Universal Service in the provision of transport to small ... · Universal Service in the provision of transport to small islands A travel choice approach to the evaluation of levels

Universal Service in the provision of transport

to small islands

A travel choice approach to the evaluation of levels of service

Sara Correia de Oliveira Levy

(Licenciada)

Dissertação para obtenção do Grau de Mestre em

Engenharia do Ambiente

Júri

Presidente: Prof. Doutor António Jorge Gonçalves de Sousa, DEMG, IST

Orientação: Prof.ª Doutora Maria do Rosário Maurício Ribeiro Macário, DECivil, IST

Co-orientação: Prof. Doutor Costas Panou, STT, University of the Aegean

Vogais: Prof. Doutor José Manuel Caré Baptista Viegas, DECivil, IST

Setembro de 2009

ii

“No man is an island, entire of itself; every man is a piece of the continent, a part of the main; if a clod

be washed away by the sea, Europe is the less...any man's death diminishes me, because I am

involved in mankind...”

(John Donne, 1624)

iii

ACKNOWLEDGEMENTS

I would like to thank all the people that made this project possible. First, I would like to thank Professor

Costas Panou for welcoming me in Greece, for scientific support and for having introduced me to a

new reality - the reality of Greek islands, filled with both beauty and hardship.

To Professor Rosário Macário goes my gratitude and admiration for her work.

I owe an enormous deal to the group of students of the MSc in Shipping in Transport and International

Trade of the University of the Aegean: Filipa, Kostas, Nikos, Maria, Kyriakos, Vicky, Giannis, Navsika,

Sofia, Eleny and Theodora. They worked hard to obtain the data used in this dissertation.

I would also like to express my gratitude to the Transportnet team, both Professors and colleagues. I

learned a lot from them and we had some fun together.

Additionally, I would like to thank Michel Bierlaire, David S. Bunch, Amalia Polydoropoulou and André

Duarte for precious advice on the mechanics of discrete choice models.

I would also like to thank Instituto Superior Técnico (IST), Fátima Figueira and the coordinators of the

Environmental Engineering Master for being flexible and allowing me to develop this master thesis

from abroad.

Finally, and most importantly, I would like to thank Tiago for unconditional support, love and

enthusiasm over long methodological debates.

Ευχαριστώ

iv

ABSTRACT

The transport system serving the islands is of lifeline importance to the islanders, providing access to

essential goods and services that people elsewhere take for granted. Notwithstanding, remote islands

are often transport deprived.

The concept of Universal Service can provide the basis for a new approach to the island transport

problem. Universal Service refers to the provision of adequate transport services to all users,

irrespective of their geographical location, at a specified quality and reasonable price. The goal of this

research is to understand how different levels of service fulfil the ideal of “adequate transport to every

user”. We develop an integrated framework for evaluating transport opportunities based on the

analysis of travel choices, and apply it to the case of a Greek island.

Our findings show that the mode choice and the travel decision are different although inter-related

decisions. Mode choice depends on price, travel time, frequency and the characteristics of the user.

Travel choice depends on the preferences of individuals in terms of mode choice. In case the

preferred alternative is not available, individuals who prefer less expensive modes will judge other

modes mainly based on price. If price is considered too high, the islanders will prefer to cancel the trip,

independently of any compensation in terms of travel time. Individuals who prefer the most expensive

alternatives will judge the disutility of travelling mainly based on the possibility to return home as soon

as the activity is finished.

These results have implications for Universal Service. They imply that replacing less expensive

services, such as ferry boats, with faster services on the basis of the trade-offs implied by mode

choice models might not be appropriate.

Key Words: transport, islands, peripherality, universal service, discrete choice modelling, multinomial

logit model

v

RESUMO

O sistema de transportes que serve as ilhas é de uma importância vital, dado que fornece acesso a

bens e serviços essenciais. Contudo, as alternativas de transporte disponíveis são muitas vezes

consideradas insuficientes pelos ilhéus.

O conceito de Serviço Universal pode servir de base a uma abordagem alternativa à questão do

transporte para as ilhas. Serviço Universal define-se como o fornecimento de serviços de transporte

adequados, de dada qualidade e a preços razoáveis, a todos os utilizadores independentemente da

localização geográfica. O objectivo deste estudo é avaliar o nível de serviço fornecido às ilhas em

função do ideal de “serviços de transporte adequados para todos os utilizadores”. Neste sentido,

desenvolveu-se uma metodologia para a avaliação das opções de transporte disponíveis nas ilhas

baseada na análise das escolhas dos residentes das ilhas. Esta metodologia foi aplicada ao caso de

um ilha grega do mar Egeu.

Os nossos resultados mostram que a escolha do meio de transporte e a decisão de viajar são

escolhas diferentes embora relacionadas. A escolha do meio de transporte depende do preço,

duração da viagem, frequência e das características do utilizador. A decisão de viajar depende das

preferências dos indíviduos em relacão ao meio de transporte. Caso a alternativa preferida não esteja

disponível, indíviduos que preferem alternativas mais económicas, vão decidir viajar ou não com base

no preço. Caso o preço seja considerado demasiado alto, os ilhéus vão preferir cancelar a viagem.

Indivíduos que preferem alternativas mais caras, vão julgar as alternativas dísponíveis

essencialmente com base em quão cedo podem regressar a casa.

Estes resultados têm implicações importantes em termos do fornecimento de serviços de transporte

às ilhas. Substituir serviços mais económicos, como os ferry-boats, por serviços mais caros, com

base nos mecanismos de compensação sugeridos pelos modelos de escolha de modo de transporte

pode não ser apropriado.

Palavras Chave: transporte, ilhas, periferalidade, acesso universal, modelos de escolha discreta,

modelo logit multinomial

vi

TABLE OF CONTENTS

1. INTRODUCTION ......................................................................................................................9

1.1. MOTIVATION................................................................................................................................. 9

1.2. RESEARCH OBJECTIVES........................................................................................................... 10

1.3. STRUCTURE OF THIS DISSERTATION....................................................................................... 11

2. LITERATURE REVIEW ..........................................................................................................13

2.1. REPERCUSSIONS OF INSULARITY............................................................................................ 13

2.2. TRANSPORT SERVICES IN THE ISLANDS................................................................................. 15

2.3. RATIONALE FOR UNIVERSAL SERVICE.................................................................................... 17

2.4. STUDIES OF ISLAND TRANSPORT DEMAND ............................................................................ 21

2.5. STUDIES OF INSULAR ACCESSIBILITY..................................................................................... 22

3. METHODOLOGY ...................................................................................................................24

3.1. METHODOLOGICAL APPROACH ............................................................................................... 24

3.2. CHOICE MODELLING.................................................................................................................. 28

3.2.1. DISCRETE CHOICE MODELLING........................................................................................................... 28

3.2.2. MODEL STRUCTURE............................................................................................................................. 29

3.3. ASSUMPTIONS AND LIMITATIONS ............................................................................................ 32

3.4. LOCAL CONTEXT ....................................................................................................................... 32

3.5. OPERATIONALIZATION OF VARIABLES.................................................................................... 34

3.6. SURVEY DESIGN ........................................................................................................................ 37

4. RESULTS...............................................................................................................................40

4.1. SAMPLE DESCRIPTIVES ............................................................................................................ 40

4.2. MODEL ESTIMATION .................................................................................................................. 43

4.2.1. MODE CHOICE....................................................................................................................................... 43

4.2.2. VALUE OF TIME ..................................................................................................................................... 52

4.2.3. TRAVEL CHOICE.................................................................................................................................... 53

5. DISCUSSION OF RESULTS...................................................................................................62

5.1. MODEL RESULTS ....................................................................................................................... 62

5.2. EVALUATION OF TRANSPORT OPPORTUNITIES...................................................................... 64

6. CONCLUSION AND OUTLOOK .............................................................................................67

6.1. CONCLUSION AND CONTRIBUTIONS........................................................................................ 67

6.2. FURTHER WORK ........................................................................................................................ 68

vii

LIST OF TABLES

Table 1 Frequency of ferry services in the Aegean islands (August vs January).................................16

Table 2 Distribution of traffic by trip purpose on samples from studies made in the Aegean islands ...34

Table 3 Variables for mode and travel choice models........................................................................35

Table 4 Survey: scenarios for different trip purposes .........................................................................39

Table 5 Results of Choice experiments A and B................................................................................42

Table 6 Specifications of utility of alternatives for the Models A1 and A2............................................43

Table 7 Relevant statistics and parameter values for Models A1 and A2............................................44

Table 8 Specifications of utility for Models A3 to A5...........................................................................45

Table 9 Relevant statistics and parameter values for Models A3 to A5 ..............................................45

Table 10 Specifications of utility for Models A6 to A8.........................................................................46

Table 11 Relevant statistics and parameter values for Models A6 to A8 ............................................47

Table 12 Specifications of utility for Models A9, A10 and A11............................................................48

Table 13 Relevant statistics and parameter values for Models A9 to A11...........................................49

Table 14 Specifications of utility for Model A12..................................................................................50

Table 15 Relevant statistics and parameter values for Model A12 .....................................................50

Table 16 Relevant statistics and parameter values for the Mode Choice Model .................................51

Table 17 Values of Time based on the Mode Choice Model ..............................................................52

Table 18 Value of a day wait based on Model A11 ............................................................................52

Table 19 Specifications of utility for Models B1, B2 and B3................................................................53

Table 20 Relevant statistics and parameter values for Models B1 to B3 ............................................54

Table 21 Specifications of utility for Models B4 and B5......................................................................55

Table 22 Relevant statistics and parameter values for Models B4 and B5..........................................55

Table 23 Specifications of utility for models B6 and B7......................................................................56

Table 24 Relevant statistics and parameter values for Models B6 and B7..........................................57

Table 25 Specifications of utility for Models B8 to B10.......................................................................58

Table 26 Relevant statistics and parameter values for Models B8 and B10........................................59

Table 27 Specifications of utility for Model B11..................................................................................59

Table 28 Relevant statistics and parameter values for Model B11 .....................................................60

Table 29 Relevant statistics and parameter values for the Travel Choice Model ................................61

Table 30 Probability to travel depending on Price and Household income..........................................64

Table 31 Probability to travel depending on Price and Age ................................................................65

Table 32 Probability to travel depending on Price and Travel experience...........................................65

Table 33 Probability to travel depending on trip and islanders characteristics ....................................66

viii

LIST OF FIGURES

Figure 1 Population change between 1990 and 1999 in average annual percentage change. ............14

Figure 2 Framework for the islander’s travel decisions.......................................................................26

Figure 3 Comparison between the Normal and the Gumbel (or type I Extreme Value) distributions:...30

Figure 4 Location of Greece in Europe. .............................................................................................33

Figure 5 Location of Chios Island in the Aegean Sea.........................................................................33

Figure 6 Survey: Choice experiments A and B...................................................................................38

Figure 7 Histogram of age of respondents.........................................................................................41

Figure 8 Distribution of respondents regarding level of education ......................................................41

Figure 9 Distribution of respondents by activity..................................................................................41

Figure 10 Distribution of respondents per monthly household income................................................42

LIST OF ACRONYMS

ASC - Alternative Specific Constant

i.i.d. - identically and independently distributed

MNL - Multinomial Logit Model

PSO - Public Service Obligations

SP - Stated Preference

USO - Universal Service Obligations

-

9

1. INTRODUCTION

1.1. MOTIVATION

Insularity can be seen as an extreme form of peripherality. Islands differ markedly, however, from

other remote places, since their peripherality is a trait of permanent character. According to the

European Islands Association, Eurisles (1997), “the result is that these territories suffer an overall

handicap which makes illusory any hope that they might, without a deliberate policy on the part of the

European Union and the Member-States concerned, be able to face up to the challenges posed by the

single European space on a relatively equal footing”.

Islands are usually highly dependent economies due to a weak domestic market and dominant role of

external trade (Eurisles, 1997). Baldacchino (2006) takes it further, arguing that small islands “are by

definition open economies and their survival strategies are intimately connected with the ability to

source and obtain income, transfers and ‘‘rents’’ from beyond their shores”.

In this context, the transport system serving the islands assumes a crucial character. Small and

remote islands often fail to provide the necessary demand to be of commercial interest to private

transport operators. In a context of full liberalization of maritime and air transport, remote islands might

not be served at a sufficient level (Panou, 2007).

The provision of transport services to the islands is an issue of particular importance to Greece.

Greece has circa 6 000 islands, of which 227 are inhabited, but only 78 of those have more than 100

inhabitants. The islands represent about 18% of the Greek territory, and are home to approximately

12% of the population. Some authors (Chlomoudis, Pallis, Papadimitriou and Tzannatos, 2007) have

alerted to the need “to point out those islands communities that face special difficulties that need to be

addressed and, where possible, sustainable solutions must be designed and implemented in the light

of available resources”. Lekakou (2007) stresses the need to provide “uniform and consistent levels of

service all year round; offering reasonable prices for both passengers vehicles and freight; and

providing a minimum of socially acceptable service to non-commercial destinations”.

The European Union (EU) has acknowledged the necessity to protect island routes that are

considered of lifeline importance to the regions concerned. In this context, island routes assume the

status of public service, and Member States are allowed to impose Public Service Obligations (PSO) -

obligations to provide service in non-commercial routes - upon shipping operators, as a prerequisite to

allow them to provide island cabotage services. Typically, fares and frequencies are appointed on the

basis of cost benefit analysis, historical entitlements, or as the result of competitive tendering

procedures. Nevertheless, both academics and island representatives (Eurisles, 2003; Chlomoudis,

Pallis, Papadimitriou and Tzannatos, 2007; Panou, 2007), keep alerting to fact that current levels of

PSO do not guarantee a socially acceptable minimum level of transport provision, since they are not

implemented universally or consistently. Until now, PSO have been signed up on a case-by-case

10

basis, with no uniform rule as to the establishment of minimum frequencies, mandatory ports, the

affordability of fares, and the obligation to provide continuity of service.

Panou (2007) suggests that “the application of Universal Service Obligations (USO) could significantly

improve the quality of transport services” to the islands. Universal Service refers to the provision of

adequate transport services to all users, irrespective of their geographical location, at a specified

quality and reasonable price. Thus, Universal Service corresponds to the extension of the public

service to the whole territory (of a State), and its application under a uniform rule, to guarantee socially

accepted minimum levels of service provision. USO are a regulatory policy tool commonly applied in

network industries such as telecom, postal services, water, gas and electricity. Ultimately, the

implementation of USO requires setting target levels of service and a financing mechanism. However,

the Universal Service goal can provide the basis for a new approach to the evaluation of levels of

service in the provision of transport to the islands.

1.2. RESEARCH OBJECTIVES

In this research, we focus in understanding how different levels of service fulfil the Universal Service

ideal of “adequate” transport to every user. We argue that this concept can provide the basis for an

alternative approach to the problem of island transport1, shifting the focus from the supply system to

the user - in this case, the islander. It is important to note that, although the definition of Universal

Service implicates all users, it ultimately concerns the improvement of the transport opportunities

available to the islanders.

The main goal of this research is to devise a methodology to evaluate the transport opportunities

available to the islanders in light of the Universal Service concept. The key idea is to evaluate the

transport opportunities available to every islander in terms of their “adequacy, quality and price”. The

adequacy of transport opportunities is evaluated from the islander’s perspective, and expressed

through the impact of different levels of service on the travel choices made by the islanders. Quality, in

the context of this research, refers to the travel time and frequency of the services. The expected

results are the following:

� To develop a methodology to evaluate the transport opportunities available to the islanders.

This methodology should:

� Reflect the users’ perspective: The evaluation of transport services should comprise criteria

1 This was thoroughly discussed in earlier versions of this research, presented at the 49th European Congress of the Regional Science Association International, in Łódź, Poland 25-29 August (Levy, 2009a) and at the European Transport Conference, 5-7 October, 2009 (Levy, 2009b).

11

that are relevant to the user, and thus avoid the full-fledged exogenous imposition of what

constitute adequate transport services. On the other hand, these criteria have to be

balanced with supply system considerations, in order to avoid the risk of excessive

specification of provision against a very low price.

� Be universal: It should be able to incorporate the views of different types of users. This

requirement points to the use of a disaggregate approach, based on the individual

preferences of the users, and considering different socio-demographic groups.

� Relate to objective characteristics of the transport services: The evaluation of transport

opportunities should functionally depend on objective characteristics of the transport

services. It would be of little use to measure, for instance, the degree of satisfaction of

every user of the transport system, without having any insights about what this satisfaction

depends upon. On the contrary, a methodology that can relate the users’ perceptions of the

transport services to operational parameters such as the price of the fare or the duration of

the trip, can serve as a useful policy or forecast tool.

� Be trip-purpose specific: The adequacy of transport opportunities is bound to differ

according to the purpose of the trip. Furthermore, there is an on-going debate over whether

Universal Service should encompass only basic access (access to merit goods or services)

or all of the islander’s trips, independent of purpose.

� Be able to evaluate the impact of transport policies on the population, and specifically on

particularly vulnerable parts of the population, such as low-income groups or the elderly.

� To apply this methodology to the case of an island. The methodology is applied to the case of

the Greek island of Chios, in the Aegean Sea.

1.3. STRUCTURE OF THIS DISSERTATION

This dissertation is structured in six chapters. Chapter 2 reviews the corpus of relevant academic

literature on the subject of islands and island transport. First, it provides a brief overview of the

implications of insularity in the economy of the island and the islander’s quality of life. Second, it

discusses the question of island transport and how it has been dealt with in the context of liberalization

of maritime cabotage. Third, the possible rationale for the imposition of Universal Service Obligations

in the provision of services to the islands is examined. Chapter 2 closes by presenting studies of

islands travel demand and accessibility.

Chapter 3 presents the methodology for the evaluation of transport opportunities available to the

islanders. Chapter 4 presents the results and Chapter 5 discusses them. Chapter 6 presents the

concluding remarks and a description of how this research can be further developed.

12

13

2. LITERATURE REVIEW

2.1. REPERCUSSIONS OF INSULARITY

Islands that are dependent from a mainland State face a set of specific challenges to their economic

growth and prosperity. Adding to their insularity, most islands are small in size and peripheral in terms

of access to the main economic centres. Their openness, weak domestic market and limited resource

availability usually result in a high degree of specialization in export niche markets (Armstrong and

Read, 2004; Baldacchino, 2006). Some islands add to the burden of insularity problems related to

being mountainous and scattered across the sea (archipelagos). These are mutually reinforcing

factors that typically result in great vulnerability of island economies.

Research has suggested that these challenges do not necessarily imply worse than the average

economic performances. According to the work of Armstrong and Read (2004), while many EU islands

lacking autonomy or sovereignty are poor performers (relatively to their EU neighbours), there is also a

significant number of better than average performers - mainly, islands in the Mediterranean where

summer tourism abounds. Insularity seems to be a necessary but not sufficient condition for poor

economic performance. This does not contradict the fact that islands do face serious challenges to

economic growth. Some, however, have found mechanisms and policies that allowed them to

overcome these challenges successfully. Promoting tourism has often been the dominant strategy,

and in many cases, the sector has grown to constitute a disproportionately large part of the islands’

economy.

Depopulation and out-migration can also constitute serious challenges to island development. There is

great diversity among the islands with respect to population trends. According to Eurisles/Gederi

(2004) data, we find in the European islands two opposite trends: small and peripheral islands usually

tend to present negative demographic balances, strong migratory dynamics and an ageing population,

whereas islands with strong touristic inclination have grown at higher rates than the corresponding

country average. Figure 1 shows the migratory and natural population growth rates for most European

islands, in terms of average rates during the 1990s decade. The negative or only slightly positive

average growth rates (in any case, inferior the country average) of Gotland and Saaremaa (Sweden),

Bornholm (Denmark), Orkney and the Western Islands (Scotland, UK), Sardinia (Italy) and the

Northern Aegean Islands (Greece) contrast with the positive demographic balances of Gozo (Malta),

the Balearic Islands (Spain), the Southern Aegean Islands, the Ionian Islands and Crete (Greece), the

Åland Islands (Finland) and Corsica (France).

Cross and Nutley (1999) surveyed the inhabitants of nine islands off the west coast of Ireland in order

to test the hypothesis that depopulation in the islands was associated with poor accessibility and

service deprivation. In an earlier paper, Cross (1996) had used the same survey to test the relation

between depopulation and service deprivation, expressed through the islanders perceptions about the

14

general adequacy of the service levels available, and of the change in the standards of service

provision in the last 20 years. Although neither one of the two papers established a correlation

between accessibility, service deprivation and population change, the study still pioneers in what

concerns the relation between insularity, accessibility and depopulation.

Figure 1 Population change between 1990 and 1999 in average annual percentage change. Source: (Gederi, 2004), adapted

There is another interesting side to Cross and Nutley’s (1999) research. The data collected made it

possible to test if trip rate correlated with in-island service availability on the one hand and

characteristics of the supply system on the other hand. The first hypothesis was that trips to the

mainland would be encouraged by poor facilities on the island, or conversely, that a good range of on-

island services would make trips to the mainland less necessary. Therefore, lack of service availability

would be a driving force for travelling to the mainland. However, the authors found no evidence to

support this, suggesting that “it appears more likely that trips away from the island are influenced

mainly by transport opportunities” (Cross and Nutley, 1999). The characteristics of the supply system

(such as travel time, frequency, safety and others), acting as an impedance to the movements

between islands and mainland, appeared to be more influential in determining trip rates between the

islands and the mainland.

The fact that no correlation was found between what can be termed the need to travel and trip rates

should not discourage further investigation. In fact, virtually all transport theory is based on the

assumption that travel is a derived demand, set off by the desire (or need) to pursue certain activities,

and offset by the balance between the expected utility arising from the fulfilment of the activity and the

cost of travelling to where it takes place. Therefore, one might speculate that, in the case of the

islands, the perceived (or generalized) cost of travelling is too high, thus constraining the islander’s

displacements. Cross and Nutley (1999) suggest that in order to test this hypothesis it would be useful

15

to compare trip rates in the islands with those of rural areas on the mainland.

Hernández Luis (2004) makes a similar point, when analysing the quality of regular inter-island air

transport in the Canary Islands. He observes that demand for air and sea travel per capita is much

higher for the most remote islands, a fact that he attributes to the lack of other services, which are

available for the inhabitants the more central islands. In addition, he notes that air transport is used far

more heavily than shipping between islands where there are no high-speed ships, or where sailing

time is over 2h.

Kitrinou, Polydoropoulou and Bolduc’s (2009) are interested in the factors affecting the residential

relocation decision to island areas. Amongst other variables, the authors test the implications of

several policy scenarios pertaining to housing prices and the transport supply system on the

residential relocation choice. According to their results, travel cost and travel time of trips to/from the

islands were amongst the most significant variables affecting the residential relocation decision.

In some islands, especially those dependent on summer tourism, seasonality adds to the issues of

population decline and out-migration causing asymmetric demand for transport and other services.

Kizos (2007) describes the several seasonal dynamics of the Northern Aegean Islands: the

seasonalities associated with tourists and tourism workers that invade the islands from May to

October; the seasonal movements of university students arriving at the start of the academic year and

returning to their homes on holidays; and the diverse seasonalities associated with teachers, doctors,

nurses and military officers, that are “imported” from the mainland and stay in the islands for variable

periods of time, but almost always with a temporary character. This seasonal dynamics have

consequences for the transport system serving the islands, having to respond to high peaks in the

summer and holiday periods and peaks of low demand in the winter.

2.2. TRANSPORT SERVICES IN THE ISLANDS

The situation of the islands concerning infrastructure and service availability in European islands is

well documented by Eurisles/Gederi (2004), but lacks a comparison to the urban context and to the

rural peripheries. Larger islands tend, of course, to be better equipped, while it is increasingly difficult

to find more costly infrastructures, such as hospitals and universities, in islands where the number of

inhabitants is below 6000 and 10000, respectively.

Whether for financial reasons or because of the lack of a hinterland with sufficient population, small

islands tend to have limited infrastructure and service availability. The case of the islands would be no

different from the case of rural peripheries, if not for a critical difference, which is the almost total

irrelevance of car ownership for inter-island travelling. Studies of access to services made on mainland

peripheries always fall back on issues of mode split between public and private transport, or study

accessibility based on car distances, despite having warned us that this indicators ignore scattered

pockets of inaccessible locations for the car-deprived (see, for instances, Escalona-Orcao and Diez-

Cornago, 2007).

16

In this context, the transport system serving the islands assumes a crucial character. Illustrating this

point, ferry services connecting to the mainland are often termed the “lifeline” of the islands. The

ferries bring in essential goods and provide access to services that people elsewhere take for granted.

In some of the smallest islands, ferries transport children to mainland schools, act as occasional

emergency services, providing special sailings to take sick islanders to mainland hospitals.

Small and/or remote islands often fail to provide the necessary demand to be of commercial interest to

private transport operators. The situation is especially acute in the winter, when due to both lack of

demand and unfavourable weather conditions, transport supply exhibits a strong decrease. In the

Aegean Islands, for instances, at least 21% of daily connections from the islands are reduced to a less

than daily frequency, and at least 13% of the winter connections are made no more than once a week.

Table 1 Frequency of ferry services in the Aegean islands (August vs January)

(Source: Chlomoudis, Pallis, Papadimitriou and Tzannatos, 2007, adapted)

Service frequency Summer Winter %∆ = (W - S)/S

At least once a day 50,5% 40,0% - 21%

2 to 6 times per week 40,0% 47,0% + 18%

once a week 9,5% 13,0% + 37%

The European Union has gradually been bringing about changes in the legislative corpus concerning

the provision of sea and air transport services, engaged in a process of growing liberalization.

Liberalization was expected to increase competition, generating benefits for passengers in the form of

reduced fares and improved frequency and service levels.

Sambracos and Rigas (2007) examine the evolution of the transport market in the Aegean area,

following deregulation of air transport (1999) and maritime transport (2004). The authors observe that

in the air passenger market, competition seems to have reduced fare levels by between 20% and 30%

in a first period but increased slightly after 2000. In the case of maritime transport, as the regulatory

framework changed, the incumbent companies expanded their operations. According to the authors,

after 2004, passengers had options between two operators on most major lines, although the market

did not become fully competitive. The authors subscribe to the view that, in the long-term, islands with

high travel volumes are expected to profit from improvements in services in terms of destinations

served, speed, and onboard services, while islands with low travel volumes will probably remain within

the framework of Public Service Obligation lines (Sambracos and Rigas, 2007).

The imposition of PSO in maritime transport is the result of the acknowledgement of the necessity to

protect island routes which are considered of lifeline importance to the regions concerned. In this

context, the EU provides Member States, or their "relevant authorities", with two options (Hache 1996):

� to sign Public Service Contracts with shipping operators providing regular services on routes to,

from, or between islands. A public service contract is a contract passed between the relevant

authorities of a Member State and a European Union ship owner, to provide satisfactory

17

transport services. It may mention conditions such as the continuity of these services, their

regularity, capacity or quality, the provision of complementary services, a suitable adaptation to

the demand, as well as matters related to fares, with specific conditions for certain routes or

certain types of travellers.

� to impose PSO upon all shipping operators as a prerequisite to allow them to provide such

island cabotage services. PSO is the obligation to provide a service which the shipping operator

would not provide, or would not provide so fully, if it was only paying consideration to its strict

commercial interest.

However, until now, the imposition of PSO has been characterized by arbitrariness. PSO have been

signed up on a case by case basis, with no uniform rule as to the establishment of minimum

frequencies, mandatory ports, the affordability of fares, and the obligation to provide continuity of

service (Panou 2007).

France has dealt with the issue of remoteness of its island territories in a different way - by invoking

the standard of “territorial continuity allocation”. This corresponds to “theoretically abolish the distance

and the sea” (Chambre de Commerce et d'Industrie d'Ajaccio et de la Corse-du-Sud), by means of

assigning to the maritime service the same conditions of frequency and fares than those of the SNCF

(French national railway company) on the basis of comparable distances. “Territorial continuity”

constitutes an example of the application of a uniform rule to set levels of service for island transport,

but one that relies on heavy subsidizing and does not include demand considerations.

2.3. RATIONALE FOR UNIVERSAL SERVICE

In the last two decades of the 20th century, a variety of regulatory policies were put in place in order to

smooth the transition to full market liberalization of services that were previously provided in the

sphere of the welfare state. While the need for the monopolistic provision of these services was highly

questioned, the idea of public service, i.e., that certain services should be made available to all

regardless of pure microeconomic considerations, remained relatively unchallenged (Cremer, Gasmi,

Grimaud and Laffont, 1998b). As a result, at the outset of the 21st century, most public utilities are

provided through a regulated market, but one in which every form of state intervention is highly

scrutinized for economic efficiency and rationale.

Universal Service is a regulatory policy tool commonly applied in network industries such as telecom,

postal services, water, gas and electricity. The key feature is that the regulator imposes on one or

more operators the obligation to provide full geographical coverage, at affordable prices. High cost

customers (usually customers located in remote areas) will then be charged below the real costs of the

service, at the expense of either low cost customers, state subsidies or other financing mechanism.

Four main arguments are usually offered in favour of regulation of public utilities through imposition of

USO (Cremer, Gasmi, Grimaud and Laffont, 1998a): redistribution, network externalities, public/merit

good and regional policies.

18

The first refers to the USO as a form of redistribution of welfare trough prices, instead of (or in addition

to) taxes or direct transfers (Cremer, Gasmi, Grimaud and Laffont, 1998a). This redistribution occurs

by means of the “affordable prices” condition, irrespective of whether Universal Service is financed

through tax money, or through cross-subsidizing the service to costly areas through increased prices

on profitable customers.

According to recent economic literature, such policies can be optimal in a second-best setting, “that is

when the policy makers do not have the necessary information to implement (potentially) more

efficient policies like direct transfers” (Cremer, Gasmi, Grimaud and Laffont, 1998b). In practice, other

motivations may underlie the adoption of these policies. Typically, direct money transfers suffer from

low public acceptability, and alternatives involving discriminatory pricing are usually adopted.

In the case of transport to the islands, it is sometimes argued that handing out direct subsidies to

islanders is a possible alternative to imposing PSO (Eurisles, 2003), with the advantage that it

imposes no distortion of the supply side. However, it has to be taken into account that direct

subsidization of users for transport ends is worthless if it gives rise to a proportional increase in the

prices of the fares, or if the recipients of the service decide not to use the subsidy on transport,

compromising the economic viability of the service.

On the other hand, one can argue that assuming that all islanders are income deprived is a very

coarse assumption, and thus the case for Universal Service is weak if based on the redistribution

argument. This argument can more easily be used to justify fare discounts for senior citizens, students

or low-income families or even direct subsidization of low income groups within the islands.

The second archetypical situation that calls for regulation is the case when extending the service gives

rise to network externalities that the market fails to recognize. In the case of transport to the islands,

although there are no network externalities in the strict sense, one can argue that increased frequency

of departures, diversity of schedules and the extent of the network increase the value of the network.

However, this is bound to be reflected in the passenger’s willingness to pay for the trip, and increased

demand for island travelling.

Waters et al. (1996) remark that, in the case of ferries, there can be a case for subsidy due to

economies of scale in waiting times borne by the users: in the case that demand for a particular ferry

route doubles, economies of scale in ship size will determine that the ferry company will prefer to

double the capacity of the vessel, instead of offering a service with double the frequency. However, as

identified by Mohring (1972), there are increasing returns in terms of waiting time borne by the users

when the frequency is doubled (average waiting time is roughly half than before). Therefore, according

to the author, there may be grounds for subsidy to “lower the price to encourage increased use and

provision of increased capacity”.

The case has also been made for other types of positive externalities - that more transport to the

islands can stimulate commerce, tourism and favour the location of companies in the islands. Waters

et al. (1996) contrapose that, in the event that these positive externalities occur, there is only case for

19

subsidy if we assume that such benefits are not adequately being captured in market demand for the

transport services. According to the authors, development induced by transport services generally

reflects on higher land values, and thus it is unlikely that the benefits go unappreciated.

An additional hurdle is that where subsidies are given under the rationale of fostering tourism and

economic activity, one must account for the fact that the economic opportunities attracted may be

diverted from some other region rather than created. In this case, the subsidies are inducing

distortions in the market, and economic efficiency would be maximized by reducing services until they

were matched by demand (Bennett, 2006).

The third argument concerns the case of public/merit goods. This argument relies on the idea that the

provision of the service is valuable in itself (Cremer, Gasmi, Grimaud and Laffont, 1998b); that there is

an option value in the provision of these services, that does not depend on the actual demand for the

services, but on their availability (Roson, 2001). This is typically the case of health, emergency

services or education, or of road access to small and remote villages in the mainland. In this case,

subsidies are given on a social or moral basis, rather than an economic one.

Roson (2001) remarks that in these cases, “as is typical for public and semi-public goods, then, the

observed market behaviour provides little information about the consumers’ valuation of the goods

and, consequently, about the optimal level of supply”. The author used contingent valuation to

understand how users value local public transport, independently of market behaviour. The author

asked interviewees to choose from a set of alternatives, representing different balances between

taxation and service frequency on two public transport links. He finds that being a user of those

particular public transport links has only a small (although significant) impact on the willingness to pay

(taxes) for better service. Since overall willingness to trade more taxes for better service is slightly

positive, this gives an indication that non-users also value improvements of the level of service of

public transport.

The question of transport as a merit good is rather controversial. On the one hand, it can be argued

that transport systems present some characteristics of merit: they provide access to essential goods

and services; they bind the nation together (Cremer, Gasmi, Grimaud and Laffont, 1998b); they are a

means of communication and information essential for a democratic society (Cremer, Gasmi, Grimaud

and Laffont, 1998b). On the other hand, the problem of the negative externalities arising from

increased levels of mobility, such as congestion, accidents and pollution, has made it difficult to build a

case for the merit of transport based on the right to move, even despite the fact that these externalities

only occur at high levels of consumption, which is not the case in the small islands.

Panou (2007) recognizes this duality when arguing that Universal Service concept “should

acknowledge that some transport activities are particularly important to society (they are considered

merit goods), and so justifies policies that favour services to access them (those considered to provide

basic access) over others (those considered less important)”.

Preston and Rajé (2007) deal with the problem in a similar way. They argue that the policy notion that

20

transport is a merit good and that every individual deserves a basic level of mobility “fails to recognise

that too much private mobility can contribute to social exclusion through environmental degradation,

adverse public health impacts, high accident rates, declining public transport, changes in land use and

community severance. Given this, the authors argue that policy makers should focus on ensuring

basic levels of accessibility (which they define as ease of reaching) rather than mobility (which they

define as ease of moving).

Fourthly, Universal Service can also be an instrument of regional policies (Cremer, Gasmi, Grimaud

and Laffont, 1998b) and territorial cohesion (Commission, 2004). The White Paper on Services of

General Interest specifically points out the outermost regions (mainly islands) as a paradigm for US:

“the access of all citizens and enterprises to affordable high quality services of general interest

throughout the territory of the Member States is essential for the promotion of social and territorial

cohesion in the European Union, including the reduction of handicaps caused by the lack of

accessibility of the outermost regions”.

At the national level, regional policy may also aim at protecting unique cultural niches, or at

compensating inhabitants of hardship locations. For instance, reduced transport costs can be a way

to encourage households and firms to locate in the islands, avoiding out-migration and the decline of

island life. In the case of the islands, this is probably the strongest rationale for subsidized transport.

However, this is not without controversy. If, on the one hand, it is unacceptable on ethical grounds to

exclude islanders from access to certain services or goods, on the other hand, there may be perverse

effects to increased access. First, if it is true that isolation is a developmental disadvantage, it is also

true that isolation was many times the factor responsible for the uniqueness of island culture and

geography that we aim to protect. Second, regional science literature has stressed the fact that

improving transport links between two places with different economic potential usually results in the

perverse effect of impoverishing the weakest. In the words of Vickerman et al. (1999) “improving the

links between the central and the more peripheral regions may make it easier for firms in the latter to

market their products in central regions, but also enables producers in these central regions to invade

peripheral markets previously protected by their remoteness”.

From the above it becomes clear that there is a thin line in what concerns transport as a merit or

demerit good: transport is only a merit good in the extent that it provides access to other merit goods,

services or activities.

Therefore, we argue that it is appropriate to think of Universal Service as a policy tool to provide

access to certain services, where the market fails to do so. Panou (2007) argues that Universal

Service should focus on basic access, “which should be clearly defined and distinguished from

discretionary travel”. Basic access “refers to people’s ability to access goods, services and activities

society considers high value (also called essential or lifeline)”, and includes emergency services

(police, fire, ambulances, etc.); public services and utilities; health care; basic food and clothing;

education and employment (commuting); mail and package distribution; freight delivery; and a certain

amount of social and recreational activities.

21

Preston and Rajé (2007) adopt a more ambitious view. The authors argue that “social exclusion is not

due to a lack of social opportunities but a lack of access to those opportunities”; and that “in order to

avoid social exclusion, an individual requires a set of accessible facilities and social contacts”. What

makes the case of the islands special is that they are limited in resources, and it is easily the case that

the basic set of facilities, social contacts or even essential services cannot be found within the island.

Moreover, the possibilities offered by transport are usually very limited compared to the ones offered

by the almost ubiquitous mainland transport network. In order to have access to some essential

services, islanders frequently have to travel to the next island or to the mainland.

2.4. STUDIES OF ISLAND TRANSPORT DEMAND

Mode choice has dominated the island transport research agenda. Studies concerning travel demand

analysis for island destinations mainly focus on the determinants of choice between air and sea

modes. There are a few applications of discrete choice models, such as those carried out by

Polydoropoulou and Litinas (2007) and Ortúzar and Gonzalez (2002) and Román et al. (2008).

Polydoropoulou and Litinas (2007) evaluate the determinants of choice between the available

transport modes (ferry, hydrofoil2 and two airlines) for the route between the Greek island of Chios and

Athens. Their results indicate that travel cost is the most significant explanatory variable and that

travel time also plays a significant role. Socio-economic characteristics such as education level,

income, age and being a soldier are also significant to the mode choice decision. In addition, the

authors estimate values of time for the alternative modes - approximately 5€/h for the ship and 19€/h

for the aeroplane.

Ortúzar and Gonzalez (2002) study travellers’ mode choice behaviour on the route between Gran

Canaria and Tenerife, in Spain. The authors specify total travel time (including waiting times), the fare

level and the supply capacity of each model as main explanatory variables for the choice between

aeroplane, hydrofoil and ferryboat. The estimated demand elasticities in relation to travel time and fare

levels show that, for the route studied, the aeroplane and the hydrofoil are close substitutes and that

competition is mainly based on travel time. Additionally, they implement income stratification of the

sample in order to determine the effect of income on mode choice, showing that marginal utility of

income decreases with for higher income strata, as expected.

Román et al. (2008) analyze the choice of airline in the main domestic routes connecting the

archipelagos of Azores, Madeira and the Canary Islands with the mainland and in-between them. The

authors conduct Stated Preference experiments facing individuals with the choice between two virtual

2 A hydrofoil is a boat with wing-like foils mounted on struts below the hull, faster than the ferryboat.

22

airlines which differed in terms of a group of service attributes that include price, frequency, comfort

and compensation for delay among others. They conclude that price, flight frequency, quality (or

availability) of food, the penalty imposed for changes in the ticket, the compensation in case of delay

and leg room are amongst the most important factors that represent the global service.

Other studies of travel demand, using different methodologies, point to similar results, such as the

ones carried out by Sambracos and Rigas (2007) and Rigas (in Press). Sambracos and Rigas (2007)

note that distance from Athens (measured in terms of travel time by boat) affects the modal split:

“passengers seem to prefer to travel by boat to closer destinations like Paros, that takes between 4

and 6 h”; “while the air mode has more than 50% of the split on trips to Rhodes - the most distant

island from Athens, involving a more than 12 h ferryboat journey”.

Rigas (in Press) focuses on the leisure passenger segment and studies the determinants of mode

choice between boat and aeroplane for passengers of the Greek Aegean sea market. The author

estimates cross elasticities of demand for the two modes, namely, the effects on the demand of sea

transport from reductions on air fares, and the effects on the demand of air transport from reductions

on boat trip duration. The results show that a small reduction in air fares would have little impact on

boat demand, but a reduction of more than 30% would more than double air travel demand. Likewise,

it would take a reduction in trip duration of more than 30% for air passengers to consider taking the

boat.

2.5. STUDIES OF INSULAR ACCESSIBILITY

Studies of insular accessibility have drawn measures of accessibility focused on the transport supply

side. Hernández Luis (2002) compares inter-island accessibility in the Canary Islands in two different

years, five years apart. The author chooses total travel time (including access and egress times, travel

time for the crossing and travel time over land) and time available at the destination (for different trip

purposes) as the main measures of temporal accessibility. According to the author, “the maximum

availability of time for passengers in certain places on the destination island is a very important

requirement of inter-island transport systems”; “this is because if the return trip cannot be completed

by ferry within one day, the costs increase considerably by either having to use air transport, if

available, or having to pay for a hotel room and losing part of the next working day”. The author found

that existing trip schedules do not allow people to take full advantage of the public administration and

commercial working hours. In some cases, a return trip on the same day was not possible.

Rutz and Coull (1996), in a study of the inter-island shipping network of Indonesia, quantify the

“efficiency of contacts in space” by calculating the overall journey time and weighted average speed

from the primary central node of the network to the most important ports in the outer islands. The

weighted average speed is the actual direct sea distance divided by the overall journey time (including

time spent at intermediate ports). According to the authors, the weighted average speed is a measure

of efficiency in spatial terms, accounting for the differential speeds of the vessels, the time spent at

intermediate ports and the detours necessary on various routes to provide a comprehensive service.

23

24

3. METHODOLOGY

3.1. METHODOLOGICAL APPROACH

According to the definition of Universal Service, its purpose is to provide every user with adequate

transport services. However, the particular criteria through which each user evaluates the adequacy of

transport services are unknown to the analyst. Nevertheless, although we might be unaware of the

criteria, we do know that the user bases its travel-related decisions on this evaluation.

The choice to travel, for a particular purpose, is the outcome of a decision process based on a

judgement made on (among other things) the available trip alternatives. If the outcome of this process

is the decision not to travel, this implies a negative judgement of the trip alternatives available to the

user. On the contrary, the choice to travel implies that there is an admission on the part of the user

that at least one of the trip alternatives available fulfilled his/her needs.

We develop an integrated framework for the evaluation of transport opportunities based on the

analysis of travel related choices. This framework is illustrated in Figure 2. Two decision processes are

relevant. The first is the choice of whether to travel or not, for a specific purpose and a given set of trip

alternatives. The second is the mode choice, since it yields important information on the trade-offs

between different trip attributes.

Destination choice is also relevant in a broader context. However, here we assume that a given trip

purpose already dictates a destination. This is not far from reality especially in the context of the

Aegean islands. Because of the radial shape of the transport network and a centralized country,

Athens can safely be assumed to be the relevant destination for almost any trip purpose.

The framework presented is based on Rational Choice theory and Random Utility Maximization theory,

in the line of what has been presented by Moshe Ben-Akiva, Daniel McFadden and many others.

According to Ben-Akiva and Lerman (1985), a framework for choice analysis should define the

following elements:

1. The decision-maker: In the framework of this research, the decision-maker is the islander.

Islanders have different socio-economic characteristics, which translate into different tastes.

Hence, different individuals value attributes in different ways.

2. The alternatives: The set of alternatives that are available and known to each individual. In

Figure 2, the mode choice is a choice between two trip alternatives (alt 1 and alt 2); and travel

choice corresponds to the choice between the alternatives “to go” and “not to go”.

3. The attributes of the alternatives: The alternatives are characterized by a vector of attribute

values. The attributes of alternatives may be generic (apply to all alternatives equally) or

alternative-specific (apply to one or a subset of alternatives). Typically, the attributes that play

25

a major role in the mode choice process are travel time, travel cost, frequency, comfort, safety

and reliability. In this case, we limit our analysis to travel time, travel cost and frequency, plus

purpose of the trip.

4. The decision rule: The decision rule is the mechanism that the decision-maker invokes in

order to process the available information that leads to a unique preference/ choice.

The decision rule may include random choice, habit, variety seeking, “follow the leader” behaviour or

other processes which we refer to as being irrational (Koppelman and Bhat, 2006). The process is

said to be rational when it satisfies two fundamental constructs: consistency and transitivity.

Consistency implies the same choice in repeated choice experiments under identical circumstances.

Transitivity implies a unique ordering of alternatives on a preference scale (if alternative A is preferred

to B and alternative B is preferred to C, then alternative A is preferred to C) (Koppelman and Bhat,

2006).

Discrete choice models are based upon the Utility Maximization rule. The Utility Maximization rule

posits that individuals will select the alternative with the highest utility value. This assumes that the

attractiveness of an alternative to an individual can be expressed in terms of a scalar value - utility.

Also, it implies commensurability (Ben-Akiva and Lerman, 1985), i.e., that there is a compensatory

decision process where "trade-offs" among attribute values are possible. Moreover, the utility

maximization rule implies that there is an objective function expressing the attractiveness of an

alternative in terms of the value of its attributes (Ben-Akiva and Lerman, 1985) and the characteristics

of individuals.

The utility function is not, however, necessarily deterministic. Experience with utility theory has pointed

towards the development of Random Utility theory. The key idea is that, if deterministic utility models

described behaviour correctly, we would expect similar individuals to make the same choices when

faced with the same set of alternatives (Koppelman and Bhat, 2006). However, we observe variation

in behaviours. Random Utility theory provides a reasonable representation of these unexplained

variations in travel behaviour. As with deterministic choice theory, the individual is assumed to choose

an alternative if its utility is greater than that of any other alternative. However, Random Utility theory

recognizes and accommodates our lack of information or understanding about the decision-making

process by describing preferences and choices in terms of the probability of choosing an alternative.

Utility is thus seen as a stochastic variable.

26

Figure 2 Framework for the islander’s travel decisions.

To better describe the framework illustrated in Figure 2, take the case of an islander living in one of

the Aegean islands, whose family has moved to Athens and is now inviting him/her to spend the

weekend there, enjoying the sight of the Acropolis. The islander will search for information on the

alternative trip modes available for his/her trip to Athens. If none of the trip alternatives are fully

satisfying, the islander will probably consider the different implications of 1) not going, thereby missing

out on a good time with his family, and 2) going, using either one of the available trip alternatives,

thereby supporting the associated monetary and time costs and general inconveniences.

The decision process illustrated in Figure 2 is not necessarily sequential. Most probably there is a joint

evaluation of all possible alternatives. However, this does not imply that the decision to travel and the

mode choice decision are the same. It implies only that the decision to travel depends on the available

trip alternatives.

Mode choice is the outcome of a comparison between the utility of the alternative modes available.

Utilities are latent constructs unobservable and unknown to the analyst, but reflected in the decision

outcome. According to our framework, the comparison between the utilities of the two trip alternatives

will depend on:

� The attributes of the trip alternatives

In Figure 2, alternative 1 and alternative 2 represent two different trip alternatives (for instances,

airplane and ferry boat), characterized by different set of attribute values. Relevant attributes include

price, travel time, departure and arrival schedule, frequency, comfort, safety and reliability. On a

brighter note, utility may also depend on the extent to which each mode allows the user to enjoy the

trip. The utility that each user derives from each of the alternatives will depend on these and other

unobservable variables. It is worth noting that attributes of the trip alternatives can only influence the

mode choice decision provided they are known (or inferred) by the decision maker. It is important to

take this into account especially when models are built on stated preferences over unlabelled

alternatives.

27

� The nature of the activity that compels the islander to travel (trip purpose)

Consider the case of an islander that, due to long endured back problems, has scheduled an

appointment with a very renowned orthopaedist in Athens. Due to the urgent and inflexible character

of the trip (it is very difficult to get an appointment with this doctor), this islander is bound to be less

sensitive to the price of the fare than when he is travelling to visit friends or family. In general terms,

users have been known to be more sensitive to price when travelling for leisure, while more time

conscious when travelling for work.

� Individual characteristics

Socio-economic characteristics of the decision makers have been proved to influence the way in

which individuals value the attributes of the different modes. In Polydoropoulou and Litinas’ (2007)

study on mode choice in the islands, age, education level and income level were found significant to

the mode choice decision.

The choice of whether to travel is the outcome of a comparison between the utility of travelling and the

utility of not travelling. Consider the case of an islander examining his alternatives for a particular trip

to the mainland. He/she may find he is not willing to support the costs associated with any of the trip

alternatives available, and therefore decide not to go. This decision is bound to depend not only on the

characteristics of the trip alternatives, but also on the characteristics of the activity motivating the trip.

For instances, it is intuitive to expect that leisure trips will be more easily cancelled or postponed than

health-related trips and work trips. Systematizing, the comparison between the utilities of the two

alternatives will depend on:

� The maximum utility derived from the available trip alternatives

The utility of travelling depends on the evaluation the islander makes of the trip alternatives available.

The islander will choose to travel if he finds that he is willing to support the costs associated with at

least one of the trip alternatives available.

� Trip purpose: the nature of the activity that compels the islander to travel

We expect that leisure trips will be more easily cancelled or postponed than health trips and work trips.

Additionally, the importance, urgency and degree of substitutability of the activity may influence the

decision process. Consider again the case of the islander with an appointment with the orthopaedist.

He may instead resort to his general medical practitioner working on the island. The islander will

compare the utility derived from seeking more specialized medical advice, while causing more hurt to

his back when sitting on the boat, with the utility of saving the inconvenience of travelling while not

getting as good a medical advice from the island doctor as from the Athenian physician.

The key idea is that islanders travel because they want to perform a specific activity, available on

mainland, but not available or less available (available in a degree that provides less satisfaction)

within the island. The islander will embark on a trip to another island if the difference in satisfaction

28

derived from performing the activity out of the island is higher than the costs of travelling.

� Individual characteristics

Socio-economic characteristics of the decision makers influence the travel decision. Besides the

expected effect of income on the travel decision, other effects might be considered. For instances,

older people may be more inclined not to travel since the inconvenience of travelling (independently of

the mode chosen) may be more decisive in the case of the elderly.

3.2. CHOICE MODELLING

3.2.1. DISCRETE CHOICE MODELLING

Discrete choice models are used to model choices over discrete alternatives, as opposed to models

built to describe continuous variables. To use a more comprehensive definition, discrete choice

modelling refers to a group of statistical procedures and techniques used for describing the choice of

one among a finite set of mutually exclusive and collectively exhaustive alternatives (Ben-Akiva and

Koppelman, 1974; Ben-Akiva and Lerman, 1985; Koppelman and Bhat, 2006).

The last decades of the XX century were prolific in theoretical developments concerning discrete

choice models, especially since the work of Nobel Laureate Daniel McFadden on the Multinomial Logit

Model (MNL) and on the general structure of the Generalized Extreme Value (GEV) class of models in

the seventies, later generalized by Moshe Ben-Akiva and Bernard François in the eighties (1983).

Since then, a number of such structures have been derived, such as the Nested MNL (Daly and

Zachary, 1976; Ben-Akiva and Lerman 1977; Williams 1977; McFadden 1978), the Tree Extreme

Value (McFadden, 1981), and the Ordered GEV (Small 1987). The first three are used for categorical

discrete choice problems (e.g., choice of mode to work), whereas the OGEV model is intended for use

with ordinal discrete choices (e.g., satisfaction response scales).

Discrete choice models have been widely used in the Transportation field, predominately applied to

problems of mode choice but also to destination choice, route choice, activity participation, auto

ownership and residential location, between others. The emergence of these models corresponds to a

change in paradigm in the Transport demand studies, from the aggregate approach (of which the

Four-step model is the main expression) to the disaggregate approach. Disaggregate models are

considered by most authors to be superior to aggregate models, or at least to have significant

advantages over it, since they are causal in nature, and explain behaviour at the relevant level - that of

the decision maker. Additionally, aggregation leads to considerable loss in variability, thus requiring

much more data to obtain the same level of model precision than the disaggregate models

(Koppelman and Bhat, 2006).

Ultimately, these models can be applied to a countless number of different problems, such as to

predict behavioural responses to policies, prices, trends or events, to estimate market shares for

alternative transport modes or to provide estimates of useful indicators such as Value of Time or

29

Willingness to Pay.

Notwithstanding this realm of possibilities, most applications of discrete choice modelling to the

transport field are largely concerned with questions of mode choice in the urban context, usually

involving the choice between private (i.e., car) and public transit. The focus is essentially on the home-

based work trip, while the modelling of home-based non-work trips and non-home-based trips has

received less attention in the urban travel mode choice literature (Koppelman and Bhat, 2006).

Intercity travel mode choice models are usually segmented by purpose (business versus pleasure),

day of travel (weekday versus weekend) and party size (travelling individually versus group travel).

The travel modes in such models typically include car, rail, air, and bus modes (Koppelman and Bhat,

2006).

There have been few applications of discrete choice modelling to the insular context. Polydoropoulou

and Litinas (2007) have presented a mode choice Multinomial Logit (MNL) model for island travelling.

The authors evaluate the choice between four alternative modes serving the island of Chios:

conventional ship, new technology ship, and the two airlines Olympic Airlines and Aegean Airlines.

The data was based on Stated and Revealed Preferences of Chios’ residents travelling from/to

Athens. Additionally, the authors declare to have tested other model structures to ensure that the

alternative choices of ship or plane were independent (not a justifying a Nested Logit), and that there

were no biases introduced by repeated observations (not a justifying a Mixed Logit Model).

Román, Espino et al. (2008) have used an MNL to model the choice between two virtual airlines which

differed in terms of a group of service attributes that include price, frequency, comfort and

compensation for delay among other.

Kitrinou, Polydoropoulou and Bolduc (2009) have developed a behavioural framework to describing

the factors affecting the residential relocation decision in island area, integrating within the discrete

choice model latent variables for capturing the decision makers’ attitudes and perceptions about

quality of life and transport on the islands.

3.2.2. MODEL STRUCTURE

According to Random Utility theory, the utility itU of each alternative i for an individual t is composed

of a systematic part itV , consisting of observable attributes of the alternative and characteristics of the

decision-maker, and a random component itε , usually called disturbance, representing the

unobservable portion of the utility.

ititit εVU += Eq. 1

The systematic part of the utility of an alternative is a mathematical function of the attributes of the

alternative and the characteristics of the decision maker.

The random component can be represented by a wide range of distributions. Different assumptions

30

will lead to different model structures. The assumption that the error term is normally distributed leads

to the formulation of the Multinomial Probit (MNP) probabilistic choice model. However, the use of the

MNP has been limited, due to its mathematical complexity.

The Gumbel distribution closely approximates the Normal distribution (see Figure 3) while

simultaneously generating a model structure that is easy to estimate, interpret and predict - the

Multinomial Logit Model (MNL).

Figure 3 Comparison between the Normal and the Gumbel (or type I Extreme Value) distributions: Probability density function (left) and cumulative distribution (right)

Source: (Koppelman and Bhat, 2006)

The MNL is based on the following assumptions: 1) the error terms are Gumbel distributed, 2) the

error terms are identically and independently distributed (i.i.d.) across alternatives, and 3) the error

terms are identically and independently distributed across observations/individuals.

If the error terms are independent and identically Gumbel distributed, the probability that a given

individual choose alternative i is given by:

( )( )∑

∈

=

Jj

j

ii

Vexp

VexpP

Eq. 2

Where jV is the systematic part of the utility function for alternative j.

This formulation implies that the probability of choosing an alternative increases monotonically with an

increase in the systematic utility of that alternative and decreases with increases in the systematic

utility of each of the other alternatives (Koppelman and Bhat, 2006).

Moreover, the S shape of the logistic function has gradual slope at extreme values of utility and much

steeper slopes at the centre of the graphic (where the utility of the alternative is close to the utility of

the remaining alternatives). This implies that when the utility of the alternatives is very close, marginal

increases in utility can induce large changes in the probability of the alternative being chosen. On the

other hand, when utility of an alternative is high enough or low enough, marginal changes in utility will

not have a significant effect in the choice probabilities (Koppelman and Bhat, 2006).

31

Another fundamental property of the MNL is that the choice probabilities depend only on the

differences in the systematic utilities of different alternatives and not their actual values. This relates to

the fact that the choice probability equation (Eq. 2) is unchanged if the same incremental value is

added to the utility of each alternative (Koppelman and Bhat, 2006).

The assumptions made on independence of the error terms leads to the property known as

Independence of Irrelevant alternatives (IIA), which states that the ratio of the choice probabilities of

any two alternatives is entirely unaffected by the systematic utilities of any other alternatives (Ben-

Akiva and Lerman, 1985). In other words, the remaining alternatives are irrelevant to the decision of

choosing between the two alternatives in the pair. How ever handy this property might come in terms

of model flexibility and computation, the assumption of independence of the error components may

sometimes not be applicable3. In those cases, use of the MNL will lead to erroneous results4. This

shortcoming of the MNL has lead to the development of other types of GEV models, such as the

Nested MNL, that relax the independence of error terms assumption.

The procedure for maximum likelihood estimation involves developing a joint probability density

function of the observed sample (Koppelman and Bhat, 2006), called the likelihood function (Eq. 3):

( ) ( )∏∏∈ ∈

∂=Tt Jj

jtjt βPβL Eq. 3

i if alternative j was chosen

=∂0

1jt

otherwise Eq. 4

Where jtP is the probability that individual t chooses alternative j; and β is the vector of parameters of

the model.

In order to find the parameter values which maximize the likelihood function, we should derivate the

likelihood function (Eq. 3), or instead, the log-likelihood function (Eq. 5), which yields the same

maximums, while being easier to compute.

3 This is true, for instances, when two alternatives share characteristics that were not made explicit in the systematic part of the model, such as the case of transit modes competing with a private mode. 4 Consider a mode choice model for which the probability of choosing between going by boat or taking Olympic Airlines or Aegean Airlines (two Greek airlines) yields 50%, 25%, 25%, respectively. If Olympic Airlines goes out of business (as it eventually happened) and stops being an alternative, it is counter-intuitive to think that their market share will now be 66% for the boat and 33% for Aegean Airlines. Most probably, people that used to favour Olympic will now favour Aegean more than the maritime alternative. The final market shares will probably be closer to 50% / 50%.

32

( ) ( )( ) ( )( )∑∑∈ ∈

β∂=β=βTt Jj

jtjt PlnLlnLL Eq. 5

3.3. ASSUMPTIONS AND LIMITATIONS

The choice models presented in this dissertation are estimated as Multinomial Logit models. MNL

models are based on the assumption of identically and independently distributed (i.i.d.) error terms

across alternatives and observations and individuals. Therefore, we do not take into account possible

biases do to taste heterogeneity and repeated observations. A Mixed Logit structure was also

estimated to evaluate the magnitude of these biases. This resulted in a statistically insignificant Sigma

parameter (that captures the unobserved heterogeneity within the sample population) and no

significant increase of goodness-of-fit.

Stated Preference (SP), as a method for data collection, is often criticized. Some authors argue that

there are discrepancies between stated and actual behaviour. Respondents are often overoptimistic in

estimating their own ability to, for instances, change modes. On the other hand, SP methods allow the

collection of more data with less effort, and more flexibility in the range of attribute levels treated, since

the alternatives do not have to correspond to currently existing alternatives.

In our models, the utility of the alternative not to travel (to cancel or to postpone the trip) has been

described in an oversimplified manner. We can posit that the utility of not travelling also depends on

characteristics of the activity (its urgency and importance) and on the alternative ways in which the

islander can substitute the activity carried out outside the island by an activity carried out within the

island.

3.4. LOCAL CONTEXT

We apply the methodology described in section 3.1. to model the travel related choices of the

inhabitants of the island of Chios, in Greece. Greece has circa 6000 islands, most located in the

Aegean Sea (Figure 4). The Aegean Sea is enclosed by continental Greece to the north and west,

Turkey to the east and the island of Crete to the south. Of the Aegean islands, 53 have more than 50

inhabitants (Kizos, 2007). Chios is the sixth largest of the Greek islands. It is located in the North-East

Aegean Sea, seven km off the Asia Minor coast (Figure 5). The island has a population of

approximately 52 000 people. The island is noted for its strong merchant shipping community, its

unique mastic gum and other traditional agricultural products.

Chios island is served by three shipping companies, that offer a total of two connections per day to

Athens, 5 days a week, and a single connection on the remaining two days. In the winter, this

frequency is reduced. The distance between the ports of Chios and Athens (Piraeus) is 153 naval

miles. Travel time from Chios to Piraeus varies from 7 to 9 hours using conventional ship. Prices for

economy class usually round 60 Euros for the round trip. Until November 2008, the trip could between

Chios and Piraeus could also be made by hydrofoil (fast boat), allowing for travel times of 5 to 6.5 h.

33

Figure 4 Location of Greece in Europe. Source: Wikipedia (Quizimodo)

Figure 5 Location of Chios Island in the Aegean Sea. Source: Google maps

Chios is served by two airlines, Olympic and Aegean Airlines. They offer a total of five flights daily from

Chios to Athens. The flight lasts 30 to 40 min. Prices round 200 Euros for the round trip.

Passenger traffic on Chios port varies between 10.000 passengers per month (in winter months) and

almost 80.000 passengers per month in the summer. Air traffic from Chios to Athens moves between

10.000 passengers per month in the winter and a little less than 30.000 passengers/month in the

summer (Polydoropoulou and Litinas, 2007).

Studies of travel demand carried out in the Aegean islands have showed that the main trip purposes

34

are work or business and leisure. Additionally, Sambracos and Rigas’ (2007) show that there is a

great difference in the relative proportions of these volumes between high and low season.

Table 2 Distribution of traffic by trip purpose on samples from studies made in the Aegean islands

Low season (2005)

(Sambracos and Rigas, 2007)

High season (2005)

(Sambracos and Rigas, 2007)

2001 - 2005

(Polydoropoulou and Litinas, 2007)

Trip purpose ferry airplane ferry airplane undifferentiated

Work or business 46% 26% 25% 13% 31%

Leisure 30% 34% 60% 67% 23%

Education 1% 5% 2% 5% 23%5

Personal 20% 35% 13% 15%

other purposes 3% 23%6

3.5. OPERATIONALIZATION OF VARIABLES

The operationalization of the framework described in Figure 2 is carried out in two stages. The first

stage corresponds to the estimation of the mode choice model, and the second stage corresponds to

the estimation of the travel choice model.

Each alternative, in each model, is described by a utility function composed of a systematic

component and a random component. In every case, the random component is assumed to be

Gumbel and i.i.d. across alternatives and observations, as described above. The systematic

component of the utility of the alternatives is a mathematical function of the attributes of the

alternatives, the attributes of the decision maker, and the interactions between attributes of

alternatives and the characteristics of the decision maker (Eq. 6).

( ) ( ) ( )ittiit X,SVSVXVV ++= Eq. 6

where:

tiV is the systematic component of utility

( )iXV is the portion of utility of alternative i associated with the attributes of alternative i,

( )tSV is the portion of utility associated with characteristics of individual t, and

5 Includes military 6 Includes health

35

( )it X,SV is the portion of the utility which results from interactions between the attributes of alternative i and the characteristics of individual t.

The mode choice model is derived from a choice experiment between two trip alternatives. Travel cost

and travel time are the central attributes of the trip alternatives in mode choice models. Frequency is

also many times included, usually in the form of average waiting time. In this research, frequency is

expressed by the number of days wait for the return journey (ret variable), and it is operationalized as

an attribute in its own rights, independent of travel time. The reason this is so is that we are interested

in the effect of frequency of departures on the evaluation of the transport opportunities available to the

islanders. The travel choice model contrasts the above characteristics of the trip with the utility of not

travelling, described by a constant.

Trip purpose is also central to our research. According to the local context, work or business, leisure

and other trip purposes (including health related) are the main motivations for islander’s trips. We

exclude education since it is not applicable to all islanders.

Empirical analysis has shown that people with different personal characteristics have different

preferences among sets of alternatives. Besides income, characteristics that have been shown to

influence choices are: age, gender, education, activity and prior travel experience (Koppelman and

Bhat, 2006; Polydoropoulou and Litinas, 2007).

The operationalization of each of the variables that constitute each of the components of utility above

is described in Table 3.

Table 3 Variables for mode and travel choice models

Attributes of the alternatives, Xi

price1 price, in Euros, of the return trip in alternative 1 Price

price2 price, in Euros, of the return trip in alternative 2

tt1 duration of the return trip, in hours, of alternative 1 Travel Time

tt2 duration of the return trip, in hours, of alternative 2

ret1 number of days wait for the return journey in alternative 1

ret2

number of days wait for the return journey in alternative 2

ret takes the values one (if it is possible to return on the same day that the activity took place), two or three, if it is possible to return the following or the next day, respectively.

Frequency

dret1 dummy variable, takes the value 0 if the return trip is possible on the same day the activity took place, 1 otherwise

Trip purpose

Trip purpose pheal dummy variable, takes the value 1 if the purpose of the trip is health related, 0 otherwise

36

pwork dummy variable, takes the value 1 if the purpose of the trip is work related, 0 otherwise

pleis dummy variable, takes the value 1 if the purpose of the trip is leisure, 0 otherwise

Characteristics of individual St

inc average of the income class

incl<1000 dummy variable, takes the value 1 if household income is less than 1000 €, 0 otherwise Income

incl<2000 dummy variable, takes the value 1 if household income is less than 2000 €, 0 otherwise

Gender gend dummy variable, takes the value 1 if respondent is female, 0 otherwise

age the age of the respondent

Age agegr

a categorical variable, taking the values: 0 if age ≤ 25; 1 if 25 < age ≤ 50;

2 if 50 < age ≤ 75; and 3 if age > 75

Education edu dummy variable, takes the value 1 for respondents with more than 12 years of school, 0 for respondents with less than 12 years of school

Frequent traveller freq

dummy variable, takes the value 1 for respondents that in the past 6 months made more than 4 round trips out of the island, 0 otherwise

Chios town chtw dummy variable, takes the value 1 if respondent lives elsewhere on the island, 0 if respondent lives in Chios town

Discount disc dummy variable, takes the value 1 if respondent stated he had more than 25% discount on boat trips, 0 otherwise

house dummy variable, takes the value 1 if respondent is a house worker, rural worker or fisherman, 0 otherwise

mili dummy variable, takes the value 1 if respondent works in the military, 0 otherwise

student dummy variable, takes the value 1 if respondent is a student, 0 otherwise

worker dummy variable, takes the value 1 if respondent works for the public administration, works for private company, has a liberal activity or is retired, 0 otherwise

flex dummy variable, takes the value 1 if respondent is unemployed, a house worker, rural worker or fisherman or works in the military, 0 otherwise

Activity

flex2 dummy variable, takes the value 1 if respondent is unemployed, a house worker, rural worker or fisherman, a student or works in the military, 0 otherwise

Mode choice choiceA dummy variable, takes the value 1 if respondent’s mode choice was alternative 1, 0 otherwise

Interaction terms St Xi

price1 / ln (inc) Price and Income

price2 / ln (inc)

price divided by the natural logarithm of the household income of the respondent

37

price x pheal variable that takes the value of price if trip purpose is health related, 0 otherwise

price x pwork variable that takes the value of price if trip purpose is work related, 0 otherwise

Price and Trip purpose

price x pleis variable that takes the value of price if trip purpose is leisure, 0 otherwise

Interaction terms St Xi (cont.)

Travel time and Trip purpose

tt1 x pleis variable that takes the value of travel time if trip purpose is leisure, 0 otherwise

Frequency and Trip purpose

ret1 x pleis variable that takes the value of number of days wait for the return trip if trip purpose is leisure, 0 otherwise

3.6. SURVEY DESIGN

Two different types of choice experiments were made for each trip purpose. In Choice experiment A,

the respondents were asked to choose between two alternatives, for which the following attributes

were given: cost of round trip (in Euros), travel time for round trip (in hours), availability of return trip

(day of the week of the next available return trip). Choice experiment A is similar to a mode choice

experiment, except for the fact the alternatives are unlabelled, i.e., they do not correspond to any

specific mode of transport.

For Choice experiment A, we adopted a non-factorial7 design. There are three attributes (Round trip

price, Total travel time and Return day possible), each with multiple levels. Round trip price levels vary

from 20 € to 250 €. This range is designed to contain the range of typical prices practiced in the

maritime and air connections to Athens. Travel times range from 2 h to 20 h, which contains the

normal travel times for the air trip (two times 30 minutes plus access and egress times will round up to

a minimum of 2 hours); and for the ferry trip (a maximum of 9 hours each way, plus access and egress

times will round up to a 20 hours). The Return day ranges from 0 days (return trip possible as soon as

the activity is completed) to 3 days later.

There are 6 versions of the survey, designed in order to vary on the levels of the attributes of each

alternative. Within each version of the survey, the order of the alternatives has been randomised to

control for order effect.

7 A factorial design is one in which each level of each attribute is combined with every level of every other attribute.

38

Figure 6 Survey: Choice experiments A and B

In Choice experiment B, respondents were asked if they would travel or not using the other alternative,

in case the alternative chosen was no longer available. The objective of making the alternative chosen

in Choice experiment A not available has to do with two facts:

1) In Choice experiment A, respondents choose between two trip alternatives. That choice is

already an admission that they would travel if the alternative chosen was available. To ask

next if they would or not in fact travel might be understood as a boycott to the survey.

2) The conditional choice (only one trip alternative available) is a better indicator of the

judgement of the islander about the available trip opportunities than the unconditional choice.

If the decision is to use the “other” (non-preferred) alternative, it implies that both alternatives

were judged adequate.

The sequence Choice experiment A, Choice experiment B was repeated twice per each of three trip

purposes. For each trip purpose, the interviewer would describe a situation involving a trip (see Table

4). Athens was always the destination, which allowed for price comparability. The description of the

situation had to:

1) Convey the necessary information

2) Leave no room for different interpretations.

3) Provide time and inspiration for the interviewee to imagine him/herself before the actual

choice.

Additionally, the situations described had to be applicable and appealing to every respondent. This is

one reason why the Trip purpose education was excluded from the survey.

39

Table 4 presents the text with the description of the scenarios for each trip purpose.

Table 4 Survey: scenarios for different trip purposes

Trip purpose Scenario description

health Imagine that you have a doctor appointment in Athens on Monday morning. You travel to

Athens on Sunday night.

leisure

Now imagine that you are thinking of going to a cultural event (choose your favourite:

concert, theatre, football match) in Athens on Saturday night. You travel on Saturday

morning.

work Now imagine that you are going for a business meeting in Athens on Wednesday

afternoon.

40

4. RESULTS

4.1. SAMPLE DESCRIPTIVES

The data collection methodology involved a survey addressed to the residents of Chios Island. The

interviews were carried out between the 11th and the 26th of May 2009. During this 16-day period, 412

questionnaires were collected. These questionnaires provided 2403 stated preferences.

People were inquired about their preferences on hypothetical scenarios concerning trips from Chios to

Athens for three different purposes: health, culture and/or leisure and work. Additionally, the survey

collected socio-demographic data and data concerning the respondent’s travel habits (such as

average travel frequency) and characteristics (travel discount beneficiary).

In a first stage, a pilot survey was conducted in order to test survey design and the amount of working

hours needed to have a sufficiently large sample size. The pilot survey was conducted by phone

interviews. The survey was based on face-to-face interviews in which each respondent was asked the

questions and the interviewer would complete the survey accordingly.

The interviewees were randomly chosen amongst the population. Randomness of the sample is

needed in order to guarantee unbiased estimations of the model parameters. Simple random sampling

without replacement was used, such that each individual had the same probability of being interviewed

and no individual was interviewed twice. A team of interviewers approached random people on the

streets of Chios town and villages, keeping away from sampling from the same streets or towns. The

sample descriptive statistics were analysed throughout the sampling process, to determine if there

was need for stratified sampling techniques. Variables analysed were Gender and Place of residence

(Chios town versus elsewhere on the island). In the survey sample, 51% of the respondents were

male. 67% of the respondents lived in Chios town, the islands’ capital city, while the remainder lived

elsewhere on the island.

Only people over 16 years old were interviewed, to ensure that the respondents were mature enough

to answer the survey, and that they were able to make their own transportation choices. The

distributions of the respondents regarding age group, education level and activity are represented in

Figure 7, Figure 8 and Figure 9 respectively. There is no data available on the age distribution of the

Chios population. There is a prevalence of respondents between 25 and 30,and of highly educated

people. However, this can be due to the presence of the University on the Island, which may attract an

unusually high share of young post-graduates.

41

Figure 7 Histogram of age of respondents

Figure 8 Distribution of respondents regarding level of education

Figure 9 Distribution of respondents by activity

42

The use of household income instead of individual income is typical in most transportation surveys

(Koppelman and Bhat, 2006). This has both advantages and disadvantages. On the one hand, it

raises issues in the interpretation of the price and income parameters, and also in the estimation of

value of time. However, it is also true that a large part of our sample were non-workers that probably

do not have any individual income, and finance their trips with the household income.

Figure 10 Distribution of respondents per monthly household income

The average household size was 2,8 people per household8, and the distribution of monthly household

income showed almost 50% of the sample falling in the 1000 to 2000 Euros category (see Figure 10).

In Choice experiment A, relative to the choice between trip alternatives, almost one third of the

respondents chose Alternative 1. In Choice experiment B, concerning the choice of whether to travel

or cancel the trip in case the preferred trip alternative was not available, about half the respondents

admitted to switch to the other mode, while the other half preferred cancelling the trip.

Table 5 Results of Choice experiments A and B

Alternative 1 Alternative 2

Choice experiment A (mode choice) 64% 36%

Travel Cancel or

8 The survey question regarding household size was formulated so as to minimize the possibility of misunderstanding as to what constitutes an individual household. The question was: “How many people live in your house (share house and meals), including you? Note that if you live in a student residence, household size is 1” (See Annex)

43

postpone the trip

Choice experiment B (travel choice) 49% 51%

4.2. MODEL ESTIMATION

4.2.1. MODE CHOICE

The first stage consists of using the survey data to build an MNL from the results of Choice

experiment A. The objective is to estimate the parameters of a model for the choice between trip

alternatives, which will integrate a larger framework model of travel choices in the islands. As a by-

product, it allows us to estimate the Value of Time for the islanders.

Choice experiment A refers to the choice between two alternative combinations of the following

attributes: price of return trip (in Euros), travel time for return trip (in hours), availability of return trip

(day of the week of the next available return trip). The data is organized so that Alternative 1 is the

less expensive alternative, independently of the values of the other attributes. Alternative 1 is taken as

the reference alternative.

Starting from a minimal specification, we introduce incremental changes to the alternatives’ utility

functions and re-estimate the model, in an effort to improve the model in terms of its behavioural

realism and its empirical fit to the data, while avoiding excessive complexity. Model estimation is

carried out using the software package BIOGEME 1.8 (Bierlaire, 2003).

Alternative Specific Attributes: Price, Time and Frequency

Travel cost, travel time and frequency are alternative-specific attributes, i.e., they take a different value

for each alternative. These attributes influence the utility of each alternative for all the individuals in the

population. Table 6 reports different specifications for utility of Alternative 1 (U1) and utility of

Alternative 2 (U2), and Table 7 reports the relevant statistics for these models.

Table 6 Specifications of utility of alternatives for the Models A1 and A2

U1 = BETAP * price1 + BETAT * tt1 + BETAR * ln (ret1) Eq. 7 Model

A1 U2 = BETAP * price2 + BETAT * tt2 + BETAR * ln (ret2) Eq. 8

U1 = BETAP * price1 + BETAT * tt1 + BETAR * ln (ret1) Eq. 9 Model

A2 U2 = ASC2 + BETAP * price2 + BETAT * tt2 + BETAR * ln (ret2) Eq. 10

The parameter BETAP, associated with price, is statistically significant in all the models tested, which

confirms the importance of the variable Price on the mode choice decision. The parameter BETAP is

44

negative, indicating that, all other things being equal, less expensive alternatives will be chosen more

likely than more expensive alternatives.

The parameter BETAT, associated with travel time, is also negative and statistically significant. Again,

all other things being equal, alternatives with shorter travel times are preferred over alternatives with

longer travel times.

Table 7 Relevant statistics and parameter values for Models A1 and A2

Model A1 Model A2

Parameters Value Robust t-stat Value Robust t-stat

ASC2 -0,817 -5,18**

BETAP -0,0120 -8,48** -0,00542 -2,93**

BETAT -0,0884 -5,74** -0,105 -7,86**

BETAR -0,411 -3,21** -0,292 -2,21*

Goodness-of-fit

N. obs. 2408 2408

N. Ind. 404 404

LL ratio 273,666 338,980

Adjusted ρ2 0,080 0,099

LL [A1/A2] = 65,314**

** Statistically significant at 0,99 level of confidence * Statistically significant at 0,95 level of confidence

The parameter BETAR, associated with frequency, is statistically significant and negative, confirming

the intuitive notion that low frequencies have a negative impact on the utility of the alternatives. The ret

variable is introduced in the form of natural logarithm, which results in a higher significance level for

the parameter BETAR and a marginally better goodness-of-fit of the model.

The Alternative Specific Constant (ASC) is usually included in mode choice models. It measures a

bias towards one of the alternatives that cannot be explained by the observed attributes of the

alternatives. In other words, the ASC represents the average effect of attributes that were not made

explicit in the model. In the present case, the ASC2 is also acting like a dummy variable identifying the

more expensive alternative. Thus, the sign of the ASC2 indicates a bias towards the less expensive

alternative, beyond what is captured by the price variable.

In order to compare the two models, and since Model A1 is a restricted version of Model A2, we use a

likelihood ratio test. The likelihood ratio obtained is sufficiently large to reject the null hypothesis that

that the Model A1 is the true model, with a rejection confidence higher than 99,9%.

45

Household income

Household income is an attribute of the individual. In Model A3, the parameters GAMINC<1000 and

GAMINC<2000 represent the effect of household income being, respectively, “less than 1000 €” and

“less than 2000 €”. In Model A4, income is introduced as a deflator of price by dividing the price of an

alternative by the natural logarithm of household income, translating the idea that high-income

travellers place less importance on cost. In Model A5, the same term is introduced in addition to a

linear effect of price (Table 8).

Table 8 Specifications of utility for Models A3 to A5

U1 = MODEL A2 Eq. 11 Model

A3 U2 = MODEL A2 + GAMINC<1000 * incl<1000 + GAMINC<2000 * incl<2000 Eq. 12

U1 = BETAT * tt1 + BETAR * ln (ret1) + BETAGAMA * price1 / ln (inc) Eq. 13 Model

A4 U2 = ASC2 + BETAT * tt2 + BETAR * ln (ret2) + BETAGAMA * price2 / ln (inc) Eq. 14

U1 = MODEL A2 + BETAGAMA * price1 / ln (inc) Eq. 15 Model

A5 U2 = MODEL A2 + BETAGAMA * price2 / ln (inc) Eq. 16

The statistics for the abovementioned specifications of household income are reported in Table 9.

Household income proved to be statistically significant in all specifications. In Model A3, the

parameters GAMINC<1000 and GAMINC<2000 have a clear negative effect on the utility of the more

expensive alternative. Everything else being equal, the utility of alternative 2 will be approximately half

a point (-0,526) lower for respondents belonging to the income category “1001 to 2000 €” than for

respondents with income higher than 2000 €; and approximately one point lower (0,526 + 0,366 =

0,89) for people with income “1000 € or less” than for those with income higher than 2000 €.

Table 9 Relevant statistics and parameter values for Models A3 to A5

Model A3 Model A4 Model A5

Parameters Value r. t-stat Value r. t-stat Value r. t-stat

ASC2 -0,395 -2,34* -0,739 -4,77** -0,846 -5,34**

BETAP -0,00463 -2,50* 0,0208 2,64*

BETAT -0,105 -7,82** -0,109 -7,61** -0,106 -7,69**

BETAR -0,264 -2,08* -0,317 -2,43* -0,278 -2,19*

GAMINC<1000 -0,366 -2,03*

GAMINC<2000 -0,526 -3,05**

BETAGAMA -0,0482 -3,57** -0,189 -3,28**

Goodness-of-fit

46

N. obs. 2392 2392 2392

N. Ind. 402 402 402

LL ratio 388,282 350,074 372,083

Adjusted ρ2 0,113 0,103 0,109

Horowitz Φ[-5,9] < 0,001 Φ[-3,8] < 0,001

** Statistically significant at 0,99 level of confidence * Statistically significant at 0,95 level of confidence

In Models A4 and A5, BETAGAMA, the parameter for the composite price-income variable is

statistically significant and negative, as expected. In Model A5, BETAP is positive. This may be due to

the fact that the parameters ASC2 (indicates the more expensive alternative) and BETAGAMA are

already covering the bulk of the negative impact of price, where as BETAP may be conveying some

preference for more expensive alternatives due to an association with higher quality and comfort.

In order to compare the three models we use the non-nested hypothesis test proposed by Horowitz in

1982, as presented by Ben-Akiva and Lerman (1985). The result obtained (see last line of Table 9)

indicates that we can reject the hypothesis that Model A4 and A5 are the true models, with a rejection

confidence higher than 99,9%.

Trip Purpose

Three dummy variables - pheal, pwork and pleis - express the purpose of the trip. In each of the model

specifications in Table 10, a term expressing the interaction between price and purpose was added.

Table 10 Specifications of utility for Models A6 to A8

U1 = MODEL A3 + BETAPHEAL * price1 * pheal Eq. 17 Model

A6 U2 = MODEL A3 + BETAPHEAL * price2 * pheal Eq. 18

U1 = MODEL A3 + BETAPWORK * price1 * pwork Eq. 19 Model

A7 U2 = MODEL A3 + BETAPWORK * price2 * pwork Eq. 20

U1 = MODEL A3 + BETAPLEIS * price1 * pleis Eq. 21 Model

A8 U2 = MODEL A3 + BETAPLEIS * price2 * pleis Eq. 22

The parameter identifying leisure trips is highly significant and while the parameters associated with

work or health related trips are also significant but positive. Health trips and work trips are not

statistically different from one another, and so only one of the purpose dummy variables should be

included in the model. Moreover, for a leisure trip, the parameter associated with price increases

magnitude almost two and half times ([BETAP + BETAPLEIS]/ BETAP = 2,4) in comparison to the

average for work and health related trips. This implies that, everything else being equal, individuals

are less willing to pay for leisure trips than for work or health related trips.

47

Table 11 Relevant statistics and parameter values for Models A6 to A8

Model A6 Model A7 Model A8

Parameters Value r. t-stat Value Value r. t-stat r. t-stat

ASC2 -0,435 -2,30* -0,436 -2,32*

BETAP -0,00621 -3,20** -0,00558 -2,91** -0,00527 -3,29**

BETAT -0,103 -6,85** -0,106 -7,81** -0,104 -6,94**

BETAR -0,262 -2,05* -0,263 -2,07* -0,300 -2,37*

GAMINC<1000 - -1,88 (ns) - -1,88 (ns) - -1,89 (ns)

GAMINC<2000 -0,528 -3,04** -0,528 -3,05** -0,701 -4,62**

BETAPHEAL 0,00406 5,57**

BETAPWORK 0,00252 3,69**

BETAPLEIS -0,00730 -7,95**

Goodness-of-fit

N. obs. 2392 2392 2392

N. Ind. 402 402 402

LL ratio 409,370 396,338 433,791

Adjusted ρ2 0,119 0,115 0,127

** Statistically significant at 0,99 level of confidence * Statistically significant at 0,95 level of confidence

(ns) not statistically significant

Individual Characteristics

The dummy variables gend, agegr, edu and freq describe respectively the gender of the respondent,

his/her level of education and if the respondent is a frequent traveller. Additional specifications for age

included the natural logarithm of the age of the respondent - translating the idea that a one year

difference at an early age weights more than one year difference at a more advanced age - but this

resulted in worse fit and significance.

Additionally, we included a dummy variable for people living outside Chios town; and another

expressing if the respondent holds a transport discount card. The parameters related to living outside

Chios town and having a discount card have no statically significant effect on mode choice.

48

Table 12 Specifications of utility for Models A9, A10 and A11

U1 = BETAP *price1 + BETAT *tt1 + BETAR * ln(ret1) + BETAPLEIS *price1 *pleis Eq. 23

Model

A9

U2 = BETAP *price2 + BETAT *tt2 + BETAR * ln(ret2) + BETAPLEIS *price2 * pleis

+ GAMINC<2000 * incl<2000 + GAMAGE * agegr + GAMEDU * edu + GAMGEND *

gend + GAMFREQ * freq Eq. 24

U1 = BETAP *price1 + BETAT *tt1 + BETAR * ln(ret1) + BETAPLEIS *price1 *pleis Eq. 25

Model

A10

U2 = BETAP *price2 + BETAT *tt2 + BETAR * ln(ret2) + BETAPLEIS *price2 *pleis

+ GAMINC<2000 * incl<2000 + HOUSEDUM * house + MILIDUM * mili + STUDUM *

student Eq. 26

U1 = BETAP *price1 + BETAT *tt1 + BETAR * ln(ret1) + BETAPLEIS *price1 *pleis Eq. 27

Model

A11

U2 = BETAP *price2 + BETAT *tt2 + BETAR * ln(ret2) + BETAPLEIS *price2 *pleis

+ GAMINC<2000 * incl<2000 + GAMAGE * agegr + GAMEDU * edu + STUDUM *

student + FLEXDUM * flex Eq. 28

The parameters GAMFREQ and GAMGEND, associated respectively with frequent travellers and

women are not statistically significant, although depending on which other variables are included in the

model. This variation is due in part to some correlation between parameters. The hypothesis that

these two parameters are zero can be tested through a likelihood ratio test. The likelihood ratio

obtained is LL [A9unrest/ A9restr] = 28,77**, which sufficiently large to reject the null hypothesis that the

abovementioned socio-demographic variables have no effect on mode choice. Notwithstanding,

keeping the two variables in the model will add to model complexity and data requirements with

adding much to its goodness-of-fit. Therefore, we choose to leave them out.

In what concerns the effect of activity on mode choice, Model A10 reveals that students prefer the less

expensive alternative more often than any other respondents do. Respondents in the category “house

worker, rural worker or fisherman” also shows less tolerance to the expensive alternative than the

remaining categories, as do the military.

The parameters related to activity also seem to present some degree of correlation between them. For

this reason, we group the activities together, taking into account the parameter’s estimated values, the

correlations between pairs of parameters and the hypothesis that activities are expressing the effect of

individual income. Three groups were created: the “Workers”, comprising the categories “works for the

public administration”, “works for private company”, “liberal activity” and “retired”; the “Flex”,

comprising the categories “unemployed”, “house worker, rural worker or fisherman” and “military”; and

the “Students”. The hypothesis motivating these groups is that individuals in the “Workers” category

receive their own fixed salary or pension, which translates into having higher individual income or

liquidity; individuals in the “Flex” have lower individual income than the previous; the “Students” have

the lowest individual income.

49

Table 13 Relevant statistics and parameter values for Models A9 to A11

Model A9 Model A10 Model A11

Parameters Value r. t-stat Value Value r. t-stat r. t-stat

BETAP -0,00328 -1,73 (ns) -0,00466 -2,88** -0,00389 -2,07*

BETAT -0,112 -7,73** -0,109 -7,17** -0,114 -7,68**

BETAR -0,230 -1,72 (ns) -0,304 -2,45* -0,240 -2,43*

BETAPLEIS -0,00718 -7,96** -0,00740 -7,99** -0,00729 -7,90**

GAMINC<2000 -0,645 -3,85** -0,645 -4,41** -0,509 -3,17**

GAMAGE -0,246 -3,40** -0,276 -3,25**

GAMEDU 0,447 2,06* 0,418 2,23*

GAMFREQ 0,0934 0,57 (ns)

GAMGEND 0,159 1,04 (ns)

HOUSEDUM -0,864 -3,41**

MILIDUM -0,663 -2,38*

STUDUM -1,02 -4,13** -1,30 -5,09**

FLEXDUM -0,686 -3,43**

Goodness-of-fit

N. obs. 2366 2366 2366

N. Ind. 397 397 397

LL ratio 453,070 497,445 528,461

Adjusted ρ2 0,133 0,145 0,154

Φ [-8,3] < 0,001 Φ [-5,5] < 0,001

** Statistically significant at 0,99 level of confidence * Statistically significant at 0,95 level of confidence

(ns) not statistically significant

The parameter FLEXDUM for the “Flex” category is statistically significant, indicating that individuals in

this category of activities prefer the less expensive alternative more often than the “Workers”. The

parameter STUDUM associated with the “Student” category is also statistically significant, and 2 times

in magnitude the FLEXDUM, indicating thus an even more pronounced preference for the less

expensive alternative, relative to the “Workers” category.

Model A11 is superior to Models A9 and A10, as indicated by the Horowitz test, that predictably leads

to the rejection of Models A9 and A10.

The ASC looses significance when all the variables are added to the model. In theory, an ASC should

have no place in this model since the alternatives are abstract (unlabelled), and thus there are no

alternative specific unobservables. Therefore, an insignificant ASC is in accordance to theory.

50

Relaxation of equal alternative specific parameters assumption

Another possible way to refine the model is to admit that the values of the alternative specific

parameters differ according to the alternative. Relaxing the constraint of equal parameters for the price

and the frequency variables resulted in non-significant parameters and worse fit of the model. In the

case of the variable travel time, relaxing the assumption that BETAT is the same for both alternatives

resulted in significant parameters and marginally better fit.

Table 14 Specifications of utility for Model A12

U1 = BETAP *price1 + BETAT1 *tt1 + BETAR * ln (ret1) + BETAPLEIS *price1 *pleis Eq. 29

Model

A12

U2 = BETAP *price2 + BETAT2 *tt2 + BETAR * ln (ret2) + BETAPLEIS *price2 *pleis

+ GAMINC<2000 * incl<2000 + GAMAGE * agegr + GAMEDU * edu + GAMGEND *

gend + GAMFREQ * freq Eq. 30

The hypothesis that the parameters BETAT1 and BETAT2 are in fact different can be tested through a

likelihood ratio test. The likelihood ratio obtained is LL [A12unrest/ A12restr] = 2,5 (ns), which is not

sufficiently large to reject the null hypothesis that the abovementioned parameters are the same.

Therefore, we adopt Model A11 as the final model describing mode choice.

Table 15 Relevant statistics and parameter values for Model A12

Model A12

Parameters Value r. t-stat

BETAP -0,00409 -2,13*

BETAT1 -0,117 -7,83**

BETAT2 -0,166 -4,68**

BETAR -0,240 -1,93 (ns)

BETAPLEIS -0,00726 -7,88**

GAMINC<2000 -0,447 -2,71*

GAMAGE -0,165 -2,02*

GAMEDU 0,614 2,67*

STUDUM -1,15 -4,54**

FLEXDUM -0,691 -3,45**

Goodness-of-fit

N. obs. 2366

N. Ind. 397

LL ratio 530,961

Adjusted ρ2 0,155 ** Statistically significant at 0,99 level of confidence * Statistically significant at 0,95 level of confidence

(ns) Not statistically significant

51

Final Mode Choice Model

On the basis of the previous considerations, we present the final mode choice model in Table 17.

Table 16 Relevant statistics and parameter values for the Mode Choice Model

Mode Choice Model

Parameters Value r. t-stat

Price -0,00400 -2,36*

Travel Time -0,115 -7,69**

Number of days wait for return trip -0,249 -2,49*

Monthly household income < 2000 € -0,492 -3,03**

Price x Purpose Leisure -0,00760 -6,90**

Price x Purpose Health - 1,30 (ns)

Age group -0,283 -3,31**

Education level (>12 years of school) 0,382 2,02*

Frequent traveller - 0,94 (ns)

Student -1,33 -5,19**

House worker, rural worker, fisherman, military, unemployed

-0,673 -3,35**

Goodness-of-fit

N. obs. 2384

N. Ind. 400

LL ratio 531,745

Adjusted ρ2 0,154 ** Statistically significant at 0,99 level of confidence * Statistically significant at 0,95 level of confidence

(ns) Not statistically significant

52

4.2.2. VALUE OF TIME

Models of Mode Choice are often used to estimate the Value of Time of populations. The Value of

Time is the marginal rate of substitution between travel time and cost, and is equal to the ratio

between the derivative of utility with respect to time and the derivative of utility with respect to cost (Eq.

31). When the utility function is linear in both travel time and cost and neither time nor cost is

interacted with any other variables, the Value of Time can be calculated as the ratio of the parameter

for travel time over the parameter for travel cost (Eq. 32).

∂

∂

∂

∂

=

Cost

V

TTime

V

TIMEofVALUEi

i

Eq. 31

BETAP

BETATTIMEofVALUE = Eq. 32

In Table 17 we present Values of Time based on the results of the final mode choice model. This

model allows us to differentiate between the Value of Time for a leisure trip and the Value of Time of a

health or work-related trip.

Table 17 Values of Time based on the Mode Choice Model

Value of Time

Work and health related trips 29 €/h

Leisure trips 10 €/h

Similarly, the value of one day wait for the return trip can be calculated from Eq. 32:

BETAP

BETARWAITDAYaofVALUE = Eq. 33

Table 18 Value of a day wait based on Model A11

Value of a day wait

Work and health related trips 62 €/day

Leisure trips 21 €/day

53

4.2.3. TRAVEL CHOICE

Choice experiment B corresponds to the choice between embarking on a trip or not for a particular

purpose, given a specific price, travel time and day for the return trip. In Choice experiment B,

respondents were asked if they would travel or not using the other alternative, in case the alternative

chosen was no longer available. Alternative 3 represents the choice not to travel, and alternatives 1

and 2 represent the choice of travelling with either one of the alternatives in Choice experiment A.

In what follows, the utility of the alternatives 1 and 2 is described by attributes related to the trip

alternatives (price, travel time and frequency), and the utility of Alternative 3 is described by a

constant. This constant is meaningless, since it is the difference between the utilities (and not the

absolute value) of each alternative that determines the choice probability. Although this Choice

experiment is characterized by three alternatives, both conceptually and in practice, there are only two

alternatives, since for every respondent, either alternative 1 or alternative 2 was not be available,

depending on which was chosen in Choice experiment A.

Mode Choice

Mode Choice refers to the alternative chosen by the respondent in Choice experiment A. The

alternative chosen in Choice experiment A conditions the alternative available in Choice experiment B.

Therefore, it is crucial to understand if the alternative specific parameters for alternative 1 and 2 are

the statistically different from one another. Table 19 describes different specifications, two in which the

alternative specific parameters are different for alternatives 1 and 2, and another where alternatives 1

and 2 share common parameters.

Table 19 Specifications of utility for Models B1, B2 and B3

U1/2 = BETAP * price1/2 + BETAT * tt1/2 + BETAR * ret1/2 Eq. 34 Model

B1 U3 = ASC3 + GAMINC<2000 + MODE * Choice A Eq. 35

U1 = BETAP1 * price1 + BETAT1 * tt1 + BETAR1 * ret1 Eq. 36

U2 = BETAP2 * price2 + BETAT2 * tt2 + BETAR2 * ret2 Eq. 37 Model

B2

U3 = ASC3 + GAMINC<2000 Eq. 38

U1 = BETAT1 * tt1 + BETAR1 * ret1 Eq. 39

U2 = BETAP2 * price2 Eq. 40 Model

B3

U3 = ASC3 + GAMINC<2000 Eq. 41

Mode Choice influences the travel decision, as implied by the parameter MODE, associated with

choosing the less expensive alternative in Choice experiment A. MODE is significant and positive,

implying that respondents who choose the less expensive alternative, experience a larger decrease in

the utility of travelling using their second choice than respondents who have chosen the more

54

expensive alternative.

Model B2 relaxes the constraint that the alternative specific parameters are the same in Alternatives 1

and 2. This hypothesis can be tested through a likelihood ratio test. The likelihood ratio obtained is

sufficiently large to reject the null hypothesis that the abovementioned parameters are the equal for

alternatives 1 and 2.

Model B3 corresponds to constraining the parameters BETAP1, BETAT2 and BETAR2 to zero. Model

B3 achieves almost the same fit with less parameters, and a better adjusted ρ2. The likelihood ratio

between B3 and its unrestricted version (which is B2) is not sufficiently large to reject the null

hypothesis that the parameters BETAP1, BETAT2 and BETAR2 are zero. Therefore, Model B3 is

chosen over Models B1 and B2.

Table 20 Relevant statistics and parameter values for Models B1 to B3

Model B1 Model B2 Model B3

Parameter Value r. t-stat Value r. t-stat Value r. t-stat

ASC3 -1,02 4,42** -0,628 -2,73* -0,687 -4,00**

BETAP - -1,71 (ns)

BETAT - -1,31 (ns)

BETAR - -0,50 (ns)

MODE 0,667 3,35**

GAMINC<2000 - 1,78 (ns) - 1,77 (ns) - 1,79 (ns)

BETAP1 - 0,49 (ns)

BETAP2 -0,00410 -3,62** -0,00446 -5,19**

BETAT1 - 1,66 (ns) - 1,75 (ns)

BETAT2 - -0,17 (ns)

BETAR1 -0,420 -2,34* -0,428 -2,43*

BETAR2 - 0,23 (ns)

Goodnes-of-fit

N. obs. 2392 2392 2392

N. Ind. 402 402 402

LL ratio 92,954 97,566 96,961

Adjusted ρ2 0,024 0,025 0,026

LL[B2r/ B2u] = 25,50** LL[B3r/ B3u] = 0,604 (ns)

** Statistically significant at 0,99 level of confidence * Statistically significant at 0,95 level of confidence

(ns) not statistically significant

55

Price, Travel Time, Frequency and Household income

In Model B3, the effects of income and travel time are not statistically significant. In Model B4, we

change the form of the specification of the variables price and income. Household income is included

as a deflator of the price of the trip. In Model B5, we change the form of the specification of the

variables travel time and frequency. Table 21 reports different specifications for utility of the

alternatives and Table 22 reports the relevant statistics for these models.

Table 21 Specifications of utility for Models B4 and B5

U1 = BETAT1 * tt1 + BETAR1 * dret1 Eq. 42

U2 = BETAGAMA2 * price2 / ln (inc) Eq. 43 Model

B4

U3 = ASC3 Eq. 44

U1 = BETAT1 * tt1 + BETAR1 * dret1 Eq. 45

U2 = BETAGAMA2 * price2 / ln (inc) Eq. 46

Model

B5

U3 = ASC3 Eq. 47

Table 22 Relevant statistics and parameter values for Models B4 and B5

Model B4 Model B5

Parameter Value r. t-stat Value r. t-stat

ASC3 -0,422 -2,77* -0,517 -2,98**

BETAGAMA2 -0,0299 -4,72** -0,0275 -4,35**

BETAT1 - 1,41 (ns) - 1,55 (ns)

BETAR1 -0,812 -2,74* -0,829 -2,78*

GAMINC - 1,42 (ns)

Goodness-of-fit

N. obs. 2392 2392

N. Ind. 402 402

LL ratio 103,591 107,963

Adjusted ρ2 0,029 0,030

** Statistically significant at 0,99 level of confidence * Statistically significant at 0,95 level of confidence

(ns) not statistically significant

The parameter BETAGAMA, associated with price of alternative 2 over household income is

statistically significant. BETAGAMA is negative, indicating that, in the choice of whether to cancel the

trip or switch to a more expensive alternative, the price of the trip decreases the utility of travelling.

The higher the price, the more individuals will choose to cancel. Introducing the variable income as a

56

deflator of price increases the goodness-of-fit of the model.

Travel time appears not be statistically significant to the travel choice. If the respondent has chosen

the less expensive alternative in Choice experiment A, he/she will base his decision to travel or cancel

mainly on price. If the respondent has chosen the more expensive alternative in Choice experiment A,

it appears that his/her decision will be made mainly on the possibility to return the same day, and not

so much on travel time. The parameter BETAR1, associated with the return trip, is significant. Not

surprisingly, not being able to return home on the same day decreases the utility of travelling.

Trip Purpose

Again, three dummy variables - pheal, pwork and pleis - express the purpose of the trip. Work related

trips are taken as reference, and therefore WORKDUM is omitted. Table 23 reports the specifications

of utility of the alternatives and Table 24 reports the relevant statistics for these models.

Table 23 Specifications of utility for models B6 and B7

U1 = BETAT1 * tt1 + BETAR1 * ret1 Eq. 48

U2 = BETAGAMA2 * price2 / ln (inc) Eq. 49 Model

B6

U3 = ASC3 + HEALDUM * pheal + LEISDUM * pleis Eq. 50

U1 = BETAT1*tt1 + BETAT1LEIS*tt1*pleis + BETAR1*ret1 + BETAR1LEIS*ret1*pleis Eq. 51

U2 = BETAGAMA2 * price2 / ln (inc) + BETAGAMA2 * price2 / ln (inc) * pleis Eq. 52 Model

B7

U3 = ASC3 Eq. 53

LEISDUM, the parameter for the dummy variable denoting a leisure motive behind the trip, is positive

and highly statistically significant, increasing the utility of the choice not to travel, indicating that leisure

trips will be more easily cancelled or postponed than health or work trips. The parameter HEALDUM is

not significant, implying that, everything else being equal, health related trips are not significantly

different from work related trips.

A different specification of trip purpose involved associating trip purpose with the variables price, travel

time and frequency. To determine if the later are significantly different for leisure trips, we test the null

hypothesis that the parameters BETAG2LEIS, BETAT1LEIS and BETAR1LEIS are zero through a

Likelihood ratio test. The likelihood ratio is sufficiently large to reject the restriction.

57

Table 24 Relevant statistics and parameter values for Models B6 and B7

Model B6 Model B7

Parameter Value r. t-stat Value r. t-stat

ASC3 -0,923 -4,46** -1,09 -5,12**

BETAGAMA2 -0,0318 -4,04** -0,0351 -4,21**

BETAT1 0,372 2,21* - 1,94 (ns)

BETAR1 -1,21 -3,25** -1,46 -3,60**

LEISDUM 1,35 12,27** 1,71 7,23**

HEALDUM - -0,47 (ns)

BETAG2LEIS - 0,57 (ns)

BETAT1LEIS - 0,34 (ns)

BETAR1LEIS - 1,50 (ns)

Goodnes-of-fit

N. obs. 2392 2392

N. Ind. 402 402

LL ratio 325,181 339,522

Adjusted ρ2 0,094 0,098

LL[B7r/ B7u] = 14,514**

** Statistically significant at 0,99 level of confidence * Statistically significant at 0,95 level of confidence

(ns) not statistically significant

Individual characteristics

For Choice B we test the same socio-demographic characteristics as in Choice A. The variables

describing gender, education and frequent travellers are dummy variables. The variable age, in Model

B6, is introduced as the natural logarithm of age of the respondent - translating the idea that a one

year difference at an early age weights more than one year difference at a more advanced age. In

Model B7 dummy variables describing the activity of respondents were introduced. In Model B8 these

variables were grouped.

Table 25 reports different specifications for utility of alternatives 1, 2 and 3, and Table 26 reports the

relevant statistics for these models.

58

Table 25 Specifications of utility for Models B8 to B10

U1 = BETAT1 * tt1 + BETAR1 * ret1 Eq. 54

U2 = BETAGAMA2 * price2 / ln (inc) Eq. 55 Model

B8 U3 = ASC3 + LEISDUM * pleis + GAMAGE * ln (age) + GAMFREQ * freq + GAMEDU* edu + GAMGEND * gend

Eq. 56

U1 = BETAT1 * tt1 + BETAR1 * ret1 Eq. 57

U2 = BETAGAMA2 * price2 / ln (inc) Eq. 58 Model

B9 U3 = ASC3 + LEISDUM * pleis + GAMAGE * ln (age) + GAMFREQ * freq + GAMEDU* edu + HOUSEDUM * house + STUDUM * student + MILIDUM * mili

Eq. 59

U1 = BETAT1 * tt1 + BETAR1 * ret1 Eq. 60

U2 = BETAGAMA2 * price2 / ln (inc) Eq. 61 Model

B10 U3 = ASC3 + LEISDUM * pleis + GAMAGE * ln (age) + GAMFREQ * freq + GAMEDU* edu + FLEXDUM * flex

Eq. 62

Like in the case of Choice A, the parameters associated with place of residence (Chios town versus

rest of the island) and a holding a transport discount card were not statistically significant.

The parameters GAMGEND and GAMEDU, associated with the gender and education level of the

respondents, are not statistically significant.

The parameter associated with age, GAMAGE, is positive and statistically significant. According to the

estimates, age lowers the utility of travelling. Older respondents are less prone to travel using an

option they do not favour. This effect is not linear: a one year difference at early ages has more effect

than a one year difference at older ages.

The parameter associated with frequent travellers, GAMFREQ, is negative and statistically significant.

Everything else being equal, frequent travellers will more probably decide to travel than non-frequent

travellers. Of course, maybe that is the reason why they are frequent travellers.

The parameters associated with the respondent’s activities (HOUSEDU, MILIDUM and STUDUM,

associated respectively with the activity categories “house worker, rural worker or fisherman”, “military”

and “student”) are not statistically significant. In order to evaluate if these parameters have any

significant contribution to the model, we use a likelihood ratio test for the null hypothesis that the three

parameters are zero. The result is LL [B9unrest/ B9restr] = 11,982**, which is sufficiently large to reject

the null hypothesis. Since we know now that the three parameters are different from zero, we can

group them. Two groups were created: the “Workers”, comprising the categories “works for the public

administration”, “works for private company”, “liberal activity” and “retired”; and the “Flex”, comprising

the categories “unemployed”, “house worker, rural worker or fisherman”, “military” and the “students”.

The parameter for the flex group is also not statistically significant.

59

Table 26 Relevant statistics and parameter values for Models B8 and B10

Model B8 Model B9 Model B10

Parameter Value r. t-stat Value r. t-stat Value r. t-stat

ASC3 -3,13 -3,17** -4,16 -4,33** -4,22 -5,25**

BETAGAMA2 -0,0352 -4,54** -0,0363 -4,67** -0,0356 -4,61**

BETAT1 - 1,98 (ns) - 1,80 (ns) - 1,84 (ns)

BETAR1 -1,27 -3,21** -1,23 -3,12** -1,24 -3,12**

LEISDUM3 1,42 13,30** 1,43 13,33** 1,43 13,35**

GAMAGE3 0,688 3,02** 0,856 3,33** 0,874 4,18**

GAMEDU3 -0,373 -1,44 (ns)

GAMFREQ3 -0,315 -2,03* -0,335 -2,15* -0,326 -2,10*

GAMGEND3 0,0227 0,16 (ns)

HOUSEDUM3 - 1,16 (ns)

MILIDUM3 - 1,56 (ns)

STUDUM3 - 0,89 (ns)

FLEXDUM3 - 1,91 (ns)

Goodnes-of-fit

N. obs. 2366 2378 2378

N. Ind. 397 399 399

LL ratio 381,635 388,933 387,945

Adjusted ρ2 0,111 0,112 0,113

Horowitz

** Statistically significant at 0,99 level of confidence * Statistically significant at 0,95 level of confidence

(ns) not statistically significant

The parameter associated with the group “flex” is also not statistically significant, and therefore, for the

sake of simplicity, we remove it. Table 27 presents the structure of Model B11, which will be the base

for further calculations, and Table 28 reports the relevant statistics.

Table 27 Specifications of utility for Model B11

U1 = BETAT1 * tt1 + BETAR1 * dret1 Eq. 63

U2 = BETAGAMA2 * price2 / ln (inc) Eq. 64 Model

B11

U3 = ASC3 + LEISDUM * pleis + GAMAGE * ln (age) + GAMFREQ * freq Eq. 65

60

Table 28 Relevant statistics and parameter values for Model B11

Model B11

Parameters Value r. t-stat

ASC3 -3,81 -4,78**

BETAGAMA2 -0,0348 -5,12**

BETAT1 0,00909 0,37 (ns)

BETAR1 -0,716 -2,26*

LEISDUM3 1,41 13,28**

GAMAGE3 0,800 3,79**

GAMFREQ3 -0,348 -2,25*

Goodness-of-fit

N. obs. 2378

N. Ind. 399

LL ratio 372,784

Adjusted ρ2 0,109

** Statistically significant at 0,99 level of confidence * Statistically significant at 0,95 level of confidence

(ns) not statistically significant

61

Final Travel Choice Model

On the basis of the previous considerations, we present the final travel choice model in Table 29.

Table 29 Relevant statistics and parameter values for the Travel Choice Model

Travel Choice Model

Parameters Value r. t-stat

ASC for alternative “cancel the trip” -3,60 -3,30**

Price (alt 1) over ln (income) - -0,69 (ns)

Price (alt 2) over ln (income) -0,0427 -4,55**

Travel time (alt 1) -0,000714 -0,03 (ns)

Travel time (alt 2) 0,0303 1,07 (ns)

Return possible the same day (alt 1) -0,646 -1,98*

Return possible the same day (alt 2) - -1,71 (ns)

Leisure trip 1,40 12,20**

Health trip - -0,63 (ns)

Age 0,753 2,94**

Education level (>12 years school) - -1,09 (ns)

Frequent traveller -0,310 -1,99*

Student - -0,38 (ns)

House worker, rural worker, fisherman, military, unemployed

- 1,65 (ns)

Goodness-of-fit

N. obs. 2378

N. Ind. 399

LL ratio 390,001

Adjusted ρ2 0,110 ** Statistically significant at 0,99 level of confidence * Statistically significant at 0,95 level of confidence

(ns) Not statistically significant

62

5. DISCUSSION OF RESULTS

5.1. MODEL RESULTS

Mode choice depends on the price, travel time and frequency of the trip alternatives. All other things

being equal, less expensive alternatives will be chosen more likely than more expensive alternatives.

Moreover, individuals are less willing to pay for leisure trips than for work or health related trips. This is

not a surprising result. On the contrary, most mode choice models have confirmed it. We estimated

that, for a leisure trip, the parameter associated with price increases magnitude almost triples (in the

final mode choice model [BETAP + BETAPLEIS]/ BETAP = 2,8) in comparison to the average for work

and health related trips.

Similarly, alternatives with shorter travel times are preferred over alternatives with longer travel times.

The elasticity of demand to travel time varies according to the purpose of the trip. For work or health-

related trips, the islanders in this sample are willing to pay about 30 € more for each hour less in the

duration of their trip. This value goes down to 10 € for leisure trips. The Values of Time implied by our

model are somewhat larger than those calculated by Polydoropoulou and Litinas (2007). The authors

estimate Values of Time for the alternative modes - approximately 5 €/h for the ship and 19 €/h for the

aeroplane. However, these authors’ estimates are based on data collected between 2001 and 2005. A

1% per year increase in these values during six years leads to values of the same magnitude as the

ones estimated here.

Low frequencies have a negative impact on the utility of the trip alternatives. For an islander travelling

to Athens to engage in a given activity, the possibility to return home as soon as the activity is over

increases the utility of the trip mode. We estimated that, for work and health-related trips, the islanders

are willing to pay 62 € to avoid a day wait for the return trip. On leisure trips, the value goes down to

21 €. We have no basis for comparison of these values. It is known however, that users penalize in-

vehicle time more than other types of travel time. A day wait in the destination cannot be considered

travel time or waiting time in the strict sense. The disutility of a day wait in Athens is bound to be much

less than the equivalent in travel or waiting time. Many islanders will, for instances, adapt the purpose

of the trip to accommodate for the extra day in Athens. Therefore, these values do not seem

unreasonable.

Individuals from high-income households are less sensitive to price than individuals from low-income

households. Lower household income has a clear negative effect on the utility of the more expensive

alternative. The income category “less than 1000 €/month” suffers the most disutility, followed by the

income category “1000 to 2000 €/month”. There are no statistically significant increases in utility for

income categories above 5000 €/month. The effect of income appears not to be linear, but the true

functional for is difficult to estimate, since our data does not include income values but only income

categories. According to our estimates, the decrease in utility of the more expensive alternative

63

caused by income is higher when income is lowered from above 2000 € to less than 2000 €, than it is

when income is lowered from above 1000 € to less than 1000 €.

According to our results, choices made concerning health related trips are statistically no different from

the choices made work trips. We are not aware of any other study that has estimated the influence of

a health-related purpose on the utility of trip alternatives. However, the fact that health-related trips are

comparable to work trips, and different from leisure trips, is not surprising.

Concerning the decision of whether to travel or to cancel the trip in case the preferred alternative is not

available, it seems that individuals who prefer the less expensive alternative will judge the disutility in

travelling using the other alternative mainly based on price. Conversely, individuals who have

preferred the most expensive alternative will judge the disutility of travelling mainly based on the

possibility to return home as soon as the activity is finished.

Our most striking result relates to the fact that travel time appears not to play a very significant role in

the decision of whether to travel or not. This result has serious implications for the set up of Universal

Service and the provision of transport services to the islands. It implies, first of all, that the mode

choice decision and the decision of whether to travel, although interrelated, are not the same.

Mode choice models are based on the assumption that trade-offs between different attributes of the

alternatives are possible. For instances, the mode choice model here presented may suggest that

most islanders are willing to pay a fare 30 € higher for a one hour decrease in travel time. However,

the results to Choice experiment B show that, at best, only 49% of the respondents are willing to make

that trade (see Table 5). If, for a given island, less expensive alternatives are replaced with more

expensive but faster ones, n the basis of this trade-off, this may lead in fact to lower demand for

transport, since for the actual decision to travel, the users will not feel the increased utility of a shorter

travel time in the same way. For islands that have long been served by ferry boats, the substitution of

these services by hydro-foils or air travel, with increases in price, may result in an impoverishment of

the transport opportunities of the islanders. This is especially true for the low-income classes and the

elderly, since these groups are more prone to cancel their trip if a less expensive alternative is not

available.

Household income also affects the choice of whether to travel. For low-income households, the

disutility in travelling caused by an increase of price is higher than for high-income households.

The effect of socio-economic characteristics of individuals differs from the mode choice model to the

travel choice model. Age and education level of the individuals play a part in the mode choice

decision, confirming the results of Polydoropoulou and Litinas (2007). Age lowers the utility of

expensive alternatives while education has the opposite effect. However, education does not play

influence the choice to travel. Age also lowers the utility of travelling. This may be due to the fact that

the inconvenience or physical effort of travelling increases with age. Frequent travellers do not have

distinct tastes in terms of mode choice, but this variable affects the choice to travel. Frequent travellers

are less prone to cancel their trips. Alternatively, it can be said that frequent travellers are frequent

64

travellers exactly because they are less prone to cancel their trips.

In what concerns the effect of activity on mode choice, our estimates reveal that students are a

particularly sensitive group when it comes to price of the trip. The categories “house worker, rural

worker or fisherman” and “military” also show less tolerance to the expensive alternative than the

remaining activities. We interpret these results on the basis of an hypothesis: that, further than the

effect of household income, there is an effect of individual income that, lacking a more adequate

variable, expresses itself through the variables associated with activity. According to this hypothesis,

students have the lowest individual income, while the military and the house workers, rural workers

and fishermen individuals have lower incomes than the remaining categories. The reference

categories are mainly liberal workers, public servants and individuals working for private companies,

that receive a fixed salary or pension, which translates into having higher individual income or liquidity.

5.2. EVALUATION OF TRANSPORT OPPORTUNITIES

Take the case of an islander who prefers the least expensive alternative to travel. If this alternative is

no longer available, for this islander, the choice of whether to travel will mainly depend on the price of

the available alternative. In Table 30 we illustrate this case, showing how the probability to travel is

affected by price of the trip and household income of the islander. In Table 31, we describe how the

probability to travel is influenced by the age of the islander, and in Table 32 we describe the behaviour

for frequent and non-frequent travellers.

Table 30 Probability to travel depending on Price and Household income

Income = 1000 € Income = 2000 € Income = 5000 €

Price Work or health

Leisure Work or health

Leisure Work or health

Leisure

40 € 62% 28% 62% 29% 63% 29%

80 € 57% 24% 58% 25% 59% 26%

120 € 52% 21% 53% 22% 55% 23%

160 € 47% 18% 49% 19% 51% 20%

200 € 42% 15% 44% 16% 47% 18%

240 € 37% 13% 40% 14% 43% 15%

Mode choice = 1 (least expensive) Age = 50 Non frequent traveller

65

Table 31 Probability to travel depending on Price and Age

Age = 25 Age = 50 Age = 75

Price Work or health

Leisure Work or health

Leisure Work or health

Leisure

40 € 74% 41% 62% 29% 54% 22%

80 € 70% 37% 58% 25% 50% 19%

120 € 66% 33% 53% 22% 45% 17%

160 € 62% 29% 49% 19% 41% 14%

200 € 58% 25% 44% 16% 36% 12%

240 € 53% 22% 40% 14% 32% 10%

Mode choice = 1 (least expensive) Income = 2000 € Non frequent traveller

Table 32 Probability to travel depending on Price and Travel experience

Frequent Traveller Non Frequent

Price Work or health

Leisure Work or health

Leisure

40 € 70% 36% 62% 29%

80 € 66% 32% 58% 25%

120 € 62% 28% 53% 22%

160 € 57% 25% 49% 19%

200 € 53% 21% 44% 16%

240 € 48% 19% 40% 14%

Mode choice = 1 (least expensive) Income = 2000 € Age = 50

This time, take the case of an islander who prefers the most expensive alternative to travel. If this

alternative is no longer available, for this islander, the choice of whether to travel will mainly depend

on if the return trip is available the same day he finishes his activity. In Table 33 we illustrate this case.

66

Table 33 Probability to travel depending on trip and islanders characteristics

Return Work or health Leisure

same day 66% 33% Non frequent traveller not on the same day 49% 19%

same day 74% 41% Frequent traveller not on the same day 58% 25%

Age = 50

67

6. CONCLUSION AND OUTLOOK

6.1. CONCLUSION AND CONTRIBUTIONS

The aim of this research was to develop a methodology to evaluate the transport opportunities

available to the islanders. The methodology developed is able to reflect the users’ perspective on the

transport system, and to relate it to the most relevant characteristics of the transport system. The

transport opportunities available to islanders are evaluated on the basis of how they affect the

islander’s behaviour in terms of his/her travel related choices. The methodology developed stresses

the effect of trip purpose on travel related decisions. Furthermore, it is able to evaluate the impact of

transport attributes on particularly vulnerable groups of the population, such as low-income classes or

the elderly. Hence, the methodology developed corresponds to the proposed objectives.

Our most important result relates to the fact that the mode choice decision and the travel decision are

ruled by different parameters. In the case of the islanders of Chios island, the mode choice decision is

founded on an evaluation of relevant trip attributes such as price of the trip, travel time and frequency.

Additionally, the evaluation of this attributes varies according to the socio-economic characteristics of

the decision maker. Namely, income, age, education level and activity of the individual affect the

preferences in terms of mode choice. The travel choice decision depends on the preferences of

individuals in terms of the mode choice. Individuals who prefer less expensive alternatives will judge

the disutility in travelling using other alternatives mainly based on price. If price is considered too high,

the islanders will prefer to cancel the trip, independently of any compensation in terms of travel time.

Individuals who prefer the most expensive alternatives will judge the disutility of travelling mainly

based on the possibility to return home as soon as the activity is finished. If this is not possible,

individuals may cancel the trip, independently of the price savings.

These results have important implications for Universal Service. They imply that replacing less

expensive alternatives with more expensive ones on the basis of the trade-offs implied by mode

choice models, might not be appropriate. Mode choice models are based on the assumption that

choices between modes are based on trade-offs between the different attributes of a mode. For

instances, that higher prices are compensated with shorter travel times. Our results indicate, however,

that this is only true to some extent. Some islanders will simply decide not to travel if the preferred

alternative is not available.

The methodology developed was applied to the Greek island of Chios, in the Aegean Sea. The results

show that the extent to which the decision to travel varies according to the attributes of the trip

alternatives available, and the socio-economic characteristics of the islanders.

68

6.2. FURTHER WORK

We have developed an integrated framework in which to evaluate both the mode choice decision and

the decision of whether to travel or not, for a specific purpose and a given set of available trip

alternatives. This model yielded useful information concerning the determinants of travel-related

decisions.

This information can be used to design a random utility-based accessibility measure. Random utility-

based accessibility measures are being increasingly adopted in transport literature. These measures

are usually derived from discrete choice models of mode and/or destination choices. Notwithstanding,

with the results of the models herewith developed, we can explore a new approach to an accessibility

indicator, one that focus on the decision of individuals of whether to travel, for a set of specific

purposes and a given set of available trip alternatives. Accessibility would be thus measured on the

basis of the expected maximum utility that an individual can derive from a set of travel-related

alternatives, including the option to cancel or postpone a trip. We believe that this alternative approach

can yield a richer measure of accessibility and more adequate to the context of inter-island travelling.

Another issue that can be further explored is the manner in which we explain Choice B. In our models,

the utility of the alternative not to travel (to cancel or to postpone the trip) has been described in an

oversimplified manner. We can posit that the utility of not travelling also depends on characteristics of

the activity (its urgency and importance) and on the alternative ways in which the islander can

substitute the activity carried out outside the island by an activity carried out within the island. In other

words, we hypothesize that the Utility function of the alternative “cancel or postpone the trip” can be

made dependent upon:

� Trip purpose

� The attitudes of individuals concerning the specific purpose of the trip (the importance attributed

to the activity), which will vary depending on the individual.

� The degree to which the individual perceives he/she can get satisfaction from engaging in the

same activity within the island.

Methodologically, this hypothesis can be tested by integrating latent variables in the discrete choice

model. These latent variables would capture the decision makers’ attitudes and perceptions about the

trip purpose in question.

69

REFERENCES

Armstrong, H.W. and R. Read (2004). Insularity, remoteness, mountains and archipelagos: A

combination of challenges facing small States? Europe at the Margins: EU Regional Policy,

Peripherality and Rurality, Angers, France, 15-16 April 2004.

Baldacchino, G. (2006). Small Islands versus Big Cities: Lessons in the Political Economy of

Regional Development from the World’s Small Islands, The Journal of Technology Transfer 31 (1):

91-100.

Ben-Akiva, M.E. and F.S. Koppelman (1974). Multidimensional Choice Models: Alternative

Structures of Travel Demand Models, Transportation Research Record (149): 129-142.

Ben-Akiva, M.E. and S.R. Lerman (1985). Discrete choice analysis: theory and application to travel

demand, MIT Press. Cambridge, Massachussetts.

Bennett, P. (2006). Competing for the Island Lifeline: European Law, State Aid and Regional Public

Services, Regional Studies 40 (8): 953 - 966.

Bierlaire, M. (2003). BIOGEME: A free package for the estimation of discrete choice models.

Proceedings of the 3rd Swiss Transportation Research Conference, Ascona, Switzerland.

Chambre de Commerce et d'Industrie d'Ajaccio et de la Corse-du-Sud A Summary of the

Corsican/Continental France Maritime Territorial Continuity, Commission des Iles.

Chlomoudis, C.I., P.L. Pallis, S. Papadimitriou and E.S. Tzannatos (2007). The liberalisation of

maritime transport and the island regions in EU. Evidence from Greece., European Transport \

Trasporti Europei (37): 1-15.

Commission, E. (2004). White Paper on Services of General Interest.

Cremer, H., F. Gasmi, A. Grimaud and J.J. Laffont (1998a). The Economics of Universal Service:

Practice, The Economic Development Institute of the World Bank.

Cremer, H., F. Gasmi, A. Grimaud and J.J. Laffont (1998b). The Economics of Universal Service:

Theory, The Economic Development Institute of the World Bank.

Cross, M. (1996). Service Availability and Development among Ireland's Island Communities - the

Implications for Population Stability, Irish Geography 29 (1): 13.

Cross, M. and S. Nutley (1999). Insularity and Accessibility: the Small Island Communities of

Western Ireland, Journal of Rural Studies 15 (3): 317-330.

Escalona-Orcao, A. and C. Diez-Cornago (2007). Accessibility to basic services in one of the most

sparsely populated areas in Europe: the province of Teruel (Spain), Area 39 (3): 295-309.

Eurisles (1997). Statistical Indicators of Regional Disparities Caused by Insularity and Peripherality.

Summary report. EURISLES. Greek Edition: Mytilini, University of the Aegean.

70

Eurisles (2003). Island transport and the EU Situation in 2003. Study realized under the initiative of

CPMR Islands Commission.

Gederi (2004). Depopulation in Islands. GEDERI Seminar Which strategies for maintaining people to

remain living in, or attracting them to, islands suffering from depopulation?, Stornoway, Western Isles,

1-3 December 2004.

Hernández Luis, J.Á. (2002). Temporal accessibility in archipelagos: inter-island shipping in the

Canary Islands, Journal of Transport Geography 10 (3): 231-239.

Hernández Luis, J.Á. (2004). The role of inter-island air transport in the Canary Islands, Journal of

Transport Geography 12 (3): 235-244.

Kitrinou, E., A. Polydoropoulou and D. Bolduc (2009). Development of Integrated Choice and

Latent Variable (ICLV) models for the residential relocation decision in island areas. International

Choice Modelling Conference 2009, Institute for Transport Studies, University of Leeds, 30 March - 1

April 2009.

Kizos, T. (2007). Island Lifestyles in the Aegean Islands, Greece: Heaven in Summer, Hell in Winter?

Seasonal Landscapes, Springer Netherlands. 7: 127 - 149.

Koppelman, F.S. and C. Bhat (2006). A Self Instructing Course in Mode Choice Modeling:

Multinomial and Nested Logit Models, U.S. Department of Transportation Federal Transit

Administration.

Lekakou, M.B. (2007). Chapter 8 The Eternal Conundrum of Greek Coastal Shipping, Research in

Transportation Economics 21: 257-296.

Levy, S. (2009a). Universal Service in the provision of transport to the islands: challenges and

proposals. 49th European Congress of the Regional Science Association International, Łódź, Poland,

25-29 August 2009

Levy, S. (2009b). Universal Service in the provision of transport to the islands: theoretical framework

and application. European Transport Conference, Noordwijkerhout, the Netherlands, 5 - 7 October

2009.

Ortúzar, J.D.D. and R.M. Gonzalez (2002). Inter-Island Travel Demand Response with Discrete

Choice Models: Functional Form, Forecasts, and Elasticities, Journal of Transport Economics and

Policy 36: 115-138.

Panou, C. (2007). Universal service obligations in insular areas. 11th World Conference on Transport

Research WCTR, Berkeley, California.

Polydoropoulou, A. and N. Litinas (2007). Chapter 9 Demand Models for Greek Passenger

Shipping, Research in Transportation Economics 21: 297-322.

Preston, J. and F. Rajé (2007). Accessibility, mobility and transport-related social exclusion, Journal

of Transport Geography 15 (3): 151-160.

Rigas, K. (in Press). Boat or airplane? Passengers' perceptions of transport services to islands. The

71

example of the Greek domestic leisure market, Journal of Transport Geography In Press, Corrected

Proof.

Román, C., et al. (2008). Analyzing Mobility in Peripheral Regions of the European Union: The Case

of Canarias-Madeira-Azores, Networks and Spatial Economics 8 (2): 141-160.

Roson, R. (2001). Assessing the Option Value of a Publicly Provided Service: The Case of Local

Transport, Urban Studies 38 (8): 1319-1327.

Rutz, W.O.A. and J.R. Coull (1996). Inter-island passenger shipping in Indonesia: development of the

system : Present characteristics and future requirements, Journal of Transport Geography 4 (4): 275-

286.

Sambracos, E. and K. Rigas (2007). Passenger reactions to market deregulation: First results from

the experience of the Greek islands market, Journal of Air Transport Management 13 (2): 61-66.

Vickerman, R., K. Spiekermann and M. Wegener (1999). Accessibility and Economic Development

in Europe, Regional Studies 33 (1): 1 - 15.

Waters, W.G., J. Evans and J. Caravan (1996). Subsidy policy on low volume ferry routes : British

Columbia's coastal services, Transport Policy 3 (3): 111-121.

72

ANNEX

73

SURVEY VERSION A

FOR THE INTERVIEWER

Date: dd – mm – yyyy

Time:

Interviewer name:

SURVEY INTRODUCTION

Hello, we work for the University of the Aegean and we are doing a survey on transportation. We

would like to ask you some questions about the trips you make to other islands or to mainland Greece.

The survey will only take about 15 minutes.

The answers you provide are anonymous and will only be used for academic purposes.

The survey is divided in three main parts. First, we will ask some questions about your social and

demographic environment. Then, we will ask some questions about your perceptions on the

availability of some services in the island of Chios. In the final part, we will ask you about your

preferences over alternative scenarios for trips.

Would you mind taking this survey?

Thank you

74

1. SOCIO-DEMOGRAPHIC DATA

Question Answer Observations Code

1.1 Do you live in Chios? Yes No If answer is No, stop survey. Y or N

1.2 Do you live in Chios town? Yes No Y or N

1.3 Age: In what year were you born? 19__ If after 1993, stop survey.

1.4 Gender Male Female M or F

1.5 Where is your main activity (work/study)

located? in Chios

else-

where

If answer is “in Chios” go to

1.8

1 2

If not in Chios, in which part of Greece do you work/study?

B. Ellada 1

S. Ellada 2

Peloponese 3

1.6

Islands 4

1.7 If you answered islands, which island? name

1.8 Choose the option that best describes your main activity

Work for private company 1 Student 6

Work for public administration 2 Military 7

Liberal activity 3 Unemployed 8

House worker, fisherman, rural worker 4 Retired 9

Sailor 5 None of the above 10

1.9 Size of household: How many people live in your

house (share house and meals), including you?

If lives in a residence

household size is 1

number

≥ 1

75

1.10 What is the average monthly income of your household?

less than 1000 € 1

1001 € – 2000 € 2

2001 € – 5000 € 3

more than 5000 € 4

1.11 What is your education level? (tick mark only the highest education level completed)

Less than 6 years of school 1

Complete basic level (6 or more but less than 12 years of school) 2

Complete high school (12 or more years of school but not graduate) 3

Graduated from university or technical university or more 4

1.12 Do you benefit from any personal discounts in maritime transport? (military, under-18, students, pensioners)

(Please tick mark the box corresponding to the average discount)

100% (I travel for free) 1 If answer is 100%, Stop survey

99%-50% 2

49%-25% 3

24% - 1% 4

No discount 5

1.13 Since the beginning of the year 2009, how many trips did you make from Chios? (consider every trip departing

from Chios to any destination outside the island, return trip not included) (Please tick mark the appropriate box)

0 1 11 – 20 5

1 – 3 2 21 – 30 6

4 – 5 3 more than 30 7

6 – 10 4

1.14 Do you have an internet connection at home? Yes No Y or N

76

2. PERCEPTIONS ON ISLAND SERVICES

2.1 How satisfied are you with the availability of the following services in the island of Chios?

very

satisfied satisfied neutral dissatisfied

very

dissatisfied

2.1a health services (availability of hospital,

doctors, nursing, medical exam labs,…)

2.1b leisure and culture opportunities (theatre,

concert, sports events, …)

2.1c job and business opportunities (jobs,

suppliers, customers, …)

1 2 3 4 5

2.2 When you go to the doctor/ leisure / work or business, do you often travel to another island or mainland?

never rarely sometimes always frequently

2.2a I have to travel (out of Chios) in order to

get the health care I need

2.2b I have to travel (out of Chios) to go to

leisure, sports or cultural events

2.2c I have to travel (out of Chios) on work or

business

1 2 3 4 5

2.3 Please tell us how you feel about the following sentences

strongly

disagree disagree

don’t agree

nor disagree agree

strongly

agree

2.3a I have all the health care I need in Chios

2.3b I have a sufficient cultural and leisure

offer in Chios

2.3c I have enough job or business

opportunities in Chios

1 2 3 4 5

77

2.4 Rank the following activities in order of how important you think they are for you

write first, second or third

2.4.a getting adequate health care 1, 2 or 3

2.4.b going to cultural and leisure activities 1, 2 or 3

2.4.c having a good job or a successful business 1, 2 or 3

78

3. STATED PREFERENCES

3.1a Imagine that you have a doctor appointment in Athens on Monday morning. You travel to Athens on

Sunday night. Tick mark the alternative you were most likely to choose

Alternative 1 Alternative 2

Round trip price (economy class) 50 Eur 150 Eur

Total Travel Time (round trip) 18h (9h + 9h) 4h (2h + 2h)

Frequency allows you to be back on Wednesday afternoon Monday afternoon

1 2

3.1b Now, imagine that the alternative you chose is no longer available

I would choose the other

alternative

I would not go or try to

reschedule

1 2

3.1c For the same scenario as above, now choose between the following two alternatives

Alternative 1 Alternative 2

Round trip price (economy class) 40 Eur 100 Eur

Total Travel Time (round trip) 6h (3h + 3h) 2h (1h + 1h)

Frequency allows you to be back on Tuesday afternoon Monday afternoon

1 2

3.1d Now, imagine that the alternative you chose is no longer available

I would choose the other

alternative

I would not go or try to

reschedule

1 2

79

3.2a Now imagine that you are thinking of going to a cultural event (choose your favourite: concert, theatre,

football match) in Athens on Saturday night. You travel on Saturday morning. Tick mark the alternative you

were most likely to choose

Alternative 1 Alternative 2

Round trip price (economy class) 50 Eur 150 Eur

Total Travel Time (round trip) 18h (9h + 9h) 4h (2h + 2h)

Frequency allows you to be back on Tuesday afternoon Sunday afternoon

1 2

3.2b Now, imagine that the alternative you chose is no longer available

I would choose the other

alternative

I would not go or try to

reschedule

1 2

3.2c For the same scenario as above, now choose between the following two alternatives

Alternative 1 Alternative 2

Round trip price (economy class) 40 Eur 100 Eur

Total Travel Time (round trip) 6h (3h + 3h) 2h (1h + 1h)

Frequency allows you to be back on Monday afternoon Sunday afternoon

1 2

3.2d Now, imagine that the alternative you chose is no longer available

I would choose the other

alternative

I would not go or try to

reschedule

1 2

80

3.3a Now imagine that you are going for a business meeting in Athens on Wednesday afternoon.

Alternative 1 Alternative 2

Round trip price (economy class) 60 Eur 150 Eur

Total Travel Time (round trip) 18h (9h + 9h) 4h (2h + 2h)

Frequency allows you to be back on Friday night Wednesday night

1 2

3.3b Now, imagine that the alternative you chose is no longer available

I would choose the other

alternative

I would not go or try to

reschedule

1 2

3.3c For the same scenario as above, now choose between the following two alternatives

Alternative 1 Alternative 2

Round trip price (economy class) 40 Eur 100 Eur

Total Travel Time (round trip) 6h (3h + 3h) 2h (1h + 1h)

Frequency allows you be back on Thursday night Wednesday night

1 2

3.3d Now, imagine that the alternative you chose is no longer available

I would choose the other

alternative

I would not go or try to

reschedule

1 2