway stations in Stockholm, Sweden. The study also assesses whether the variation in suicide rates is related to the station environments by controlling for each station’s location and a number of contextual factors using regression models and geographical information systems (GIS). Data on accidents are used as references for the analysis of suicides. Findings show that suicides tend to occur during the day and in the spring. They are concentrated in the main transportation hubs but, interestingly, during off-peak hours. However, the highest rates of suicides per passenger are found in Stockholm’s subway stations located in the Southern outskirts. More than half of the variation in suicide rates is associated with stations that have walls between the two sides of the platform but still allow some visibility from passers-by. The surrounding environment and socioeconomic context show little effect on suicide rates, but stations embedded in areas with high drug-related crime rates tend to show higher suicide rates. ã 2015 Elsevier Ltd. All rights reserved.

Keywords: Metro Underground Lethal accidents Public transport systems GIS Sweden

1. Introduction Suicides are rare events. In Sweden, the number of suicides has declined by almost 50 percent since 1980 (2237 suicides). In 2008, 1467 suicides were recorded in Sweden (Karolinska Institutets Folkhälsoakademi, 2010). Yet suicide is one of the main public health issues in Sweden (Rådbo et al., 2005). With around 9 million inhabitants, Sweden registered a yearly average of 1500 official suicides over the past 30 years (Socialstyrelsen, 2006). In 2008, suicides in transportation environments (including trains and other objects in motion) accounted for about 5 to 10 percent of all suicides in Sweden; the main cause is poisoning, at 30 percent (Karolinska Institutets Folkhälsoakademi, 2010: 30). In Stockholm county, 284 suicides were recorded in 2008 – 17 percent of Sweden’s total. The most common method was hanging or suffocation, followed by jumps from a height or in front of a moving object (Fig. 1). Although the figures for suicides committed in transportation environments might be perceived as small, suicide in public places, such as subway stations and rail lines, has a stronger impact on society as a whole than more private means of suicide do. Train suicides lead to high costs as a result of driver and bystander trauma, as well as service delays (O’Donnell and Farmer,

* Corresponding author. E-mail address: [email protected] (A. Uittenbogaard). http://dx.doi.org/10.1016/j.aap.2015.03.043 0001-4575/ ã 2015 Elsevier Ltd. All rights reserved.

1994). Suicides in transportation environments are often covered by the media, which might influence subsequent suicide cases (Sonneck et al., 1994) or produce ‘an uncomfortable image’ of transportation locations by many travellers who use them on a daily basis. Railway suicides in Sweden tend to happen during the daytime in more densely populated areas like Stockholm, with a weak increase during the warmer half of the year (Rådbo et al., 2005). Those findings provide an interesting starting point. However, railway stations differ from subway stations in terms of size and design. Railway stations are open, large structures found anywhere in the country, whereas subway stations are generally underground structures concentrated in large cities. Moreover, although Rådbo et al. checked for time variations of suicide events, they did not investigate the effect of differences in the railway environment on suicide rates. Here, we postulate that the time and the station's environment can affect decisions to commit suicide. The objective of this article is to identify potential temporal and spatial patterns of suicide events in subway stations. The study also assesses whether the environments of these stations and their contexts can help explain variations in suicide rates. Investigation of the temporal and environmental aspects of suicides might ultimately help to prevent them. This study makes a direct contribution to suicide prevention by presenting specific patterns over time and space and identifying the environmental aspects of subway stations that may influence

A. Uittenbogaard, V. Ceccato / Accident Analysis and Prevention 81 (2015) 96–106

80 Hanging & suffocaon

Number of cases reported

70

60 50

Jump from height or in front of moving object

40

Medicaon

30 Shoong 20 Alcohol & drugs

10 0 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Year

Fig. 1. Number of cases for certain methods of suicide (ages 15–24) in Stockholm county from 1997 to 2008. Source: Karolinska Institutets Folkhälsoakademi (2010).

the choice of location for suicide. These results may help transport, police, health, and security authorities to improve safety in public places, such as subway stations, and decrease the societal and psychological costs resulting from suicide in railway environments. In this study, fatal accidents are used as reference for the analysis of suicides at subway stations. Because accidents may have different causes than suicides, they should show a different profile in terms of temporal and spatial patterns. Also, one of the difficulties in working with secondary sources for suicide data, such as the ones used in this study, is untangling ‘suicide’ events from ‘accident’ events. Although transportation agencies differentiate between accidents and suicides, no clear-cut distinction can be established in some cases. In Stockholm county, about 19–25 percent (for man and for women, respectively) of the cases were unsure reports of suicide, which is high compared to the national average of 20 percent (Karolinska Institutets Folkhälsoakademi, 2010). For this reason, this study reports and analyses data on suicides in parallel to data on accidents. Accidents in transportation environments are expected to differ from suicides in both geography and in temporal patterns because they are caused by other factors. Therefore, accidents are not further analysed; they function here as a ‘background’ or ‘benchmark’ for the data on suicides. The study combines different types of geographical information systems (GIS) to geographically map event and socioeconomic data. GIS is also used for buffer and distance analyses regarding the spatial location of events and psychiatric care centres. Statistical software is used to visualise time variations and analyse relationships between suicide rates and the environmental variables. This unique analysis combines both temporal and spatial data to uncover the environmental patterns of suicides in Stockholm’s subway stations. This article first discusses the existing literature on the temporal aspects of suicides in stations and transportation settings. It then focuses on the spatial and contextual aspects of the environment where suicides take place. Section 3 presents the framing of the case study, with data description and methods. The results are presented in Section 4, focusing first on the temporal analysis, followed by the spatial analysis and modelling. The results are discussed in Section 5, and final conclusions are presented in Section 6. 2. Theory and hypotheses of the study Many international studies show an overrepresentation of specific population groups committing suicides. Men are overrepresented; more specifically, younger men between 20 and 40 years old. Swedish studies also show that young men are much more likely to commit suicide than women (Sonneck et al., 1994; Rådbo

97

et al., 2005). In Germany, Ladwig and Baumert (2004) found that most suicides were committed by people between the ages of 20 and 29 and that the median age of the men was lower than that of the women. For people under the age of 49, most suicides were related to the subway, as compared to other methods (Ladwig and Baumert, 2004). Schmidtke (1994) also found that nearly half of the recorded railway suicide events in Germany occurred among men between 20 and 39 years old; for women, no such significant difference was found. Similarly, they found that railway suicides in Germany were more common among younger age groups. In the southern hemisphere, results are similar; in Australia, De Leo and Krysinska (2008) found that railway suicides peak among young men, with men between the ages of 15 and 24 making up nearly 30 percent of all cases. A peak among younger males and females was also observed in a study on the London Underground system; in 40 percent of the cases, the person was between 15 and 34 years old (O’Donnell and Farmer, 1994). Men with alcohol and drug consumption problems also dominate suicide statistics (Krysinska and De Leo, 2008). Although this study does not focus on the characteristics of those who commit suicide, it is important to note that there are features of an individual, such as age and gender, which are particularly relevant to explain general patterns of suicides. 2.1. Temporal factors affecting suicides in transit stations 2.1.1. Daily variations Researchers have shown contrasting findings regarding the timing of suicide. Previous studies in Germany, Australia, and the Netherlands suggest that most suicides are committed during the afternoon hours (Erazo et al., 2004; Ladwig and Baumert, 2004; Houwelingen and Beersma, 2001; Emmerson and Cantor, 1993) as certain events may build up during the day and culminate in suicide in the afternoon, especially around sunset (De Leo and Krysinska, 2008; Houwelingen and Beersma, 2001; Schmidtke, 1994). Furthermore, during the darker hours, a person is less easily detected and may have a better chance of success when committing suicide. In a later study, Erazo et al. (2005) argue that the highest rate of occurrence is at night, whereas Schmidtke (1994), also studying a German case, found the lowest numbers during the night. Contrary to previous evidence, O’Donnell and Farmer (1994) suggest in their study on the London underground that suicide peaks occur outside rush hours at mid-day between 11:00 and 16:00. Rush hours may be avoided by people aiming to commit suicide because of the unwanted crowds, attention, and possible intervention during these hours. These findings are contested by Houwelingen and Beersma (2001), who found a peak around the rush hours in late afternoon in the Netherlands, but still most events happen during the day. 2.1.2. Weekly variations Most studies indicate that weekends (Saturday and Sunday) have fewer suicides as compared to weekdays (De Leo and Krysinska, 2008; Erazo et al., 2004). Several studies in Germany found that most suicides occur at the beginning of the week (Monday to Wednesday) (Erazo et al., 2004; Ladwig and Baumert, 2004; Schmidtke, 1994; Emmerson and Cantor, 1993). It is thought to be a threshold for depressed individuals because it is the start of a new week in which they will, again, see themselves ‘failing’. Yet De Leo and Krysinska (2008) showed that in Australia, most acts of suicide at railways happen on Thursdays and Fridays. They also found a strong relationship between suicides and drug and alcohol use, which may relate suicides to unstructured activities (leisure time, for example) outside ones’ routine.

98

A. Uittenbogaard, V. Ceccato / Accident Analysis and Prevention 81 (2015) 96–106

2.1.3. Seasonal variations Seasonal shifts in suicide levels have been detected in Germany and the Netherlands by Erazo et al. (2004) and Houwelingen and Beersma (2001), respectively. They showed that the peak hours of suicide shift according to season: during the winter suicides tend to occur earlier in the day; during summer, they tend to occur later in the day. This follows the sunset time, which is earlier in winter than in summer. In terms of seasonal changes, other researchers have found a peak in suicide acts during the autumn months (Veress and Szabó, 1980; Deisenhammer et al., 1997; De Leo and Krysinska, 2008; Erazo et al., 2005, 2004; Schmidtke, 1994). Overall, studies have shown that the winter and summer months contain the fewest suicide events (Erazo et al., 2005, 2004; Schmidtke, 1994; Krysinska and De Leo, 2008). More interestingly, in both the northern and southern hemispheres (represented by Germany and Australia, respectively), spring has shown high suicide levels (De Leo and Krysinska, 2008; Erazo et al., 2004). 2.2. Factors affecting suicides in transit environments 2.2.1. The place of suicide Schmidtke (1994) indicates that suicide events occur mainly within stations, yet these attempts result in less serious injuries compared with suicides committed outside along open tracks. However, the author does not provide more details on this. Several factors may contribute to increased suicidal behaviour in particular parts of stations. Studies from London find that within stations, most acts occur at the beginning of platforms (Clarke and Poyner, 1994), where fewer passengers are likely to be found than at the middle of the platform. This provides a more isolated place where a suicidal person can remain unnoticed and evade possible interruption of the suicide attempt, resulting in higher success rates. Contrary to findings of Schmidtke (1994), other researchers indicate that acts of suicide along rail-bound traffic are more likely to be committed along open tracks (Erazo et al., 2005), places outside the stations (Krysinska and De Leo, 2008), and places where tracks are publicly accessible (Clarke and Poyner, 1994). Easy access to tracks provides an uncomplicated way for a person to commit suicide by stepping on the tracks in front of the train. 2.2.2. Traffic density and crowdedness Erazo et al. (2005) present results from Germany suggesting that bigger, main-line stations and stations through which fast train lines run account for more suicides than other station types. Faster trains do have a longer braking stretch and will hit a person at a higher speed, making it more probable for the suicide to be successful. Similarly, De Leo and Krysinska (2008) suggest that in Australia, highly trafficked stations are subject to increased suicide attempts. A station with high traffic provides more opportunities

for a person to throw him- or herself in front of a train than a station that sees only a couple of trains each day. This is also an advantage for spontaneous suicide attempts because no planning is required to be at a station at a specific time when the train passes by. According to psychological theory, central stations exert a specific attraction. According to Lindh (2014), a central location attracts people willing to commit suicide because of several factors. Crowds at a central location allow an audience and can provide the attention desired by the victim. Also, more trains pass through central stations, which make a suicide attempt more likely to succeed. Furthermore, the presence of more people in central stations allows possible last-minute intervention that the suicidal person may unconsciously desire. 2.2.3. The influence of surrounding environment of the station The surrounding neighbourhood of stations and transit systems may also be an influencing factor in the selection of suicide place. De Leo and Krysinska (2008, page 774) argue that individuals with low socioeconomic status show a higher vulnerability to suicide. Long-term economic and social problems may cause certain people to be depressed or may cause an overall social pressure on a person that could result in suicide. Moreover, several studies in the literature indicate that suicides tend to occur close to psychiatric care centres (Krysinska and De Leo, 2008; Kerkhof, 2003; Clarke and Poyner, 1994; O’Donnell and Farmer, 1994). On the basis of the above literature, this study investigates five hypotheses: 1) Suicide patterns at subway stations vary over time: daily,

weekly, and seasonally. 2) The environment at subway stations affects people’s decisions

to commit suicide at a particular station (e.g. CCTV and a good overview of the platform may deter suicide by encouraging surveillance and increasing the chance of intervention, whereas objects that block the view may promote suicide attempts). 3) Suicides tend to happen in main transfer stations, centrally located stations, and stations with high passenger flows. 4) Stations in neighbourhoods of low socioeconomic status show higher rates of suicide because of private and societal pressures. 5) Stations near psychiatric care centres show higher rates of suicide than those placed elsewhere.

3. Stockholm as a case study: data and methods This study uses the Stockholm subway stations as a case study. The city has 897,700 inhabitants (Stockholm City, 2014a) and is located on several islands. This article starts by exploring all the subway stations in the Stockholm system – a total of 100 stations

Table 1 Description of the data sets used in the study. Modelling type

Variable

Description

Time period

Source

Regression modelling (spatial analysis)

Y1 = suicide rate

Suicide rate at subway station: number of suicides/ (passengers a day/1000) Overall event rate at subway station: number of suicides and accidents/(passengers a day/1000) Percentage of overall suicides at each station. Containing individual data on times, place, and type of event. Percentage of overall suicides at each station. Containing individual data on times, place, and type of event.

2000–2013

Stockholm Public Transport Stockholm Public Transport Stockholm Public Transport MTR Stockholm AB

Y2 = all event rate Statistical analysis (temporal analysis)

Suicide events

Statistical analysis (spatial analysis)

Suicide events

2000–2013 2000–2013 2010–2014

A. Uittenbogaard, V. Ceccato / Accident Analysis and Prevention 81 (2015) 96–106

on three lines (green, red, and blue) (Appendix A). According to data from MTR Stockholm AB (42 records over four years), the green line had the highest number of suicides (19) followed by the red and blue line (12 events each). Within the stations, 42 percent of the recorded suicides or accidents took place from the platform, whereas 14 percent of the incidents occurred at the entrance tunnels to the stations (MTR Stockholm AB, 2010–2014). Eighteen stations are located outside of the Stockholm municipality and were not included in the modelling analysis because of data limitations: socioeconomic data was only available for locations within the municipality boundaries. Of the remaining 82 stations, 42 are subterranean, and the rest are above ground. The total subway system carries around 1.2 million passengers per day. The principal station is Central Station (T-Centralen) receiving 214,737 passengers daily. It is located in the central business district of Stockholm and connects to other regional and national railway lines, commuter trains, trams, and busses. The data used comes from the main operators of the public transport system in Stockholm: Stockholm Public Transport (Storstockholms Lokaltrafik, SL) and MTR Stockholm AB (MTRS) (Table 1). The MTRS database consists of suicide records at stations and contains of a total of 42 events, including both attempted and fulfilled suicides (not, however, classified as such). The data contain information about the station, metro line, date, and time, as well as an immediate description of events. The MTRS database covers four years, 2010–2014, and is used in this paper to identify the locations of suicides within stations. The MTRS data has been used as counts of the events for basic statistics in Section 4.1. Total events and percentages of overall suicides for each station were calculated in excel. The SL data set consists of 250 suicidal actions and accidents on the tracks and is used in this paper for statistical analysis and regression modelling in Sections 4.2 and 4.3. The SL data was transformed into rates. Suicide rates per station were calculated by standardizing the number of suicides by passenger flow. The passenger flow information was gathered from an SL database of the number of people going in and out of the station on a daily basis (24 h). The classification of these events was done by SL and is not without problems. The line between accident and attempt is blurred; an accident may be classified as a suicide if no information is recorded, whereas a suicide attempt may be classified as accident. The SL database covers 13 years, from 2000 to 2013, and contains the station name, date, day of the week, time, direction of metro line, and classification as suicide or accident. The suicide category may include both attempted and fulfilled acts. The accident category may include, for instance, a person falling on the tracks, presumably with no intention to commit suicide. The fairly superficial evaluation of the event made by health authorities and/ or police implies that in certain cases, it is difficult to be completely certain whether the event was suicide or an accident. This paper deals with this uncertainty by using accident records as a background analysis for the suicide records. Thus, reported suicides are compared with reported accidents, and possible divergent results are discussed. A regression model using ordinary least squares (OLS) was set up in SPSS statistical software to test the relationship between suicide rates at a station and the influence of the local environment at that station. Both dependent variables – suicide rate and overall event rate – were skewed toward the lower end and therefore, transformed into a normal distribution using the natural log (ln) in order not to violate OLS regression assumptions (Burns and Burns, 2009). The independent variables were selected from current literature and findings from previous studies (see Section 2). Environmental variables were gathered during previous fieldwork using a checklist (Ceccato et al., 2013) to assess the presence of

99

specific features of the station environments. This was done by observation of all the stations’ environments at different times of the day and year. The surveyors looked at whether CCTVs were visibly placed, how crowded the platform was, whether any hiding places could be detected, and the presence of surveillance and visibility opportunities, among other factors. A few variables were added manually, such as distance to psychiatric care centres and whether the station is a main transfer station. The psychiatric care centres were located through a search for “psykiatrisk mottagning” (psychiatric clinic) on the public website Vårdguiden 1177.se. Later, the location was manually geocoded in GIS and buffers of 100 m (about a block) and 500 m around the stations were created (Psych 100 m and Psych 500 m, respectively) (Fig. 2). A buffer larger than 500 m would include areas too far away from the stations and not make a significant contribution to the analysis. In fact, other buffer sizes were tested but did highly correlate (Pearson value >0.6) with many of the other independent variables in the modelling and were therefore omitted. Drug use rates at the stations were collected from police reports and based on a 100-m buffer around the stations. They were standardized by the total population within the buffer. The socioeconomic variables were calculated from a database from Stockholm City and based on 100-m buffers. The modelling strategy has three stages, for both suicide rates and overall rates (suicides and accidents). The first models include all the subway stations and the following independent variables, using environmental and socioeconomic factors: ‘main station’, ‘underground’, ‘crowdedness’, ‘early detection’, ‘overview’, ‘CCTV’, ‘hiding spots’, ‘walls’, ‘drug rates’, ‘distance to city centre’, ‘young male population’, ‘population density’, ‘foreign background’, and ‘average income’, as well as the variables for proximity to psychiatric care centres: ‘Psych 100 m’ and ‘Psych 500 m’. The second and third models uses split models to analyse the city centre and suburbs separately using the same independent variables in order to check whether any differences between these areas would affect the modelling results. The city centre was defined as a circle within a 3-km radius of the central station. A total of 22 stations are located within this area; 78 stations are located in the periphery. In each step, we checked for potential correlation between the independent variables. In the first step, the variables ‘foreign background’ and ‘distance to centre’ were eliminated because of high correlation results. Obviously, in both models representing the city centre and periphery, the variable ‘distance to centre’ had to be excluded because it is already embedded in the geographical selection of the data. Moreover, in the model analysing only the

Fig. 2. Stations and psychiatric care centers (red crosses) in buffers of 100 m (light grey) and 500 m (dark grey). Subway stations (black dots) and lines (grey) are included for context. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.).

100

A. Uittenbogaard, V. Ceccato / Accident Analysis and Prevention 81 (2015) 96–106

city centre, the variables ‘CCTV’ and ‘overview’ had to be excluded because they were highly correlated (Pearson value >0.6) with several other variables (Appendix B).

50%

4. Results

40%

4.1. Temporal analysis

30%

In Stockholm, suicide in subway stations is concentrated during the day and shows a peak in the afternoon hours from 3 pm to 4 pm (Fig. 3). These findings are in line with earlier findings in Germany (Erazo et al., 2004) and Australia (Emmerson and Cantor, 1993). In the Netherlands, too, Van Houwelingen and Beersma (2001) observed a major drop in suicidal acts during the night, which they suggest is related to most people sleeping at night and to the traffic intensity being lowest during night hours. However, based on German findings, Schmidtke (1994) argues that individuals who wish to commit suicide do look for the hours around dusk; this may in turn explain the peak around 4 pm in Stockholm. This study found no strong distinction between suicide events on weekdays and on weekends in Stockholm’s subway stations (Fig. 4), as shown in previous studies (De Leo and Krysinska, 2008; Erazo et al., 2004). The number of suicides was slightly higher on weekdays (57%) versus weekends (43%), which contradict the findings of German studies (Erazo et al., 2004; Ladwig and Baumert, 2004; Schmidtke, 1994) and the Australian study by Emmerson and Cantor (1993). Broken down by days of the week, Wednesdays and Fridays showed a slightly higher percentage of suicidal acts and attempts than other days in subway stations. This finding coincides with people’s social unstructured activities, for example, weekend days are common times for evenings out, and people engage more often in leisure activities on weekends than on weekdays, when most people go to work or school. In Sweden, however, Wednesdays may also attract young people and students to go out, a phenomenon informally called ‘Small Saturdays’ (LillLördag). Although suicides do not show a variation during the week, a clearer pattern emerges when accident reports are used as a background reference (Figs. 3 and 4). Accidents may include fatal accidents that are not deliberate suicides. Accidents at subway stations are most frequent during the evening and night and during weekends. These findings indicate a link to passengers’

1 24 10%

2

23

3 8%

22

4

60%

20% 10% 0% Weekday

Weekend

SUICIDE 2000-2013 (n=162)

ACCIDENT 2000-2013 (n=88)

Fig. 4. Weekly variations of suicide events at Stockholm’s subway stations.

unstructured social activities, which are often carried out during evenings and weekends and during which people are at the highest risk of having an accident with a possible fatal outcome at transit systems. There are signs of seasonal variation in suicide in the Stockholm subway stations (Fig. 5). The spring months (March–May) show the highest incidence of suicides and the autumn months, the lowest. These are similar to findings in Germany, where Erazo et al. (2004) found peaks in April and September, and in Australia, where De Leo and Krysinska (2008) observed a seasonal pattern in which half of all suicides occur in both spring (southern hemisphere: September–October) and autumn (southern hemisphere: March–May). In contrast to those findings, Stockholm shows the lowest number of suicides during the autumn. Nevertheless, most accidents (possible attempts) occur during the autumn and winter in Stockholm. The results from the German study can be debated, because it uses both fatal and non-fatal suicide accidents. Similarly, it is unclear whether suicides and accidents were split in the Australian study. Several other studies have shown a peak during autumn months for suicides but do not distinguish between attempted and successful suicides. Whilst ‘accidents’ happen mostly during the colder months of the year in Stockholm, for suicides, there is no apparent variation throughout the seasons (Fig. 5). Note that the data analysed in the study relates only to subway stations and may therefore reflect seasonal changes in the use of the public subway system. Inhabitants make more use of the subway system during the colder winter months than during the summer when more people use bikes or are on vacation (AB Storstockholms Lokaltrafik, 2010). Thus, the increase in accident in the subway stations during

6% 21

5 4%

20

6

2%

19

7

0%

35% 30% 25%

18

8 20%

17

9 16

10 15

11 14

12 13

SUICIDE 2000-2013 (n=162)

15% 10% 5%

0% Winter

ACCIDENT 2000-2013 (n=88)

Fig. 3. Hourly variations of suicide events at Stockholm’s subway stations.

Spring SUICIDE 2000-2013 (n=162)

Summer

Autumn

ACCIDENT 2000-2013 (n=88)

Fig. 5. Seasonal variations of suicide events in Stockholm’s subway stations.

A. Uittenbogaard, V. Ceccato / Accident Analysis and Prevention 81 (2015) 96–106

the winter may simply be a result of having more passengers and icy, slippery platform areas. This finding corroborates findings of the studies in Germany and Australia that show peaks in spring and autumn, and underlines the thought that it is not easy to

101

disentangle the differences between a suicide and a lethal accident in a transit environment. We expected that suicide rates in the central and peripheral areas would peak at different times of the day and year. When

Fig. 6. Rates for suicides and accidents on each line of the Stockholm subway system (2000–2013) (For interpretation of the references to colour in the text, the reader is referred to the web version of this article.).

102

A. Uittenbogaard, V. Ceccato / Accident Analysis and Prevention 81 (2015) 96–106

Table 2 Results of OLS regression models for the total subway system, city centre, and peripheral areas of Stockholm. Dependent variable: Log (suicide rate) (n = 157) Independent variables

(model MST) Total SystemXXXXXR2 = 56.2

X1 = environment at Main station ( 0.331) XXXXXUnderground station ( 1.201) XXXXXCrowdedness ( 1.749)a XXXXXEarly detection (0.735) XXXXXView (0.810) XXXXXCCTV ( 0.090) XXXXXHidden spots ( 1.016) XXXXXWalls (1.729)a X2 = factors around Drug rates(1.737)a XXXXXPsych_100 m (0.940) the station XXXXXX3 = socioeconomic Young males ( 0.788) XXXXXPopulation factors density (0.172) XXXXXAvIncome ( 0.856)

(model MSC) XXXXXCity centre XXXXXR2 = 80.1

(model MSP) XXXXXPeriphery XXXXXR2 = 62.1

Main station (0.182) XXXXXUnderground ( 1.483) XXXXXCrowdedness ( 1.209) XXXXXEarly detection ( 0.374) XXXXXXXXXX- XXXXXHidden spots ( 1.083) XXXXXWalls (1.168) Drug rates(2.385)a XXXXXPsych_100 m (0.551) XXXXXPsych_500 m ( 0.436) Young males ( 0.372) XXXXXPopulation density (0.813) XXXXXAvIncome ( 1.114)

Main station ( 0.926) XXXXXUnderground ( 0.062) XXXXXCrowdedness ( 0.878) XXXXXEarly detection(1.956)a XXXXXView (1.031) XXXXXCCTV (0.480) XXXXXHidden spots (0.629) XXXXXWalls (0.196) Drug rates ( 0.207) XXXXXPsych_100 m ( 0.602) Psych_500 m ( 0.341) Young males ( 0.460) XXXXXPopulation density XXXXX 0.549) XXXXXAvIncome ( 0.118)

XXXXXDependent variable: Log (all events (suicides + accidents) rate) (n = 202) Independent variable

(model MAT) XXXXXTotal system XXXXXXR2 = 51.5

(model MAC) XXXXXCity centre XXXXXR2 = 86.8

(model MAP) XXXXXPeriphery XXXXXR2 = 63.2

X1 = environment at station

Main station ( 0.625) XXXXXUnderground ( 1.129) XXXXXCrowdedness ( 2.106)b XXXXXEarly detection (1.336) XXXXXView (0.804) XXXXXCCTV (0.199) XXXXXHidden spots ( 0.558) XXXXXWalls (1.596) Drug rates (1.681) XXXXXPsych_100 m (1.266) XXXXX-

Main station (0.245) XXXXXUnderground (-1.262) XXXXXCrowdedness ( 2.345)a XXXXXEarly detection ( 0.632) XXXXX-XXXXX-XXXXXHidden spots ( 1.113) XXXXXWalls (1.060)

Main station (0.114) XXXXXUnderground ( 0.295) XXXXXCrowdedness ( 0.906) XXXXXEarly detection (2.177)b XXXXXView (1.007) XXXXXCCTV (0.688) XXXXXHidden spots ( 0.052) XXXXXWalls (0.808) Drug rates (0.091) XXXXXPsych_100 m (0.442) XXXXXPsych_500 m ( 1.492) Young males (0.299) XXXXXPopulation density ( 0.848) XXXXXAvIncome ( 0.220)

X2 = factors around the station X3 = socioeconomic factors

Young males ( 0.181) XXXXXPopulation density (0.752) XXXXXAvIncome ( 0.509)

Drug rates (3.374)b XXXXXPsych_100 m (0.787) XXXXXPsych_500 m ( 0.063) Young males ( 0520) XXXXXPopulation density (1300) XXXXXAvIncome ( 1402)

Results from regression models showing all variables1, with t-values between brackets and significant variables in bold-italic. a 90% significance level, b 95% significance level.

tested, results from Stockholm subway stations show that there were no major differences between central and peripheral stations regarding timing of suicides and accidents. Of all recorded suicides, 34 percent happened in the centre during the day and 16 percent in the centre at night, whereas 32 percent occur in the periphery during the day and 18 percent in the periphery at night. In order to refine the analysis of suicides in subway stations, findings for counts of events (suicides and accidents) and rates per passengers are further discussed in Section 4.2. 4.2. Spatial analysis In line with findings in Germany (Erazo et al., 2005), and confirming hypothesis 1, Stockholm shows a concentration of

1 Environmental variables, recorded during fieldwork 2010–2011 (Ceccato, 2013): Main station = station is a main transfer station in the Stockholm public transport system. Underground = station’s platform is located underground. Crowdedness = platform is crowded, with more than 20 passengers at the moment of visiting. Early detection = possibility for surveillance of the whole platform (how well other passengers can see you). View = the platform has a good overall view. CCTV = CCTVs are present and easily visible. Hidden spots = places to hide are present at the platform. Walls = barriers between the two sides of the platform make a nontransparent division. Drug rates = drug offences per 1000 inhabitants within a 100-m radius of the station (2008). Psych_100 m = number of psychiatric care centers within a 100-m radius of the station. Psych_500 m = number of psychiatric care centers within a 500-m radius of the station. Socio-economic variables per 1000 inhabitants (2008): Young males = percentage of young males living within a 500-m radius of the station. Population density = inhabitants per square kilometer within a 500-m radius of the station. AvIncome = average income for population living within a 500-m radius of the station.

suicide acts at the central transfer subway stations. The green line includes several of the busiest subway stations in the system, as well as the most centrally located stations. The station with the single highest number of suicide events is T-Centralen, where SL data shows 23 classified suicides over the past 13 years, followed by Fridhemsplan,Slussen, and other large, central stations. However, after calculating the rates (by standardizing suicides by passenger flow at a particular subway station), the three central stations (T-centralen, Fridhemsplan and Slussen) lose their top position, which is taken up by stations along the green and red lines’ southern legs (Fig. 6). Peaks are found at stations in the Southern suburbs instead. Along the green line, high suicide rates can be observed around Tallkrogen and the southern end station Farsta (Fig. 6). The red line too shows a peak in the southern suburbs, around stations Fittja and Västertorp (Fig. 6). Similarly, along the blue line, a concentration can be observed in the northern suburbs around stations Husby and Hjulsta (Fig. 6). A visual interpretation of these concentrations of high rates of suicides in subway stations located in suburban neighbourhoods, such as Husby, Fittja, and Farsta, indicate a relationship of suicides to the socioeconomic conditions of the population in these areas, as proposed in hypothesis 4. These findings are reinforced by the expected signs of the variable “average income” found in the modelling results in Table 2 (although not significant). In Stockholm, these are neighbourhoods with a lower socioeconomic status; they have a lower average income and a higher percentage of inhabitants with a foreign background when compared to the rest of the city (Stockholm City, 2014b). As submitted by Krysinska and De Leo (2008, page 774), “a remarkable proportion of the victims was unemployed or not in the labour force”. Poor

A. Uittenbogaard, V. Ceccato / Accident Analysis and Prevention 81 (2015) 96–106

103

Fig. 7. Environmental features at Stockholm’s subway stations. A gate prevents trespassing at the end of platforms (left); empty platform provide anonymity and opportunities (middle); and walls create locales out of sight from other parts of the platform (right).

socioeconomic situations, such as unemployment, may motivate certain individuals see suicide as an alternative for long term depression (Nordt et al., 2015). Accident rates are more concentrated around the city centre, particularly in stations between the centre and southern periphery – for instance, in Sandsborg and Enskede Gård (green line) and Bredäng (red line) (Fig. 6). The concentration of accident rates at stations in the central-south area can be associated with the increasing popularity of these neighbourhoods during recent years, which has increased the number of passengers, as well as the fact that most stations are located above ground, which poses a greater risk for accidents on the platforms, particularly during winter when it is icy and snowy. It is interesting to note that the geography of suicide rates and accident rates are not the same: higher rates of suicide are observed in the periphery (at stations such as Hjulsta and Farsta) while higher rates of accidents are found in popular suburbs adjacent the city centre (at stations such as Västertorp and Tallkrogen). This indicates that the different environments in these type of areas influences suicide events accordingly.

one for subway stations located in the periphery (further than 3 km from the central station). The sole significant aspect influencing suicide rates only in the city centre is the rate of drug related crimes in the immediate surroundings of the stations, explaining over 80 percent of the variation in the model (‘MSC’, Table 2). Similar results were found for the model accidents plus suicides rates (model ‘MAC’, Table 2). In the periphery model, contrary to what was initially hypothesized, variables indicating the visibility show a significant positive relationship to suicide rates (models ‘MSP’ and ‘MAP’, Table 2). Thus, better surveillance opportunities, early detection, were linked to higher suicide rates at the peripheral stations (models ‘MSP’ and ‘MAP’, Table 2). Although the reasons for these results are unclear, more interestingly in these models is to highlight the signs of several remaining variables that ended up not been significant but still showed ‘expected signs’ in relation to suicides. For instance, presence of walls in the platform of subway stations (+) and crowdedness of the subway stations ( ) and the number of psychiatric care centres within a 100-m radius of the station (+).

4.3. Modelling the effect of station environment on suicides

5. Discussion of the results

Results from the regression models show that the subway stations’ environment affects suicide rates (Table 2). The models for the entire subway system show that the presence of walls, dividing the two sides of the platform, is positively related to higher suicide rates (model ‘MST’, Table 2). Dividing the sides of a platform with a wall may decrease visibility and surveillance opportunities and may create a more isolated place to commit suicide, as there may be no onlookers who could prevent it (Fig. 7). Higher suicide rates at subway stations are also associated with higher rates of drug related crimes detected around the stations (model ‘MST’, Table 2). In the case of Stockholm the mechanisms linking suicides and drug related crimes are not yet known but similar results are found in other studies elsewhere (e.g. Krysinska and De Leo, 2008). Moreover, crowded platforms seem to be related to fewer acts of suicide at the stations (Fig. 7). Overall, these three factors—crowdedness, platform walls, and drug use—explain a total of 56 percent of the variation in the ‘MST model’. This variable crowdedness becomes stronger in the model using all events (suicide and accidents), but the two other factors are no longer significant (model ‘MAT’, Table 2). The location of the stations in the city is hypothesized to influence the likelihood of suicides. Therefore, the suicide data at subway stations was split into two models: one for subway stations located in the city centre (within 3 km of the central station) and

The findings of this study confirm the hypotheses that suicide rates at subway stations vary over time and space. Stockholm subway stations show a pattern of suicides that follows some of the international trends: suicides occur mainly during the day, mostly afternoon hours, and in some of the spring months. However, in contrast to other studies (Erazo et al., 2004; Ladwig and Baumert, 2004; Schmidtke, 1994; Emmerson and Cantor, 1993), no clear distinction was found between weekdays and weekends. Although more events are registered in the subway stations of the inner city areas, the highest rates of both suicides and accidents per passenger flow were found in the a few peripheral stations of Southern Stockholm. The difference in nature of ‘suicides’ and ‘accidents’ is indicated by the fact that the stations that show highest rates in the periphery are not the same in all three subway lines. With regards to suicides, peripheral stations are preferred as they have fewer passengers, thereby allowing higher anonymity and less chance for intervention. Modelling results show strong regression models with a few significant variables across all models (Table 2). This finding indicates a homogeneous effect of the internal and external environment among Stockholm’s subway stations, with little difference between the stations. A number of features of the station environments were found to be significantly associated with suicide rates. As predicted in the

104

A. Uittenbogaard, V. Ceccato / Accident Analysis and Prevention 81 (2015) 96–106

second hypothesis, walls on the subway platforms are related to increased suicide rates (model ‘MST’, Table 2). They limit visibility and increase seclusion, making it easier to stay unnoticed during a suicide attempt. Other variables, however, such as CCTV, overview and early detection, do not confirm this assumption since they were not significant, as suggested by other studies (Kerkhof, 2003; O’Donnell and Farmer, 1994). Suicide rates and all event rates (suicides and accidents) are also significantly related to less crowded platforms (model ‘MST’ and ‘MAT’, Table 2), which go against the third hypothesis of this study but confirm the suggestion of the importance of seclusion. Similarly, in the models for the peripheral subway stations, better opportunities for surveillance are similarly significantly associated with higher suicide rates (model ‘MSP’ and ‘MAP’, Table 2). This is an unexpected finding but it may be as suggested in the literature related to the individual’s unconscious need to be rescued at the last moment or to receive attention (Lindh, 2014). Contrary to what was found by Krysinska and De Leo (2008), most of the locational factors or features of the surrounding areas of the subway stations, (e.g. being a main line station or being close to psychiatric care centres), did not show a significant impact on suicide rates at subway stations in Stockholm. The lack of effect of distance to psychiatric hospitals, may be related to the type of healthcare provided to psychiatric patients in Sweden (more often as out patients) compared with other countries. Previous studies have found a relationship between suicide rates and socioeconomic factors (such as the average income) near the stations. De Leo and Krysinska (2008), for example, suggested that socioeconomic status has an effect on suicide events (they found that most of the victims were unemployed). In Stockholm, the expected socioeconomic factor (average income of residents in the immediate stations’ surroundings) did show a negative expected relationship to the suicide rates in subway stations but it was not significant. A potential reason for these results may be the fact that some of those who commit suicide in the subway stations may not live in the same area, they travel to commit suicide. The only contextual variable that shows a significant impact on suicide rates is drug related crimes of the areas where the subway stations are located. As indicated by De Leo and Krysinska (2008) suicides are commonly associated with high levels of intoxication of the victims. This fact explains why high rates of suicides and accidents are related to high rates of drug related crimes in the city centre of Stockholm (models ‘MST’, ‘MSC’ and ‘MAC’, Table 2), but not in the models that cover these two events in peripheral subway stations (where alcohol selling premises and/or drug points are rarer compared with inner city areas) (Ceccato et al., 2002). 6. Final considerations This article sets out to investigate the potential temporal and spatial patterns of suicide events in subway stations in Stockholm, Sweden. The study also assesses whether the variation of suicide rates relates to the environments of these stations. This is done by taking into account the subway station’s location and a number of contextual factors using regression models and GIS. Unintended fatal accidents (not suicides) that happen in the subway stations are used as reference for the analysis of suicides. The results show that suicides (and accidents) in subway stations are dependent on both time and location. Certain times of the day, particularly afternoons, experience more suicides than others. Suicides are also more likely to happen during spring months while a peak of accidents take place in the autumn months. Data permitting, further testing would be needed to confirm these signs of seasonal suicide variations.

Specific subway stations were identified as preferred places to commit suicide. If only total counts are considered, three central stations stand out as the ‘chosen’ ones, namely, T-centralen, Fridhemsplan and Slussen. Note that these three stations belong to large transportation hubs – all three located in well known, often touristic parts of the city. The choice for these stations can be an indication of opportunities to commit suicide (more trains passing by at those stations), to be rescued (more crowded and thus higher potential for intervention) or to choose a subway station that is located in a known remarkable and aesthetically pleasant area, equivalent to, for instance, the Golden gate in the USA (Blaustein and Fleming, 2009). However, when suicides are considered in relation to total passengers flow, a few other stations tend to show highest rates of suicides and they are no longer part of the inner city centre, rather they are located in the Southern part of Stockholm (Hjulsta – blue line, Västertorp – red line and Enskede gård/Tallkrogen – green line). A least in the model for the inner city subway stations, drug related crime rates around the stations influence suicide rates at the stations. As previously suggested, the link between suicide and drugs may be related to the fact that central areas concentrate both alcohol selling premises and/or drug points. However, future research, should investigate data about the suicide victims, such as age, gender, and location of residence, and their whereabouts before the event will be integrated into the analysis of suicides in subway stations. Someone attempting suicide might pick a certain station just because it is near his or her home or on the daily route from/to work. An alternative is to investigate the choices for location, time and reasons from individuals that have survived attempts of suicides in subway stations. As in any study of this kind, the analysis is not free of limitations. The rarity of the events means that the results should be interpreted with caution. Even when collapsing the data for several years, the total number of suicides is less than 250 cases, of which just over 200 were usable in the modelling. A longer data series would certainly provide a more robust basis for analysis. Future studies may benefit from looking into different types of modelling, including Poisson regression models, which are potentially more appropriate for rare events such as suicides (see also Newton et al., 2014, for an alternative modelling approach). The databases used in this study contained information about both cases of suicides and cases of accidents. There is of course subjectivity in the judgement of the expert when associating a case as either an accident or a suicide. Data permitting, a follow-up study would benefit from splitting the data more precisely between suicide and accident reports in order to attempt to better disentangle the differences in spatial and temporal patterns between the two. This study indicates that there is a difference between the two categories, but for preventive actions to be in place, a deeper long-term analysis of these events is required. Furthermore, this paper only investigates subway stations, excluding railway stations, which are often larger, composed of open structures, such as tracks and bridges, where suicides may also happen. However, subway stations impose constraints to suicide: they are commonly located underground or separated from the public space. The platform is often the only place that suicide can happen. In the particular case of Stockholm, the physical environments of the subway stations follow a standard design for the platform and the area where the trains arrive – often the location of the suicides. Therefore, differences in subway suicide rates are the result of small details in the physical design of these stations, combined with opportunities for surveillance and surrounding areas. It should be kept in mind that the results presented here are context dependent; in other words, it fits Stockholm’s subway

A. Uittenbogaard, V. Ceccato / Accident Analysis and Prevention 81 (2015) 96–106

stations. Findings should be re-evaluated when applying them to other transport systems or country contexts. A follow-up study should take into account different types of stations in Stockholm, in order to better assess the effect of differences in physical environment on suicide events. Such stations could include the railway stations of regional commuter trains and national train services that have open access with bridges and high-speed trains. In spite of these limitations, this paper shows that suicides do not take place randomly in either space or time in subway stations. Although not in detail, the knowledge of these patterns can allow transportation and security companies to further consider specific suicide prevention measures when and where most suicides happen – information that is fundamental for suicide prevention and consequently directly contribute to safer trip in subway systems in Stockholm.

105

Acknowledgements We thank the Swedish Transport Administration (Trafikverket) for funding this research project. We are also grateful for the support of Christer Persson at Centre for Transport Studies, Royal Institute of Technology, in the early stages of this project. We thank Jan Ekström and Kent Martin from Stockholm Public Transport and Bengt Carlsson from MTR Stockholm AB for providing data for the analysis. Thanks also go to experts from Stockholm municipality for providing the digital maps for the spatial analysis in this study. Appendix A. Map of Stockholm’s subway system and lines (source: AB Storstockholms Lokaltrafik, 2010) (For interpretation of the references to colour in this text, the reader is referred to the web version of this article.).

106

A. Uittenbogaard, V. Ceccato / Accident Analysis and Prevention 81 (2015) 96–106

Appendix B. Correlation matrix of variables used in regression models. 1

2

3

1.Main station

–

2. Underground

0.156

–

3.Crowded platform

0.339***

0.381***

4.Early detection 5.View

0.126

0.330***

0.141

***

0.512

4

5

6

7

8

9

0.405***

–

0.245**

0.246**

0.004

0.115

0.007

0.17

–

7.Hiding spot

0.173

0.511***

0.479***

0.341***

0.375***

0.026

–

8.Walls

0.270**

0.553***

0.326***

0.214

0.505***

0.106

0.406***

9.Drug Rates Police

0.051

0.094

0.104

0.004

0.017

10.Psych 100 m

0.065

0.044

0.145

0.219**

11.Psych 500 m

0.008 0.239**

0.181 ***

0.271

0.326 ***

12.Distance city

0.355

13.Foreign background

0.145

14.Male_young pop

0.174

**

0.326 0.133 0.139

0.499

0.005 0.147

*** ***

0.480

0.159 0.129

0.153

0.124

0.192*

0.047

-0.073 ***

0.344

***

0.396

0.044 0.196*

15. Pop_density

0.093

0.430***

0.506***

0.219**

0.422***

16.Average_income

0.296***

0.201*

0.252**

0.063

0.202*

*

12

13

14

15

16

–

0.052

**

11

–

6.CCTV

0.011

10

0.095 0.088 0.052

***

0.358

**

0.273 0.077

– –

0.202 0.366

***

0.012

0.095

–

0.073

0.425***

*

–

0.215

0.265**

0.625***

–

0.215*

0.154

0.282**

0.653***

–

0.282**

0.324***

0.579***

0.549***

–

0.003

0.101

0.173

0.135

0.132

0.325***

0.306***

0.028

0.178

0.484***

0.404***

0.103

0.229**

–

0.192*

0.251**

0.13

0.132

0.429***

0.740***

0.717***

0.486***

0.175

0.084

–

p < 0.100.

**

p < 0.050.

***

p < 0.010.

References AB Storstockholms Lokaltrafik, 2010. Fakta om SL och länet 2010. (Facts about SL and the county 2010). https://www.sll.se/Global/Verksamhet/Kollektivtrafik/Fakta %20om%20SL%20och%20l%C3%A4net/fakta-om-sl-och-lanet-2010.pdf (accessed 9.02.15). Blaustein, M., Fleming, M., 2009. Suicide from the Golden Gate Bridge. Am. J. Psychiatry 166 (10), 1111–1116. doi:http://dx.doi.org/10.1176/appi. ajp.2009.09020296. Burns, R., Burns, R., 2009. Business Research Methods and Statistics Using SPSS. Chapter 24 Logistic Regression. http://www.uk.sagepub.com/burns/default. htm (accessed 26.10.14). Ceccato, V., Haining, R., Signoretta, P., 2002. Exploring crime statistics in Stockholm using spatial analysis tools. Ann. Assoc. Am. Geogr. 22, 29–51. Ceccato, V., 2013. Moving Safely: Crime and Perceived Safety in Stockholm’s Subway Stations. Lexington Books, Plymouth, United Kingdom. Ceccato, V., Uittenbogaard, A.C., Bamzar, R., 2013. Security in Stockholm’s underground stations: the importance of environmental attributes and context. Secur. J. 26, 33–59. Clarke, R.V., Poyner, B., 1994. Approaches to prevention – preventing suicide on the London underground. Soc. Sci. Med. 38, 443–446. Deisenhammer, E.A., Kemmler, G., De Col, C., Fleischhacker, W.W., Hinterhuber, H., 1997. Eisenbahnsuizide und -suizidversuche in Österreich von 1990–1994 (railroad suicides and attempted suicides in Austria 1990–1994). Nervenarzt 68, 67–73. De Leo, D., Krysinska, K., 2008. Suicidal behaviour by train collision in Queensland, 1990–2004. Aust. N. Z. J. Psychiatry 42, 772–779. Emmerson, B., Cantor, C., 1993. Train suicides in Brisbane, Australia, 1980–1986. Crisis 14, 90–94. Erazo, N., Baumert, J.J., Ladwig, K.-H., 2004. Regionale und örtliche Verteilungsmuster von Bahnsuiziden (Regional and local clusters of railway suicides). Nervenarzt 75, 1099–1106. Erazo, N., Baumert, J.J., Ladwig, K.-H., 2005. Factors associated with failed and completed railway suicides. J. Affective Disord. 88, 137–143. Houwelingen, C.A.J., Beersma, D.G.M., 2001. Seasonal changes in 24-h patterns of suicide rates: a study on train suicides in the Netherlands. J. Affective Disord. 66, 215–223. Kerkhof, A.J.F.M., 2003. Railway suicide: who is responsible? Crisis 24, 47–48. Krysinska, K., De Leo, D., 2008. Suicide on railway networks: epidemiology, risk factors and prevention. Aust. N. Z. J. Psychiatry 42, 763–771.

Ladwig, K.-H., Baumert, J.J., 2004. Patterns of suicidal behaviour in a metro subway system. Eur. J. Public Health 14, 291–295. Lindh, S., 2014. Går det att förebygga suicider genom fysisk planering? En studie om implementering av suicidprevention genom fysisk planering i Stockholms stad. (Is it Possible to Prevent Suicides Through Physical Planning? A Study on the Implementation of Suicide Prevention Through Physical Planning in Stockholm City.) Master’s Thesis in Human Geography. Stockholm University, Stockholm, Sweden. Karolinska Institutets Folkhälsoakademi, 2010. Självmord i Stockholms län och Sverige 1980–2008. (Suicide in Stockholm county and Sweden 1980–2008.). Stockholm, Sweden p. 30. MTR Stockholm AB, (2010–2014). Database acquired from MTR Stockholm AB 2014. Newton, A., Partridge, H., Gill, A., 2014. Above and below: measuring crime risk in and around underground mass transit systems. Crime Science 3, 1–14. Nordt, C., Warnke, I., Seifritz, E., Kawohl, W., 2015. Modelling suicide and unemployment: a longitudinal analysis covering 63 countries, 2000-11. Lancet Psychiatry doi:http://dx.doi.org/10.1016/S2215-0366(14)00118-7. O’Donnell, I., Farmer, R.D.T., 1994. The epidemiology of suicide on the London Underground. Soc. Sci. Med. 38, 409–418. Rådbo, H., Svedung, I., Andersson, R., 2005. Suicides and other fatalities from trainperson collisions on Swedish railroads: a descriptive epidemiologic analysis as a basis for systems-oriented prevention. J. Saf. Res. 36, 423–428. Schmidtke, A., 1994. Suicidal behaviour on railways in the FRG. Soc. Sci. Med. 38, 419–426. Socialstyrelsen, 2006. Förslag till nationellt program för suicidpreventio – befolkningsinriktade och individinriktade strategier och åtgärdsförslag. (Proposal of a national program for suicide prevention – population oriented and individual oriented strategies and prevention suggestions.) (accessed via socialstyrelsen.se, 21.01.15). Sonneck, G., Etzersdorfer, E., Negel-Kuess, S., 1994. Imitative suicide on the Viennese subway. Soc. Sci. Med. 38, 453–457. Stockholm City, 2014. http://www.stockholm.se/OmStockholm/Fakta-och-kartor/ (accessed 28.10.14). Stockholm City, 2014. Statistisk årrsbok för Stockholm 2014 (Statistical Year-book of Stockholm 2014) (accessed via http://www.statistikomstockholm.se, 02.02.15). Veress, L., Szabó, T., 1980. Die Besonderheiten der am Eisenbahnkörper verübten Selbstmorde (The specifics of bodies from suicides committed on railways). Suizidprophylaxe 7, 401–426.