Data driven improvements in public transport: the Dutch example

2.1 Passenger impacts

Service reliability, which is a result of variability in operations (Abkowitz and Engelstein 1983; Ceder 2007; Mazloumi et al. 2010), is the certainty of service aspects (such as travel time) compared to the schedule as perceived by the user (Furth and Muller 2009; Van Oort et al. 2012). Unreliability causes longer and uncertain passenger journeys (Bates et al. 2001; Noland and Small 1995; Van Oort et al. 2012). The level of service reliability affects several choices made by travellers, such as mode and route (Schmöcker and Bell 2002; Turnquist and Bowman 1980). In numerous studies, reliability-related attributes have been found among the most important service attributes in a variety of situations. Balcombe et al. (2004) even report that passengers consider service reliability twice as important as frequency. König and Axhausen (2002) conclude that research over the last decade shows that the reliability of the transportation system is a decisive factor in the choice behaviour of people.

Even though reliability is thus considered of key importance, in Van Oort et al. (2012) it is demonstrated that passengers presently experience an insufficient level of service in this field. It is revealed that service reliability impacts on passengers depend on three main aspects, as shown by Fig. 1:

Open image in new window

• How does the vehicle operate (compared to the schedule)?

• How do passengers travel through the network?

• How do passengers behave (concerning arrival at the first stop and transferring)?

Much scientific research and public discussion is being conducted concerning the impact of reduced societal costs for passengers of enhanced reliability in cost-benefit analyses (Van Oort et al. 2015a). These costs consist of passenger appreciation of a unit of time (variation) and the quantity of this time (variation). Most research (e.g. Li et al. 2010) focuses on the first part, while the latter, for instance reduction in (standard deviation of) travel time in a network context, still lacks insights (OECD/International transport forum 2010; Van Oort et al. 2015a). Contrary to car traffic (Snelder and Tavasszy 2010), calculating the quantity aspects for public transport is complex, since a schedule, waiting and transferring are involved. New data sources, as described in this paper, are helpful to calculate these aspects.

2.2 Causes and impacts of service unreliability

2.2.1 Causes

In order to improve public transport in general and its service reliability in particular, the main causes of disturbances must be understood. The next sections identify and analyse the main reasons for unwanted variation in public transport services and divides them into three components: namely driving time, stopping time and dwell time. It also distinguishes between the causes that are external and those that are internal, in order to clarify which actor is able to improve which element. This is illustrated by Fig. 2. These causes and relations are found by literature study (references will be provided in the next sections), by research and experiences of ourselves (Van Oort 2011) and by interviewing several authorities and operators via an international survey (Van Oort 2011, 2014). The fundament of our own research was data analysis (AVL and APC) of bus and tram lines in The Hague. In addition, field observations and interviews with drivers, planners and designers were performed enabling to derive the relationships.

Open image in new window

Shalaby et al. (2001) found that trip time variation not only depends on service trip time (or line length) itself, but is also affected by the number of stops made, the number of signalized intersections passed, the vehicle intensity and the capacity per lane. Abkowitz and Engelstein (1983) found that line length, passenger activity and the number of signalized intersections affect average trip time. Most researchers agree on these basic factors (Abkowitz and Engelstein 1983; Levinson 1983; Abkowitz and Tozzi 1987; Strathman et al. 2000).

Considering the trip time components individually, we observed and analysed the different processes during a vehicle trip (of bus and tram in The Hague) to gain insights into the causes of trip time variability (Van Oort 2011), in addition to performing an international survey (Van Oort 2011, 2014) and studying literature. The first process we analysed was the driving between stops. This includes accelerating, braking and unplanned stopping. Since the causes of stopping may also be responsible for slowing down vehicles, both components are presented in one overview. The causes presented below are responsible for actual driving time and stopping time variability.

2.3 Improving service reliability

Lots of research has been carried out to identify the potential of improving travel time and service reliability. One could, for instance, think of improvements of vehicles, infrastructure, planning and operations. Literature shows that in urban public transport, substantial attention is given to ways to improve services at an operational level (Vuchic 2005; Ceder 2007; Cats 2014). Concerning strategic and tactical instruments, much research is already available. A lot of this focuses on the implementation of bus lane schemes, synchronisation and traffic signal priority, which are the most common solutions, as shown by e.g. Waterson et al. (2003). Ceder (2007), Vuchic (2005), and Lee et al. (2014) all present the different methods and effects and also give an overview of the issues which need to be considered in synchronisation. Potential instruments for improving operational quality during the design of the schedule are optimising trip time determination and holding (Delgado et al. 2012; Xuan et al. 2011).

Less researched so far, but also important, are the potential instruments that are available during network design, for instance, line length and design of terminals (Van Oort and van Nes 2009, 2010). Our study of a new tram line in Utrecht, the Netherlands (Van Oort 2012), showed that about 65 % of the (societal) benefits are related to service reliability aspects, coming down to a sum of over € 200 million during the total lifetime of the tram infrastructure. On busy bus trunks in Utrecht, a reduction of 30 s of trip time per bus saves about € 100,000 to € 400,000 in operational costs per year (based on costs per hour of operation of € 100).

The first step to increase operational performance is a proper analysis of historical operations. This paper focuses on bus and tram operations. The performance for heavy railways based on track occupation data is described in (Goverde and Meng 2011). AVL systems (Strathman et al. 2000, Hickman 2004) are of great help as these provide databases of historical performance with regard to travel time and reliability. Although such data have already been available for years, it is only recently that these can be accessed and used by Dutch transit authorities, researchers and developers. In addition to facilitating analysis of performance, these data also enable forecasts of future service quality (Kanacilo and van Oort 2008; Wilson et al. 2009). In this paper we present practical examples of data to illustrate the usefulness of these kinds of analyses: several bottlenecks are identified, providing transport authorities with insight into determining investment priorities. Next step is to add data from smart cards (APC; Pelletier et al. 2011) to gain insights into passenger impacts.

3.1 Timetable data

Public transport companies have always dealt with large amounts of data when designing timetables, scheduling vehicles and staff, collecting fares and more recently tracking vehicle locations. However, it has only recently become possible to store large amounts of historic vehicle location and fare collection data, and therewith to utilize these data for analyses. Furthermore, in line with other “Open Data” initiatives in the public sector, data related to public transport is being made publicly available in more and more areas, notably in North America and more recently in certain European cities.

The first type of public transport data that became publicly available is timetable information. Besides supplying public transport route planners with timetable data, computer-readable timetable information also allows for efficient analysis and comparison of public transport networks, describing, for instance, spatial coverage, commercial speeds, frequencies, and connections to adjacent public transport networks. However, timetable information does not yet provide insights into the performance of the actual operation and hence the service reliability of public transport and the real-time timetable information.

Accurate real-time vehicle location data (AVL) has become available for public transport operators with the wide availability of GPS- and mobile phone devices. AVL-data has also become publicly available in many areas in recent years, albeit often with the condition that it can only be used for passenger information. Early examples include the transit agencies of Washington, Boston and various US bus companies. We note that these days most Western public transport operators provide some kind of real-time vehicle location (or expected vehicle arrival time) to the public, but often this information is not yet technically or legally available for storing or further processing by third parties.

3.2 The Dutch example: GOVI data

In the Netherlands, most public transport operators take part in the Borderless Public Transport Information initiative [Grenzeloze Openbaar Vervoer Informatie in Dutch, GOVI (http://govi.nu)], aimed at making a wide range of public transport information publicly available in the Netherlands, ranging from timetables to fares, vehicle location and punctuality. The data exchange interfaces are defined by the set of standards called BISON (http://bison.connekt.nl/), which are the same for all participating public transport operators, including buses, trams, metros and regional trains. These standards are connected through the BISON architecture. Each of the standards describes an aspect of public transport operations, like the timetable, punctuality, actual departure and arrival times, dynamic platform allocation or free text. Another source of open public transport information, such as a GTFS (i.e. General Transit Feed Specification) feed on the national level, is the company 9292 REISinformatiegroep BV (http://9292opendata.org/), a company owned by Dutch operators that specializes in providing passenger information. The formats of GTFS and BISON are not the same, but the data included in both formats is mostly similar. Since both formats are well defined, scripts to automatically convert data from one format to the other exist.

GOVI was designed to facilitate data communication between vehicles and the land side to enable the provision of dynamic passenger information (i.e. real-time information). As an additional benefit, the actual and scheduled vehicle positions and times are logged in a database. Typically, for one transit line this involves 3000–7000 arrival times and departure times per day. Although this database was not the main objective of GOVI, it is a helpful source when monitoring and analysing public transport performance. The arrival times and departure times of vehicles at stops are recorded with a precision of 1 s. 3–8 % of these time points are missing. This includes errors and outliers, since these are detected and denoted as missing by the GOVI organisation. Reason for missing data may be a malfunctioning GPS device or the wrong coupling of a vehicle number and a run number.

In 2012, the first Dutch public transport operator agreed to legally release AVL data via GOVI for storage and analysis by third parties, such as researchers and developers (http://webwereld.nl/cloud/56325-gvb-maakt-real-time-reisinformatie-open-data). Since then several other operators joined. The resulting data streams are publicly available via the Dutch OpenGeo Foundation (http://www.openov.nl/). The source of the data presented in this paper is the transit authority or via OpenGeo.

4.1 Data structure

This data table consists of time-stamped messages of important events on a trip. In particular, a trip starts with an INIT initial message and ends with an END message and all departures are logged with a DEPARTURE message. For some stops an ARRIVAL message is recorded, too, allowing for an estimate of the dwell time. Furthermore, in case there is no departure and arrival event taking place for a longer time duration (about a minute), an ONSTOP or an ONROUTE message is logged, including the exact location. Our data source already includes a value for delay, which comes down to the difference of the message timestamp and the planned arrival or departure time.

4.2 Line-based analysis

A commonly used visualization (Furth et al. 2006; Van Oort and van Nes 2009) of the performance of a transit line is plotting each trip as a line chart in a coordinate system of stops versus delay. Figure 3 (left) shows 1 month of bus trips of a certain line, as well as the median and 15th and 85th percentile. Such a chart reveals both the level of variations in the operation and the quality of the timetable and the systematic deviations. In detail, this chart visualizes that there is a good general on-time performance on this line, with occasional large delays but only very few buses substantially ahead of schedule. One can also observe that there are quite a few holding stops where even early buses wait for their scheduled departure time. Nevertheless, the delay slightly accumulates on average in the second part of the line. Another phenomenon that is shown by this chart is the ample time reserve used just before the last stop in order to be on time when punctuality is measured.

Open image in new window

Another way to look at the same data is to plot vehicle headways instead of delays. A high frequency line with consistent delays but regular headways remains attractive to passengers. The chart is shown in Fig. 3 (right), the scheduled headway is 10 min. This location-headway chart points out the regularity of high-frequency services along the line, as well as possible bus bunching. In detail, this chart visualizes that the headway deviations stay constant in the first two-thirds of the line, but the headway variance increases in the last third. Together with the previous chart, this indicates that interventions for punctuality and regularity are most useful towards the last part of the line.

4.3 Network-wide analysis

The ubiquitous availability of vehicle locator devices allows one to take a step further from line-based performance evaluation and investigate patterns on network level. Phenomena only visible on this level are the reliability of transfers, area-related issues and possible bunching or interference on shared sections. An example of a network-level data visualization is the average delay at each stop, including stops with several transit lines, as shown on Fig. 4 (delay averaged over all days of January 2013).

Open image in new window

In detail, the implication of this chart is that while there is a good general punctuality along the network, there are a few problem areas where a whole series of stops suffer from delays from all related bus lines, for example in the bottom left corner of the chart with lots of orange. On the other hand, there are a few isolated green points where buses tend to be often early—this problem could be solved once known by a simple adjustment in the published departure time.

4.4 Inter-operator transfers

An aspect of public transport travel that previously was invisible to the public and to operators, albeit of substantial importance to the passenger, is the reliability of transfers between operators or modes, such as between a long-distance train and a local bus. With open data, it is possible for anyone (so also for any operator) to investigate the actual reliability of inter-operator transfers and to take steps if necessary.

Figure 5 shows a discrepancy between vehicle punctuality and passenger experience, for a transfer that is scheduled to take 5 min excluding walking time [at station Hilversum Sportpark where there is one bus line and two train lines (all operated in a 30 min headway)]. The chart shows average values across all Sundays in January 2013, between 10 a.m. and 11 p.m. It is common that a public transport timetable includes a substantial time reserve before an important stop, and therefore as the left part of the figure shows, vehicles are consistently early at the transfer stop. However, this means that the passengers structurally have to wait much longer at the transfer stop than they expect from the timetable. As waiting time on the platform is perceived as much less comfortable as in-vehicle time (Van Der Waard 1988), this implies that trips including this transfer are perceived of as being of less quality than expected from the timetable.

Open image in new window

The relevance of open AVL data with regard to improving transfers is the following: open information on the reliability of inter-operator transfers enables any operator and transport authority to gain insight into the reliability of these transfers and take steps if improvement is required, such as synchronising timetables, holding vehicles in case of minor delays and informing passengers. See Sparing and Goverde (2013) for example for identifying transfers of interest and choosing which vehicles to hold in a multi-operator setting.

4.5 Enriched travel advice

As we covered in our introduction to open transit data, the original reason for opening up actual departure and arrival time information was to provide real-time passenger information. Collecting historical data, on the other hand, used to be against most terms and conditions of data providers and these are often considered sensitive business data. With the new developments, however, that also allows access to data for historical analyses, it is possible to calculate distributions of running times, arrival times and transfer waiting times, as demonstrated in the previous paragraphs. Building on these distributions, and returning to passenger information, it is possible to enrich travel advices to provide estimations on the reliability of public transport trips as follows.

Let us assume that probability distributions of departure and arrival times are available for all transit lines at all stops at all times. For example, fitting a normal distribution on all arrivals and departures in the past 3–6 months on the given type of day (working day/Saturday/Sunday) at the given time period (morning peak/midday/afternoon peak/evening) for the given line, stop and direction can be a good first estimate. Furthermore, we assume for simplicity that the times of different lines are independent from each other.

Then for a travel advice with no transfers, we can provide the passengers with the distribution of the departure and arrival times. In fact, considering what really matters to the passenger, we can summarize these two distributions using two indicators: (1) the probability the vehicle departs too early and (2) the probability the vehicle arrives on time at the destination stop (within a certain time tolerance). If necessary, we can further simplify these indicators to a “traffic light” approach: colour coding the favourable, medium and low-reliability options.

In case of a travel advice including transfers, we can add an indicator for each transfer: the probability that the transfer is possible. Note that from the passenger’s perspective, independent arrival and departure time distributions are irrelevant during a transfer, except for whether the transfer is possible or not.

See Fig. 6 for an example for such an enriched travel advice.

Open image in new window