Development of a Logit model of the transition effect to public transport

Department of Science and Technology Institutionen för teknik och naturvetenskap

Linköping University Linköpings universitet

g n i p ö k r r o N 4 7 1 0 6 n e d e w S , g n i p ö k r r o N 4 7 1 0 6 -E S

LiU-ITN-TEK-A--17/041--SE

Development of a Logit model of

the transition effect to public

transport

Therése Ziedén

2017-06-21

LiU-ITN-TEK-A--17/041--SE

Development of a Logit model of

the transition effect to public

transport

Examensarbete utfört i Transportsystem

vid Tekniska högskolan vid

Linköpings universitet

Therése Ziedén

Handledare Nikolaos Tsanakas

Examinator Anders Peterson

Upphovsrätt

Detta dokument hålls tillgängligt på Internet – eller dess framtida ersättare –

under en längre tid från publiceringsdatum under förutsättning att inga

extra-ordinära omständigheter uppstår.

Tillgång till dokumentet innebär tillstånd för var och en att läsa, ladda ner,

skriva ut enstaka kopior för enskilt bruk och att använda det oförändrat för

ickekommersiell forskning och för undervisning. Överföring av upphovsrätten

vid en senare tidpunkt kan inte upphäva detta tillstånd. All annan användning av

dokumentet kräver upphovsmannens medgivande. För att garantera äktheten,

säkerheten och tillgängligheten finns det lösningar av teknisk och administrativ

art.

Upphovsmannens ideella rätt innefattar rätt att bli nämnd som upphovsman i

den omfattning som god sed kräver vid användning av dokumentet på ovan

beskrivna sätt samt skydd mot att dokumentet ändras eller presenteras i sådan

form eller i sådant sammanhang som är kränkande för upphovsmannens litterära

eller konstnärliga anseende eller egenart.

För ytterligare information om Linköping University Electronic Press se

förlagets hemsida

http://www.ep.liu.se/

Copyright

The publishers will keep this document online on the Internet - or its possible

replacement - for a considerable time from the date of publication barring

exceptional circumstances.

The online availability of the document implies a permanent permission for

anyone to read, to download, to print out single copies for your own use and to

use it unchanged for any non-commercial research and educational purpose.

Subsequent transfers of copyright cannot revoke this permission. All other uses

of the document are conditional on the consent of the copyright owner. The

publisher has taken technical and administrative measures to assure authenticity,

security and accessibility.

According to intellectual property law the author has the right to be

mentioned when his/her work is accessed as described above and to be protected

against infringement.

For additional information about the Linköping University Electronic Press

and its procedures for publication and for assurance of document integrity,

please refer to its WWW home page:

http://www.ep.liu.se/

Abstract

The importance of traffic planning has, throughout the years, been in-creased, providing sustainable developments of traffic and infrastructural investments. The analysis of the current traffic situation and the evalua-tion of the effects of a future investment are crucial for the socio-economic benefits maintenance. These analyses and evaluations are most commonly done using traffic simulation models. One of the main traffic planning aims, nowadays, is to increase the number of public transport users against the number of private car users. This change in mode choice is called transition effect and could be beneficial both from an environmental and socio-economic perspective.

This thesis aims to evaluate and improve the macroscopic traffic demand and transition model, used fot the city of Norrköping. Additionally, the thesis investigates if a general transition Logit model can be developed and which parameters are the most important to be included in a modal choice estimation. For the needs of this study, the traffic planning software Visum is used. The travel mode distribution is calculated by Logit models coded in Python-scripts integrated in Visum. Then, a traffic assignment is performed by Visum, computing new travel times as inputs to the Logit model and this iterative procedure continues until the system reaches an equilibrium. The thesis aims for a more reliable prediction of the transition effect by correcting the Python-scripts and estimating the parameters of the Logit model using data from surveys.

The study shows that travel times is the most important factor for realistic results generation. However, the data used for the estimation of the Logit model parameters did not include sufficient information of travel times. The travel times had to be calculated, using two different methods, in order to be included in the estimation of new parameters. Although these methods could not provide any positive effects on the transition, they did prove the importance and significance travel time have when developing a traffic model. The result of the study invokes the importance to further develop the method of calculating travel times, when the input data is not sufficient, and shows that the travel time parameters are case specific.

vi

Sammanfattning

Vikten av trafikplanering har genom åren ökat för att kunna tillhandahål-la hållbara trafik- och infrastrukturlösningar. Analyser av befintliga tra-fiksituationer och utvärdering av effekterna av framtida investeringar är nödvändiga för hållbara socioekonomiska samhällen. Dessa analyser och utvärderingar görs vanligen med hjälp av trafiksimuleringsmodeller. Ett av målen med dagens trafikplanering är att öka andelen kollektivtrafikanvän-dare i förhållande till andelen bilanvänkollektivtrafikanvän-dare. Detta trafikslagsbyte kallas överflyttningseffekt, vilket är en positiv effekt i både miljö och socioekono-miska perspektiv.

Detta examensarbete syftar till att utvärdera och förbättra den makro-skopiska trafikefterfråge- och överflyttningseffektsmodellen som används för Norrköping. Examensarbetet kommer ytterligare utvärdera om en generell Logitmodell för överflyttningseffekter kan utvecklas samt vilka parametrar som är viktigast att inkludera i modellerandet av färdmedelsval. Detta har gjorts med hjälp av trafiksimuleringsverktyget Visum. Färdmedelsvalsför-delningen beräknas med Logitmodeller kodade i Python-skript integrerade i Visum. Därefter görs en nätutläggning i Visum, som beräknar nya resti-der som används som indata på nytt i skripten. Denna iterativa procedur fortsätter tills systemet når jämnvikt. Examensarbetet syftar till en mer trovärdig skattning av överflyttningseffekten genom att korrigera Python-skripten samt skatta Logitmodell parametrar utifrån resvaneundersökning-ar.

Arbetet visar att restider är den viktigaste faktorn för att generera re-alistiska resultat. Dock inkluderade indatan inte tillräckligt med informa-tion för skattning av Logitmodellparametrarna. Restider behövde beräknas manuellt, vilket gjordes med två olika metoder. Även om ingen av des-sa metoder kunde tillhandahålla goda resultat på överfflyttningseffekten, kunde de bevisa vikten och signifikansen av riktigt underlag av restider när man utvecklar en trafikmodell. Resultatet av arbetet påvisar betydelsen att fortsatt utveckla en metod för att beräkna restider, då indatan inte är tillräcklig, och påvisar att restidsparametern är situationsanpassad.

Acknowledgments

First and foremost, I would like to express my gratitude to everyone in-volved in the work of my Master Thesis, and especially to those who have provided me support and information consistently throughout the project. I would like to thank my supervisor, Nikolaos Tsanakas, for you dedication and hard work to lead my on the right track in the work of my Master The-sis. Your interest, guidance and support have been crucial for this report. Further more, I want to thank Anders Peterson for your valuable feedback and support.

I would like to thank my supervisor at Ramböll, Johan Swärd, for your commitment, knowledge and help. I would also like to thank all the employ-ees at Ramböll for welcoming me to the company and providing me with support and valuable knowledge. I would especially like to thank Veron-ica Forsberg for introducing this project to me, without you, this Master Thesis would not have been conducted.

Finally, I would like to thank my family, for supporting me, through-out up and downs, and inspired me to reach my goals. I would especially thank those who have helped motivating me and constantly supporting me to follow my dreams.

Norrköping, May, 2017 Therése Ziedén

Contents

List of Figures 1

List of Tables 4

1 Introduction 5

1.1 Background . . . 6

1.2 Aim & Purpose . . . 7

1.3 Methodology . . . 7

1.4 Limitations . . . 8

1.5 Outline . . . 8

2 Introduction to traffic modelling 10 2.1 Traffic models and network flow models . . . 11

2.2 Macroscopic traffic flow models . . . 12

2.3 The four step model . . . 12

2.3.1 Trip generation . . . 14

2.3.2 Trip distribution . . . 14

2.3.3 Mode choice . . . 14

2.3.4 Route choice . . . 15

3 Estimation of discrete choice models 16 3.1 Multinomial Logit models . . . 16

3.2 The utility function and its beta parameters . . . 18

3.3 The nested Logit model . . . 19

3.4 Parameters affecting the transition between travel modes . 20 4 Computer tools used 24 4.1 Visum . . . 24

4.2 BIOGEME . . . 26

4.3 Python . . . 28 viii

5 The Visum model and scripts 29

5.1 The model set-ups . . . 29

5.1.1 Input data . . . 31

5.1.2 Parameters used in the model . . . 35

5.1.3 The result of the original model . . . 39

5.2 Changes in the model set-ups . . . 40

5.2.1 Improving the scripts . . . 41

5.2.2 Input data used to re-estimate parameters . . . 42

5.2.3 Estimation of travel times . . . 46

5.2.4 Re-estimation of parameters . . . 48

6 Numerical Experiments & Result 52 6.1 Tram line to Vrinnevi . . . 52

6.2 Tram line to Vrinnevi, new parameters . . . 58

6.3 Travel time effects . . . 62

7 Analysis 66 7.1 The new model set-uo . . . 66

7.2 Discussion of the numerical experiments . . . 68

7.3 The effect of the beta parameters . . . 69

8 Conclusions & future work 71 8.1 Conclusions . . . 71

8.2 Answers to the main questions . . . 71

8.3 Future work . . . 73

List of Figures

1 Illustration of the decisions included in the four step model. 13 2 The set-up of the Logit model, with the three alternatives

car, blue bus, and red bus. . . 19 3 The set-up of the nested Logit model with the first two

ternatives car and public transport, and the second two al-ternatives blue bus and red bus. . . 20 4 The structure of three smaller areas in Karlstad, the transit

line is illustrated as a thick black line. . . 23 5 IIllustration of the Demand, Network and Impact models,

and how the result is presented in Visum. . . 25 6 The syntax of a model file, used in this thesis, which is

exe-cuted by Biogeme. . . 27 7 The framework of the traffic model. The Visum model and

the Python-scripts exchange input util an equilibrium is reached. 30 8 The framework of the procedure sequence, most steps are

represented by a Python script. The red arrow illustrates an iteration process, where the iteration stops when a equilib-rium is met. . . 31 9 The travel mode distribution of the trips within the

munic-ipality of Norrköping, based on the travel survey reportcon-ducted by Trivector. . . 33 10 The travel mode distribution of the trips within the

munic-ipality of Norrköping, according to the travel survey from Östgötatrafiken. . . 33 11 The travel mode distribution of based on both travel surveys

from Östgötatrafiken and Trivector. . . 34 1

2 LIST OF FIGURES 12 The travel mode distribution of based on both travel

sur-veys from Östgötatrafiken and Trivector, weighted with the passenger scale 1.4 (passengers per vehicle). . . 35 13 The transformation of distance using the natural logarithm. 38 14 The Box-Cox transformation of distance. . . 38 15 The modal choice distribution from the Original model, only

including internal trips. . . 39 16 The weighted modal choice distribution from the Original

model, only including internal trips. . . 40 17 The travel mode distribution when neglecting the alternative

specific constants. . . 40 18 The errors found in the modal choice script, which are marked

with arrows. . . 41 19 The travel mode distribution of the internal trips within the

municipality of Norrköping, without weight. . . 42 20 The travel mode distribution from the unfiltered travel

sur-vey from Trivector. . . 44 21 The travel mode distribution from the unfiltered travel

sur-vey from Sweco. . . 44 22 The travel mode distribution from the unfiltered

combina-tion of both travel surveys. . . 45 23 The travel mode distribution from the filtered input data, as

a combination of both travel surveys, used in the estimation of new parameters. . . 45 24 The travel mode distribution from the filtered input data, as

a combination of both travel surveys, used in the estimation of new parameters. . . 46 25 An example of how the curve fitting tool provided by Matlab,

uses the travel time and distance measures (black dots) to create a relation/function between the measures. . . 47 26 The internal travel mode distribution, unweighted, using the

curve fit method. . . 51 27 The internal travel mode distribution, unweighted, using the

matching method. . . 51 28 The original public transport network in the municipality of

Norrköping, where the blue bars are bus lines and the red bars are tram lines. . . 53

LIST OF FIGURES 3 29 The new public transport network, with a tram line going

through the residential area of Vilbergen (in the south area of Norrköping instead of buses, where the blue bars are bus lines and the red bars are tram lines. . . 54 30 The transition to public transport, using the original beta

parameters. . . 55 31 The transition to public transport using the original beta

parameters but no alternative specific alpha. . . 56 32 The transition to public transport with the new estimated

parameters, with travel times calculated in the curve fitting tool in Matlab. . . 57 33 The transition to public transport with the new estimated

parameters, with travel times calculated using a matching method in Matlab. . . 57 34 The transition to public transport, using public transport

travel time parameters for the . . . 61 35 The transition to public transport, using public transport

travel time parameters. . . 62 36 The transition to public transport, using only the in–vehicle

travel time for the public transport utility calculations. . . . 63 37 The transition to public transport using weighted travel times

for public transport. . . 64 38 The travel mode distribution using the curve fit method,

for all tests in the forecast model for 2035, both the base scenarios and the scenarios with the new tram line. . . 65

List of Tables

1 The total number of users changing travel mode to public transport. . . 21 2 Input data used to develop the original model. . . 32 3 The original parameters used for the modal choice decision.

The β parameters are the estimated values found in the ta-ble. . . 36 4 Alternative specific constants found in the script for the

travel mode decision. . . 37 5 The difference in number of working trips and working trips

with car as travel mode, before and after correcting the scipts. 42 6 The new estimated parameters, based on the curve fit method. 49 7 The new estimated parameters, based on the matching method. 50 8 The total number of users changing travel mode to public

transport. . . 58 9 The estimated parameters with the travel time for public

transport included, where the travel time have been esti-mated in the curve fitting tool in Matlab. . . 59 10 The estimated parameters with the travel time for public

transport included, where the travel time have been matched using Matlab. . . 60

Chapter 1

Introduction

Traffic engineering and planning has an important role in urban develop-ment and infrastructural investdevelop-ments. The aim of traffic planning can be to prepare for a future travel demand and maintain accessibility regard-less of travel mode. Cities and communities change and grow over time, and therefore traffic investments should be able to adapt to future changes. Traffic planning solutions must also progress to maintain sustainable cities and communities in the future, as well as increase citizens quality of life.

Mathematical models and computer tools can be used in traffic planning processes to identify problems, generate and evaluate possible solutions, and develop plans for final solutions. To produce reliable traffic planning solutions corresponding to reality, models must be calibrated. Over the years, several traffic planning models have been introduced, implemented, and calibrated. These models have now become more advanced and can accurately describe real traffic situations. However, the focus has previously primarily been on the private transportation sector.

In the past years increased car ownership, urban sprawl, and the need for personalized mobility, have led to a higher share of private vehicle trips [1]. This has caused congestion problems in some cities. Therefore, the focus of traffic planning has moved towards the public transport system, with the aim to decrease congestion and air pollution, and maintain socio-economical sustainability. Public transport is considered to offer mobility enhancement and energy conservation.

With the new focus on the public transport system, new models must be developed and calibrated to facilitate the traffic planning processes. How-ever, the models describing the effect an investment can have on a travel mode choice, are usually associated with several uncertainties. The effect depends both on the traffic assignment and the level of detail of the modal

6 Introduction choice model.

1.1

Background

Because of the growing interest in maintaining sustainable and environ-mental cities, the attention to the public transport system is increasing. The public transport system is described to be sustainable from an eco-nomic, social and environmental point of view. Traffic investments require analysis of the current traffic situation and an estimation of the future traf-fic situation, to predict possible outcomes. The public transport system has therefore become an important aspect in traffic analysis, maintaining sustainable investment results.

Nowadays, it is of interest to achieve a higher share of public transport users, and develop urban areas that promotes the usage of public transport. Fewer private vehicles in a city have positive a effect on the environment and can decrease congestion. When travellers change their travel mode thanks to a change in the infrastructure is called the transition effect. The trend today is to straighten public transport lines, to achieve shorter travel times and a higher share of public transport users. In this thesis the tran-sition from mainly private vehicles to public transport is in focus. The transition effect has previously been calculated differently for each case, since no general model exists. This means that modal choice parameters must be estimated for each case, which is time-consuming. It is therefore of interest to investigate the possibility of developing a general model that can be used in more than one specific case.

The transition effect is affected by several factors, and is perceived dif-ferently by every individual. It can therefore be hard to take account for all the important parameters in a traffic analysis. The transition effect can be modelled on a macroscopic simulation level, using computer soft-wares such as Visum and Emme. Depending on the simulation tool, there are different ways of implementing parameters affecting the users travel mode choice. One of the most common ways to describe the modal choice decision is to use a Multinomial Logit (MNL) model, which is a tool for modelling discrete choices. The Logit model includes a utility factor which has been estimated for each available alternative. Estimation of a utility factor includes measurable parameters, such as travel time and costs, and users perceptions of the available travel modes. A modal choice decision can also be modelled using a Nested Logit (NL) model, in cases where similar alternatives are available.

1.2 Aim & Purpose 7 therefore an appropriate method or model is required to analyze the tran-sition effect. How the model is developed and which parameters that are used have an impact on the result. The outcome of this report is a pro-posal on how to describe the modal choice using a Logit model and which parameters are most important to be included. The report will also discuss how a general transition effect model can be developed and adjusted for different cases.

1.2

Aim & Purpose

One aim of this thesis is to evaluate and possibly improve the modelling of the transition effect of an already developed traffic simulation model. The thesis focuses on the transition from private vehicles to public transport in the specific case of the municipality of Norrköping. The project aims to reach a method that is applicable in more cases, other than for Nor-rköping, and in different scenarios. How the logit model can be adjusted and which parameters are the most important to include will analyzed be reviewing related work and reestimating the parameters for this specific case. The conclusion aims to answer if the developed model is appropriate to use, if adjustments are required and how or why the model have to be reconstructed. The main questions to be answered in the conclusion are:

• How can the logit model be adjusted in the modelling of the transition effects?

– How can important parameters to include in a transition effect analysis be found?

– Can some estimated beta parameter values be used in more than one case and how are they used?

• How can a nested Logit model be used to evaluate the transition effect?

My contribution with this project is a suggestion on how to improve the developed model and how the Logit model can be adjusted in order to analyze the transition effect of a traffic investment.

1.3

Methodology

The development of a general transition effect model requires input data for the estimation of the parameters. The traffic simulation model has

8 Introduction already been developed and calibrated and cover the city of Norrköping. The input data used in the original model will be used for re-estimating beta parameters used in to improve the simulation model and modify the used logit model. The software BIOGEME will be used for the estimation of the beta parameters. The new parameter values will be compared to the beta parameter values used in the original model, which were estimated using Stata. The new parameter values will also be compared to result in related studies, found in the literature review. The traffic simulation model is implemented in the traffic simulation tool Visum, which uses Python scripts for the traffic assignment procedure. The modal choice Logit model is calculated in one of the Python scripts, and therefore Python will be a part in the evaluation of the transition effect model.

The transition effect model will be evaluated using a forecast scenario. The forecast case is an extension of a tram line from the inner city to the hospital in Vrinnevi in Norrköping. Today, the transition effect result has been questioned for this case using the simulation model of the city of Norrköping.

1.4

Limitations

The focus of this thesis is on the third step of the four-step model, which is the modal choice. As mentioned earlier, a developed macroscopic traffic model developed has been provided, where the modal choice have been implemented using Python scripts. The development of a transition effect model is therefore going to be implemented in Python scripts and in Visum. The evaluation will only be based on the traffic model of Norrköping and the estimation of new parameters will be based on travel surveys from the municipality of Norrköping. The estimation of the beta parameters will be preformed using BIOGEME, this was chosen since it is an open source software. The main focus is to evaluate how the re-estimation of beta parameter affect the result.

1.5

Outline

The report begins by generally introducing traffic modelling in Chapter 2, whee traffic flow and assignment are presented. This is followed by a deeper description of multinomial Logit models ,and the estimation and calibration of parameters included in the Logit model, in Chapter 3. The computer tools used in this project are presented in Chapter 4. The simula-tion model is presented in Chapter 5, where a descripsimula-tion of the model, the

1.5 Outline 9 modal choice script, and my contribution, are presented. The numerical experiments are presented in Chapter 6, followed by a discussion regarding the experiments and the literature review in Chapter 7. Finally, in Chapter 8, the conclusions are drawn together with suggestions on further work.

Chapter 2

Introduction to traffic

modelling

Models are used in many different areas to describe the reality in a simplified way and can also be used to visualize the outcome of a project, such as infrastructural investments. Designing a model requires observations of the reality to be used as input data. The input data can, for example, be the design of the infrastructure in a city, the number of vehicles in the network, and the population size of the investigated area and surroundings. A model describes the reality with limitations, and cannot be a perfect reflection of a real situation. Model limitations are used to filter out unimportant information. For instance, it is not necessary to know how many floors there are in a building next to a street, when evaluating the capacity of the street. On the other hand, it is important to include parameters that affect the outcome of the model, such as one way streets in a macroscopic simulation or the signal scheme of a junction in a microscopic simulation. Models can be a useful tool to describe and simplify a real situation, to estimate future outcomes, or to solve eventual problems in a network.

Traffic models are an effective tool when planning and analyzing traffic situations and investments. Models can be used to generate measures, set up investment scenarios, and optimize or improve traffic situations. To maintain a reliable traffic simulation model, the model must be calibrated. The calibration is done by collecting as much data as possible from the concerned area, and comparing the input data with the output from the model. A traffic simulation model use links, nodes and zones, for example, to represent roads, junctions, and areas, as a network within a city. There are several mathematical models implemented in a traffic simulation model. These mathematical models are used to generate travel demand between

2.1 Traffic models and network flow models 11 an origin and a destination (OD), the route choices and the link traffic flows. There are also mathematical models to estimate the proportion between available travel mode alternatives. These mathematical models can use measured data as input to generate traffic flow between zones and the average travel speed on the links.

Depending on the size of the analyzed traffic network, different types of traffic simulation models can be applied. Large traffic networks, such as districts and cities, can more efficiently be modeled at a macroscopic simulation level, described in section 2.2. Smaller traffic networks such as a few blocks or even a single junction can be modelled at a microscopic simulation level. How the network should be modelled also depends on if the traffic should be simulated using flow rates, average speeds, and density rates, or using time headways, individual speeds, and distance headways. Macroscopic models use the average properties of a system and microscopic models uses individual properties of drivers, vehicles, and can model inter-actions between vehicles [2].

2.1

Traffic models and network flow models

In most traffic flow models, traffic flow is sometimes described as fluid. Traffic, just like fluid, can be described as a relation between flow, speed and density, known as the fundamental relation [3, 4]. The fundamental relation is a relation between flow, density and speed. The majority of microscopic and macroscopic traffic flow models were developed with the fundamental relation as a basis. Traffic simulation models are also built based on flow network models. A flow network has directed graphs where traffic flows. These has a maximum capacity, just as streets have in reality. When the maximum capacity has been reached, congestion occurs, and the traffic is jammed. Traffic flow networks are used to model traffic in a road system, both on a macroscopic and a microscopic level. The maximum capacity constraint is included in the calculation of route choices [5].

The traffic network in a macroscopic traffic model mainly consists of three parts: links, such as roads and sidewalks; nodes, such as junctions and roundabouts; and centroids, which are zones where trips starts and ends. The links can only contain one specific property, meaning that a street with varying number of lanes, must be divided into several links and connected with nodes.

12 Introduction to traffic modelling

2.2

Macroscopic traffic flow models

Macroscopic simulation models are used to describe how traffic flows in a large network, such as a city. The traffic is simulated using aggregated and averaged input data, such as density and speed [6]. A macroscopic model does not contain details of individual vehicles or interactions between vehi-cles, which can be simulated using a microscopic simulation model instead. Microscopic simulation models includes a higher level of detail, which fa-cilitates the modelling of individual vehicles and drivers behaviour in the traffic [2]. Microscopic models can be used to analyze smaller areas, such as junctions, or a few blocks. Thanks to a lower level of detail, one advantage with macroscopic models are the relatively simple calculations. Speed and density within a network, are calculated as an average on each link. Macro-scopic models are therefore a useful tool for traffic and travel forecasting, when evaluating large networks.

Since traffic plans take years before they can be fully implemented, infrastructure investments, for example, must last for a long time-period. This requires a strategic plan, which usually describes the current and future traffic situation in a stationary condition, during a specified time-period, most commonly during peak hour. The stationary condition and specified time-period simplifies the traffic flow calculations, resulting in average travel times and traffic flows to be evaluated. The assumption that no or very little information is lost when individual details are averaged is one of the disadvantages of macroscopic traffic models [2].

2.3

The four step model

When evaluating the potential beneficial outcome of a traffic investment and the transition from private vehicles to public transport, the possible changes in the network need to be predicted. A common transportation forecast tool is the four-step model, which will be described in detail in this chapter. The fundamental theory behind the four step model and travel forecasting is presented by: de Dios Ortuzar, Juan and Willumsen; Immers and Stada; Hydén; and WSP [7, 8, 9, 10]. These are the main references used in this section. Travel forecasting predicts how the traffic will change in the future according to the four steps in Figure 1.

2.3 The four step model 13

Figure 1: Illustration of the decisions included in the four step model.

The basic concept of the four-step model, is that a user is presented to a new set of alternatives in each step, where the probability of choosing an alternative is calculated, and is multiplied with the result from the previous step. The steps can be performed as Figure 1 presents, but nowadays it is more common that the first three steps are calculated simultaneously and the fourth step is performed separately. The first two steps result in a travel demand matrix, with information of how many trips are made between each pair of zones. This matrix is later used in step three to estimate the travel mode share in each OD-pair. The fourth and final step also uses the traffic demand matrix produced in step two.

The four-step model can be used to investigate changes in a travel pat-tern, and is a useful tool for analyzing the effects from new road designs, alternation of the public transport supply, or implementation of tolls, for example. Travel forecasting makes it possible for the decision maker to determine which investment can be the most beneficial in the future, and which changes can lead to the most efficient transition between personal vehicles to public transport, for example.

In this report the focus is to evaluate the transition from private vehicles to public transport and therefore the most important step in the four step model will be the mode choice. For deeper descriptions of the four-step model see de Dios Ortuzar & Juan Willumsen and Immers & Stada [7, 8].

14 Introduction to traffic modelling

2.3.1 Trip generation

The first step of the four-step model predicts the total number of trips that begin or end in every zone included in a studied area. The trips can be categorized depending on the travellers purpose. These trip purposes can be work trips, school trips or shopping trips, for example. In trip generation, socio-economic attributes such as income, car ownership, family size or accessibility to urban areas, is accounted for. The land use is included as well, since zones containing workplaces, industries, schools, sport centers and shopping centers attract trips. Socio-economic input data, such as employment statistics for each zone, is also included in the attraction model. The trip generation can be estimated including physical measurements [11]. The trip generation step results in the total number of trips produced and attracted from each zone, and the total number of trips in the entire system.

2.3.2 Trip distribution

In this step the produced and attracted trips, are connected between a start zone to an end zone. From Figure 1 we can see that the destination decision is followed by the initial question to travel or not. How a traveler choose a destination in a system depend on the trip purpose. The result of the trip distribution step is a proportion of how much traffic that flows between each pair of zones. The distribution is presented as a travel demand matrix or an origin to destination matrix. Depending on the amount of information available, the complexity of the OD-estimation varies. If we have input data including estimations of both the number of trips produced and attracted the model must satisfy both conditions, and is therefore double constrained.

2.3.3 Mode choice

The purpose of this step is to estimate travel mode share in a system, which is the step with most importance for this report. Depending on the investigation area, the available travel mode alternatives have to be adapted. There are several parameters affecting the mode choice, these can be both qualitative and quantitative factors. Qualitative factors are individual and consists of comfort, convenience and safety for example. Quantitative factors can be travel time and travel cost. The parameters that affect the transition between private vehicles and public transport are described more in detail in Chapter 3. The travel mode choice can also be constrained depending on the user. The constraints of choosing an alternative could be of legal regulation, such as not owing a driver’s license, or not afford a car, or other individual characteristics. Factors such

2.3 The four step model 15 as purpose of the journey, what time of the day the journey is taking place, if the user is traveling alone or with others, also affects the mode choice.

To maintain a reliable simulation model, the most important factors must be included in the mode choice estimation. A valid assumption, in the mode choice model, is that trips are made on a tour basis, meaning that a traveler will go both forth and back with the same travel mode. A common way of calculating the distribution between the modal alternatives is to use the multinomial Logit model.

2.3.4 Route choice

Knowing the travel demand and which travel modes are going to be used, the final step is to distribute the travellers on the available routes or lines in the system. The assignment is done for each mode separately, since there is large variation in the transportation systems for different travel modes. The traffic assignment depends on the users behaviour and preferences, such as minimizing travel distances or travel times. Although, in traffic modelling it is assumed that all users have the same information about a network and they all have the same goal of minimizing travel time or distance. A route assignment is most often solved by using a user equilibrium principal. The user equilibrium principle states that no traveler wants to change its route, since the route cost or travel time is the same for all travellers, when equilibrium is achieved [12].

The route assignment for private vehicles and public transport does not work exactly in the same way. This is important to have in mind when working towards a higher transition effect from private vehicles to public transport. Public transport assignment models are used to predict the distribution of passengers between the lines [13]. The public transport lines between two nodes in an OD-pair can be used by different public transport modes, such as bus, tram or metro. The line services are maintained by a pre-defined number of vehicles, and the capacity of a line is linked to the passenger capacity of a vehicle and the frequency of a service.

Both the private vehicle route assignment and the public transport as-signment uses an impedance function when determining the travel time. The impedance calculates the travellers unwillingness to cunduct a trip, which increases when the travel time increases. The impedance is simi-lar for the route assignment and the transit assignment, but with some differences.When a traveler has decided to use public transport, the de-mand is assigned to different transit lines. How the dede-mand is distributed among the transit lines does not depend on the modal choice model in the developed model used in this thesis.

Chapter 3

Estimation of discrete choice

models

Discrete choice models are used to describe and predict choices between two or more alternatives. These alternatives can for example be which travel mode to chose, which route to travel, or the purpose of a trip. A discrete choice model uses statistics to relate a choice made by a person to the attributes of the person, and the attributes of the available alternatives. For example, choosing car as travel mode is related to a persons income, age, the distance between the origin and the destination, the cost of owing a car, the travel time, and so on. This chapter contains a presentation of the discrete choice models Multinomial Logit models and Nested Logit models, and how to estimate them using a maximum likelihood estimation method. Finally the most important parameters for travel mode decisions are presented and connected to the transition to or from public transport.

3.1

Multinomial Logit models

The distribution between different routes or modes can be calculated using the multinomial logit model [14]. Since the focus is on the mode choice, modal choice aspects will be considered in this chapter. To calculate the probability that an individual chooses a specifictravel mode alternative, several aspects needs to be accounted for. The logit model uses the utility of an alternative to calculate the distribution probabilities. An individual’s utility of an alternative, is a function of the characteristics of the alter-native and the users individual characteristics [8]. This means that two individuals facing the same identical choice situation, could choose differ-ent alternatives.There are several aspects accounted for when modelling a

3.1 Multinomial Logit models 17 decision, these can be:

• Available alternatives.

• The characteristics of the decision maker. • The characteristics of the alternatives. • Formation of the utility function.

These are described more in detail in Koppelman and Bath [14].

Depending on where we live, there are different travel mode alternatives available. The available alternatives in a city is the first limitation when choosing a travel mode. Even though there are many travel mode alterna-tives in a city, they might not be available for everyone, to travel by car requires availability of a owing a car or going as a passenger, for example, and a drivers licens if the user is driving. The travel mode choice is also dependant on how a user actually perceives a travel mode or the knowledge of the available travel modes. All these limitations affects the choice of a user and is used in the utility function, described later on, to decide how a user choses its travel mode.

The concept of the Logit model is that the decision maker must choose exactly one alternative from a set of alternatives, including the above lim-itations. The model is based on parameters on how a individual makes a decision. All individuals are assumed to choose the alternative that benefits themselves the most. The difficulty is to consider that all individuals have different preferences and conditions when choosing a travel mode, which implies that everyone acts and chooses differently. The characteristics of every individual is therefore important to consider. This is used to es-timate the parameters correctly and for the model to describe as many unique choices as possible.

The alternatives can be categorized depending on their attributes and there are two principals that can be used for this categorization. The first category contains attributes which are the same for all alternatives and the second contains of attributes that only is valid for one or a subset of alternatives. In-vehicle travel time, for example, is only applicable on mo-torized vehicles, and not bicycle or walking. The perceived value of an alternative specific attribute is described using parameters. The parameter value is negative if the attribute is perceived negatively, such as the travel time or travel costs. The differences in the parameter value for the alterna-tives indicates how unappreciated an alternative is compared to the other alternatives.

18 Estimation of discrete choice models

3.2

The utility function and its beta parameters

As described in the section above, several aspects are included when facing a decision. These steps are all considered when formulating the utility of the alternatives. This means that the model assumes that an individual evaluates the trade-off between the alternatives, by knowing all the available information, and always chooses the alternative with the highest utility. The utility can be divided into a measurable/observable part and into an unmeasurable/unobservable part. Both are important to be include, since a decision can be effected by the decision makers’ individual characteristics and preferences. When all the data has been collected, and a model has been developed, it is common to realize that the outcome is unexpected and erroneous. This could be because of the following:

• Unobserved characteristics – It is possible to miss important charac-teristics of an individual or mode, due to lack of data for example. • Measurement errors – Such as travel times, bus frequency, car

avail-ability and more.

• Incorrect formulation of the utility function – The utility function is commonly formulated as a linear function, but in some cases, the characteristics should be combined to achieve an accurate utility. Knowing that these errors can occur, the result could include a range of variations. This is corrected for by rewriting the utility into a stochastic variable Ua, for alternative a, that consists of the non-stochastic utility Va and a stochastic error term εa. Sheffi [15] presents a formulation of the choice function and the multinomial Logit model, using the utilities and errors. The alternative specific stochastic utility function is formulated Equation 3.1, where Va represents the observable characteristics of a travel mode, such as travel time, gender, ticket costs for public transport and so on.

Ua= Va+ ea (3.1)

The observable characteristics function is formulated as in Equation 3.2, where Xa can denote, for example, travel time, gender and income. The

β are the parameters to be estimated for travel mode a, and which are in focus in this project. α is a constant separating the alternatives if all X parameters were to be zero, for example.

3.3 The nested Logit model 19 The choice probabilities can now be calculated using the linear multi-nomial logit model formulated, which is shown in Equation 3.3 where Pa is the probability of choosing alternative a, A is the set of all available alternatives and Va is the utility of alternative a.

P(a) = e Va

q

j∈AeVj

(3.3) More variables and parameters can be introduced to fit the traffic sim-ulation model to be used.

3.3

The nested Logit model

The multinomial Logit model sometimes produces unrealistic results. Ac-cording to Koppelman & Bhat, [14], the multinomial logic model has been criticized for its Independence of Irrelevant Alternatives (IIA) property. In the logit model the different choices are independent of each other, meaning that two similar alternatives (or almost the same) would be treated as two unique alternatives by the model. Koppelman & Bhat, among others, ex-plains the effect of the IIA property with the red bus/blue bus paradox. In the paradox, we consider the case were a commuter has the choice of going by car or the red bus to work, assuming the same probability of choosing both travel modes. Now assume that a blue bus is added to the alterna-tives, the blue bus has the same characteristics as the red bus, except for the color. This gives us the choice set up as illustrated in Figure 2.

Figure 2: The set-up of the Logit model, with the three alternatives car, blue bus, and red bus.

If commuters are indifferent to the color of the bus, the relative prob-ability between the buses would be pRedBus = pBlueBus. Because of the

20 Estimation of discrete choice models IIA property, and the relative probability between car and the red bus, we would then have that pCar = pRedBus = pBlueBus.

Knowing that the red and the blue bus have the same characteristics, except for the color, the share between the choice probabilities is not re-liable. The most logical outcome is that the commuters, that previously was choosing bus, would split equally between the red and blue bus, and that no one chooses to change the travel mode from car to bus. This can be modeled using a nested logit model. The nested logit model is divided into one or more subsets, giving the user the opportunity to first do a main choice, and one or more sub choices, see Figure 3.

Figure 3: The set-up of the nested Logit model with the first two alternatives car and public transport, and the second two alternatives blue bus and red bus.

In this case, a user first has two mode choices, either to go by car or by public transport, the probability share is pCar = pP ublicT ransport =

1

2. If the

user choose public transport, there is a new decision, to go by the red or blue bus. The probability share of choosing either of the buses is now pRedBus =

pBlueBus=

1

4. This is a more likely outcome when considering that the two

buses are almost identical. The nested Logit model set-up requires some changes in the utility functions, to maintain the result presented above. We must add the utility for the public transport alternative. This utility is a combination of the utility of the two bus alternatives and the utilities of the two buses are also modified to include the utility of going by public transport.

3.4

Parameters affecting the transition between

travel modes

Holmberg [16] discusses measures to increase the public transport share, and describes that everyday trips are determined by habits. It is likely

3.4 Parameters affecting the transition between travel modes 21 to go to work with the same travel mode, on the same route, every day. These habits are usually hard to breach with minor measures, the focus should, according to Holmberg [16], be on situations, such as, when people change jobs, or move to new places. Information about public transport can be sent out at these situations to influence people to change their travel habits, since they often must reevaluate their everyday travel when moving, for example.

Dickinson & Wretstrand [17] presents some of the most common pa-rameters which affect the utility of public transport. There are factors that are directly connected to public transport, such as travel time, frequency between buses, reliability, ticket cost, comfort, and so on. The focus of this report is on the travel time parameter. How much these parameters affects the choice is not the same for every traveler. Travel time one of the most important factor affecting the mode choice, which is showed in several studies, according to Persson [18]. The travel time for a public transport journey can be divided into sveral parts, such as walking time to and from the bus/tram stop, waiting time, in-vehicle time, and eventual transfer time [17]. The walking time to the bus stop and the waiting time at the bus stop is considered twice as demanding for the passenger, than the actual travel time on board. If delays occur, this is considered four times as demanding, than the travel time. Hydén [9] also discusses how travel time components for public transport are perceived. In Table 1 some travel time components are presented together with relative weights, that represents an individu-als perceived travel time. As seen in the table, the perceived travel time can vary depending on the weight used. These weights will be used in the analysis of the transition effect.

Table 1: The total number of users changing travel mode to public transport.

Travel time component Relative weight

In-vehicle time, seated 1

In-vehicle time, standing 1.5–5

Walking time to bus stop 2–5

Waiting time at bus stop 1–10

Transfer time 2–4

Delay time 9–19

Other factors that affects the public transport passenger share are pay-ment systems, ticket fares, departure frequencies, and the route network design [17]. Some successful public transport factors are route networks

22 Estimation of discrete choice models with a smaller number of transit lines, high departure frequencies and low ticket fares. The environment at the bus stops and transfer points has an impact on the decision to travel by public transport. Factors such as weather protection, lights, information systems, seating, access to service etc., have a positive effect on the decision.

There are some factors that affect the transition, from private transport to public transport, that are not directly connected to the public transport. The availability of parking lots near working places, parking fees and gas prices, indirectly affect the transition to public transport. Other positive transition effect factors can be longer travel time for private vehicles and shorter travel times for public transport by implementing bus lanes in the infrastructure. The land use, such as closeness to working places, transit stops and population density, has an indirect effect on the transition to public transport [17].

Other factors that affect the modal choice are individual perceptual experiences. The motivation of choosing, or not choosing, to go by private transport could be the experience of driving, perceived stress of driving, excitement, uncertainty, safety, enjoyment and more [19]. Elderly people tend to use motorized transport modes more often than non-motorized, due to limited mobility for example. In a study by [20] they found that travellers from high-income families in Asia tend to choose car because of comfort, privacy and status considerations. The travel mode can also be affected by our surroundings, such as the quality of walking and bicycle paths, the public transport services and the access to public transport. A car is also perceived to provide freedom, flexibility and convenience [19].

The design of the transit network has an important role in the opera-tional planning process [21]. It is easier to maintain a high quality transit line if the formation of communities, or urban areas, has a high density and a high population. Communities with low densities and larger distances to each other, tend to have a higher share of car users, and are harder to supply with a high quality public transport.

SOU [22] presents how the structure of smaller areas should be ac-counted for in the transit network design, see Figure 4. The first area, Råtorp, requires the shortest route to provide a high accessibility for the area population. From a safety point of view it is also of importance to consider the walkways to and from the bus stops when designing the transit network. In the experiments, a new public transport line is introduced and a old line is removed. The new line is drawn the same way as for Råtorp, replacing the line drawn as in Rud.

3.4 Parameters affecting the transition between travel modes 23

Figure 4: The structure of three smaller areas in Karlstad, the transit line is illustrated as a thick black line.

According to Persson [18], straightening of transit lines can have a posi-tive effect on travel times, and can lead to shorter waiting times and higher departure frequencies. These improvements have a higher positive effect on the decision, than the negative effect of eventually longer walking times to the transit stops.

Chapter 4

Computer tools used

In this chapter the computer tools used in this thesis is presented together with a description on how they work and which role they have in the work with this study.

4.1

Visum

PTV Visum, developed by the PTV group, is a software for traffic analysis and travel forecasting. The software use geographical information system (GIS) based data to visualize traffic systems. Visum is used to plan and analyze transportation systems including travel demand of both private and public transport, and can also be used for bicycling and walking. Visum is also used to evaluate traffic situations and investments, both with todays travel demand and a future travel demand, estimating the effects traffic can have on a system. A traffic model requires information of the travel demand within a current system, the road network and which methods are going to be used to produce a result.

The main focus in the project is to evaluate how the transition effect can be modeled more realistic. This is achieved by evaluating, and improving, an already developed macroscopic simulation model for the municipality of Norrköping. The model is built using external Python scripts. As explained previously, Visum includes the four step model for travel forecasting. In the model used in this project, this is instead done outside of Visum, ex-cept for the user equilibrium/route assignment. After a transport network have been built in Visum, the network is assigned with traffic. This can be done in several ways, depending on the network, what is evaluated or implemented and so on. It is therefore required to tell Visum what has to be done and in which order. Running a simulation in Visum is done from

4.1 Visum 25 the P rocedureSequence. Here it is defined what sort of procedures Visum have to perform to assign traffic. The procedure sequence of the developed car network model used in this thesis, is shown in Figure 5.

Figure 5: IIllustration of the Demand, Network and Impact models, and how the result is presented in Visum.

As seen in the figure, it is in the Procedure Sequence the Python scripts are called from. Visum uses the scripts to assign traffic to the network and the result of an assignment can be visualized in many different ways. Two methods, used in this thesis, is the volume on the links in the car network and the number of passengers per line between stops in the public transport network. To visualize the transition to public transport in different zones,

26 Computer tools used an attribute was created an added to the zones list. The values of these attributes was calculated using Excel, and are the difference in the total number of public transport users in two scenarios. The zones are given a color, depending on which range the attribute value belongs to.

4.2

BIOGEME

The implementation of discrete choice models are widely used in several sci-entific areas, such as transportation demand analysis. These days, models are expected to describe reality with high accuracy. This requires estima-tions of the model parameters based on input data from the real situation, which are used in the calculation of the Logit model. Biogeme (BIerlaire’s Optimization package for GEV Models Estimation) is an open source package, designed to estimate discrete choice model parameters. Since Biogeme is free to use this software was chosen to estimate new beta pa-rameters used in this thesis. The background of Biogeme and how it works is presented in an article and a conference paper by Michel Bierlaire [23, 24] and will only be briefly described in this thesis.

Biogeme requires two types of input files to be executed. A data file, of the format .dat or .txt, and a model specification file, of the format .mod. Biogeme uses the data file to estimate the parameters according to the model specification file. The data file contains observed or measured data such as travel times, distances, gender, income, and so on. The data file also contains the information of which travel alternative a user have chosen in a real situation, which is used as a base in the model specification file. The model specification file is built with sections, and there are five sections that must be included. These are: Choice; Beta; Utilities; Expressions; and Model. In Figure 6, the Biogeme model specification file, used to re-estimate the parameters in this thesis, is shown.

4.2 BIOGEME 27

Figure 6: The syntax of a model file, used in this thesis, which is executed by Biogeme.

28 Computer tools used The first section, Choice, describes the choice variable, in this case it is the travel mode. The Beta section describes a list of the parameters that must be estimated. In the next section, U tilities, the utility function for each available alternative is specified. In the Expressions section, there is information on how to compute attributes that might not be available in a data file. The last section, M odel, tells biogeme which type of model is going to be used for the estimation, in our case we want to work with a multinomial logit model.

When executing a model with Biogeme, several output files are gener-ated. The estimated beta parameters are presented in a report output file. The t-test is an important parameter in the output file. The t-test is a statistically hypothesis test, and in this case it is used to test whether or not the beta parameters have a significant effect on the result or not. If a beta parameter is significant, it is larger than |1.96|. From the report file, the new beta parameter values are provided and these are the values implemented in the modal choice script later on.

4.3

Python

Python is a general purpose programming language, which is designed to offer a high readability of codes and allows fewer lines of code than Java and C++, for example. This is possible since statements doesn’t need to be closed, using end statements. Python is instead using whitespace to indicate which blocks code belongs to.

A script is a smaller piece of code that executes something and usually don’t benefit from being compiled, they are instead being interpreted. In-terpreting a script in Visum can, for example, be a way of editing matrices, or as in the case of this project, calculating the four step model. Python is not designed to be a scripting language, but it does also work as a nice scripting language.

Chapter 5

The Visum model and

scripts

The traffic simulation model used for the transition effect evaluation, has already been developed and implemented in the macroscopic traffic simula-tion tool Visum. The model is based on the four-step model and is built to include several travel modes. This makes it possible to evaluate the effect of investments or measures, with respect to the mode choice. The model does not include chain trips, meaning that a user is assumed to only travel to one single destination from home and back. It is also assumed that the user travels with one mode during the trip, and uses the same travel mode both from home and back, except for walking to the bus stop, which is included in the model. The original model is described in the section below and is based on a memorandum (MEMO) of the development of the model, which is provided by Sweco. In Section 5.2 suggestions on how to improve the model are presented.

5.1

The model set-ups

In this section the original model is described, the model is complex and contains of three version files, eight python scripts and ten distance and travel time matrices. The description is relevant for the work of this thesis, and is not a complete description of the entire model. How the model works and how all parts are connected is illustrated in Figure 7.

30 The Visum model and scripts

Figure 7: The framework of the traffic model. The Visum model and the Python-scripts exchange input until an equilibrium is reached.

From Figure 7 we can see that the Visum model and the Python-script exchange input iteratively throughout an execution. The connection be-tween Visum and Python is described more in to detail in Figure 8.

5.1 The model set-ups 31

Figure 8: The framework of the procedure sequence, most steps are represented by a Python script. The red arrow illustrates an iteration process, where the iteration stops when an equilibrium is reached.

5.1.1 Input data

The data used in the model was provided by the municipality of Norrköping, Östgötatrafiken and from travel surveys. The aim was to recreate the trip purpose choice and travel mode choice as accurate as possible, using the provided data. The travel surveys were conducted during 2010, by Trivector, and 2014, by Östgötatrafiken. The travel survey from Trivector was sent to 5 800 persons in the age 16 – 84 in Norrköping municipality and the response rate was 42 %.

Östgötatrafiken sent their survey to 30 000 persons in the county of Östergötland, from age 16 and up, with a response rate of 39 %, where 2 174 of the answers came from residents in Norrköping municipality. The surveys have been combined and analyzed together because of the low response rate. A small response rate could lead to large varieties, and not reach the statistical accuracy required. The persons not answering the surveys could also have a different travelling behavior, which can affect the result

32 The Visum model and scripts from the model. The surveys also contained some erroneous results, for example, the number of trips completed by the response group does not correspond to the number of places visited in the surveys. Sometimes the origin or destination, or both, where missing for a trip in the surveys. This erroneous data was removed from the data set.

Information on the land use is another important input to the model, this information is mainly of the population in the municipality, the city for-mation and structure, geographical forfor-mation and job opportunities. The statistics used is from 2014 and was provided by the municipality of Nor-rköping. The main input data used in the model is shown in Table 2.The gender share in the municipality is even, as seen in the table, but the share is varying in the different districts.

Table 2: Input data used to develop the original model.

Population in the municipality 135 256

Job opportunities 61 400

Inhabitants with drivers license 81000

Inhabitants owning one car (or more) 45 900

Gender distribution in the municipality 50 % males & 50 % females

Statistics about the household types were used in the model and divided into houses and apartments. Attraction points, shopping centers and recre-ation areas is of interest to locate in a traffic forecasting model. However, these were hard to locate because of the low details in the land use data, therefore variables were implemented to represent these attractions.

Statistics of trips into Norrköping, such as commercial traffic and pri-vate traffic, either by car or public transport, was collected from Sampers. Sampers is a forecast model, which estimates the travel demand on a na-tional or regional level. Sampers models the travel demand for individuals, with the travel modes: car; public transport; train; flight; bicycle; and walk. Commercial traffic is divided into three groups; with private vehicles (such as cabs and craftsman), truck without trailer, and trucks with trailer (the last two referred to as heavy traffic).

Since a focus of this project is connected to modal choice, the dis-cussion of the result from the experiments will be based on a comparison between the modal choice distribution from the travel surveys and the orig-inal model. The travel survey conducted by Trivector, was presented in a report which presents the travel mode distribution shown in Figure 9.

5.1 The model set-ups 33

Figure 9: The travel mode distribution of the trips within the municipality of Norrköping, based on the travel survey reportconducted by Trivector.

The travel survey from Östgötatrafiken resulted in the travel mode dis-tribution shown in to Figure 10.

Figure 10: The travel mode distribution of the trips within the municipality of Norrköping, according to the travel survey from Östgötatrafiken.

These figures only include car trips without passengers, the public trans-port mode includes bus and trams, no trains, and all other travel modes except for walk and bike are neglected. Since the two travel surveys did not include the same options regarding travel mode alternatives, the re-sult had to be filtered to be compared on a fair basis. They both have a similar distribution of car travellers, which is around 60 %, and cyclists, which is between 11 – 13 %. Although, there are big differences between the public transport users and pedestrians. The travel survey from Trivec-tor resulted in 7 % public transport users, while Östgötatrafiken presented

34 The Visum model and scripts 20 %. Trivector measured 19 % pedestrians and Östgötatrafiken presented 11 %. Because of these differences, the results from both reports where combined and the travel mode distribution based on both travel surveys are presented in Figure 22.

Figure 11: The travel mode distribution of based on both travel surveys from Östgötatrafiken and Trivector.

The Visum model does not generate car trips including passengers. These have to be weighed in manually after simulating the model. Weigh-ing the modal choice distribution, with the scale factor 1.4 [25] passengers per vehicle, for the combinations of both surveys results in the distribution presented in Figure 12. The scale factor 1.4 was chosen since Norrköping is a smaller city, where people tend to go by private vehicles more than public transport.

5.1 The model set-ups 35

Figure 12: The travel mode distribution of based on both travel surveys from Östgötatrafiken and Trivector, weighted with the passenger scale 1.4 (passengers per vehicle).

Figure 11 and Figure 12 will be used in the analysis for comparisons of the experiments with the original model and input data.

5.1.2 Parameters used in the model

The assignment of traffic follows the four step model, where the calculations are based on the multinomial Logit model, explained in Chapter 3.

The parameters connected to the modal choice is divided into param-eters that are specific for an alternative and those that are common for all alternatives. The travel mode decision, as well as the entire four-step model, is implemented in Visum using Python scripts. The scripts are ex-ecuted from the procedure sequence, but can also be exex-ecuted separately. The travel mode decision script is included in the user equilibrium assign-ment, meaning that the script re-calculates the travel mode probabilities for each new user equilibrium solution until a convergence criterion is met. The travel mode probability calculations have a high complexity since it depends on each OD-pair, and each trip purpose, resulting in a four-dimensional matrix, P (i, j, tp, tm), where i and j are the zones, tp is the trip purpose, and tm is the travel mode alternative. There are 189 zones available in the network, six trip purpose alternatives and four travel mode alternatives. This high complexity results in high run times, on a computer with an Intel core i5 processor it takes about an hour to execute the Visum model. To be able to generate a travel mode distribution, several factors are accounted for. The distribution is calculated using a multinomial Logit model, as described in Chapter 3. In the original model they have used the