Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

ALTERNATIVE NESTED LOGIT MODELS: STRUCTURE, PROPERTIES

AND ESTIMATION
FRANK S. KOPPELMAN* and CHIEH-HUA WEN
Department of Civil Engineering and Transportation Center, Northwestern University, 2415 Sheridan Road, Evanston,
Illinois 60208, U.S.A.
(Received 19 March 1997; in revised form 30 October 1997)
AbstractTwo distinctly dierent nested logit models have been widely used in both research and applica-
tions. The dierences, not widely recognized, between these models will substantially inuence estimation
results, behavioral interpretation and policy analysis. The McFadden nested logit model is derived from
random utility theory; the Daly or non-normalized nested logit model is based on probability relationships
and is not consistent with utility maximization. This paper describes and compares the model structure and
properties of these dierent nested logit models identifying important dierences between the dierent model
structures. An empirical application demonstrates that the dierent nested logit models produce dramatically
dierent results with respect to nesting structure and the relative importance of utility components. Thus, the
selection of one or another of the nested logit models has important consequences for model interpretation
and prediction with consequent impacts on policy analysis. The authors prefer the McFadden model because
of its basis in utility theory, intuitively reasonable elasticity relationships and a clear interpretation of utility
function parameters across alternatives. # 1998 Elsevier Science Ltd. All rights reserved
Keywords: nested logit, discrete choice, travel demand, intercity rail
1. INTRODUCTION
Discrete choice analysis is used to model the choice of one among a set of mutually exclusive
alternatives. The multinomial logit (MNL) model (McFadden, 1973), the most widely used dis-
crete choice model, is based on principles of utility maximization and has the advantages of simple
mathematical structure and ease of estimation. However, it has the property that the relative
probabilities of each pair of alternatives are independent of the presence or characteristics of all
other alternatives. This property, known as the independence of irrelevant alternatives (IIA),
implies that the introduction or improvement of any alternative will have the same proportional
impact on the probability of each other alternative. This representation of choice behavior will
result in biased estimates and incorrect predictions in cases which violate the IIA property.
Extreme examples of non-IIA cases include the addition of a second bus service identical to an
existing service, except for a change in vehicle color, to existing competition between a single bus
alternative and private rail (the red bus, blue bus problem) or the addition of a classical CD to an
existing choice between another classical CD and a jazz CD. Less extreme, but more realistic
examples include mode choice between public transit (bus and light rail) and private vehicle (drive
alone and shared ride) (Ortuzar, 1983) or route choice among alternatives which include common
road segments (Yai et al., 1997).
The most widely known relaxation of the MNL model is the nested (NL) or hierarchical (HL)
logit model which allows interdependence between the pairs of alternatives in a common group
(McFadden, 1978; Ben-Akiva and Lerman, 1985; Bo rsch-Supan, 1990a). The utility maximizing
nested logit (UMNL) model is a special case of the generalized extreme value (GEV) model
(McFadden, 1978, 1981) which ensures that it is consistent with utility maximization, provided
that the logsum parameters are bounded appropriately. An alternative, is the non-normalized
nested logit (NNNL) by Daly (1987), which is identical to the McFadden model except that the
model does not include the inverse of each logsum parameter in the utility function of the
Transpn Res. B, Vol. 32, No. 5, pp. 289298, 1998
# 1998 Elsevier Science Ltd. All rights reserved
Pergamon Printed in Great Britain
PII: S0191-2615(98)000803-4
0191-2615/98 $19.00+0.00
289
*Author for correspondence. Fax: (847) 491-4011; e-mail: f-koppelman@nwu.edu
alternatives in the corresponding nest. This apparently small dierence has a dramatic impact on
the properties and behavior of this model. Most importantly, the NNNL model is not consistent
with utility maximization.* Estimation and application of these dierent NL models will, in gen-
eral, result in dierent utility function parameters, dierent nesting structures and dierent travel
forecasts; these dierences may be substantial enough to impact capital investment and operating
decisions.
This paper compares these alternative NL models to identify their important dierences and
presents an empirical example which demonstrates the extent of these dierences which are not
widely recognized by users of NL models. The remainder of this paper is organized as follows.
Section 2 presents the formulation and properties for the alternative NL models. Section 3
describes the estimation structure for both NL models. Section 4 discusses the empirical example
and demonstrates the magnitude of the empirical dierences in one case. Section 5 provides a
summary and implications for model application.
2. ALTERNATIVE NL MODELS
2.1. Model formulation
2.1.1. The utility maximizing nested logit (UMNL) model. McFadden's UMNL model is a
member of the GEV family, and is consistent with utility maximization (McFadden, 1978). A
GEV model can be derived from any function, G(Y
1
; Y
2
; . . . ; Y
n
), Y
1
, Y
2
; . . . ; Y
n
50, which is a
non-negative, linear homogeneous function
,
which approaches innity as any Y
i
, i = 1, 2; . . . ; n
does and has mth cross-partial derivatives which are non-negative for odd m and non-positive for
even m. If G satises these conditions, the probability of choosing alternative i is given by
P
i

Y
i
G
i
Y
1
; Y
2
; . . . ; Y
n

GY
1
; Y
2
; . . . ; Y
n

1
where G
i
is the rst derivative of G with respect to Y
i
. A two-level UMNL model is obtained from
the G function:
GY
1
; Y
2
; . . . ; Y
n

X
M
m1
X
n2N
m
Y
1
m
n
!

m
2
where N
m
is the set of alternatives in nest m and
m
is an index of the dissimilarity of alternatives
included in nest m.
{
The UMNL model is consistent with utility maximization if the conditions,
0 <
m
41,

are satised for all


m
. If
m
= 1 for all m, the NL collapses to the MNL.
To ensure positive Y
i
, the transformation Y
i
= exp(V
i
) is adopted, where V
i
represents the
observable components of the utility for each alternative (U
i
= V
i
"
i
, i = 1; 2; . . . ; n). The prob-
ability that alternative n is chosen can be written as
P
n
P
n=m
P
m
3
where
P
n=m

e
V
n
=
m
P
n
0
2N
m
e
V
n
0 =
m
4
*Except for the special case when the logsum parameters are equal across all nests at the same level.
,
McFadden's original formulation requires the function to be linear homogeneous of degree one; however, Ben-Akiva and
Lerman (1985) refer to an unpublished report by Ben-Akiva and Francois (1983) which demonstrates that the function
can be homogeneous of any degree.
{
Dissimilarity implies the absence of correlation between the error distribution of pairs of alternatives within a nest. The
correlation between pairs of alternatives equals (1
2
m
).

Values of logsum parameters greater than one can be consistent with utility maximization over a limited portion of the
real space which may be relevant in cases where the range of attribute variables is limited during both estimation and
prediction (Bo rsch-Supan 1990a; Kling and Herriges, 1995; Herriges and Kling, 1996).
290 Frank S. Koppelman and Chieh-Hua Wen
P
m

e

m
P
M
m
0
1
e

m
0 =
m
0
5
and

m
ln
X
n
0
2N
m
e
V
n
0 =
m
6
V
n
is the utility associated with alternative n in nest m, P
n=m
is the conditional probability of
choosing alternative n conditional on choosing nest m, P
m
is the marginal probability of choosing
nest m of which n is a member; N
m
is the set of all alternatives included in nest m,
m
is the logsum
variable of nest m* and
m
is the logsum or inclusive value (IV) parameter. Thus, V
m
, the utility of
nest m is equal to
m

m
.
,
All alternatives may be considered to be in a nest; in the case of a single alternative in a nest, the
nest is considered degenerate and the logsum parameter,
m
, will equal one. If only one alternative
is available in each nest, the model collapses to the multinomial logit
P
i

e
V
i
P
M
m1
e
V
m
7
2.1.2. The Daly non-normalized nested logit (NNNL) model. Daly's (1987) NNNL model is
dened by the following equations:
P
n
P
n=m
P
m
8
where
P
n=m

e
V
n
P
n
0
2N
m
e
V
n
0
9
P
m

e

m
P
M
m
0
1
e

m
0
m
0
10

m
ln
X
n
0
2N
m
e
V
n
0
11
where the notation is the same as that used in the McFadden NL model. The only dierence
between the models is that the Daly model excludes the inverse of the IV parameters,
m
, in the
utility function of the elemental alternatives in each nest [as can be seen by comparing eqns (9) and
(11) to eqns (4) and (6)].
{
The NNNL model is equivalent to the UMNL model only when the
m
= , for all m in both
models. In this case, the utility function parameters in both models will be equal up to the scale
*
m
represents the expected value of the maximum of the random utilities of alternatives in nest m.
,
Additional variables which apply to all nested alternatives may be included in V
m
however, this can be identically repre-
sented by adding these variables to each alternative in the nest.
{
Also, the inclusion of variables common to all alternatives in a nest at the level of the elemental alternatives is not
equivalent to including these variables at the level of the nest in the Daly formation.
Alternative nested logit models 291
factor, . However, when
m
,= , for any m, the NNNL model is neither equivalent to the
UMNL model nor consistent with utility maximization.* As a result, the behavioral interpretation
of the NNNL model is based solely on the logic of marginal and conditional choice probabilities
and the interpretation of the utility parameters across alternatives is obscure. In particular, the
utility value of generic variables, such as time or cost, for dierent alternatives will depend on the
IV parameter associated with the nest of which the alternative is a member. While there are con-
ditions under which it would be interesting to consider dierent utility values of some attributes
across alternatives (e.g. the value of travel time on a mode may depend on the extent to which the
traveler can use his/her in-vehicle time for other purposes), it is always interesting to test such
model specications against the hypothesis that the utility value of time is constrained to be equal
across alternatives. This constraint is even more likely to be relevant for the utility value of money
spent on travel by each alternative.
2.2. Relationship among alternatives
Both nested logit models allow the user to consider the existence of dierent competitive rela-
tionships between groups of alternatives in a common nest. Such dierences indicate that the eect
of a change in an attribute of an alternative on the probability of that alternative depends on
whether it is or is not in a nest and on other alternatives depends on whether they are or are not
in a common nest. One way to describe these dierences is in terms of direct- and cross-elasticities
of the choice probabilities of the same or other alternatives with respect to a change in the attri-
butes of an alternative. The direct- and cross-elasticities for the MNL and the two NL models are
shown in Table 1. All of the elasticity terms are proportional to the magnitude of the variable
which changes and the parameter associated with that variable. The direct-elasticities are dier-
entiated between alternatives which are or are not in a nest and the cross-elasticities are dier-
entiated between pairs of alternatives which are or are not in a common nest. The direct- and
cross-elasticities of the MNL model depend exclusively on the probability of the mode which is
subject to change. Thus, the MNL cross-elasticities are identical across alternatives and produce
the commonly observed equal proportional eect of the addition, deletion or change of any
alternative on all other alternatives.
The direct elasticities for alternatives not in any nest are structurally identical to the MNL
direct-elasticities for both NL models.
,
The UMNL model direct-elasticities for alternatives in a
nest are greater than the corresponding direct elasticity for alternatives not in a nest; this rela-
tionship is consistent with the idea that alternatives in a nest face a more competitive choice con-
text than do alternatives not in any nest. However, the NNNL direct-elasticities are smaller for
alternatives in a nest than the corresponding direct-elasticity for alternatives which are not in any
nest. These NNNL model dierences are counter-intuitive in that they imply that similarity of an
alternative to other alternatives reduces the eect of a change in that alternative on its own choice
probability relative to the case where that alternative is not in any nest. The magnitude of the
dierence in direct-elasticities between alternatives in or not in a nest increases as the IV para-
meter,
m
, decreases from one to zero for both models.
The cross-elasticities between pairs of alternatives for the two NL models are dierentiated
between pairs which are or are not in the same nest. The cross-elasticities for pairs of alternatives
which are not in the same nest are structurally identical to that for the MNL case for the UMNL
model. The corresponding cross-elasticities for the NNNL model dier from the MNL model by
the scale parameter,
m
. The cross-elasticities between pairs of alternatives in a common nest are
greater for both NL models than the corresponding cross-elasticities for alternatives not in the
same nest. This result is consistent with the notion that cross-elasticities between pairs of alter-
natives within a nest should be larger than cross-elasticities between pairs of alternatives not in the
same nest. In both cases, the cross-elasticities are equal to those for the MNL when
m
is equal to
one.
*The NNNL model does not satisfy the necessary condition that the addition of a constant value to all elemental alternatives
has no eect on the choice probabilities of the alternatives (Daly and Zachary, 1978; McFadden, 1981).
,
The empirical value of the elasticities will dier from the MNL model if the utility function parameters change as a result of
a change in the nesting structure. For this reason, Section 4 includes empirical comparison between the direct-elasticities
for alternatives in or not in a nest and the cross-elasticities between alternatives which are or are not in the same nest
within each model.
292 Frank S. Koppelman and Chieh-Hua Wen
3. MODEL ESTIMATION
Maximum likelihood techniques are used to estimate parameters of the MNL and both NL
models. The log-likelihood for discrete outcomes is of the form
L
X
q
X
i

qi
ln P
qi
12
where
qi
is 1 if individual q chooses alternative i and 0 otherwise, P
qi
is the probability that indi-
vidual q chooses alternative i. The likelihood function for the two-level NL model is
L
X
q
X
i

qi
ln P
qi

X
q
X
M
m1

qm
ln P
qm

X
n2N
m

q;nm
ln P
q;n=m
!
13
where
qm
is 1 if individual q chooses nest m and 0 otherwise, P
qm
is the probability that individual
q chooses nest m,
q;nm
is 1 if individual q chooses alternative n in the nest m and 0 otherwise, and
P
q;n=m
is the probability that individual q chooses alternative n in the nest m.
The log-likelihood function of the UMNL model is dened by the following equation:
L
X
q
X
m1

qm

m

m
ln
X
m
0
1
e

m
0
m
0
! " #

X
n2N
m

q;nm
V
n

m
ln
X
n
0
2N
m
e
V
n
0
m
! " # ( )
14
The log-likelihood function of the NNNL model is
L
X
q
X
M
m1

qm

m

m
ln
X
m
0
1
e

m
0
m
0
! " #

X
n2N
m

q;nm
V
n
ln
X
n
0
2N
m
e
V
n
0
! " # ( )
15
The dierence between these log-likelihood functions is that the log-likelihood of the conditional
probability (the last term in each equation) includes the inverse of the scale parameter in the
UMNL model but not in the NNNL model. The dierences between these equations will result in
dierent estimations for all of the model parameters. The simplication in the NNNL eqn (15),
due to the elimination of
1
m
, appears to make the NNNL model less dicult to estimate than the
UMNL model. This enables software for the NNNL model to accommodate a large number of
levels in the tree structure, an important capability in some contexts.
Estimation of both NL models should use full information maximum likelihood to increase
estimation eciency and allow the user to impose constraints on utility function parameters in
dierent nests in the structure (Brownstone and Small, 1989; Hensher, 1991). Commercial soft-
ware packages have been developed to estimate either the UMNL model, the NNNL model or
both. These include ALOGIT (Hague Consulting Group, 1988, 1992 and 1995), HieLoW
Table 1. Direct- and cross-elasticities of the MNL, UMNL and NNNL models
Model
structure
Direct-elasticity
(change in P
n
due to change in X
n
)
Cross-elasticity
(change in P
n
/ due to change in X
n
)
MNL model (1 P
n
)X
n
P
n
X
n
McFadden
NL model
n not in nest
(1 P
n
)X
n
n and n
/
not in same nest
P
n
X
n
n in nest m

(1 P
n
) (
1
m
1)(1 P
n=m
)[X
n
n and n
/
in nest m

P
n

1
m
1

P
n=m

X
n
Daly NL
model
n not in nest
(1 P
n
)X
n
n and n
/
not in same nest, n in nest m

m
P
n
X
n
n in nest m
[(1 P
n
) (
m
1)P
n=m
(1 P
m
)[X
n
n and n
/
in nest m
[P
n
(1
m
)P
n=m
(1 P
m
)[X
n
Alternative nested logit models 293
(Bierlaire, 1995; Bierlaire and Vandevyvere, 1995), HLOGIT (Bo rsch-Supan, 1990b) and LIM-
DEP (Econometric Software, 1996). The model equations included in software documentation
should provide the user with a clear statement of the formulations which are estimated by each
package. It is particularly important to recognize that software designed to estimate the NNNL
model may be used to estimate the UMNL model by inclusion of dummy nodes and links in the
NNNL structure and constraining the IV parameters at such nodes appropriately.* This
approach, illustrated in the Fig. 1, where the addition of a dummy link (dashed line) and node
and imposition of equality constraints on the IV parameters, enables the estimation of the UMNL
model. The resultant utility function parameters will be scaled by . The utility function scaling to
match the original tree structure can be calculated by multiplication of the parameters by the scale
adjustment. The standard errors of the corrected parameter can be obtained by using Cramer's
approximation (Ben-Akiva and Lerman, 1985, p.298). This approach can be applied to any
tree structure by adding dummy links and logsum parameters so that the product of the logsum
parameters between the root and each elemental alternative is identical.
4. EMPIRICAL EXAMPLE
The following empirical example demonstrates the dierences between the estimation results for
the UMNL and NNNL models.
,
The data used in this study was assembled by VIA rail in 1989 to
estimate the demand for high-speed rail in the TorontoMontreal corridor and to support future
decisions on rail service improvements in the corridor (Forinash and Koppelman, 1993). The
choice set includes four intercity travel modes; air, train, bus and car.
{
2779 individuals who have
all four modes available were selected. The distribution of choices in the data set is: train (463,
16.66%), air (1039, 37.39%), bus (10, 0.36%) and car (1267, 45.59%).
The utility function specication includes mode-specic constants, frequency, travel cost, and
out- of-vehicle and in-vehicle travel times. The estimation results for the MNL and the two NL
Fig. 1. Estimation of single node UMNL model using software designed to estimate the UMNL model (a) or software
designed to estimate the NNNL model (b).
*Software packages dier in their ability to incorporate this approach. The user should ensure that he/she can use parti-
cular software for this purpose through examination of the documentation, testing and/or verication from the software
provider.
,
The reported models were estimated using Gauss (Aptech Systems, 1993) based programs developed at Northwestern
University's Transportation Center and veried with HLOGIT for the McFadden model and ALOGIT for Daly model.
{
Screening of observations without a full set of alternatives does not impact the substantive results reported in this paper.
294 Frank S. Koppelman and Chieh-Hua Wen
models with a common utility specication are reported in Table 2. Using the UMNL model, four
of the 25 possible nested structures had all dissimilarity parameters within the zero-one
range, signicantly rejected the MNL model and were not rejected by other UMNL struc-
tures. Of the four models, two include train and car, with and without bus, and two include air
and car, with and without bus, in a single nest. In each case, the exclusion of bus in the nest has
very little eect on the overall goodness of t of the model so only the results for the models with
train-car and with air-car are reported in columns 2 and 3, respectively, of Table 2.
Using the NNNL model, four structures had all dissimilarity parameters within the zero-one
range, signicantly rejected the MNL model and were not rejected by other NNNL structures. Of
these, two included air and train in a nest, with and without bus; as for the McFadden model, the
exclusion of bus has almost no eect on the model goodness of it and this model is dropped. The
remaining models include train and bus, and air and car in separate nests (model 4), air and train
(model 5) and air and bus (model 6) in a single nest. The model with air-bus in a nest has the best
t; however, the air-bus nest is illogical and the value of in-vehicle time implied by this model, $3
per hour, is unreasonably low. The log-likelihood values of the other NNNL models are somewhat
greater than those for all of the McFadden models. This result raises the question of whether the
NNNL model should be used despite the problem with respect to utility maximization and
Table 2. The estimation results of the MNL, UMNL and NNNL models
Variables Estimated parameters (standard errors)
MNL model UMNL model NNNL model
Traincar
nested
Aircar
nested
Trainbus,
Aircar nested
Airtrain
nested
Airbus
nested
1 2 3 4 5 6
Mode constants
Air 7.7876 7.5014 6.6186 7.0342 7.2664 11.8620
(0.599) (0.549 (0.666) (0.815) (0.659) (0.869)
Train 5.8532 5.9698 5.2677 5.7050 5.6265 7.4589
(0.348) (0.334) (0.382) (0.377) (0.439) (0.302)
Car 4.1751 4.3288 3.9762 3.8376 3.5905 5.8149
(0.414) (0.388) (0.397) (0.463) (0.446) (0.270)
Bus (base)
Frequency 0.0940 0.0932 0.0731 0.0967 0.0976 0.1631
(0.005) (0.004) (0.007) (0.005) (0.005) (0.014)
Travel cost 0.0455 0.0423 0.0312 0.0372 0.0422 0.0639
(0.004) (0.004) (0.005) (0.006) (0.004) (0.004)
In-vehicle time 0.0100 0.0094 0.0094 0.0126 0.0097 0.0027
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Out-of-vehicle-time 0.04265 0.0400 0.0366 0.0512 0.0496 0.0471
(0.003) (0.003) (0.003) (0.004) (0.004) (0.004)
Logsum 80.8204 0.7400 0.7267
a
0.7955 0.4764
(0.068) (0.071) (0.069) (0.041) (0.049)
0.7896
b
(0.083)
Log-likelihood
At convergence 1983.3 1980.6 1977.6 1976.5 1970.1 1938.8
At market share 2903.4 2903.4 2903.4 2903.4 2903.4 2903.4
At zero 3852.5 3852.5 3852.5 3852.5 3852.5 3852.5
Likelihood ratio index
vs market share 0.3169 0.3178 0.3189 0.3192 0.3215 0.3322
vs zero 0.4852 0.4859 0.4867 0.4870 0.4886 0.4967
Value of time (hours)
In-vehicle time C$13 C$13 C$17 C$20 C$14 C$3
Out-of-vehicle time C$56 C$57 C$70 C$83 C$71 C$44
a
Logsum parameter for train and bus net.
b
Logsum parameter for air and car nest.
Alternative nested logit models 295
interpretation. Our view is that preference be given to adopting the model with a strong theoretical
foundation* and interpretability.
The estimation results for the McFadden and Daly models are very dierent. The preferred
models in each case have dierent nesting structures (see Fig. 2) which implies dierent patterns of
competitiveness among the alternatives. The direct elasticities of train and the cross- elasticities of
air, car and bus with respect to a change in train cost and in-vehicle time are reported in Table 3
for UMNL model 3 and NNNL model 5. The emphasis in this table is placed on changes in train
service because the original study was directed toward consideration of major improvements in rail
service in the corridor. The direct elasticities for train and the cross- elasticities of each other mode
with respect to changes in train service dier substantially between these models. These dierences
are due to dierences in the preferred nesting structures, scale parameters and utility function
parameters including dierences in the relative values of level of service parameters (as indicated
by the dierent values of time reported in Table 2). The cross- elasticities of the air, car and bus
modes to changes in train service, an indication of the extent to which train improvements will
impact air and bus revenues and reduce congestion at air terminals and on roadways, also dier
substantially across models.
These results conrm that the use of the dierent NL model structures may lead to substantial
dierences in estimation results and the subsequent forecasts of train ridership and the sources of
train riders under proposed futures. These dierences may lead to very dierent decisions about
whether and what service improvements to make.
Fig. 2. Comparison of nesting structures obtained using the McFadden and Daly NL models.
*Screening of observations without a full set of alternatives does not impact the substantive results reported in this paper.
296 Frank S. Koppelman and Chieh-Hua Wen
5. CONCLUSIONS
Two alternative NL models have been developed and used in research and applications. The
McFadden UMNL model is consistent with utility maximization; the Daly NNNL model, which
is similar to the McFadden model except that it excludes the inverse of the scale parameter in the
utility function of the alternatives in the corresponding nest, is not consistent with utility max-
imization. The models have substantially dierent properties with respect to competitiveness
between pairs of alternatives.
An empirical application showed that the dierent NL models produce dierent results with
respect to both nesting structure and the relative importance of utility components. This conrms
the important dierences which can result from the selection of the UMNL or NNNL models. We
recommend use of the McFadden model because it is based on utility maximization, has intuitively
reasonable elasticities and provides a clear interpretation of parameters across alternatives.
AcknowledgementsThis research was supported, in part, by NSF Grant DMS-9313013 to the National Institute of
Statistical Sciences and, in part, by a Dissertation Year Fellowship to the second author from The Transportation Center,
Northwestern University. The quality of this paper was enhanced by suggestions, comments and critiques by Chandra
Bhat, Christopher Fornash, James Ryan, Andrew Daly, Tom Adler, Carlos Daganzo and three anonymous referees.
REFERENCES
Aptech Systems (1993) GAUSS: User's Guide. Aptech Systems.
Ben-Akiva, M. and Francois, B. (1983) Homogeneous generalized extreme value model. Working paper, Department of
Civil Engineering, MIT, Cambridge.
Ben-Akiva, M. and Lerman, S. R. (1985) Discrete Choice Analysis: Theory and Application to Travel Demand. The MIT
Press, Cambridge.
Bierlaire, M. (1995) A Robust Algorithm for the Simultaneous Estimation of Hierarchical Logit Models. Technical Report,
Department of mathematics, FUNDP, Namar, Belgium.
Bierlaire, M. and Vandevyvere, Y. (1995) HieLoW: the Interactive User's Guide. Transportation Research Group, FUNDP,
Namur.
Bo rsch-Supan, A. (1990a) On the compatibility of nested logit models with utility maximization. Journal of Econometrics
32, 371387.
Bo rsch-Supan, A. (1990b) HLOGIT Documentation.
Brownstone, D. and Small, K. A. (1989) Ecient estimation of nested logit model. Journal of Business and Economic Sta-
tistics 7, 6774.
Daly, A. (1987) Estimating `tree' logit models. Transportation Research 21B, 251267.
Daly, A. and Zachary, S. (1978) Improved multiple choice models. In Determinants of Travel Choice, eds. D. A., Hensher
and M. Q., Dalvi. Saxon House, Sussex.
Forinash, C. V. and Koppelman, F. S. (1993) Application and interpretation of nested logit models of intercity mode
choice. Transportation Research Record 1413, 98106.
Econometric Software (1996) LIMDEP V7.0. Econometric Software, Bellport NY.
Hague Consulting Group (1988) ALOGIT: User's Guide. HCG, The Hague.
Hague Consulting Group (1992) ALOGIT: User's Guide. HCG, The Hague.
Hague Consulting Group (1995) ALOGIT: User's Guide. HCG, The Hague.
Hensher, D. A. (1991) Ecient estimation of hierarchical logit mode choice model. Proceedings of the Japanese Society of
civil Engineering 425, 1728.
Table 3. Direct- and cross-elasticity comparisons
Train level of
service attribute
MNL model UMNL model 3
Air-car nested
NNNL model 5
Air-train nested
(a) Train direct-elasticity in response to improvement in train service
Cost 2.045 1.402 1.744
In-vehicle time 1.809 1.555 1.614
(b). Air cross-elasticity in response to improvement in train service
Cost 0.486 0.334 0.603
In-vehicle time 0.430 0.371 0.558
(c) Car cross-elasticity in response to improvement in train service
Cost 0.486 0.334 0.392
In-vehicle time 0.430 0.371 0.362
(d) Bus cross-elasticity in response to improvement in train service
Cost 0.486 0.334 0.392
In-vehicle time 0.430 0.371 0.362
Alternative nested logit models 297
Herriges, J. A. and Kling, C. L. (1996) Testing the consistency of nested logit models with utility maximization. Economics
Letters 50, 3339.
Kling, C. L. and Herriges, J. A. (1995) An empirical investigation of the consistency of nested logit models with utility
maximization. American Journal of Agricultural Economics 77, 875884.
McFadden, D. (1973) Conditional logit analysis of qualitative choice behavior. In Frontiers in Econometrics, ed. P. Zarembka.
Academic Press, New York.
McFadden, D. (1978) Modeling the choice of residential location. Transportation Research Record 672, 7277.
McFadden, D. (1981) Econometric models of probabilistic choice. In Structural Analysis of Discrete Data, eds. C. F.
Manski and D. McFadden. The MIT Press, Cambridge MA.
Ortuzar, J. D. (1983) Nested logit models for mixed-mode travel in urban corridors. Transportation Research 17A, 283299.
Yai, T., Iwakura, S. and Morichi, S. (1997) Multinomial probit with structured covariance for route choice behavior.
Transportation Research 31B, 195207.
298 Frank S. Koppelman and Chieh-Hua Wen

You might also like