You are on page 1of 16

An Investigation of the Willingness to Pay to Reduce Rail

Overcrowding
Gerard Whelan and Jon Crockett
MVA Consultancy

Abstract
Overcrowding is becoming an increasingly important issue for the UK rail industry. The demand for
passenger rail travel is at a 50 year high and is growing at a rate of 4.5% per annum. In many cases, the
network is at, or close to, capacity and there is limited scope to price-off demand as principal commuter
fares are regulated. The UK Department for Transport (DfT) is responsible for both rail industry strategy
and for the specification of franchises; strategies for managing demand and capacity are integral to both
of these. To be able to demonstrate value for money, it is essential for the DfT to have robust estimates
of both the economic value of reducing overcrowding and the demand effects of it.

Evidence on the valuation of overcrowding contained within the rail sector’s Passenger Demand
Forecasting Handbook (PDFH) has a number of limitations. Existing values only extend to a certain level
of crowding which is already exceeded in a number of situations, and without mitigating action, will be
exceeded in many others in the future. In addition, the level of disaggregation, particularly in terms of
geographical coverage, may not be sufficient to accurately reflect both current and future situations. The
DfT has therefore commissioned this study to obtain robust estimates of the valuation of overcrowding
across the range of contexts in which it may occur, so that it can take appropriate decisions on policies to
mitigate overcrowding and demonstrate value for money.

The methodological approach adopted in this research is based on a large scale programme of market
research including a series of six focus groups, 50 in-depth cognitive interviews, and the distribution of
more than 10,000 web and paper based questionnaires across 14 sites aimed at collecting transfer price
(TP), revealed preference (RP) and stated preference data. In this paper, we provide a description of the
development and analysis of the stated preference survey.

An important aspect of the study was to undertake a qualitative investigation of how rail users consider
overcrowding, how they change their behaviour to mitigate its effects, and how to describe/ present
different levels of crowding as “show material” in the questionnaire. This investigation helped shape the
specification of the strength-of-preference experiments and overall questionnaire design.

The questionnaire returns were analysed under choice modelling framework involving the estimation of a
range of advanced choice models. The focus of model estimation was to examine how the value of travel
time savings for rail users is influenced by both the duration and level of overcrowding, and how these
(non-linear) values vary across individuals and market segments. In defining the analysis we pay close
attention to how values of crowding vary with journey duration and the travelling environment (especially
standing, or sitting) and develop a “family of curves” showing the relationship between crowding
penalties and the level of crowding.

1 Introduction

Overcrowding is becoming an increasingly important issue for the UK rail industry. The demand for
passenger rail travel is at a 50 year high and is growing at a rate of 4.5% per annum. In many cases, the
network is at, or close to, capacity and there is limited scope to price-off demand as principal commuter
fares are regulated. The UK Department for Transport (DfT) is responsible for both rail industry strategy
and for the specification of franchises; strategies for managing demand and capacity are integral to both
of these. To be able to demonstrate value for money, it is essential for the DfT to have robust estimates
of both the economic value of reducing overcrowding and the demand effects of it.
Evidence on the valuation of overcrowding is relatively scarce with almost all of the available evidence
contained in confidential consultancy reports to government departments or train operators1. A summary
of this evidence is presented in the UK rail sector’s Passenger Demand Forecasting Handbook (ATOC,
2005). Existing values, however, only extend to a certain level of crowding which is already exceeded in a
number of situations, and without mitigating action, will be exceeded in many others in the future. In
addition, the level of disaggregation, particularly in terms of geographical coverage, is limited. The DfT
has therefore commissioned this study to obtain robust estimates of the valuation of overcrowding across
the range of contexts in which it may occur, so that it can take appropriate decisions on policies to
mitigate overcrowding and demonstrate value for money.

The methodological approach adopted in this research is based on a large scale programme of market
research including a series of six focus groups, 50 in-depth cognitive interviews, and the distribution of
more than 10,000 paper based questionnaires across 14 sites aimed at collecting transfer price (TP),
revealed preference (RP) and stated preference data.

In this paper, we provide a description of the development and analysis of the stated preference survey.
An important aspect of the study was to undertake a qualitative investigation of how rail users consider
overcrowding, how they change their behaviour to mitigate its effects, and how to describe/ present
different levels of crowding as ‘show material’ in the questionnaire. This investigation helped shape the
specification of the preference experiments and overall questionnaire design.

The questionnaire returns were analysed under choice modelling framework involving the estimation of a
range of advanced choice models. The focus of model estimation was to examine how the value of travel
time savings for rail users is influenced by both the duration and level of overcrowding, and how these
(non-linear) values vary across individuals and market segments. In defining the analysis we pay close
attention to how values of crowding vary with journey duration and the travelling environment (especially
standing, or sitting) and develop a ‘family of curves’ showing the relationship between crowding penalties
and the level of crowding.

The remainder of this paper is structured as follows. In Section 2 we provide details of the
methodological framework covering the multi-stage SP design process. Section 3 describes data
collection. Section 4 describes model estimation and results. Finally Section 5 provides our conclusions
and recommendations.

2 Stated Preference Experimental Design

The SP experiment was subject to a multi-stage design process including six focus groups and detailed
cognitive testing, to ensure that respondents were presented with scenarios which were not just easily
interpretable, but also enabled the key attributes of interest to be presented in a quantifiable manner.
Different aspects of the design are discussed in turn below.

(a) The Choice Context

The SP experiments involved presenting respondents with situations in which they can choose between
two alternatives that involve a trade-off between crowding and other travel attributes. Possible choice
contexts included:

1
Accent, 2006; Accent and Hague Consulting Group, 1997; Faber Maunsell, 2007; Faber Maunsell and Mott
Macdonald, 2007; Hague Consulting Group, 1999; LUL, 1988; MVA, 1989; MVA and ITS, 1989; MVA, 2000; Passenger
Focus, 2006; TCI-OR, 1996.
 travelling on a fast crowded train versus slower un-crowded train;

 travelling earlier or later than desired in order to avoid crowding;

 travelling in First or Standard class (note in this context we would need to be careful to
account for the other benefits of travelling first class as well as crowding); and

 standing near the front of the train or sitting at the rear.

To ensure a credible variation in the level of crowding and to keep the non-crowding related aspects of
travel as close to respondents’ current travel conditions as possible we adopted a choice context based on
a faster more crowded train versus a slower less crowded train.

(b) Response Method

We explored a series of response method options in the focus groups and during cognitive testing, and
rather than simply ask respondents to state a preference between the two travel alternatives our sample
valued response methods that allowed them to express their strength-of-preference for each alternative.
To keep modelling relatively straightforward we adopted a response method with a four point scale:
Stongly Prefer A, Prefer A, Prefer B, and Strongly Prefer B.

(c) Choice Attributes

A number of choice attributes were considered for inclusion within the SP exercises including: (i) the level
and duration of crowding, (ii) how frequently they have to stand, (iii) fare, (iv) journey time, (v) time of
departure, and (vi) rolling stock quality/layout. Given the choice context, the SP exercise was designed to
capture the value of standing / sitting at a range of loads in different geographic, temporal and journey
purpose contexts by presenting respondents with scenarios described in terms of the level of crowding
level, the ability to get a seat, the journey time and the fare.

(d) Units of Measurement for the Attributes

All attributes need to be described and presented in an objective and quantifiable way with minimal scope
for differences in interpretation. A key objective of the focus groups was therefore to identify the most
reliable means of presentation of each of these attributes. For some of the attributes listed (e.g. fare and
journey time) the units of measurement are natural and straightforward. However, this is not the case for
others (e.g. crowding, requirement to stand, rolling stock layout), where no natural units exist. The focus
groups helped shape our design and descriptions to make them relevant and meaningful to respondents.

A significant technical issue that needed to be addressed was how to present the attribute of crowding,
such that respondents and analysts have a clear and consistent understanding of what the levels
presented for this attribute actually represent. To this end, we developed a combination of verbal and
graphical stimulus material for use in the SP study. This material was tested and further developed
through the Focus Groups. Key issues identified include: (i) what defines the base (uncrowded) level of
crowding (ii) what is the smallest incremental change in crowding levels that respondents can distinguish,
and (iii) what impact does rolling stock have on perceptions of crowding?

In developing the stimulus material we concluded that it will be necessary to develop customised sets of
graphical stimuli for different stock types. This was firstly to enable respondents to identify with crowding
in the context in which they normally experience it. Secondly, it was not appropriate to compare the
experience of standing on a stock type that is designed to carry large numbers of standing passengers
with one that is not.

We defined three generic stock types: Intercity, Interurban, and Commuter. Intercity encompasses stock
with 2+2 seating and doors at the carriage ends, such as HSTs, Pendolinos, Class 444s, Voyagers and
their derivatives. Interurban represents stock with 2+2 seating, but door at 1/3 and 2/3 of the carriage
length, such as Class 185s and Class 170s. Commuter stock also has 1/3 and 2/3 doors, but involves
3+2 seating, such as Class 450, Class 165/6, and Class 150.

Our initial thinking was to include as many as eight levels of crowding, including two relatively uncrowded
levels. These levels were defined in terms of thresholds where the availability of specific types of seating
or standing space change. For example, on a train with 2+2 seating, if 50% of seats are taken this
represents a threshold where individual travellers are likely to have to sit immediately next to another
passenger in a pair of seats.

Table 2.1 summarises the crowding levels that were tested for commuter stock. Definitions for other
stock types are similar but with some adjusted to suit that particular stock.

Table 2.1: Levels for Crowding Attribute tested in focus groups

Level Seat Standing Physical Threshold


Occupancy

1 25% Nil Single traveller can have cluster of four seats to self

2 50% Nil Single traveller can have pair of seats to self

3 80% Nil Middle seats on banks of three start to fill

4 90% Up to four in each door area Some passengers choose to lean in vestibules in
preference to taking middle seats

5 100% Up to eight in each door area Standees can all lean on bulkheads

6 100% up to twelve in each door area, a few Some standees will need to start using seats for
in aisles support

7 100% all door and aisle areas 'comfortably Everyone still has personal space
full', not touching

8 100% uncomfortably full, standees touching Personal space is compromised

We represented these levels graphically through plan views of the carriage, populated by representations
of individual passengers Consideration was also undertaken with photographic images of carriage
interiors, manipulated to show different numbers of passengers and hence different levels of crowding.
However, it proved difficult to construct images that could show relatively small differences in crowding
levels convincingly and consistently, so this idea was abandoned. An example of the layout of the SP
experiment is shown in Figure 2.1 below – it is noted that rolling stock layouts were shown to scale to
ensure that an objective crowding metric could be used during analysis.
Figure 2.1: SP Layout

(e) Levels of Attributes

Within an SP design it is necessary to include a sufficient number of attribute levels to enable an


examination of non-linearities in response. In the main, this will involve three attribute levels for each
attribute. However, for the crowding variable it was necessary to include more levels, in order to
understand non-linearities over the whole range of crowding experiences. The selection of values for
attributes levels was guided by the need to ensure realism in the representation of the choice problem
and the need to include values relevant for policy testing. Other factors to consider include the range of
any interest in non-linearity, and the variability (and hence significance) of attribute coefficients.

Definition of a suitable base level, where passengers feel ‘uncrowded’, and thereby no crowding penalty
occurs, is critical. Undoubtedly such levels can vary quite significantly according to characteristics of the
traveller, rolling stock, geographic context etc. For example, one passenger’s perception of (the
commencement of) crowding may be the inability to choose their favourite seat, whilst another’s may be
the requirement to stand for their entire journey. Table 2.2 shows the range of crowding levels modelled.
Table 2.2: Crowding Levels

Graphic ID Seats Occupied Passengers Standing

1 25% 0

2 50% 0

3 60% 0

4 70% 0

5 80% 0

6 80% 5

7 90% 0

8 90% 5

9 100% 0

10 100% 10

11 100% 20

12 100% 30

13 100% 40

14 100% 50

15 100% 60

16 100% 70

In order to capture the full range of effects of crowding on choice behaviour it was necessary to have
both a large number of sub-designs of the main design, to allow the survey to be tailored to individual
circumstances, and only present a number of subsets to each respondent (in order to reduce the
cognitive burden).

The designs were initially based upon an orthogonal design permitting the identification of the main
effects and key interactions (e.g. the interaction between journey time and the level of crowding). This
design was then modified in order to maintain realism in the choice context for the respondent. As the
crowding levels used in the experiment contained significant variation, in order to investigate the full
spectrum, we had to depart from orthogonality to match whether the respondent was ‘sitting’ or
‘standing’ with the load factor presented. For example, we could not present a respondent with a
scenario where they were told they were standing in a load factor of 90%. We accepted the sacrifice in
statistical terms with our blocking approach due to the gain in respondent experience.

3 Data Collection

Prior to piloting, the SP designs were subjected to a series of computer simulations to ensure that a full
range of attribute values were recoverable and the questionnaires ‘sense checked’ via a series cognitive
interviews with a sample of rail passengers at Manchester Piccadilly station. The questionnaire was then
successfully piloted.

The main data collection exercise was undertaken during February and March 2008 across 21 commuter
flows and two inter-city flows in Great Britain. Questionnaires were distributed at the ‘origin’ stations
during the morning peak period.
The objective of the data collection exercise was to gather information on rail users’ preferences in order
to develop a model to infer values of overcrowding. The surveys were administered in the form of a self
completion questionnaire distributed across a comprehensive range of short and long-distance rail routes
(a copy of the questionnaire is available from the authors on request). A total of 10,000 questionnaires
were distributed with a sample of 2,314 completed forms returned by freepost. Details of the composition
of the sample are presented in Table 3.1.

Table 3.1: Characteristics of the Sample

Income Percent Age Percent Gender Percent Purpose Percent

< £10K 2 16-24 10 Male 51 Commute 83

£10K-£20K 7 25-34 31 Female 49 Employer 8

£20K-£35K 19 35-44 24 Total 100 Education 4

£35K-£50K 17 45-54 22 Other 5

£50K-£75K 21 55-64 11 Total 100

£75K-£100K 11 65-74 1

>£100K 13 75+ 0

Prefer not to say 11 Prefer not to say 1

Total 100 Total 100

4 Model Estimation

4.1 Modelling Framework

This section describes the methodology employed in the analysis of the Stated Preference data, the types
of model fitted and the processes by which the final values were derived. The approach adopts a
traditional choice modelling framework in which decision-makers are assumed to make choices based on
the concept of utility maximisation. Respondent preferences were analysed using a binary logit model in
the first instance, followed by mixed logit (random parameters) and ordered response approaches. As the
Stated Preference experiment involved a binary choice there is no need to consider methods that
accommodate correlation between alternatives, such as the nested logit model.

It is useful to think about crowding in terms of differences in passengers’ values of time in different travel
conditions. As the level and duration of crowding increases, the disutility of travel also increases and
passengers’ willingness to pay to save travel time is likely to increase. The benefits of adopting this
approach include:

 The framework is consistent with economic theory;

 The framework is tried and tested and widely accepted on theoretical and practical
grounds;

 The framework is flexible, allowing for differences in behaviour and values; and
 The framework is consistent with UK Department for Transport recommendations on
modelling and appraisal.

The analysis began with the estimation of a series of relatively general dummy variable
models, as specified in equation 1 below, in which separate values of time were identified for
the range of crowding levels specified in the SP experiments.

C
U = α 0Fare + ∑ α cDc IVT (1)
c =1

Where:

U is the utility of travel


Fare is the single fare equivalent (pence)
IVT is the in-vehicle time (minutes)
Dc is a vector of C dummy variables representing sitting in 16 different levels and standing at 11
different levels of crowding
α is a vector of coefficients to be estimated

The value of time spent sitting or standing in different levels of crowding can be easily estimated via the
ratio of marginal utilities to time and cost (e.g. α c α 0 ). Alternatively, the value of time in crowded

conditions can be expressed relative to values in un-crowded conditions to express the value of crowding
as a multiplier of the base value of time (e.g. α crowded α uncrowded ). Expressing the value of crowding in

terms of a value of time multiplier provides an easily understandable, transferable and simple way to
capture the behavioural influences and benefits that arise from a reduction to the level of crowding.

A number of points emerged from the preliminary dummy variable estimation as follows: (a) The value of
time is stable up to a given load factor and thereafter it increases as the load factor increases; (b) the
value of time is higher for standing than for sitting; (c) the value of time is higher for respondents in
London and the South East than it is for regional respondents; and (d) the value of time (and crowding
VOT multiplier) is higher for longer distance journeys than for short distance journeys.

4.2 Functional Form

The next stage was to examine a range of alternative functional forms to establish if the findings from
the dummy variable estimation can be captured using a more parsimonious form. Four alternative utility
function specifications were put forward for consideration including linear piecewise, power, exponential
and (s-shaped) Gompertz.

The linear and power functions are shown in equation 2 below:

U= α 0Fare +
α5
α 1IVTsit + α 2 IVTsit ⋅ CROWD sit + (2)
α6
α 3 IVTstand + α 4 IVTstand ⋅ CROWD stand

Where:

IVTsit is the time spent sitting


IVTstand is the time spend standing
CROWDsit is the level of crowding for those sitting
CROWDstand is the level of crowding for those standing
Within this specification there is an underlying value of time spent either seated or standing and these
values of time increase as the level of crowding increases beyond a given minimum threshold. As such,
the crowding variable CROWD is specified as:

CROWDsit = max(load factor − θsit ,0)


(3)
CROWDstand = max(load factor − θstand ,0)

The values of the crowding thresholds (θ values) are set to maximise the goodness of overall model fit
using a manual ‘grid-search’ method.

The power function is a more general specification of the linear piecewise model in which the disutility
arising from crowding is specified to be non-linear. When the crowding power terms (coefficients α5 and
α6) are less than 1 the disutility of crowding increases with the level of crowding with the rate of increase
diminishing as crowding increases. Conversely, with crowding power terms greater than 1 the rate of
increase in crowding disutility increases with the level of crowding.

The exponential function is similar to the power function in that it allows the rate of increase in crowding
disutility to increase with the level of crowding.

U= α 0Fare +
α 1IVTsit ⋅ exp(α 2 CROWD sit ) + (4)
α 3 IVTstand ⋅ exp(α 4 CROWD stand )

The Gompertz function is arguably the most complex function tested. This function shows the rate of
increase in crowding disutility to follow an s-shaped curve in which the rate of increase in crowding
disutility is initially strong then diminishes approaching a maximum value.

U= α 0Fare +
α1IVTsit + α2 IVTsit ⋅ exp(α3 exp(α 4 CROWD sit )) + (5)
α5 IVTs tan d + α6 IVTs tan d exp(α7 exp(α8 CROWD s tan d ))

All four models were successfully estimated to data from the whole sample, however, as Figure 2 shows,
there was very little difference between them across load factor levels. Indeed, the adjusted Rho Squared
values for the models were 0.228, 0.228, 0.227 and 0.23 for the linear, power, exponential and
Gompertz function respectively. Given that the models are not significantly different from each other it is
sensible to progress with the simple piecewise linear function.
3.00

2.50
Value of Time Multiplier

Linear (Sit)
2.00
Exponential (Sit)
Gompertz (Sit)
1.50 Power (Sit)
Linear (Stand)
Exponential (Stand)
1.00
Power (Stand)

0.50

0.00
0 50 100 150 200 250
Load Factor (%)

Figure 4.1: Functional Form Comparison

4.3 Load Factors and PPM

Having identified a suitable base functional form for the model the next stage is to compare model
performance using alternative measure of crowding, and in particular to compare Load Factors with a
crowding metric based on the number of people standing per square metre of available carriage space –
recall that this is possible as the rolling stock layout images were shown to scale. The estimated model
results are shown in Table 4.1. It is noted that for all models presented hereafter, fare is specified as a
natural logarithm. This transformation provides a remarkable improvement to the level of model fit and
implies that the underlying value of time increases as the fare increases. It is also noted that the models
do not take account of the repeat observations nature of the data.

Table 4.1: Base Model Results

Load Factor Model pass/m2 Model

Coefficient Value t-test Value t-test

Fare (log) -4.07 -34.31 -4.31 -35.99

IVTsit -0.0565 -26.72 -0.0552 -25.85

IVTsit*CROWDsit -0.0004 -29.04 -0.0058 -26.84

IVTstand -0.0846 -27.25 -0.0847 -25.58

IVTstand*CROWDstand -0.0005 -12.33 -0.0047 -8.31

Rho-Squared 0.229 0.221

Final Log-likelihood -7754.201 -7827.808

Number of 14504 14504


Observations
The base models results are very positive showing a good fit to the data and precisely estimated
coefficients parameters with plausible values of time and values of crowding. For the load factor model
the crowding threshold is set at 90% for seated passengers, whereas in the pass/m2 specification
crowding penalty incurs for both seated and crowded passengers at a 100% load factor.

Table 4.2: Crowding Value of Time Multipliers

Load Factor Sit Stand pass/m2 Sit Stand

80 1.00 1.50 0 1.00 1.53

100 1.08 1.50 1.0 1.11 1.62

120 1.23 1.67 2.0 1.21 1.70

140 1.38 1.85 3.0 1.32 1.79

160 1.53 2.02 4.0 1.42 1.87

180 1.68 2.20 5.0 1.53 1.96

200 1.83 2.37 6.0 1.63 2.04


2
Please note: The Load Factor and pass/m (passengers per square meter) estimates vary by rolling stock type. The
rows in this table are therefore do not match across different crowding metrics.

For trains on which there are no people standing, the load factor (people/seats) is the
appropriate measure (we assume that if seats are available on a train then any person
standing perceives no penalty – they are certainly no worse off, and may even obtain a benefit
from standing). For trains where people have to stand, we use as the measure of crowding
passengers/ square metre of standing space (pass/m2). Load factors cannot be sensibly used
in this case, due to the wide variety of rolling stock, which have very different relative spaces
allocated to seating and standing. As the value of crowding across market models estimated
was zero up to between 90% and 120% moving forward with models based on a crowding
measure of passengers per m2 has benefits in simplicity without yielding very much statistical
significance.

4.4 Market Segmentation

The analysis undertaken thus far has identified the piecewise linear model as a suitable functional form
and the number of passengers per square metre as a suitable crowding metric. The next stage to the
analysis is to identify variation in the value of crowding across key market segments including: journey
purpose (business, non-business) and geographical location (London and the South East, Regional, Inter-
urban).

Table 4.3 shows the estimated model results based on journey purpose. The model shows a good fit to
the data with coefficients showing expected signs and magnitudes and all but one being statistically
significant. It should be noted when interpreting the results that business travel accounts for only 9% of
the sample.
Table 4.3: Model Results Segmented by Journey Purpose

Non-Business Business

Coefficient Value t-test Value t-test

Fare (log) -4.42 -35.24 -3.6 -5.5


IVTsit -0.06 -25.43 -0.033 -6.39
IVTsit*CROWDsit -0.0062 -24.29 -0.0044 -11.39
IVTstand -0.0889 -25.11 -0.063 -8.68
IVTstand*CROWDstand -0.0058 -8.97 -0.0014 -1.45

Rho-Squared 0.222

Final Log-likelihood -7809.801

Number of 14504
Observations

Table 4.4 shows the estimated model results based on Geographical Location.

Table 4.4: Model Results based on Geographical Location

London & SE Regional Inter-Urban

Coefficient Value t-test Value t-test Value t-test

Fare (log) -4.71 -23.36 -4.13 -17.41 -6.22 -5.89


IVTsit -0.0814 -24.08 -0.0702 -10.93 -0.0287 -6.91
IVTsit*CROWDsit -0.0073 -16.09 -0.0169 -18.25 -0.0033 -14.49
IVTstand -0.116 -20.54 -0.0941 -10.43 -0.0508 -10.15
IVTstand*CROWDstand -0.0107 -9.75 -0.0191 -9.61 -0.0011 -2.23

Rho-Squared 0.246

Final Log-likelihood -7561.269

Number of 14504
Observations

The implied value of time multipliers derived from the market segmentation models are shown in Table
4.5 where it can be seen that business travellers have a higher penalty associated with time spent either
sitting or standing in crowded conditions, and that they are particularly reluctant to stand in any level of
crowding. Value of time multipliers are highest for regional traffic, albeit starting from a low base value,
and broadly similar for LSE and interurban flows. As might be expected, standing crowding value of time
multipliers are high for all levels of crowding on interurban flows and show a steep increase and standing
densities increase on LSE and regional flows.
Table 4.5: Crowding Value of Time Multipliers by Journey Purpose and Location

Pass/m2 Non-Business Business LSE Regional Interurban

Sit Stand Sit Stand Sit Stand Sit Stand Sit Stand

0 1.00 1.48 1.00 1.91 1.00 1.43 1.00 1.34 1.00 1.77

1 1.10 1.58 1.13 1.95 1.09 1.56 1.24 1.61 1.11 1.81

2 1.21 1.68 1.27 1.99 1.18 1.69 1.48 1.88 1.23 1.85

3 1.31 1.77 1.40 2.03 1.27 1.82 1.72 2.16 1.34 1.89

4 1.41 1.87 1.54 2.08 1.36 1.95 1.96 2.43 1.46 1.92

5 1.52 1.97 1.67 2.12 1.45 2.08 2.20 2.70 1.57 1.96

6 1.62 2.06 1.81 2.16 1.54 2.21 2.44 2.97 1.69 2.00

4.5 Distance and Income Effects

Distance and income effects are examined using a more general specification of the linear piecewise
model incorporating distance and income interaction terms as follows:

U= ( )
α0 Fare * Dλ F * I λI +
α (IVTsit * D ) + α (IVTsit * CROWDsit * D ) +
1
λ1
2
λ2
(6)
α (IVTstand * D ) + α (IVTstand * CROWDstand * D )
3
λ3
4
λ4

Where:
D is journey distance (miles)

I is annual income (£’000s)

λ is a vector of income and distance related coefficients.

The value of time sitting in crowded conditions is:

α1 (λ1 − λ F ) − λ I α2
VoT = D I + CROWDsit * D(λ 2 − λ F )I− λ I (7)
α0 α0

and the value of time multiplier is:

α2
VoTmultiplier = 1 + CROWDsit * D(λ 2 − λ1 ) (8)
α1

The value of time multiplier increases with the level of crowding and with distance if λ2 > λ1.

The model results, presented in Table 4.6, show a strong positive relationship between income and the
value of time and distance and the value of time. For relatively un-crowded conditions, the elasticity of
the value of time to income is 0.57 and the elasticity of the value of time to distance is 0.61. Whilst the
income elasticity is plausible, the journey distance elasticity is a little high with past evidence presented
in the UK (Whelan and Bates, 2002) and Swiss (Axhausen et al., 2008) value of time studies suggesting
elasticities values around 0.3.
Table 4.6: Model Results based on Income and Journey Distance

Coefficient Value t-test Coefficient Value t-test


Fare -0.792 -7.16 λI -0.568 -16.49
IVTsit -0.178 -9.22 λF -0.949 -25.0
IVTsit*CROWDsit -0.0684 -10.48 λ1 -0.335 -10.83
IVTstand -0.25 -9.07 λ2 -0.588 -21.65
IVTstand*CROWDstand -0.14 -6.28 λ3 -0.298 -10.39

λ4 -0.83 -15.68

Rho Squared 0.255

Final Log-Likelihood -7478.746

Number of 14504
Observations

The values of time multipliers implied by the model in Table 4.6 vary by income, distance and the level of
crowding. Holding annual income constant at £40K, the range of values of seated time multipliers by
distance and level of crowding are presented in Figure 4.2.

5
6-7
4 5-6
VoT Multiplier 4-5
3 3-4
2-3
2 1-2
0-1
1 6
4
0 Crowding
2 (Pax/m2)
10 20 30 40 50 60 0
70 80
Distance (miles)

Figure 4.2: Values of Time Multiplier by Distance and Level of Crowding

Interestingly, the introduction of an explicit distance effect has the effect of dampening the value of time
multipliers as distance increases, perhaps reflecting a ceiling to the value of time. Given that this effect is
marginal, it seems sensible to omit a distance effect on the value of time multiplier.

5 Conclusions and Recommendations

The objective of the study was to derive a set of monetary values of the wiliness to pay for reductions in
rail overcrowding. The analysis presented has:
 Advanced a method for valuing changes in rail overcrowding in a theoretically consistent and
practical way based on the economic concept of utility maximization;

 Meticulously crafted the design of a SP experiment to make it relevant to the respondent and at
the same time provide an objective measure of crowding either using load factors or passengers
per square metre;

 Undertaken a large-scale market research exercise including focus groups, cognitive interviews,
pilot and main surveys;

 Estimated robust choice models and generated plausible values which are easy to understand and
apply within existing modelling frameworks;

 Examined a range of model forms and identified a linear piecewise specification showing an
increase in the value of time as the level of crowding increases; and

 Assessed the variation in value of time across key market segments including journey purpose
(business and non-business), geographical location (London and the South East, Regional and
Inter-urban), journey distance and income.

In recommending a set of values, consideration is given to the practicalities of implementation, the


plausibility and statistical robustness of the results as follows:

 We recommend the adoption of passengers per square metre as the appropriate unit by which to
measure crowding levels over 100% load factor. This reflects the increasing diversity of internal
layouts available in different rolling stock types, some of which have considerably more dedicated
standing space than other stock types. In these circumstances, load factors cease to be
comparable between stock types;

 We furthermore recommend that crowding penalties should be expressed as multipliers of Value


of Time (VOT), rather than the existing practice of monetary valuation;

 Although there is a clear difference between business and non-business travel, the business
travel sample size is small and correlated with inter-urban journeys. Much of the business travel
effect will therefore be accounted for within interurban market segments and we therefore feel
that a separate set of values for business travel is not needed;

 There is clearly a strong relationship between the value of time and income but the relationship
between the value of time multiplier and distance is less strong. We therefore recommend that a
distance effect should not be included in the value of time multipliers;

 The level of income influences the value of time but not the value of time multiplier and therefore
there is not a requirement to include the influence of income in our recommendations; and

 The recommended values to take forward are therefore based on the model with geographical
segmentation with values shown in Table 4.5.

Acknowledgements

We wish to acknowledge the support of the Department for Transport in undertaking this project,
together with assistance from the Train Operating Companies who facilitated the market research. We
are also grateful to our colleagues John Segal, Chris Pownall, Sekai Seawell and Laura Hunt for input to
the project. The thoughts expressed in this paper are entirely those of the authors, however, and do not
necessarily reflect those of the Department for Transport.
References

Accent (2006) Rail Customer Valuations of Sitting, Standing and Crowding. Report for Transport for
London.

Accent and Hague Consulting Group (1997) Customer Valuation of Different Load Factors. Report for
First Great Western.

ATOC (2005) Passenger Demand Forecasting Handbook v4.1. Passenger Demand Forecasting Council,
ATOC, London UK.

Axhausen, K., Hess, S., Konig, A., Abayc, G., Bates, J. Bierlaire, M. (2008) Income and distance
elasticities of values of travel time savings: New Swiss results. Transport Policy, Vol 15, pp 173-185

Faber Maunsell (2007) Demand Management Techniques – Peak Spreading. Report for the Department
for Transport, Transport for London and Network Rail.

Faber Maunsell and Mott macdonald (2007) Rail Overcrowding, Reliability and Frequency. Report for
Centro, the West Midlands Passenger Transport Executive.

Fosgerau, M. and Bierlaire M. (2008) A practical test for the choice of mixing distribution in discrete
choice models. Transportation Research Part B

Hague Consulting Group (1999) Evaluating Enhanced Service Packages and Crowding. Report for the
PDFSS.

Hess, S., Bierlaire, M., Polak, J. (2005) Estimation of value of travel-time savings using mixed logit
models. Transportation Research Part A 39 (3), 221–236.

LUL (1988) Crowding Penalties for London Underground. Report for London Underground Limited.

MVA (1989) Network Southeast Quality of Service Research. Report for Network Southeast, British Rail.

MVA and ITS (1989) Regional Railways Overcrowding Study. Report for Regional Railways, British Rail.

MVA (1991) Market Research for ORCATS. Report for British Rail.

MVA (2000) Valuation of Crowding Improvements on Rail Services. Report for the Strategic Rail
Authority.

Passenger Focus (2006) Overcrowding: a passenger perspective. [Available from:


http://www.passengerfocus.org.uk/news-and-publications/document-search/; Accessed: 31/10/07].

TCI-OR (1996) Crowding Costs. Rail OR Report Number SSF002/02, 03, 04.

Whelan G.A. and Bates J. (2002) ‘UK Value of Time Working Paper 5: Market Segmentation Analysis’.
Working Paper 565. Institute for Transport Studies, University of Leeds.

You might also like