Scholarly article on topic 'An Individualized Travel Information System for Optimizing Mode and Route Choice behavior of Commuters'

An Individualized Travel Information System for Optimizing Mode and Route Choice behavior of Commuters Academic research paper on "Civil engineering"

CC BY-NC-ND
0
0
Share paper
OECD Field of science
Keywords
{Commuters / "travel behavior analysis" / "descrete choice modelling" / "transport information system" / "GPS data"}

Abstract of research paper on Civil engineering, author of scientific article — Oliver Roider, Christian Rudloff, Markus Ray

Abstract Urban sprawl and the increasing mobility demand cause serious problems for the transport systems of urban areas, in particular during peak hours. Accidents or roadwork sites worsen the situation for transport users. New technical developments in the field of satellite based localization systems interacting with intermodal and interoperable online recognition of the actual traffic situation make it possible to transmit customized information for transport users who wish to optimize their mobility behavior.On behalf of the Austrian Federal Ministry for Transport, Innovation and Technology, the research project PROVET focused on the basic conditions for an individualized transport information system (ITIS). Based on the user's mobility profile and with the help of GPS data, ITIS should be able to determine the most likely mode, route, and destination of a transport user in real-time. By comparing this information with the actual traffic situation, possible delays on the predicted route can be identified and better (intermodal) alternatives can be suggested during a trip or even prior to the beginning of the trip.PROVET mainly focused on the technical feasibility of collecting mobility data to predict the mode and the route but also on the requirements users might have in regard to the system. Based on a test sample, algorithms were developed by using logistic regressions which achieved a detection rate of about 80%. In order to identify alternatives which offered maximum benefit for the user a discrete choice model with known class approach was calibrated, taking variables such as travel time, travel costs, gender specific aspects, or basic requirements for certain modes into account. It showed that the mode as well as the route normally used for the daily trip to work has an essential impact; particularly car commuters consider using their car as a high basic benefit.

Academic research paper on topic "An Individualized Travel Information System for Optimizing Mode and Route Choice behavior of Commuters"

Available online at www.sciencedirect.com

SciVerse ScienceDirect PrOC6d ¡0

Social and Behavioral Sciences

Procedia - Social and Behavioral Sciences 48 (2012) 1948 - 1957 —

Transport Research Arena- Europe 2012

An individualized travel information system for optimizing mode and route choice behavior of commuters

Oliver Roidera*, Christian Rudloffb, Markus Rayc

a Institute for Transport Studies, University of Natural Resources and Applied Life Sciences (BOKU),

Peter-Jordan-Str. 82, 1190 Vienna, Austria b'c Mobility Department (Dynamic Transportation Systems), Austrian Institute ofTechnology (AIT) Giefinggasse 2, 1210 Vienna, Austria

Abstract

Urban sprawl and the increasing mobility demand cause serious problems for the transport systems of urban areas, in particular during peak hours. Accidents or roadwork sites worsen the situation for transport users. New technical developments in the field of satellite based localization systems interacting with intermodal and interoperable online recognition of the actual traffic situation make it possible to transmit customized information for transport users who wish to optimize their mobility behavior.

On behalf of the Austrian Federal Ministry for Transport, Innovation and Technology, the research project PROVET focused on the basic conditions for an individualized transport information system (ITIS). Based on the user's mobility profile and with the help of GPS data, ITIS should be able to determine the most likely mode, route, and destination of a transport user in real-time. By comparing this information with the actual traffic situation, possible delays on the predicted route can be identified and better (intermodal) alternatives can be suggested during a trip or even prior to the beginning of the trip.

PROVET mainly focused on the technical feasibility of collecting mobility data to predict the mode and the route but also on the requirements users might have in regard to the system. Based on a test sample, algorithms were developed by using logistic regressions which achieved a detection rate of about 80%. In order to identify alternatives which offered maximum benefit for the user a discrete choice model with known class approach was calibrated, taking variables such as travel time, travel costs, gender specific aspects, or basic requirements for certain modes into account. It showed that the mode as well as the route normally used for the daily trip to work has an essential impact; particularly car commuters consider using their car as a high basic benefit.

©2012 Published by Elsevier Ltd. Selection and/or peer review under responsibility of the Programme Committee of the Transport Research Arena 2012

Keywords: commuters; travel behavior analysis; descrete choice modelling; transport information system; GPS data

* Corresponding author. Tel.: +431-476545307; fax: +431-476545344 E-mail address: oliver.roider@boku.ac.at

ELSEVIER

1877-0428 © 2012 Published by Elsevier Ltd. Selection and/or peer review under responsibility of the Programme Committee of the Transport Research Arena 2012

doi:10.1016/j.sbspro.2012.06.1168

1. Introduction

In the last decades European settlement patterns have changed significantly towards more urban sprawl. Economic growth as well as investment in the transport infrastructure fostered this development in recent years. As a negative consequence, urban areas with a high population density and compact cities have been replaced by sparsely built-up areas with more than a doubling of the space consumed per inhabitant, whereas many workplaces remain in the center conurbations [European Environment Agency 2006]. Thus, the kilometers traveled to get to work have increased tremendously [Herry 2007]. Fig. 1 illustrates the catchment areas of the agglomerations in Austria showing municipalities with 30 % or more of their population commuting daily to the central city. Particularly the population living in the south and southeast of the city of Vienna (Wien) has increased by more than 25 % from 1991 to 2006.

Fig. 1. Catchmentareas of larger agglomerations in Austria in 2001 (Source: [Statistik Austria 2008])

This development of decentralized settlements causes serious problems for the transport system of urban areas, in particular during peak hours. For example, more than 200,000 people commute daily to work in Vienna, many of them by car [Statistik Austria 2007]. Particularly in the east and the southwest of the city of Vienna the share of public transport in all trips is partially less than 20 %. Accidents or roadwork sites worsen the situation for road users. However, for long distance trips, commuters prefer to use a train (Fig. 2). Regions in the far northwest of Vienna for instance are well connected to the city centre by train with a travel time of 1.30 hours or more. Interruptions of operation or breakdowns without any information for travelers lower the attractiveness of these connections.

□□Forrest, Alpine mountains, wasteland rL

30 to < 50 %

gSjk. H 50 to < 75 %

.rv^- ■ 75 to < 100 %

% of public transport users

0to< 10% 10 to <20% 20 to < 30 %

Fig. 2. Mode share of public transport in commuter trips to the city of Vienna (Source: [OIR 2010])

2. Objectives of an individualized transport information system

The individualized information system (ITIS) is meant to provide customized travel information for each situation transport users might have to face. This helps them to optimize their daily mobility with the help of satellite based location systems (GPS, GALILEO) or cell phone recognition as well as intermodal and interoperational real life transport information systems. Dynamic data are used to determine the most likely mode (mode detection), the route or the public transport line used at the time, as well as the destination (profiling) of the user. By comparing these data with the actual traffic situation, delays on the predicted mode or route are identified and better (intermodal) alternatives are suggested during the trip or even prior to the trip (situation awareness). For this purpose, real-time traffic data are used which are provided by external sources like the telematic system of the road traffic authority, on-street loop detections, floating car data of the taxi fleet, or online timetable data of the public transport operators (Fig. 3).

real time information systems

private

public

transport

online

information

1 Mode Detection [ [ Profiling ]

[ Situation Awareness ]

dynamic datq of the user

Fig. 3. Concept of the individual transport information system

Based on these data the travel situation of the user at the time of request is predicted and individual information provided for two stages: (1) based on the usual mobility behavior, ITIS forecasts the upcoming trip and provides information on the best departure time and ideal route. (2) ITIS optimizes the route during the trip. For example, the system might suggest to use the train in the morning because of congestions on the motorway (information prior to the trip) or to change to another public transport connection because of a breakdown of a line normally preferred by the user (information during the trip).

The research project PROVET, funded by the Austrian Federal Ministry for Transport, Innovation and Technology, defined the framework of the system from the users' point of view and tested the technical feasibility of collecting data on individual mobility behavior.

3. Requirements from the user's point of view

3.1. SurveyDesign

Data to analyze the requirements from the user's point of view were collected in a two-stage survey among commuters to the city of Vienna. The screening was conducted by phone (using randomly selected telephone numbers) to identify commuters and to ask which mode the respondent used for the trip to work on a certain reference day. In the second phase, face-to-face interviews were conducted in two parts (see e.g. [Axhausen, Sammer 2001]: (1) in a revealed preference survey information was collected about the mobility behavior at the time, and (2) in the stated preference part hypothetical scenarios were simulated and respondents were asked to state their preferences in case of congestion or breakdown on their usual

route, assuming that information is provided prior to the trip (scenario 1) or during the trip (scenario 2). To each respondent 4 options were proposed: (1) to stay on the usual route and wait; (2) to choose an alternative route by car; (3) to change to public transport or (4) to use park and ride. A similar approach was used for public transport users. To achieve a high level of orthogonality, for each variable up to 3 different attributes were defined in advance; one of them was selected at random. However, all attributes were based on observed values (revealed preference); to provide an example: based on the usual travel time, an extension of this time by 30, 50 or 100 % was possible. This approach ensures that the new travel time suggested in the scenarios seems plausible to the respondents (Fig. 4).

HK/ Scenario 1 - information received prior to the trip p„.„

| The information systems tells you that the route which you normally use is congested and that you will nee i-i minutes for your trip to work. The information is provided 10 minutes befor your planned dep d an additional

Date:______ Departure Time:__:__hours |jhe system proposes the following alternatives

Usai HBI route alternative route-1 alternative ^^^^^ route 2 Bjjyj alternative route 3 • •

Explanation You use same the route as on the reference day. Take a small detour. Go to the next park-and-ride facility and changeto public transport. Take public transport.

travel time □ Min- □O -DO □ hQ Min.

km of motorways I-1 km | km

mainmod. _______' I

number of interchanges average waiting time at interchanges - mm - mm

costs € € € €

Arrival time —■-hours -■— hours -■— hours —■-hours

Iwhich one would you H Ichoose? □ □

□ Others:

Fig. 4. Example of the design of the stated preference survey 3.2. Travel behavior of commuters

Almost 70 % of the respondents work flexible hours and to a certain extent they are flexible regarding their departure and arrival time (± 15 minutes). This means that they are able to change their usual travel behavior; they might start a trip a few minutes earlier or later if they receive the information that this might improve travel conditions (e.g. no congestion). However, it showed that commuters are currently not flexible in the choice of their mode and route. On average, the same mode is used 4.6 days a week; only 12 % of the respondents decide the previous day which mode to use for the trip to work, all others always use the same mode. The situation for public transport users is similar: on average, the respondents use the same route or public transport line(s) 4.2 days a week; 72 % use the same route / line(s) every day. On average, car users know 5.5 alternative routes, but only 15% of them use such an alternative occasionally. The decision for a route is mainly based on a trial-and-error procedure. In order to determine variables for a utility function, the participants were asked to rank the attributes of a route. For car trips, most important decision criteria are a short travel time, low probability of congestion, and a short travel distance, whereas public transport users prefer a high share of railroad, reliability, and a short travel time. Information about the traffic situation prior to or during the trip plays a minor role, only few people use the internet or a navigation system, the majority listens to travel information on the radio. In case of a breakdown, most of the respondents wait on the route or public transport line 10 minutes or longer before they start to look for alternatives. Thus, the potential market for an individual information system seems to be big.

3.3. Development ofa discrete choice modelfor optimizing route and mode choice

To provide information about alternative modes or routes for users in case of an accident or a breakdown of the public transport service on the usual route is one of the main objectives of the individualized travel information system. For this reason, a discrete choice model was developed considering a variety of variables, such as travel time, travel costs, use of the train etc. (see [Ben-Akiva et. al 1997] or [Maier et al 1990]). Moreover, it has to be ensured that benefits are not only calculated for the same mode but also considering the possibility of changing the mode, if appropriate. The analysis is focused on an individual person with more than one feasible alternative. The utility of each alternative is calculated by using the following utility function:

U* = V„, +^with Fm= fi fiS x ^ x

Utility of alternative i ofperson n Vni Systematic component of the utility eni Random component fit Constant of alternative i

fin coefficient of alternative-specific variable I of alternative i Pmi coefficient ofperson-specific variable m of alternative i

alternative-specific variable I of alternative i and person n ^nm person-specific variable m of person n L total number of alternative-specific variables M total number of person-specific variables

As most of the commuters observed use the same mode and take the same route every time, a discrete choice model with known classes seems to be appropriate. Therefore the model approach explained above is extended with a conditional term as follows

Pn (J,) = X Pn (ck) x pn (J, | ) with ^ ^ ! Ct} =

and Vnik = +X fit, xXmii +X fiPmik X7

Pfj () Probability of person n to choose alternative i

Pn(c^ ) Probability of person n to belong to class k

Pn (\ck ) Probability of person n to choose alternative i on condition that person n belongs to class k

The term Pn(ck) is 1 (person belongs to class k) or 0 (person does not belong to class k). The calibration of the coefficients can be undertaken for each alternative and for each class separately or independently of both, where appropriate. The software package Latent Gold Choice 4.5 was used to estimate the coefficients by applying the maximum likelihood procedure [Vermunt, Magidson 2005]. The calibration of the coefficients takes the results of the revealed preference as well as the stated preference survey into account. Including observed data in the analysis offers the advantage that one can consider people's behavior in case of a breakdown related to the mode or route normally used (reference level). Classes are defined by car commuters and by commuters using public transport for the daily trip to work. With the help of the calibrated coefficients the individualized travel information system is able to

calculate the utility of each alternative and to indicate the one which maximizes the utility for the user (Table 1).

Table 1. Coefficients of the utility fonction

Independent variables Coeff. ß car-commuter Coeff. ß ptuser p-value *) p-value **)

Main Mode = car (C) 16,877 1,755 0,000 0,002

Main Mode = Train (T) 1,949 1,949 0,000

Stat. Pref. Scenario x car use (SC) -3,063 5,062 0,160 0,063

Travel time by car without delay (TC) [min] -0,133 -0,045 0,000 0,070

Delay of car and pt travel time (D) [min] -0,155 -0,038 0,000 0,004

Cost of public transport use (CP) [Euro] -0,610 -0,610 0,000

Gender = male x car (MC) -2,541 -2,541 0,016

Season ticket x car (STC) -2,618 -2,618 0,066

*) significance of the coefficient

**) significance of the diversity of the two classes

Goodness of fit index Car commuters PT users All

R2 0,6448 0,2966 0,4646

Thus, the utility function is defined as follows:

Vt = PscxSC + pcxC + pTxT + pTcxTC + pDxD + pCPxCP + pMCxMC + pSTCxSTC

All algebraic signs of the coefficients of the utility function are plausible and show a high level of significance. Driving variables of the calculation are the modes (routes) which are normally used, the costs of the trip, and some socio-demographic characteristics. Clearly recognizable is the high value of the car for car commuters as shown in the coefficient of the variable "main mode = car" (C), which means that a mode shift of car commuters to public transport is difficult to achieve unless the advantage of this alternative is extraordinary high in comparison to the usual behavior. As already stated in the revealed preference survey, the train has a positive effect on the attitude to the public transport system (T). The interaction variable of observed behavior and car use shows a reduction of the utility in case of a delay on the car route, and the benefit of the car use for public transport users in case of a breakdown of the public transport supply (SC). The variable representing the delay on the route (D) is calculated for both classes separately. It shows that car commuters rate one minute delay as less unpleasant than public transport users. The cost of car trips does not seem to have any significant influence on the utility of a car; however, costs of a public transport trip play an important role (CP). Costs of a car trip do not seem to be taken much into consideration, but public transport users are well aware of the cost of a public transport trip, particularly if they have no season ticket. By using two interactive variables, the model takes socio-demographic characteristics into account. The negative sign of the coefficient of the variable "gender = male x car" (MC) shows that males are more likely to shift mode from car use to public transport than females. It goes without saying that owning a season ticket reduces the benefit of using the car (STC). The length of a trip (car as well as public transport) and the possible use of a motorway for the trip have no significant impact and are not considered in the utility function.

3.4. Expectedfeatures

Apart from the technical conditions required to calculate the utility of the alternatives provided, the system has to provide particular features requested by the users. More than 75 % of the respondents believe that the system can suitably optimize their daily trip to work in any case and wish information about travel time and delays, regardless of existing alternatives. Saving time by using alternative connections is one of the most important bits of information requested, but travel time by using an alternative mode is of minor interest. The preferred medium for the provision of information is the mobile phone, where applications for modern multimedia phones seem to be useful. Users require a flexible system. Information should only be provided on demand and the user should be able to adjust all settings.

4. Technical approach of the individualized transport information system

4.1. Data collection

For the development of mode detection algorithms large amounts of GPS tracks are required that include all available modes of transport. The data collection has to include all kinds of modes in different transport situation. In total, 294 hours of GPS data were collected at 2 second intervals, resulting in 792 trajectories to be used to test mode detection algorithms. In contrast to the data requirements for the mode detection analysis, determining individual mobility behavior requires long term motion data with recurring behavior. Four commuters were asked to record their trips to work for about six weeks using GPS devices. Additionally, they documented detours (e.g. collecting children from the kindergarten) and unusual situations (e.g. congestions). In total, information about 323 commuting trips was collected. The users were chosen in such a way to ensure that they have a reasonably large set of at least four different routes (including different modes) for their trips to work.

4.2. Profiling and route detection

User profiles are used to detect a commuter's route on a certain day. Based on a profiling algorithm raw GPS data are transformed into these profiles. To achieve this, GPS trajectories are split into (1) routes (trips between home and work) and (2) points of interest (e.g. interchange stations or a stopover for shopping purposes). Moreover, the routes are split into trajectories representing a single mode, and finally the single mode trajectories are collected by classes, where each corresponds to a route of a trip to work, to a certain time of day, to a certain mode, and the sequence of stops [Rudloff, Ray 2010]. Points of interest (POI) are identified in three different ways: (1) Signal loss of at least 180 seconds (time spent in buildings), (2) Stops of at least 180 seconds, where the stop detection algorithm follows a method proposed in [Hariharan, Toyama 2004], and (3) start and end-point of a walking segment identified as period of at least 180 seconds with speeds below 10 km/h. Since the algorithm is only concerned with routine behavior, only repeated POIs are of interests which are detected by using a clustering algorithm.

To correct faulty GPS tracks, the GPS data are connected according to the following approach: If two consecutive points result in unrealistically high speeds, the first point is deleted from the trajectory (due to GPS drifts). This is repeated until a trajectory contains only realistic speeds. A self-organizing-network algorithm (growing neural gas (GNG), see [Fritzke 1995]) is used to reduce the GPS positions in the collected routes to a set of spatial nodes together with a topology of edges that represent routes between the calculated nodes. The number of nodes is further reduced by a clustering algorithm, replacing the nodes in the cluster by their mean position. Points on trajectories are assigned to the network nodes by finding the nearest one, and transferred into classes by clustering trajectories whose corresponding nodes

are in each other's topological neighborhood, (see [Bauer et al. 2008]) For each class, a sequence of POIs (stops and transfer points) is determined. The profile then contains all routes which the commuter used multiple times, defined by the corresponding sequences of POIs, the modes used in between the POIs, as well as the network nodes corresponding to each segment between POIs. Once a profile including the user's routines has been designed, the system starts detecting the most likely route in real-time according to (1) actual time and day of the week (2), proximity of a user's GPS track to nodes at the time of contact as well as the sequence of close nodes, and (3) the sequence of modes used so far while commuting.

4.2.2Mode detection

Many different types of public transport have to be considered in the Viennese region (busses, trams, underground, interurban, and commuter trains). Individual traffic includes not only private car traffic but also walking, cycling, or using taxis in and outside of the city. Problems arise with the distinction between busses and trams as both have similar patterns in speed and acceleration. To overcome these difficulties, a high number of variables was calculated based on the GPS data:

• 95% quantile, median and standard deviation of speed and 95% quantile of acceleration/deceleration

• maximum change of orientation (angle between three consecutive points of smoothed GPS trajectory)

• percentages of time traveled at over 16 km/h, below 2 km/h and between 4-40km/h

• maximum signal loss time greater than 60 s and distance traveled during maximum signal loss time

• ratio of the distance from start to end-point and the traveled distance on GPS trajectory

Speed and acceleration-related variables are an obvious choice when distinguishing traffic modes. The speed percentage variables are meant to help to differentiate between modes with different speed profiles (see Figure 1), e.g. a speed of less than 2 km/h gives a rough estimate of the time the user is not moving at all. This helps to distinguish between public transport with regular stops and individual traffic. The other speed intervals provide an idea of the duration of the movement at 'fast' speed over 16 km/h or at medium speed between 4 and 40 km/h, which might help to distinguish between slower moving trams or busses and commuter trains with constant and relatively high speeds between stops. Fairly long signal losses indicate an underground train. The change-of-orientation variable is used to distinguish between transport on the road and in streets, where relatively sharp turns are possible, and transport on tracks which lias larger radii.

Speed profile bike Speed profile commuter train

0 100 200 300 0 100 200 300 time [sec] time [sec]

Fig 5. Examples of speed profiles for different modes of traffic in Vienna

4.3. Results of mode detection and profiling

Based on the result of the speed profile, walking and cycling segments of a trip are clearly identifiable. The detection results for busses and underground trains are poor due to the small number of available trajectories for these modes. The distinction between bus and streetcar without checking the availability of bus and streetcar lines by geo-coding is unlikely to achieve the same accuracy as the detection of the other modes, due to only minor differences in their speed and acceleration profiles. But bus segments can easily be identified as on street public transport (bus or streetcar) which might be sufficient for many applications of mode detection (Table 2).

Table 2. Mode detection results for a logistic regression model comparing real life data and detected modes

Detected as Used Mode Percentage detected correctly Walk Bike Streetcar Bus Underground train Commuter Train Car outside cities Car inside cities

Walk 100 % 85 0 0 0 0 0 0 0

Bike 95.46% 2 42 0 0 0 0 0 0

Streetcar 82.35 % 0 2 14 1 0 0 0 0

Bus 30% 0 2 10 6 0 0 0 2

Underground train 75 % 0 0 0 0 3 1 0 0

Commuter Train 75 % 0 0 1 0 1 18 1 3

Car outside cities 92.31 % 0 0 0 0 0 1 12 0

Car inside cities 63.33 % 1 0 3 1 0 4 2 19

For 80% of all observations, the algorithm for the prediction of routes was able to identify the route within 5 minutes from the start of the commuting trip. This result was achieved despite the fact that only data from two weeks could be used to calculate the profile and unexpected detours, e.g. to go shopping on the way, had to be taken into account.

5. Conclusions

By using current GPS data or cell phone recognition mobility patterns considering modes, routes and destinations of the users can be indentified in a proper way. The potential market for an individual information system seems to be big, as more than 75 % of commuters to Vienna asked in a survey responded that such a system is suitable to optimize their daily trip to work. A discrete choice analysis clearly showed variables to be used for calculating the utility of different modes or route, like the mobility behavior normally preferred by the user, travel time, the use of railroads etc.

Further research work will concentrate on the development of an application for modern smart phones and on the integration of the utility function into this prototype of an individualized travel information system. As the discrete choice model is not yet part of a prototype which might suggest an alternative

route to the commuter, further research on the technical feasibility should concentrate on users' behavior ensuring the algorithm to be included in the system. While real time data are already available, there are still problems with the integration of these data in an individualized travel information system due to technical problems when trying to combine different sources (e.g. different public transit agencies, road transport authorities) and different providers. In this regard, the development of common interfaces has to be solved in the near future.

Acknowledgements

The authors gratefully acknowledge the support of the Austrian Federal Ministry for Transport, Innovation and Technology (BMVIT) which funded the project PRO VET within the national funding framework I2VSPlus. Moreover, the authors acknowledge with gratitude the close collaboration with all project partners and their contribution within this project.

References

AxhausenK. W., G. Sammer (2001). Stated responses: Überblick, Grenzen, Möglichkeiten, Arbeitsbericht Nr. 73 of the IVT at the ETH Zürich, Zurich

Bauer, D., M. Ray, N. Brändle, H. Schrom-Feiertag (2008). On extracting commuter information from GPS Motion Data, Proceedings oflntern. Workshop on Computational Transportation Science, Dublin.

Ben-Akiva M., S. R. Lerman (1997) Discrete Choice Analysis, Theory and Application to Travel Demand, MIT Press Cambridge, London 1997

Commission of the European Communities (2007). Green Paper, Brussels

European Environment Agency (2006). Urban sprawl in Europe, The ignored challenge, EEA Report Nr.10/2006, Luxembourg

Fritzke B. (1995). A Growing Neural Gas Network Learns Topologies", in G. Tesaurso, D. Touretzky, and T. Leen (Eds.), Advances in Neural Information Processing Systems, pp. 625-632., MIT Press Cambridge

Hariharan R. and K. Toyama (2004). Project Lachesis: Parsing and Modeling Location Histories, in

Proceedings ofthe Third International Conference on GIScience, Adelphi, MD, USA.

Maier G., P. Weiss (1990) Modelle diskreter Entscheidungen, Theorie und Anwendungen in der Sozial-und Wirtschaftswissenschaft, Springer-Verlag Wien New York (ed), Vienna

Rudioff, C. and M. Ray (2010), „Detecting Travel Modes and Profiling Commuter Routes Solely Based on GPS Data" in Prceedings ofthe TRB 89th Annual Meeting, Washington DC

Statistik Austria (2007): "Volkszählung 2001", Vienna

Statistik Austria (2008) Pendlereinzugsbereich der Agglomerationsräume 2001 nach Gemeinden, at http://www.statistik.at/web_de/statistiken/bevoelkerung/volkszaehlungen_registerzaehlungen/pendler/ind ex.html (20 September 2011), Vienna

ÖIR, Österreichisches Institut für Raumplanung (2010). Analysis of commuters of the project Pendo" at http://www.pendo.at (20 September 2011), Vienna

Herry et. al (2007): Verkehr in Zahlen 2007, Österreich, Ausgabe 2007 on behalf of bmvit (ed), Vienna

Vermunt J. and J. Magidson (2005): Latent Gold ® Choice, User's Guide, Statistic Innovation Inc. (ed), Belmont