EVALUATION OF PASSENGER TRAIN CONCEPTS – Practical Methods for ...

EVALUATION OF PASSENGER TRAIN
CONCEPTS – Practical Methods for
Measuring Travellers’ preferences in
Relation to Costs
K. Kottenhoff a
a
Inst. of Infrastructure
Transportation & Logistics
KTH (The Royal Institute of Technology)
SE-100 44 Stockholm, Sweden
Abstract : This paper presents practical methods in these areas; costs, valuation methods
and handling of package effects and finally how to trade off passengers’ valuations to costs.
The methods are used as a toolbox for identifying the measures that would increase the
value for rail passengers and the society and increasing the profitability for rail operators.
”Practical methods” means that the procedures for data collection and analysis should be
efficient in terms of research time and cost and effective in terms of high quality results. Both
costs and valuations are described as per cent change. These changes can be traded off.
Beneficial measures are those where the difference between passengers’ marginal valuation
and operators’ marginal cost is positive. Examples of such measures are higher train speeds
and many on-board comfort measures.
Key Words: Passenger Train, Valuation, Cost, Trade-off, Passengers, Comfort

Introduction
It is often difficult to operate rail traffic with sufficient profitability. For many years, the railway
industry has failed in some respects to choose rolling stock based on the end users’ preferences
and willingness to pay for using the railways. The reasons for this include the lack of
established, practical methods in the railway sector 1 to evaluate the relationship between supply/train
properties and travellers’ preferences. The methods should be interpretable and affordable
to have a better chance of being adopted by the industry.
This paper will present practical methods in the areas; costs, valuation methods and package
effects and finally trade off. A cost model will be introduced first. Then follows the two sections
about valuation methods. Finally a trade-off method is presented.
1. Method to
identify
cost reduction
potentials
2. Valuation
methods:
SP with CASI
interviewing
3. Method to
circumvent
package
effects
4. Trade-off
method
costs/
valuations
Figure 1: This paper presents practical methods grouped into four chapters.
1
Stated preference studies has though been conducted by some railways, for example BR, DSB and NS.
2

1 Model for the identification of cost reduction potentials of
various measures
In order to calculate the cost of passenger rail services, an economic model 2 has been developed.
The aim is to analyse how much different measures influence the cost of train operation.
Outputs from the economic model are SEK/passenger kilometre for the actual combination
and an earlier reference combination of traffic and vehicle types. The structure of the
model is shown in figure 2. The model output is train service cost per passenger-km using
different input values, such as vehicle investment, maintenance, staff, energy costs and rail
charges. The trains are supposed to operate according to one of a number of traffic type alternatives.
In the examples further on, InterCity services 1 have been used.
Purchase
unit prices
Rail fees
fixed fee
track use fee
etc
Maintenance
km–dep.
time-dep.
Energy
traction
heating etc
Cleaning
time-dep.
Train staff
time-dep.
*
annuity
*
no. of
vehicles
/train
*
no. of
vehicles
/train
*
km
/run
*
no. of
vehicles
/train
*
hours
/run
*
staff
size
Fixed vehicle costs
Distance costs/run
Time costs/run
*
shopspare
*
runs/year
Cost/year/trainset
Type of
traffic
Mariginal costs
(for extra train-sets)
/
km/year
Expenses
(/ traveller)
*
1/2
Km-costs
+ X % exp.
/no. of seats
* occupancy rate
Cost of traffic/pass.km
Figure 2: Economic model used to calculate train service costs and elasticities. The three cost types,
fixed, distance and time, add to cost/year/trainset. Expenses are anticipated to be roughly 20-25% extra.
These expenses should represent marginal costs for terminal services as ticketing, train shunting
and so on. Finally the cost of traffic per passenger kilometre is calculated.
2
The economic model has been named "Tåganalys" in Swedish
3

The real cost figures for the main operator in Sweden, SJ AB, is not official and can not be
used explicitly. The economic model is therefore based in part on cost assumptions in the
“Handbook on Investment Cost Benefit Analysis” by the Swedish Rail Administration [1]
The cost levels has been checked against other sources (e.g. Baumgartner [4]) and various
railway journals.
1.1 Cost elasticities
Mathematically elasticity is defined by the formula:
Elasticity
=
∆y
/ y
∆x1 / x
1
, where
y =
f ( x 1
, x2....)
Traffic planners often use a similar measure to express how much travelling is altered marginally
for a marginal change in travelling cost or travelling time. That measure can be called
"travelling elasticity with respect to (for example) travel cost". The measure used to identify
how sensitive train traffic costs are can be called "train-cost elasticity with respect to input
factor (vehicle investment, energy, interest rate etc.)".
Cost elasticity rates of some of the factors one often tries to influence have been summarised.
Figure 3 shows (absolute values of) some of these elasticity rates for a Swedish train operator.
This shows that the elasticity rate for space utilisation is comparatively high – higher than that
of many other factors, which are frequently addressed.
Space utilisation
Vehicle investment
Vehicle kilometres
Staff costs
Average speed (

1.2 A calculation example – space lean seating
In this example, the travellers’ comfort valuation is not included, because the perception of
comfort should be constant. Examples of measures that can make a seat space lean are:
– an intelligent recliner mechanism design
– a thinner seatback
– rounding the seat in front off for your legs
Space lean adjustment
movement

2. Practical SP based methods
Attractiveness can be examined in different ways. Stated preference (SP) methods have been
used for investigating passengers' valuations of certain aspects of trains and train travelling.
In qualitative studies, answers can be obtained to questions like ”what?” and ”how?”, but conflicts
between objectives are more difficult to resolve. One such conflict can be between low
ticket prices and a high quality or standard level. This trade-off is dealt with in Section 4. To
make these trade-offs, it is necessary to quantify the values travellers ascribe to various measures.
To be able to use valuations for planning, it is important to know exactly what is being valued.
It is, for example, not so useful to know that ”price” is more important than ”travelling
time”. What price and what time? Is reducing the price by one Euro, for example, a more important
measure than shortening the travelling time by one hour?
It is not even particularly useful to use a simple grading scale for the measures that are being
examined. Does, for example, four points mean that something is somewhat more important
or twice as important as three points?
Stated Preference methods are usually able to present valuations on a useful scale. Theory and
practice of SP methods are described by Hensher [7, 8] and a number of other researchers. I
will just give an intuitive example.
Table 2 shows the fictitious result of a stated preference experiment. The factors are described
as differences in the level of attributes; attribute swings [6], and the weights are measured on
a relative scale. It can be seen that 1 Euro was given weighting of 0.10. Another difference in
attribute level, in this case 10 cm of legroom, is weighted at 0.05 and must therefore be valued
as 0,50 Euro.
Table 2: Results from a fictitious SP experiment. See text
.
Attribute swing Relative weight Value
10 min shorter journey 0,07 Euro 0,70
1 Euro lower price 0,10 Euro 1
10 cm larger leg room 0,05 Euro 0,50
The important thing is to obtain these weights with satisfactory precision. When choosing
methods, I have tried to find and use methods, which are quick and cost effective to perform
but still reasonably accurate. CASI interviewing and also two "listings" methods will be presented
in this paper.
2.1 Computer assisted self interviewing (CASI)
At the end of the 1980s KTH Traffic Planning conducted their first SP studies. These were
made with cards describing various alternatives. The respondents had to sort the cards.
Around 1990 a 5 kg portable computer were bought and interviewing with pairwise choices
with computer generated alternatives started. The first SP study I organised, in 1992 on the
Swedish east coast main line, were made with this heavy "portable" computer and to make the
electric charge last we had a lead acid battery that we put at the floor by the seat of the pas-
6

senger. The computer was placed so as to allow both the interviewer and the respondent to
view the screen. The respondents who wanted to do so, were invited to do the typing by themselves.
Later in 1992 I found 0,5 kg pocket computers (Poquet) that could run DOS, on the market.
About four of these were used in my next study with train passengers at the local railway Roslagsbanan.
The small computers were handed out to the passengers and they were given a
very short instruction. ("Please read the questions on the screen and press a digit and
when answering.") The battery life for our "Poquet" computers is two weeks, more than for
any competitor.
Later we complemented our interview tools with a number of 1,3 kg HP Omnibook computers.
They have larger and more distinct, but still reflective, screens and no hard disks,
which makes the batteries last 8 hours. Modern backlit laptops often do not have the necessary
battery time and most of them are still to heavy.
One CASI interview normally takes about 10-15 minutes and in trains we do about 100 interviews
per day with 5-8 computers in use. (In February 1999 a KTH student managed to conduct
over 400 interviews in two days!) The answer data files can be saved on PC-cards and
from these cards we copy the results to a master computer. Using a laptop in the field allows
us to make statistics even in the field.
Figure 5: Self completed computer-assisted interview (CASI) with pocket size computers on
board a Swedish train. (MINT interview program) The paper sheet on the table shows alternative
train designs.
Decline is a comparatively small problem when doing CASI interviews in trains. People are
in the train and often welcome the interviewer offering them something to do while travelling.
Regarding refusals from old people and other persons feeling uncomfortable handling the
computer by themselves, they have often been offered assistance from the interviewer. Sometimes
this has been necessary during the full interview, but often a good introduction has relieved
their discomfort.
7

Comparison of stated preference elicitation procedures
In 1992-1993, a series of SP studies at KTH were supervised by Widlert [19] to test different
SP methods. The SP interviews were made on long-distance trains.
One interesting finding in these studies is that the use of very small hand-held computers and
self-interviewing produced results similar to those produced by the method used previously,
when an interviewer was present and a computer with a larger screen was used. Lundin &
Steen [15] reports the following results for the same SP experiment carried out with and without
interviewer assisting (table 3).
Table 3: Results (Parameters, t-values and valuations) for assisted and non-assisted computer interviewing
(CASI).
Attribute
Portable computer with
interviewer assisting the
respondents.
Pocket computers, self assisted
by respondents
(CASI interview)
Cost (%) -0,093 (8,6) -0,088 (8,5)
Time (%) - 0,085 (9,5) -0,074 (9,2)
Frequency (min) -0,0091 (8,6) -0,0079 (7,9)
Modern coach +0,36 (3,2) +0,38 (3,7)
X2000-coach +0,19 (1,8) +0,29 (2,9)
Time value 0,91 % fare /% time 0,85 % fare /% time
Frequency 0,10 % /min 0,09 % /min
Modern coach +3,9 % +4,4 %
X2000 coach +2,0 % +3,3 %
Comparing the parameters of the two models shows no difference of any of the estimates on a
10% significance level. The so-called scale differs a little but the value of time and frequency
is very close. (The randomness seems to be somewhat larger for the CASI interviews, which
can indicate those respondents did their choices a little less consistent.)
Lundin & Steen conclude that the interviewers’ presence at the computer interviews did not
have much influence on the results. They also mention the big time saving there is by handing
out small computers to many passengers at a time instead of sitting next to every person being
interviewed.
2.2 Attribute based "listing" methods
I have tested a simpler approach than the conventional alternative-based SP methods. Basing
the experiment on alternatives composed of attributes and levels is also known as ”conjoint” –
the attributes are joined to form alternatives. In these alternative-based evaluations of attribute
weights, the weights are indirectly evaluated by assessing the alternatives.
8

Attributes
Alternatives
Attribute
weights
A
A
B
composition
decomposition
B
C
C
Figure 6: Alternative-based evaluation using composition and decomposition.
Instead of these alternative-based preference elicitation procedures, the alternative of eliciting
preferences directly for the attribute (levels) could be considered. An attribute-based evaluation
method involves the respondents judging attributes directly and not via assessments of
alternatives.
Attributes
A
Attribute
weights
A
B
valuation - estimation method
B
C
C
Figure 7: Attribute-based evaluation. This figure is very simplified; the valuation estimation
method may comprise several steps.
It can be easier for the experimenter and for the respondents to use an attribute-based elicitation
method. One drawback might be that interactions between attributes would be more difficult
and perhaps impossible to detect. On the other hand, how often is this done in practice?
Listings for ranking
Thurstone [17] previously showed that pairwise choices of attributes – every attribute with
every other – produces an approximately correct valuation of the attribute value if the attribute
values among people are normally distributed.
9

Attributes A B C
A - B versus A C versus A
B A versus B - C versus B
C A versus C B versus C -
Sum No of best A No of best B No of best C
Figure 8: Thurstone's method of comparing all attributes with all other in pairwise choices.
For three attributes, three pairs (choices) are sufficient. Even at only ten attributes, this
method requires (too) many choices (45) if every attribute has to be compared with every
other one once. So ranking the attributes could be considered. If, for example, B receives rank
1 from one respondent, it has been chosen in favour of both A and C and so on.
I have tested this method by simulation in a worksheet. It confirms that Thurstone’s method
approximately reproduces the input attribute values, at least when the variation in taste for the
attributes has intermediate variance.
Valuations from a study using two listing methods
A test of the two versions of ”listing” methods was conducted in a study [10] with doubledecker
trains. The same attributes and levels were used in a computer interview and in a paper
interview. In the computer interview, the respondents were asked to rank the first eight of 11
attributes 3 . The method in the paper interview instead involved marking each attribute as ”important”
or ”not important”. Figure 12 shows that the results from the two listing methods
were similar.
3
The rankings in this method can be converted to pairwise choices between (almost) all attributes and
then counted in accordance with Thurstone’s proposals (see Thurstone references)
10

5% lower ticket price
Seats that are softer
10cm more leg-room
Reclining seatback
5cm wider seat
5% shorter travelling time
Table at my seat
Dimmed lighting +
reading lamps
I can sit face-to-back
Less noise than in this car
More room for luggage
Important-list
Ranking method
I can sit face-to-face
0,0 1,0 2,0 3,0 4,0 5,0 6,0
Figure 12: Valuation results for Uppsala travellers from two methods: an important/less important
listing in paper interviews and an attribute ranking method used in the computer interviews [10].
In this case, the method for estimating the valuations from the important list was simplified to
count the numbers of important marks and assume a linear relationship to the values.
11

3. Evaluating many attributes and circumventing package effects
– a practical two level method
When working with train concepts, we are often interested in more attributes than can be included
in one regular stated preference experiment. In SP experiments with pair wise choices
of alternatives, the number of attributes seldom are more than four to five. This is often to few
when investigating new transport concepts.
3.1 Package effects
It is mostly of interest to improve trains and services using a package of measures and not just
doing one thing at a time.
Package effects are the phenomenon that a specific attribute often receives a lower value
when it is included in a package together with other attributes (packages of measures). The
valuation of the package tends to be lower than the sum of the valuations of the included attributes.
This appears especially to be the case when it comes to secondary attributes.
Monetary values for time, frequency and other primary factors have been acceptably consistent
in the different studies [3]. However, the valuations of secondary factors such as many
comfort and on-board service attributes have been less consistent, particularly when viewed
as components of an improvement package. The value of comfort attributes should perhaps be
reduced by as much as 50% when three to four of them are included in a package.
Methods for circumventing the problem of package effects have been proposed and tested.
Within the framework of our research, two studies have been conducted to test circumventing
methods. Schmidt conducted one study in 1995/96 as a master’s thesis [16] under my supervision
and in 1998 I conducted a second study [12]. These studies produced some interesting
results about packaging effects [13].
3.2 Study of train attributes in packages
In general each of our SP studies has investigated the values of about three to five passenger
train related attributes. Many of the investigated attributes have received values of about 5-
10% of the ticket fare. As indicated above, it has been suspected that some of the attributes
have received to high values. One reason may be the package effect.
Bates [2] has proposed a methodology with attributes and packages of attributes at various
levels to get around the package effect. This study tests a simplified two level value estimation
method which combines binary stated choice of (package) alternatives with choices of
most important attributes inside each package. The aim is to find a practical method where a
rather big number of attributes can be evaluated without very long interviews. It is also the
aim to considerably reduce the package effect. Therefore the aggregated attributes should not
be over-valued but the trade-off between attributes in the same package may be approximate.
Interviews were made on-board X2000 trains between Malmö and Stockholm in March and
April 1998. Almost half of the travellers were business. (Usually the number of business travellers
is somewhere between, say 15 and 25 %.)
12

The design of the interviews
All interviews contained two levels of questions not counting the background variable questions.
At upper level the value of three attribute packages were investigated; a timetable package,
a comfort package and an on-board service package. At the lower level trade-offs between
the attributes in respective package were investigated. Price was included as an attribute
only at the upper level.
It is more important to use a well-established method, as pair wise choices of alternatives, at
the upper, package level than at the lower, "partworth" level. Doing so give some assurance
for not exaggerating the value of the packages and thereby the sum of the values of the included
attributes.
To make the respondents familiar with the various timetable-, comfort- and service- attributes
the lower level was conducted first. When conducting the package SP the respondents were
familiar with the content of the packages.
One objective was to test the so-called best/worst conjoint method for the lower level. The
other two were used as "listings". The way listings were used in this study was as follows;
respondents were asked to pick the best and second best from a list of five or seven attributes.
To supplement this, they were also asked to pick the worst and second worst from the inverse
level of the attributes. The listing method should be and has been modified in later studies, by
allowing respondents to rank all the attributes in a listing. In this way, the method closely correspond
to the method of pair wise choices of attributes proposed by Thurstone [17].
Packages and their levels in the pairwise SP experiments
The pair wise stated preference experiment included four factors. The price factor had 3 levels:
Lowered price (minus 25% minus 10 SEK), Today's price and Raised price (plus 20%
plus 10 SEK). The rest of the "factors", the three packages, had two levels each. The attributes
that were included in a package were such attributes that were expected to have the same
sign for most of the respondents. This applies to for example; shorter travelling time and less
shakings and noise, but not for; train stops at many stations or sitting face-to-face.
The MINT-program from Hague Consulting Group was used to produce the stated preference
pair wise choice experiment.
13

Initial questions
2. Lower level
TIMETABLE
- travel time
- delaysr
- fequency
- fixed timetable
- interchange
- pricing system
- stops often
COMFORT (in train)
- leg-room
- face-to-face/back
- adjustable seats
- shakings and vibr.
- noise
- air conditioning
- lighting
ON-BOARD
SERVICE
- restaurant
- luggage space
- music outlets
- tables
- amount of staff
Listing or B/W
Listing or B/W
Listing or B/W
1. Upper level
PACKAGE SP with pairwise choices
- Ticket price
- Time table package
- Comfort package
- On-board service package
FINAL QUESTIONS
Figure 13: Illustration of the interview-levels and the order in which the questions were put.
Package utilities from SP pairwise choice experiments
As this study is designed with two levels, the first estimation determined the values of the
packages. The next step is to divide these aggregate values into values or "partworths" for
separate attributes. The valuation of the packages is the result of estimations with Alogit
software. Four models are shown here. The parameters, t-values and number of observations
are shown in table 4.
14

Table 4: Estimation results from the pairwise choice experiment for various segments.
All trips Business trips Leisure trips Work trips
Price (%) -0.045 (23.4) -0.041 (15.8) -0.043 (16.5) -0.035 (8.1)
Timetable package 1.65 (16.9) 1.71 (15.2) 1.44 (12.9) 1.42 (7.8)
Comfort package 1.91 (22.1) 1.93 (16.3) 1.64 (14.4) 1.36 (7.7)
Service package 0.96 (13.6) 0.88 (9.5) 0.88 (8.7) 0.54 (3.5)
r2 0.362 0.321 0.315 0.200
No.of observ. 3040 1606 1447 529
The valuations for the timetable, comfort and service packages were all higher than expected.
The differences in parameters for the used segments, business, leisure trips and work+school,
is rather small. The weight for money, for 1% of the ticket price, is very much the same for all
three segments.
Table 5: Valuation (% of fare) of packages in the pair wise choice experiment
Level 1 Level 2 Valuation of level 1 in
comparison to level 2
TIMETABLE-package
Travel time is shorter:
#TIDL#, Double no of trains,
same stroke every hour, no
change, seldom delays, simple
fare system
COMFORT-package
Less shakings and noise, air
conditioning, adjustable
seats, legroom +5cm, reading
lamps
SERVICE-package
Big staff on board,
food/coffee service, bigger
room for luggage, table and
music outlet at every seat
Travel time is longer: #TIDH#,
Same no of trains, change of
train, delay every fifth journey,
today's fare system
More shakings and noise, no
air cond., fixed seats, legroom
-5cm, light tubes in the
roof
Fewer staff on board, somewhat
less room for luggage
+ 37% average for all
+ 41% for business
+ 33% for leisure
+ 41% for work trips
+ 42% average for all
+ 47% business
+ 38% leisure
+ 39% work trips
+ 21% average for all
+ 21% business
+ 21% leisure
+ 16% work trips
Valuation results – willingness-to-pay for various measures
The package values have been split up and the sub-values or partworths allocated to the separate
attributes. The valuations of the attributes are shown below, in table 6. Most of the values
are compared with values from other KTH studies.
15

Table 6: Valuation results from this study with comparisons with other studies.
Attribute swing This study Other KTH studies Comments
Timetable 37% Package value
10% ( +10 min) travelling time 2-7% 6-10%
No change of train instead of one change 8-10% 9-13%
Double train frequency 3-4% 4%
Delays are rare instead of one 15-min delay 10-11% 16%
every fifth journey
Regular departure times (One train every hour at 2-4% 1-3%
the same time instead of an irregular service)
Simpler fare but fewer discount prices instead of 0-3% –
today's fare
The train stops at few stations instead of many 3-7% –
stations
Comfort 42% Package value
10cm legroom (Your legroom is 5 cm larger instead
8% 6%
of 5 cm smaller)
More instead of somewhat less shaking and 8-10% 11%
vibration than in this train
Somewhat more instead of some-what less noise 5% 8%
than in this train
Air conditioning (Air conditioning with cool air in 5-7% ≈10%
the summer time instead of no air conditioning,
but windows that open)
Face-to-back instead of face-to-face seating 1-3% 0-2%
Your seatback is adjustable instead of not adjustable
8-12% 6-11%
Dimmed lighting and reading lamps instead of full 2-4% 4/ 11%
fluorescent lighting
On-board services 21% Package value
There is catering onboard 8% 5-11%
Music/radio outlets at your seat 1% 2- 5%
There is always staff to help instead of somewhat 3%
less staff than today
More instead of less luggage space in the coach 3% 1%
There is a table at your seat 5-7% 2%
The valuation levels are surprisingly similar to those achieved in other KTH-studies in which
secondary comfort and service attributes were estimated as separate attributes. Some of the
differences in valuation levels may be caused by different attribute descriptions. As a whole
the methodology in the package study has worked sufficiently well, with interesting support
for two things:
– The somewhat high valuation levels for comfort and service attributes obtained in other
studies are generally supported.
– It should be possible to use simplified methods similar to this one in order to assess the
relative values or partworths of a great number of secondary attributes.
16

4. A simplistic method to judge the profitability of various measures
Comparisons of costs and benefits can be made at various levels. Figure 15 illustrates relations
between a technical measure, its cost and the corresponding attribute(s) and passengers'
valuation of this. The increased (or decreased) value may lead to an increased (or decreased)
number of passengers, which in turn may call for a change in the produced supply and thereby
increasing (or decreasing) costs. Another way is that the operator changes the ticket prices in
direct relation to the changed value or willingness-to-pay (wtp). In this case the number of
passengers should not be altered.
Techn. measure
Cost
Attribute(s)
Passengers'
evaluation
Cost/
valuation
Utility - valuation -
willingness-to-pay
Increased costs
Increased
number of
passengers
Cost/
benefit
Increased
ticket prices
Figure 15: Relationships between measures, costs, willingness-to-pay, increased travelling and benefits.
In this case a restricted wtp/cost comparison is made, not a full cost/benefit analysis.
A simpler "cost/benefit" comparison is made here. The passengers' valuation of a measure is
compared with its cost. The cost/valuation model in figure 15 illustrates this.
A full cost/benefit appraisal is replaced by a cost/valuation appraisal, where valuation stands
for the passengers' valuations of various measures, estimated using stated preference methodology.
The term valuation is chosen instead of willingness-to-pay because the latter may imply
that there is an immediate opportunity to raise ticket prices and because willingness-toaccept
is sometimes more relevant.
4.1 Method for comparing marginal costs and valuations
Using an economic model, e.g. the model described in section 1, it is possible to calculate cost
changes in per cent. They can be compared with revenue changes in per cent. The prerequisite
for this comparison to be fair is that the traffic just about breaks even, that is to say costs
are covered 100%.
17

There are many opportunities for cost and revenue changes that are of interest. They include
combinations of increased and/or decreased cost and revenue. The most interesting measures
are of course those measures, which both decrease costs and increase revenues.
4.2 Summary of values and costs of various measures
The various types of measure place themselves approximately in the manner shown below in
figure 16, showing marginal values and costs. Note that the positions in figure 16 are not indisputable.
Firstly there are uncertainties in the calculations and secondly the costs may be
affected by choices of technology. Thirdly the value sets diverge in different sub markets.
COST
c/v = 1
Solely
compartments
+15 %
+10
1+2 seating
Catering
at seat
Café
Solely face-to-face
or non-facing seats
1+3 seating
-15% -10
Legroom
-5
+-10 cm
2+3 seating
in wide trains
+5
-5
-10
-15%
Reclining
Play
seat backs
Video area
AC
Music
outlets
Ind. ventilation
VALUATION
Table
+5 Reading +10 +15 %
Space lights
lean seats
-20% time
Double decks
Travel time
To be recommended
Probably profitable
Prof. in some cases
Probably unprofitable
To be avoided
Figure 16: Ball-park marks for valuation (willingness to pay) and effect on the train service cost of
different factors or undertakings.
18

One interpretation of the figure is that the principle objective should be higher speeds, doubledecker
or wide trains and items addressing comfort.
Higher speed is placed below to the right, i.e. the costs decrease as the value increases. A 20%
travelling time reduction increases willingness-to-pay by 10-20%, dependent on journey purpose
(or even more for certain business travellers). At the same time traffic service costs are
calculated to decrease by 3-5%.
As a rule, comfort and service measures are profitable, i.e. the cost rise is lower than the marginal
value.
A higher usage of space brings the costs down and appropriately done the value decreases less
than the costs. Passengers appear to value double decks in a slightly positive manner while
three-plus-two seating, which is required for the effective use of wide bodies, appears to be
slightly negative. The cost savings are of the same magnitude.
A 10% increase in seating density, corresponding to 10cm per seat at the original seating distance
of one metre, would cut costs by about 5%. This is a significant cost reduction. However,
on the other hand, the valuation of legroom is of the same magnitude.
A seemingly trivial measure, seats with space lean design, may save as much as 5 % of the
total train service costs without reducing comfort. A somewhat more expensive seat is well
compensated if the seat density can be increased about 10% with retained comfort experience,
as shown in the example in section 2.
Compartments with fewer than four seats side-by-side increase costs substantially. A few
Swedish passengers do prefer compartments, but most prefer open saloons and the average
valuation of compartments is negative. The conclusion is that compartments should be
avoided as a general form of interior.
The majority of the passengers display substantial willingness-to-pay for sitting face-to-back
(most) or face-to-face (fewer). So interiors with only one of these furnishing alternatives is a
comfort reducing mistake.
There are many other comfort and on-board service measures that should be profitable, such
as good ventilation (AC or individual air outlets), a table at your seat, reading lights and music
outlets.
There are also many measures regarding train size and traffic service concepts that have not
been analysed yet.
19

5. Summary and outlook
Models which can help design passenger rail systems with greater competitiveness and financial
strength have been developed. These include cost and valuation models/methods.
Discussion of the cost calculations
The cost model that has been developed is fairly simple. It is useful for identifying the importance
of various measures to reduce costs. The cost elasticity concept is so simple that people
may wonder why it is not used on a wider scale. The cost model has so far been used for
Swedish train services but other cost data can of course be inputted.
The idea that trains can be spacious because space is cheap in trains is contradicted by the
high cost elasticity of space utilisation. This make it important to investigate the passengers'
willingness to pay for space and comfort measures.
Discussion of computer aided interviewing
Computer aided interviewing is efficient. Performing a small-scale SP study (say 200 respondents)
with the help of computers takes just a few weeks, including the analysis. In paper interviews,
the time for coding and manually setting up assessment models is added.
The rate of useful answers is very high in computer interviews. Almost the only reason for
missing data is when people stop answering. In paper questionnaires on the other hand, the
number of missing answers can be fairly high.
We can now look forward to new types of inexpensive, lightweight computers. They should
have graphical screens, a long battery life, touch screens and an operating system that can run
modern CASI interview programs. Regrettably no such modern interviewing system has yet
been found, in spite of extensive search on the market in the spring 2003. (On the other hand,
our DOS based pocket computers together with MINT interviewing program still work.)
Transport modelling and valuation methods
Transport models are often made for travel forecasting purposes. In my work, forecasting has
not been in focus. Instead, models which can help design passenger rail systems with greater
competitiveness and financial strength have been developed. For this work, tools that are the
equivalent of or similar to those used by traffic planners have been employed.
Discussion of the attribute-based methods
Attribute-based methods were introduced as a way to simplify interviewing when we were
interested in a large number of attributes (10-30). The methods are perhaps not as effectively
underpinned in theoretical terms as the alternative-based SP methods. I do not recommend
that an attribute-based SP experiment be used alone in a study. Instead, it can be seen as a
complement to traditional SP or combined experiments, as it was in the package study presented
here.
20

Willingness to pay or to accept?
Using terms such as value, valuation or even willingness-to-pay (wtp) does not always mean
that passengers are actually willing to pay higher ticket prices if the measures are implemented.
In Swedish trains, where my studies have been conducted, many of the investigated
attributes have already been implemented at the ”high” level. For example, most of the trains
already have fairly comfortable seats with reclining seatbacks, 20 cm of legroom, a mixture of
face-to-face and face-to-back seating, tables and individual reading lamps and a café. So the
valuations in many cases should be understood as the negative value if the standard is reduced.
Validation of the valuation results
As in other research, it is desirable to validate the results. The problem as I see it is that, at
least when it comes to the valuations, there is no key, no true answers. However, there are
other studies to compare with and it is possible to obtain some idea of the quality.
The main problem that has to be investigated further is the probable over-estimation of many
comfort and on-board service attributes; so called secondary attributes. A study that really
underlines this problem is the one by Wardman [18]. According to this, the over-estimation
by most studies is very high.
The package study (see chapter 3) was designed to reduce the anticipated over-valuation normally
obtained when separate attributes are used in traditional SP experiments. However, the
result indicates almost the same, high values for train-related attributes as in earlier studies
with separate attributes. This result is good, but it contradicts the findings in other studies [2,
9, 16, 20] relating to powerful packaging effects.
Is there any example of a train service where the conditions have been improved and the
payment increased? The answer is that there are some positive indications; when the X2000
was introduced, ticket prices were raised by about as much as the value placed on the reduced
travelling time and improvement in comfort. Even so, the number of passengers increased.
Another example is the impressive increase in travel on trains in Blekinge. The luxury Danish
IC/3, marketed under the name of ”Kustpilen”, replaced simple rail-bus services in 1991. The
number of travellers increased by almost 200% in the first four years [14]. The timetable improvements
and price decrease alone cannot explain this increase. It was concluded that a
large part of the increase was due to the enhanced comfort and on-board service. The latest
Swedish example is the success by Svealandsbanan with its fast and comfortable trains [5].
Travelling has increased by about 600 %.
Conclusion
The combination of a simplistic model for the identification of cost drivers, practical methods
for the investigation of travellers' valuations and a percentage based cost benefit analysis has
produced very valuable results for the railway industry (including operators, vehicle manufacturers
and society).
Using these methods in other railways will probably show that regional rail commuters have
higher valuation of comfort than many of today’s trains fulfil. People who travel by train over
21

200 days per year are willing to pay for high train standard. It is also, in many cases, beneficial
even for the operator to run the trains at higher speeds.
Acknowledgements
I am thankful to Dr. Staffan Widlert who introduced me into the world of stated preference
methods in the 1990:s. Dr. Christer Lindh is a good discussion partner and Professor Bo-
Lennart Nelldal is our inspiring leader at the group for rail planning research at KTH.
22

References
[1] Banverket, Beräkningshandledning, (Calculation manual, in Swedish), BVH 106
[2] Bates J., Use of Stated Preference Methods to determine London Underground Passenger
Priorities
[3] Bates, J., Reflections on Stated Preference Theory and Practice, unpublished paper from John
Bates Services,
[4] Baumgartner, J.P., Orders of magnitude of costs in the railway sector, EPFL Ecole Polytechnique
Fédérale de Lausanne, Institut des Transports et de Planification, 1996, Report for
UNIFE
[5] Fröidh, O., Introduction of reginal high speed trains, PhD thesis, Div. of Transportation &
Logistics, KTH, TRITA-INFRA 03-041, ISRN KTH/INFRA--03/041--SE, 2003
[6] Gissel, Stine, Decision aid methods for railway infrastructure pallning, PhD thesis, Report 4
1999, Dep. of Planning, Technical University of Denmark
[7] Hensher, D., Stated preference analysis of travel choices: the state of practice, Transportation
vol 21, No2 May 1994, pp 107-134
[8] Hensher, D., Louviere, J., Swait, D., Stated Choice Methods; Analysis and Application,
Cambridge University Press, 2000
[9] Jones, P., Addressing the Packaging Problem in Stated Preference Studies, European Transport
Forum (PTRC), September 1997
[10]Kottenhoff, K, Utvärdering av dubbeldäckat tåg "Competence" (in Swedish), Traffic & Transport
Planning, KTH, June 1999
[11]Kottenhoff, K., Evaluation of Passenger Train Concepts – Methods methods and results of
measuring preferences in relation to costs, KTH Traffic & Transport Planning, PhD thesis,
TRITA-IP FR 99-48, ISRN KTH/IP/FR--99/48--SE
[12]Kottenhoff, K, Passenger preferences in terms of timetable, comfort and on-board service
attributes - evaluated using a stated preference method with reduced package effects, KTH
Traffic Planning, TRITA-IP AR 00-87, ISRN KTH/IP/AR--00/87--SE, 2000
[13]Kottenhoff, K., Swedish studies of reasons and scale of the package effects in stated preference
studies, Eupropean Transport Forum (PTRC), 2000
[14]Lindh, C., Kottenhoff, K., The value and effects of introducing high standard train and bus
concepts in Blekinge, Sweden, Transport Policy, 1995 Vol.2, No.4
23

[15]Lundin, A., Steen, M., Metodstudier i stated preferences. SP-metodens känslighet för olika
utformning av datainsamling, (Methodological studies in Stated Preferences. The sensitivity to
various ways of collecting SP data, in Swedish), KTH Traffic & Transport Planning, Master
thesis No.1993:6, TRITA-TPL-93-11-95, 1993
[16]Schmidt, L., Värdeminskning vid värdering av tågkoncept (Value reduction when evaluating
train concepts – a study of comfort, service and timetable packages with SP methodology, in
Swedish), KTH Traffic Planning, Master Thesis No. 1996:44, TRITA AR 96-44, 1996
[17]Thurstone, L.L., The measurement of values, The University of Chicago Press, 1959
[18]Wardman, M., Whelan, G., Rolling stock quality improvements and user willingness to pay,
ITS working paper 523, Leeds, March 1998
[19]Widlert, Staffan. Stated Preference Studies – The Designs Affects the Results, 7th int. congress
on travel behaviour, Santiago Chile, 1994
[20]Widlert, S., Gärling, T., Uhlin, S., Värdering av kollektivtrafikens standard, (Valuation of the
standard of public transport, in Swedish), TFB Report 1989:2, ISBN 91-87246-39-2, June
1989
24