You are on page 1of 16

STUDY OF A I R P O R T C H O I C E AND A I R P O R T ACCESS M O D E CHOICE

IN S O U T H E R N G E R M A N Y

Lothar Bondzio
Ruhr-University Bochum
D-44780 Bochum, Germany

1. INTRODUCTION
?

A trip results from a set of choices made by a traveller. Figure 1 illustrates one possible
choice hierarchy. First of all a decision is taken to travel to a specific destination. Then
the traveller decides which mode to use for the main part of his journey. If he decides
to travel by air, the next step is perhaps to choose the departure airport for his trip.
Afterwards he decides on the arrangements of the flight, such as airline or departing
time, as well as on the arrangements of the access to the airport. Within the airport
access the access mode choice plays an important part. This hierarchy is of course only
one possible hierarchy. Alternatively, it can be assumed that a traveller, first of all,
chooses an access mode and then decides on the departing airport which can be
reached most easily with the chosen mode.

Within this choice hierarchy the study focuses on the decisions

• airport choice and


• airport access mode choice.

Both choices depend on the quality of access to the airport. They are not independent
from each other. It can be assumed that travellers consider the quality of access as a
characteristic of a particular airport. Therefore, the quality, of access may influence the
demand for air travel. It is in the best interest both of airports and of airlines to obtain
information about the access behaviour of travellers. Sufficient information about
access behaviour is vital for the successful introduction of measures t o improve the
access to an airport and also to improve the attractiveness of an airport.

As part of the study a survey was carried out at the four German airports Frankfurt,
Munich, Stuttgart and Nuremberg collecting data from more than 2200 departing
passengers. The first goal was to gather information about the choice behaviour of
departing passengers. The second goal was to develop models that are based on the
observed behavioui" and are able to forecast the effect of changes in the quality of
airport access on the travel behaviour. As a result, a set o f different random-utility
models was developed.

The major sections of this paper summarize the theoretical framework,, describe the
data collection, present model estimation results and discuss the forecasting ability of
the models.
2. THEORETICAL FRAMEWORK

Random-utility models are widely used to model travel demand. These kinds of models
are based on the assumption that a decision-maker is faced with a set of available
choice alternatives. The decision-maker assigns a special utility to each alternative.
Then he decides on the alternative which provides the highest utility from all available
alternatives.

The utility U is a combination of two different components

• a deterministic part of the utility V and


• a random part of the utility s.

For convenience the random part ist mostly treated as an additional term to the
deterministic part of the utility. The deterministic term V consists of observed
attributes of the alternatives as well as of socioeconomic characteristics of the
individuals X i. For convenience the deterministic term V is assumed to have a linear in
parameters form:

K
Vit = B 0 + ~ Bk" Xit k (1)
k=l

where: Vit = deterministic part of the utility associated with individual t and
alternative i
Xit k = vector of attributes and socioeconomic characteristics
Bk = parameters

The random component s comprises all disturbances which cannot be described


precisely. It considers unobserved attributes as well as taste variations and
measurement errors. It can be assumed that the random component is distributed
according to a special distribution function.

On the assumption that the random components s are distributed Gumbel-Type I as


well as independent across the alternatives and identically, the resulting model is the
multinomial logit model. The model can be shown in the following form:

eVlt
Pit = j V-
(2)
'~-'~e jt
'j=l

The multinomial logit model is mostly used in cases of single-dimensional choice: For
multi-dimensional choice situations, such as mode-destination choice, a nested
structure can be identified. Consider for example a two-dimensional choice between
different airports and different airport access modes. On the assumption that the
departure airport choice is a choice on a higher level than the access mode choice, the
utility function that a individual t will choose the modej und the airport k is given as:
Ujkt = Vjt + Vkt + Vjkt + ekt + 8jkt (3)

where: V-t = the deterministic component of the utility specific to access mode j
X~Jl~t = the deterministic component of the utility specific to airport k
Vjk t = the remaining deterministic component of the utility specific to
the combination j-k
ekt = the random component of the utility specific to airport k
ejkt = the random component of the utility specific to the combinationj-k

This formulation implies that there is no random component exclusively associated


with access mode j, but all error is associated with the mode-airport combination. This
assumption is necessary to use the so-called nested logit model.

The probability that an individual t will choose the access mode j conditional on airport
k can be written as:

e(Vj+Vjk)/0
(4)
Pt(J / k ) - )_.~e(X~+,~k)/0

J
The parameter 0 represents a scale parameter and is bounded by zero and one.

The marginal probability that the individual t will choose airport k can be represented
in the upper nest of the nested logit model by the formula:

e(Vk+rk'0)
P t ( k ) = ~ e(Vk+Fk.O)
(5)
k

where: Fk = expected maximum utility of all access modes j to airport k


= in ~'~ e(Vj+Vjk 1/0 (6 /
J
The probability that an individual t will choose access mode j and departure airport k
can be represented by the product of the conditional and marginal probability (eq. 4
and 5):

Pt(J, k ) = P t ( k ) " Pt(J / k) (7)

3. D A T A COLLECTION

The investigation was performed in Southern Germany with the four international
airports of Frankfurt, Munich, Stuttgart and Nuremberg. There are some smaller
airports within the region which have only minor importance for the air travel demand.
From the airports in the neighbouring countries only Zurich influences the air traffic
demand in Southern Germany noticeably. The location of the four airports is shown in
fig. 2.

Fig. 3 shows the importance of the four airports. Frankfurt, Munich and Stuttgart are
amongst the six biggest airports in Germany. Fig. 4 illustrates that public transport has
very different market shares at the four airports. Whereas in Munich about 36 % of all
departing passengers use public transport at the airport of Nuremberg only 3 % use
this mode.

The modelling of travel behaviour requires a certain knowledge about the


socioeconomic status of the traveller as well as the circumstances of the journey. The
required data can only be ascertained by passenger surveys. Since in Germany no
appropriate data are available it was necessary to carry out a passenger survey at the
four airports under investigation.

As part of the survey more than 2200 passengers were interviewed in the departure
lounges of the airports of Frankfurt, Munich, Stuttgart and Nuremberg. The
information that was collected included trip-related information such as trip purpose
and destination as well as socioeconomic data about the individual.

4. AIRPORT CHOICE MODEL

First of all, an airport choice model was estimated. During the estimation process a
number of different specifications have been tested. The estimation results show that a
simple logit model based on the two variables access time to the airport and flight
frequency to the chosen destination provides a good approximation of airport choice
behaviour in Southern Germany. The value of Rho-squared against a model which
consists only of alternativ-specific constants pZ (c) of more than 0.9 is proof of a
superior quality of the estimated model.

It can be assumed that the airport choice behaviour varies between different groups of
passengers. Therefore, the next step was to divide the sample into the following four
groups:

• Business travellers on an outward flight


• Business travellers on a return flight
• Non-business travellers on an outward flight
• Non-business travellers on a return flight

For each group a separate model was estimated. Since the quality of the submodels
was only slightly better it is advisable to apply the overall model.

5. AIRPORT ACCESS MODE CHOICE MODEL

Four basic types of airport access mode were defined:

• Drive: The travelling party on an outward flight drives a vehicle to the


airport and leaves it parked in a lot for the duration of the trip.
The travelling party on a return flight drives a rental car and
returns it at the airport.
• Dropoff: A family member, friend or associate drives the passenger to the
airport and then removes the vehicle from the airport vicinity.
• Taxi: The passenger takes a taxi or similar personalized service to the
airport.
• Public Transport: The passengers uses public transport to get to the airport.

The data analysis shows that a single model for all travellers was not appropriate. As a
consequence, a market segmentation into the four subgroups mentioned above was
carried out. Due to the small sample size the estimation of a model for non-business
travellers on a return flight was not possible.

A number of different specifications has been tested. The major results can be
summarized as follows:

• The airport access mode choice can be modelled with multinomial logit models as
well as with nested logit models. The values of Rho-squared against a model with
alternativ-speciflc constants is only between 0.18 and 0.23.
• The access behaviour differs significantly between business and non-business
travellers as well as between travellers on outward and return flights.
• Access travel time and access travel costs are strong explanatory variables in
airport access mode choice. Business travellers are more time-sensitive and less
sensitive to cost than non-business travellers. Passengers on a return flight are less
sensitive to cost than passengers on an outward flight.
• The household income plays an important part. Travellers with higher income are
less cost-sensitive. They prefer more expensive access modes like taxi or drive.
• The attractiveness of the modes dropoffand taxi decreases with increasing distance
between the starting point of a journey and the departing airport.
• Business travellers on a single-day trip are more likely to use the alternative to
drive themselves.
• A rail rapid transit connection increases the acceptance of public transport. For
non-business travellers the number o f changes also influences the attractiveness of
public transport. The carrying of heavy luggage reduces the attractiveness of public
transport.

6. OVERALL MODEL FOR AIRPORT CHOICE AND AIRPORT ACCESS


MODE CHOICE

Subsequently, overall models for both choices have been developed. For the three
subgroups o f passengers different multinomial logit models as well as nested logit
models have been estimated. The key parameter to judge the correct model type is the
estimator of the scale parameter of the Gumbel distribution for the nested alternatives
0 which must be in the range of 0 to 1. This diagnosis condition can be expressed as
follows:

0<0<1 (8)
If the parameter 0 is within this range and significantly differs from zero, a nested logit
model will be the appropriate model structure. In cases where 0 is 1 or near 1 a
multinomial logit model will be more efficient.

Table 1 shows the results of the different models for business travellers on outward
flights. It becomes apparent that a nested logit model with the choice hierarchy mode
choice -airport choice (Model GH-NL2) is the best model. The parameter 0 is within
the range and significantly different from 1. This choice hierarchy is illustrated in fig. 5.

Business travellers on outward flights

Model GH-MNL GH-NL1 GH-NL2

Type multinomial logit nested logit model nested logit model


model

Choice hierarchy - airport choice access mode choice

I
access mode choice
I
airport choice

0 1.442 0.6742

io2 (c) 0.5635 0.5656 0.5656

Table 1: Models for business travellers on outwards flights.

These results correspond to the estimation results for the business travellers on return
flights. The model GR-NL2 with the choice hierarchy mode choice - airport choice
yields the best results. A comparison between the different models is shown in table 2.

Business travellers on return flights

Model GR-MNL GR-NL 1 GR-NL2

Type multinomial logit nested logit model nested logit model


model

Choice hierarchy airport choice access mode choice

I
access mode choice
I
airport choice

0 1.639 0.6428

(c) 0.5648 0.5675 0.5667

Table 2: Models for business travellers on return flights.


The estimation process gives different results for the non-business travellers on
outward flights. As illustrated in table 3 the parameter 0 for the model PH-NL2 with a
choice hierarchy mode choice - airport choice is near 1. Therefore, in this context a
multinomial logit model could also be an appropriate model. This model structure is
illustrated in fig. 6.

Non-business travellers on outward flights

Model PH-MNL PH-NL 1 PH-NL2

Type multinomial logit nested logit model nested logit model


model

Choice hierarchy airport choice access mode choice

I
access mode choice
I
airport choice

0 1.368 0.9655

9 z (c) 0.6008 0.6024 0.6008

Table 3: Models for non-business travellers on outward flights.

7. M O D E L TRANSFERENCE

Finally, the model transference across space was tested. The test region was the region
of Westfaiia with the two airports of Dortmund and Munster. The location of these
airports is shown in fig. 7. The situation in Westfaiia is very different from the situation
in Southern Germany. Both airpOrtS are only of minor importance for the air travel
demand in Westfaiia. The market share of public transport to both airports is only
about 1.5 %.

Due to the different spatial conditions in the test region the models could not be
transferred perfectly. Therefore, different updating procedures were carried out using
data from a survey of about 500 departing passengers at both airports. By means of
updating the aiternativ-specifie constants and the scale of the parameters a satisfactory
model transference could be ensured.

8. CONCLUSION

This paper has shown that it is possible to estimate random-utility models for the
airport choice as well as for the airport access mode choice of departing passengers. It
is demonstrated that both choices are not independent from each other but are linked
together. Therefore, multinomiai legit models as well as nested legit models have been
estimated. The models were calibrated in Southern Germany. A test of model
transference showed that by using special updating procedures the models can be
transferred across space.
The models can be used to assess the effects of airport access policies. A very simple
approach is to use logit elasticities to address specific questions about well-defined
markets. Usually aggregate forecasts are necessary in order to deal with detailed
policies. While a disggregate model is able to estimate only individual choice
probabilities, an aggregation procedure is necessary to be able to use the estimated
models in practice.

REFERENCES

Ashford, N.; Benchemam, M. (1987):


Passengers" Choice of Airport: An Application of the Multinomial Logit
Model. Transportation Research Record 1147

Ben-Akiva, M.; Lerman, S.R. (1985):


Discrete Choice Analysis. Cambridge, Massachusetts

Brilon, W.; Bondzio, L.(1996):


Entwickhng von disaggregierten Planungsmodellen ~ r die Bodenanbindung
von Flugh~en. (Disaggregate Models for Airport Choice and Airport Access
Mode Choice). Ruhr-University Bochum

Harvey, G. (1986):
Study of Airport Access Mode Choice. Journal of Transportation Engineering
112

Harvey, G (1987):
'Airport Choice in a Multiple Airport Region. Transportation Research A,
Vol. 21A, No. 6

Koppelman, F.S.; Kuah, G.; Wilmot, C.G. (1985):


Transfer Model Updating with Disaggregate Data. Transportation Research
Record 1037
Whether to travel

I
By air or other mode

I
Departure Airport

Air Travel Arrangements


I
Airport Access Arrangements
- Carrier - Access M o d e
- Fare Class - Departure Time

Fig. 1: A possible choice hierarchy for air travel.


l Cologne

Frankfurt

Nuremberg
Dresden

Strasbourg Q
Stuffgart
Munich
© Salzburg
Basle/Mulhouse

Zunch

Fig. 2: Airports in Southern Germany and in the neighbouring countries.


35

¢,,- 30
.__o.
V

~20
¢,-

n ~ 12.5

NN
i p I ,L~ , ~ , ~,~
. FRA MUC TXL CGN NUI= LEJ BRE FMO
DUS HAM $TR HAJ SXF DRS THF SCN

Fig. 3: Terminal passengers on commercial flights at German airports in 1993


40

.~ 30
,.E:
cO

°20
e--

F-
0
- 1o
-",1

0
MUC THF STR CGN BRE SCN
FRA TXL DUS HAM HAJ NUE

Fig. 4: Market share of public transpoi't at 'German airports.


MODEL GH - NL 2

Drive Dropoff Taxi Public Transport

MUC STR FRA NUE MUC $TR FRA NUE MUC 8TR FRA NUE MUC STR FRA NUE

Fig. 5: Model structure of an nested logit model for business travellers on an


outward flight.
MODEL PH - MNL

MUC MUC MUC MUC STR STR STR STR FRA FRA FRA FRA NUE NUE NUE NUE
drive dropoff taxi pt drive dropoff taxi pt drive dropoff taxi pt drive dropoff taxi pt

Fig. 6: Model structure of a multinomial logit model for non-business travellers on


outward flights.
)

©
Netherlands Bremen

Hanover

~~l~Munste r

Paderbom

Do~mund

Dusseldorf

Cologne

(
. Fig. 7: Airports in Westfalia and in neighbouring regions.
Biographical Details

Lothar Bondzio:
present position: Research Assistant at the Institute for Transportation Planning and
Traffic Engineering at the Ruhr-Um'versityBoehum (since 1993)
Diplome in Geography, Ruhr-University, 1993
PhD in Civil Engineering, Ruhr-University, 1996

You might also like