Transportation Research Part B 35 (2001) 627±641

www.elsevier.com/locate/trb

The generalized nested logit model

Chieh-Hua Wen a,*, Frank S. Koppelman b,1
a
Department of Trac and Transportation Engineering and Management, Feng Chia University, 100, Wenhwa Rd.,
Seatwen, Taichung, Taiwan, Republic of China
b
Department of Civil Engineering, Northwestern University, 2145 Sheridan Road, Evanston, Illinois, IL 60208, USA

Abstract
The generalized nested logit (GNL) model is a new member of the generalized extreme value family of
models. The GNL provides a higher degree of ¯exibility in the estimation of substitution or cross-elasticity
between pairs of alternatives than previously developed generalized extreme value (GEV) models. The GNL
model includes the paired combinatorial logit (PCL) and cross-nested logit (CNL) models as special cases. It
also includes the product di€erentiation (PD) model, which represents the elasticity structure associated with
multi-dimensional choices, and the ordered generalized extreme value model, which represents the elasticity
structure associated with ordered alternatives, as special cases. The GNL model includes the two-level nested
logit (NL) model as a special case and can approximate closely multi-level nested logit models. It accom-
modates di€erential cross-elasticity among pairs of alternatives through the fractional allocation of each
alternative to a set of nests, each of which has a distinct logsum or dissimilarity parameter. An empirical
example of intercity mode choice con®rms the statistical superiority of the GNL model to the paired com-
binatorial logit, cross-nested logit and nested logit models and indicates important di€erences in cross-
elasticity relationships across pairs of alternatives. Ó 2001 Elsevier Science Ltd. All rights reserved.

Keywords: Discrete choice; Random utility models; Travel demand; Logit; Intercity travel

1. Introduction

Choice models are used in transportation and other ®elds to represent the selection of one
among a set of mutually exclusive alternatives. The multinomial logit (MNL) model (McFadden,
1973) is the most widely used choice model due to its simple mathematical structure and ease of
estimation. However, the MNL imposes the restriction that the distribution of the random error
terms is independent and identical over alternatives. This restriction leads to the independence of

*
Corresponding author. Tel.: +886-4-4517250 ext. 4679; fax: +886-4-4520678.
E-mail addresses: [email protected] (C.-H. Wen), [email protected] (F.S. Koppelman).
1
Tel.: +1-847-491-8794; fax: +1-847-491-4011.

0191-2615/01/$ - see front matter Ó 2001 Elsevier Science Ltd. All rights reserved.
PII: S 0 1 9 1 - 2 6 1 5 ( 0 0 ) 0 0 0 4 5 - X
628 C.-H. Wen, F.S. Koppelman / Transportation Research Part B 35 (2001) 627±641

irrelevant alternatives property which causes the cross-elasticities between all pairs of alternatives
to be identical. This representation of choice behavior produces biased estimates and incorrect
predictions in cases that violate these strict conditions.
The most widely known relaxation of the MNL model is the nested logit (NL) model (Williams,
1977), which can be derived from McFadden's (1978) generalized extreme value (GEV) model.
The NL model allows the error terms of pairs or groups of alternatives to be correlated. However,
the remaining restrictions on the equality of cross-elasticities between pairs of alternatives in or
not in common nests may be unrealistic in important cases.
Other relaxations of the MNL model, which allow di€erent cross-elasticity between pairs of
alternatives, have been derived from McFadden's GEV model. These include
· the paired combinatorial logit (PCL) model (Chu, 1989;Koppelman and Wen, 2000), which al-
locates each alternative in equal proportions to a nest with each other alternative and estimates
a logsum (dissimilarity parameter) for each nest;
· the cross-nested logit (CNL) model (Vovsha, 1997), which allocates a fraction of each alterna-
tive to a set of nests with equal logsum parameters across nests;
· the ordered generalized extreme value (OGEV) model (Small, 1987), which allocates alterna-
tives to nests based on their proximity in an ordered set; and
· the product di€erentiation (PD) model (Bresnahan et al., 1997), which allocates each alterna-
tive to one nest along each of a set of pre-selected dimensions with allocation parameters asso-
ciated with each dimension and logsum parameters constrained to be equal for each nest along
each choice dimension.
This paper introduces the generalized nested logit (GNL) model, which includes these models
and the MNL model as special cases and closely approximates the NL model. The GNL ac-
commodates di€erential cross-elasticity of pairs of alternatives through the fractional allocation
of each alternative to a set of nests, each of which has a distinct logsum or dissimilarity parameter.
The remainder of this paper is organized as follows. Section 2 presents the formulation, de-
scription and estimation approach for the GNL model and shows that the NL, PCL, CNL,
OGEV and PD models are special cases. Section 3 describes the data for four intercity travel
modes in the Toronto±Montreal corridor (KPMG Peat Marwick and Koppelman, 1990) and
estimation results for the MNL, NL, PCL, CNL and GNL models. 2 Section 4 suggests further
developments in the search for a preferred structural form and directions for additional model
¯exibility. Section 5 provides a summary and conclusions.

2. The generalized nested logit model

2.1. Model formulation

The GNL model is a GEV 0 model (McFadden, 1lm 1978) derived from the function
X X
G Y1 ; Y2 ; . . . ; Yn † ˆ @ an0 m Yn0 †1=lm A ; 1†
m n0 2Nm

2
The OGEV and PD models are not included in this comparison since the alternatives neither are ordered nor fall
into categorical groupings along dimensions.
 C.-H. Wen, F.S. Koppelman / Transportation Research Part B 35 (2001) 627±641 629

where Nm is the set of all alternatives included in nest m, anm the allocation parameter which
characterizes the portion of alternative P n assigned to nest m (anm must satisfy the condition
anm P 0, the additional condition m anm ˆ 1; 8n provides a useful interpretation with respect to
allocation of each alternative to each nest), lm is the logsum or dissimilarity parameter for nest m
(0 < lm 6 1) and Yn characterizes the value for each alternative.
The function, Eq. (1) which is non-negative, homogeneous of degree one, approaches in®nity
with any Yi and has kth cross-partial derivatives which are non-negative for odd k and non-
positive for even k. The resultant GEV probability function, after substituting eVn0 to ensure
positive Yn0 , is
 P lm 1 
P Vn 1=lm Vn0 1=lm
m anm e † n0 2Nm an0 m e †
Pn ˆ P P Vn0 1=lm
lm
m n0 2Nm an0 m e †
0 P lm 1
V 0 1=lm
XB Vn 1=lm
anm e † a n 0me n †

lm C
0
n 2Nm
ˆ @P 1=l P P A: 2†
n0 2Nm an0 m e †
V m 1=l
m n0
m m
0
n 2N a n 0 m e n0 †
V m

This equation can be decomposed into components and rewritten as

X
Pn ˆ Pn=m Pm ; 3†
m

where Pm , the probability of nest m, is

P lm
Vn0 1=lm
n0 2Nm an0 m e †
Pm ˆ P  P lm 4†
a 0 e Vn0 †1=lm
m n0 2Nm nm

and Pn=m , the probability of alternative n if nest m is selected, is

anm eVn †1=lm

Pn=m ˆ P : 5†
n0 2Nm an0 m eVn0 †1=lm
The GNL model is consistent with random utility maximization if the conditions, 0 < lm 6 1, are
satis®ed. The direct elasticity of an alternative, n, which appears in one or more nests with logsum,
lm , less than one, is
P h   i
m Pm Pn=m 1 Pn † ‡ l1m 1 1 Pn=m †
bXn : 6†
Pn
The terms in the summation evaluate to zero for any nest which does not include alternative
n. The elasticity reduces to the MNL elasticity, 1 Pn †bXn , if the alternative does not share a
nest with any other alternative or is assigned only to nests for which the logsum value equals
one.
630 C.-H. Wen, F.S. Koppelman / Transportation Research Part B 35 (2001) 627±641

The corresponding cross-elasticity of a pair of alternatives, n and n0 , which appear in one or

more common nests, is
2 P 1  3
m lm 1 Pm P n=m P n 0 =m
4Pn ‡ 5bXn : 7†
P n0

In this case, the terms in the summation evaluate to zero for any nest which does not include both
alternatives, n and n0 , and reduces to the MNL cross-elasticity, Pn bXn , if the alternatives do not
share any common nest. These elasticities are independent of the elasticities for any other alter-
native or pair of alternatives.
Swait (2000) recently proposed the general logit (GenL) model, in which nest represents a
possible choice set so that the marginal probability represents the selection or availability of the
choice set and the conditional probability represents the choice of an alternative given that choice
set. The GenL model is similar to the GNL except that the allocation parameters are associated
with individuals rather than alternatives. Vovsha (1999) reports development and application of
the fuzzy nested logit model, which is identical to the GNL, except that it allows multiple levels of
nesting. While the additional levels of nesting appear to increase the ¯exibility of the model, they
raise complex problems of identi®cation since the GNL can represent the same di€erential sen-
sitivities within its two level nesting structure.

2.2. Structural relationships between the GNL and other GEV models

The PCL, CNL, OGEV and PD models are restricted versions of the GNL model. The NL
model is not a restricted case of the GNL model, but it can be approximated closely by a suitably
speci®ed GNL model.

2.2.1. The PCL model

Comparison between the GNL and PCL models requires adoption of a special case of the GNL
model that includes one nest for each pair of alternatives, as in the PCL model. Such a paired
GNL (PGNL) model has the form
2 3
" #6  V 1=lnn0 V 1=lnn0
lnn0 7
X Vn 1=lnn0
an;nn0 e † 6 a n;nn0e n† ‡ a n0 ;nn0 e n † 7
6   7
Pn;PGNL ˆ 6 P l 7: 8†
an;nn0 e n †
V 1=l nn0
‡ an0 ;nn0 e n †
V 1=lnn 0 6 1=l 1=l
ak;kk0 eVk † kk0 ‡ ak0 ;kk0 eVk0 † kk0
kk 0
7
n0 6ˆn 4 5
8k
k6ˆk 0

The PGNL model, Eq. (8), restricted so that all allocation parameters, an; nn0 † , are equal, is
equivalent to the PCL model. 3 The non-equal allocation to nests in the PGNL model allows

3
The allocation parameters equal the inverse of the number of alternatives minus one so that the sum of allocation
parameters equals one. This di€ers from, but has the same e€ect as, the original PCL model for which all allocation
parameters are equal to one.
 C.-H. Wen, F.S. Koppelman / Transportation Research Part B 35 (2001) 627±641 631

greater freedom in the magnitude of cross-elasticity than is allowed by the corresponding PCL
model. Further, the PGNL allows an allocation of zero for an alternative to a nest and the
elimination of nests for which both alternatives have zero allocation.

2.2.2. The CNL model

The CNL model is a straightforward restriction of the GNL model. That is, the restriction
that all logsum parameters, lm , are equal in the GNL model, Eq. (2), results in the CNL
model.

2.2.3. The OGEV model

The OGEV model allows cross-elasticity between pairs of alternatives in an ordered choice to
be related to their proximity in that order. Each alternative is a member of nests with one or
more adjacent alternatives. The general OGEV model allows di€erent levels of cross-elasticity
by changing the number of adjacent alternatives in each nest (and therefore the number of
common nests shared by each pair of alternatives), the allocation weights of each alternative to
each nest and the dissimilarity parameters for each nest. The choice probability for alternative
m is
2 P lm 3
Vj 1=lm
X X Vi 1=lm j2Nm wm j e †
i‡L i‡L
6 wm i e †
Pi ˆ Pi=m
Pm ˆ 4P 1=lm P  P ls 7
5; 9†
j2Nm wm j e †
V J ‡L 1=l
j2Ns ws j e †
mˆi mˆi
i Vj s
sˆ1

where L is a positive integer that de®nes the maximum number of contiguous alternatives in a
nest, w the allocation weight of the alternative to the nest and eVi =l is equal to zero for i < 1 and
i > J.
This is equivalent to the GNL model with the constraint that the weights associated with
the assignment of each alternative to a nest are associated with its ordered position in the
nest.

2.2.4. The PD model

The PD model is based on the notion that markets for di€erentiated products (alterna-
tives) exhibit increased cross-elasticity due to clustering (nesting) relative to dimensions,
which characterize attributes of the product. Such dimensions could include, in the case of
transportation modeling, mode and destination or number of cars, residential location and
mode to work. The choice probability equation for a PD model with D dimensions is given
by
0 1
P
l
XB eVi =ld k2d e
Vk =ld d

Pi ˆ @ad P P P ld 0 C
A; 10†
k2d e
Vk =ld
k 0 2d 0 e
d2D Vk 0 =ld 0
d 0 2D
632 C.-H. Wen, F.S. Koppelman / Transportation Research Part B 35 (2001) 627±641

where ad is the portion of each alternative allocated to dimension d and ld is the logsum pa-
rameter for all groups (nests) along dimension d.
This model restricts the GNL so that all alternatives have the same allocation to each di-
mension and the nests along each dimension have the same logsum parameters.

2.2.5. The NL model

As stated earlier, the two-level NL model is a special case of the GNL; that is, a GNL with
each alternative allocated to a single nest. More importantly, the GNL model can approximate
any multi-level nested logit model by including a nest, which corresponds to each node in the
nested logit. This can be seen in Fig. 1, which shows, in part (a), a three-level nested logit
structure with four nodes. Part (b) shows the corresponding GNL approximation in which the
alternatives grouped under each node in the nested logit structure are assigned to a common
nest. Alternatives, which are nested at multiple levels, are assigned to all nests represented by
nodes between the alternative and the root of the NL tree. The self- and cross-elasticities, and
substitution patterns, in the GNL model are based on the logsum parameters associated with

(b)

Fig. 1. NL approximation within GNL model structure: (a) three-level nested logit model structure; (b) GNL ap-
proximation of three-level nested logit structure.
 C.-H. Wen, F.S. Koppelman / Transportation Research Part B 35 (2001) 627±641
Table 1
Direct and cross-elasticities of the MNL, GNL, CNL and PCL models
Model Direct elasticity Cross-elasticity
MNL 1 Pn †bXn Pn bXn
GNL n assigned to a single nest with no other alternatives n and n0 not in any common nest
1 Pn †bXn Pn bXn
n in one or more nests n and n0 in one or more common nests
P h   i 2 P 1  3
m Pm Pn=m 1 Pn † ‡ l1m 1 1 Pn=m † m lm
1 Pm Pn=m Pn0 =m
bXn 4P n ‡ 5bXn
Pn Pn0

CNL n assigned to a single nest with no other alternatives n and n0 not in any common nest
1 Pn †bXn Pn bXn
n in one or more nests n and n0 in one or more common nests
P h   i 2 P 1  3
m Pm Pn=m 1 Pn † ‡ l1 1 1 Pn=m † m l
1 Pm Pn=m Pn0 =m
bXn 4P n ‡ 5bXn
Pn Pn0

PCL P h   i 2   3
1 1
n0 6ˆn Pnn0 Pn=nn 1 Pn † ‡ lnn0
1 1 Pn=nn0 † lnn0
1 Pnn0 Pn=nn0 Pn0 =nn0
bXn 4P n ‡ 5bXn
Pn Pn 0

633
634 C.-H. Wen, F.S. Koppelman / Transportation Research Part B 35 (2001) 627±641

each nest in which an alternative or pair of alternatives is (are) included. Thus, for example, the
cross-elasticity between alternatives 3 and 4 will be greater than between 3 and 5 or 6 and these
are greater than the cross-elasticity between 3 and 2. The estimation is somewhat more complex
since the GNL requires estimation of allocation parameters in addition to the four logsum
parameters.

2.3. Direct and cross-elasticities

The di€erences between the GNL model and the MNL, PCL, CNL, OGEV and PD models can
be examined further by comparison of direct and cross-elasticities of probabilities with respect to
changes in attributes of any alternative (Table 1).
The direct-elasticity formula for the MNL model is identical for all alternatives depending
only on the probability of the alternative. 4 The direct-elasticity formulae for the other models
are greater than for the MNL model for alternatives in a common nest with logsum less than
one and the same as the MNL model for other alternatives. 5 However, the similarity among the
GNL, CNL and PCL elasticities is somewhat misleading as they do not explicitly show the e€ect
of the allocation parameters which are embedded in the probabilities as shown in Eqs. (4) and
(5) for the GNL model.
The cross-elasticity formulae of the MNL model depend exclusively on the probability of the
changed mode, which gives the commonly observed equal proportional e€ect of the addition,
deletion or change of any alternative on all other alternatives. The cross-elasticity for pairs of
alternatives in the other models are greater in magnitude than for the MNL model if the pair is in
a common nest with logsum less than one and equal to the MNL model otherwise. The elasticity
increases in magnitude as lm decreases from one, with the magnitude of the impact related to the
probability of the nest and the conditional probabilities of the alternatives in the nest. As with the
direct elasticities, the similarity among the GNL, CNL and PCL elasticities is somewhat mis-
leading as they do not explicitly show the e€ect of the allocation parameters which are embedded
in the probabilities.
An alternative perspective on the relationships among pairs of alternatives is the implied
correlation between the error terms for pairs of alternatives. Table 2 reports the correlations for
di€erent combinations of allocation and logsum parameters in the CNL and GNL models. The
important point of this table is that the correlations can achieve very high values if such values are
supported by the observed behavior. However, the correlations of the CNL model are not as
¯exible as this table suggests since the logsum parameters in the CNL are limited by the re-
quirement that all logsum parameters be equal.

4
All the elasticities include the variable of change and the utility function parameter associated with that
variable.
5
Empirical experience indicates that utility function parameters are smaller in magnitude for these models than for
the MNL model so that the direct elasticities decrease for alternatives not in any nest with logsum less than one but
increase for all other alternatives. Similarly, the cross-elasticities decrease for alternatives not in a common nest and
increase for alternatives in one or more common nests with logsum less than one.
 C.-H. Wen, F.S. Koppelman / Transportation Research Part B 35 (2001) 627±641 635

Table 2
Correlation between pairs of alternatives in a nest implied by the CNL and GNL models as a function of the logsum
and allocation parameters
Allocation Logsum Allocation parameter (an0 )
parameter (an ) parameter 0.1 0.5 1.0
0.1 0.1 0.09 0.17 0.21
0.3 0.08 0.16 0.20
0.5 0.07 0.14 0.17
0.7 0.04 0.10 0.12
0.9 0.02 0.04 0.05
1.0 0.00 0.00 0.00
0.5 0.1 0.45 0.64
0.3 0.42 0.59
0.5 0.35 0.50
0.7 0.24 0.34
0.9 0.09 0.13
1.0 0.00 0.00
1.0 0.1 0.99
0.3 0.91
0.5 0.75
0.7 0.51
0.9 0.19
1.0 0.00

2.4. Estimation

The GNL model requires joint estimation of the utility, logsum and allocation parameters. This
paper employs constrained maximum likelihood (Aptech Systems, 1995) to estimate the three sets
of parameters, simultaneously, taking account of the restrictions that the logsum and allocation
parameters are bounded by zero and one and that the allocation parameters for each alternative
sum to one. The number of logsum parameters that can be identi®ed is one less than the number
of pairs of alternatives. This limitation and the general ¯exibility of the model structure require
the analyst to make judgements about the clustering of alternatives into nests. This is similar to
the problem of selecting one among a large set of alternative nesting structures when estimating a
nested logit model or imposing restrictions on the covariance matrix in the MNP model. The
GNL model requires similar judgements to be made.
Analyst judgement can be implemented in a variety of ways. First, the analyst can limit the
nesting options a priori, based on judgement about the likely elasticity or substitution rela-
tionships among pairs or groups of alternatives. Second, structural relationships can be imposed
on the cross-elasticities among pairs of alternatives to reduce the number of independent allo-
cation and/or logsum parameters. For example, logsum parameters can be constrained to be
equal for groups along choice dimensions and allocations to each dimension can be constrained
to be equal as in the PD model (Bresnahan et al., 1997). Third, the analyst can search over all or
most of the possible structures. Additional options include using various constrained versions of
the GNL model such as the PCL, CNL or PGNL to obtain preliminary estimates of the relative
636 C.-H. Wen, F.S. Koppelman / Transportation Research Part B 35 (2001) 627±641

magnitude of elasticity/substitution relationships among pairs of alternatives. Fourth, the

Hessian of the log-likelihood function for the GNL model is not negative semi-de®nite over its
whole range. It may be required to repeat optimization with di€erent starting points to locate
the global optimum.

3. Empirical analysis

The data used in this study were assembled by VIA Rail in 1989 (KPMG Peat Marwick and
Koppelman, 1990) to estimate the demand for high-speed rail in the Toronto±Montreal corridor
and to support future decisions on rail service improvements in the corridor. The data includes
4324 individuals, whose choice set includes two or more of four intercity travel modes (air, train,
bus and car) in the corridor. The fractions of the sample which had each alternative available are
train (4299, 99.4%), air (3626, 83.9%), bus (3271, 75.6%) and car (4324, 100%) and the distribution
of choices is train (623, 14.41%), air (1472, 34.04%), bus (16, 0.37%) 6 and car (2213, 51.18%). This
dataset has been used for a variety of model formulation and estimation studies including
Forinash and Koppelman (1993), Koppelman and Wen (2000, 1999, 1998), Bhat (1995, 1997a,b)
and others.
The utility function speci®cation includes mode-speci®c constants, frequency, travel cost, and
in- and out-of-vehicle travel times. 7 The estimation results for the MNL, two NL models (Ko-
ppelman and Wen, 1998) and the PCL model (Koppelman and Wen, 2000) are reported in Table 3.
The NL models have almost identical goodness of ®t, neither is able to reject the other, but they
both reject the MNL model and lead to very di€erent behavioral interpretations and di€erent
forecasts of the e€ect of changes in the alternatives. The train±car nested model represents a
higher level of competitiveness between train and car than between other modes and the air±car
nested model represents a higher level of competitiveness between air and car than between other
modes. The PCL model, which allows increased competitiveness for both the train±car and air±
car pairs, rejects the MNL model and both NL models at high levels of signi®cance as shown in
the table.
Estimation results for the CNL and GNL models are reported in Table 4. Exploratory esti-
mation, limited to a maximum of two alternatives per nest, is used to select among di€erent
nesting structures. The resultant nests, for both the CNL and GNL models (columns 1 and 2), are
bus alone, train alone, car alone, train±car and air±car. CNL Model 1 obtains a signi®cant (with
respect to one) logsum parameter that applies to both the train±car and air±car nests; the logsum
parameters for single alternative nests (train, car and bus) are set to one. This model rejects the
MNL, both NL and the PCL models at very high levels of signi®cance, in excess of 0.001, using
the nested hypothesis test for the MNL model and the non-nested hypothesis test for the NL and
PCL models (Horowitz, 1983). GNL Model 1 obtains logsum parameters (0.05 for train±car and
0.32 for air±car) that are signi®cantly di€erent from one and from each other; as with the CNL

6
The small number of cases for which bus is chosen limits the estimability of allocation and logsum parameters
associated with the bus alternative.
7
Tests of alternative model structures with di€erent utility function speci®cations, including income and city pair
indicator variables, did not substantially a€ect the comparison among model structures.
 C.-H. Wen, F.S. Koppelman / Transportation Research Part B 35 (2001) 627±641 637

Table 3
Estimation results for the MNL, NL and PCL modelsa
Variables Estimated parameters (standard errors)
MNL model NL with NL with air±car PCL model
train±car nested nested
Mode constants
Air 8.238 (0.429) 7.812 (0.450) 7.533 (0.511) 7.157 (0.430)
Train 5.412 (0.267) 5.513 (0.270) 5.061 (0.300) 5.129 (0.267)
Car 4.421 (0.301) 4.446 (0.300) 4.372 (0.303) 4.262 (0.294)
Bus (base)
Frequency 0.0850 (0.004) 0.0845 (0.004) 0.0722 (0.006) 0.0689 (0.003)
Travel cost )0.0508 (0.003) )0.0464 (0.003) )0.0420 (0.004) )0.0379 (0.003)
In-vehicle time )0.0088 (0.001) )0.0084 (0.001) )0.0080 (0.001) )0.0076 (0.001)
Out-of-vehicle time )0.0354 (0.002) )0.0339 (0.002) )0.0310 (0.002) )0.0305 (0.002)
Logsum parameters
Train±car 0.8302 (0.059) 0.5200 (0.109)
Air±car 0.8233 (0.063) 0.1922 (0.076)
Log-likelihood at )2784.6 )2781.2 )2780.9 )2769.1
convergence
Likelihood ratio index
vs. zero 0.4896 0.4903 0.4903 0.4925
vs. market share 0.3205 0.3213 0.3214 0.3243
Value of time (per hour)
In-vehicle time C$ 10 C$ 12 C$ 12 C$ 12
Out-of-vehicle time C$ 42 C$ 48 C$ 48 C$ 48
Signi®cance test rejecting ) 6.8, 1, 0.001 7.4, 1, 0.001 31.0, 2, 0.001
MNL model (v2 , DF, Sig.)
a
Note: The PCL model rejects both NL models at the 0.001 level using the non-nested hypothesis test.

model, the logsum parameters for single alternative nests are set to one. The GNL model rejects
the CNL model as well as the MNL, NL and PCL models, at the 0.001 level. Additional CNL
and GNL models with an additional air±train±car nest (columns 3 and 4) statistically reject the
corresponding models without any three alternative nests. The inclusion of train and car in the
train±car and train±air±car nests in the GNL model results in colinearity among the logsum and
allocation pa#132;rameters. Nonetheless, this model strongly rejects all the previously estimated
models. This problem is avoided in the CNL model due to the equality constraint across the
logsum parameters. Nevertheless, the ®nal GNL model appears to be superior to all models
previously estimated. Based on limited exploration, these results hold across a variety of utility
function parameters.
There are signi®cant di€erences among the di€erent structural models. These di€erences are
likely to produce important di€erences in mode forecasts under alternative scenarios for future
transportation services, possibly resulting in di€erent investment decisions. The attribute pa-
rameters in the utility function decrease in magnitude with increasing complexity in model
638 C.-H. Wen, F.S. Koppelman / Transportation Research Part B 35 (2001) 627±641

Table 4
Estimation results for the CNL and GNL models
Variable Estimated parameters (standard errors)
Without train±car±air nest With train±car±air nest
CNL Model 1 GNL Model 1 CNL Model 2 GNL Model 2
Mode constants
Air 5.746 (0.429) 5.344 (0.367) 5.476 (0.343) 6.264 (0.321)
Train 4.618 (0.286) 4.460 (0.281) 5.083 (0.256) 4.981 (0.285)
Car 4.455 (0.275) 4.300 (0.267) 4.901 (0.284) 5.133 (0.253)
Bus (base)
Frequency 0.0460 (0.006) 0.0421 (0.005) 0.0206 (0.009) 0.0288 (0.002)
Travel cost (C$) )0.0209 )0.0172 )0.0096 )0.0173
(0.004) (0.003) (0.004) (0.002)
In-vehicle time (min) )0.0059 )0.0060 )0.0023 )0.0031
(0.001) (0.001) (0.001) (0.0002)
Out-of-vehicle time (min) )0.0201 )0.0198 )0.0088 )0.0110
(0.002) (0.002) (0.004) (0.001)
Logsum parameter
Train±car 0.3141 (0.041) 0.0463 (0.019) 0.1008 (0.035) 0.0146 (0.002)
Air±car 0.3141 (0.041) 0.3159 (0.042) 0.1008 (0.035) 0.2819 (0.032)
Train±car±air 0.1008 (0.035) 0.01 ())
Allocation parameter
Train±car nest
Train 0.7032 (0.074) 0.4904 (0.046) 0.1547 (0.045) 0.2717 (0.033)
Car 0.2611 (0.047) 0.1896 (0.023) 0.1060 (0.026) 0.1057 (0.012)
Air±car nest
Air 1.0000 1.0000 0.2287 (0.065) 0.6061 (0.040)
Car 0.5163 (0.059) 0.5664 (0.054) 0.1145 (0.031) 0.4179 (0.046)
Train±car±air nest
Train 0.7409 (0.069) 0.5286 (0.031)
Car 0.6335 (0.059) 0.2741 (0.029)
Air 0.7713 (0.065) 0.3939 (0.041)
Train nest 0.2226 (0.071) 0.5096 (0.046) 0.1044 (0.037) 0.1998 (0.025)
Car nest 0.2968 (0.074) 0.2440 (0.051) 0.1460 (0.047) 0.2024 (0.032)
Bus nest 1.0000 1.0000 1.0000 1.0000
Log-likelihood at convergence )2746.6 )2736.3 )2723.1 )2711.3
Likelihood ratio index
vs. zero 0.4966 0.4985 0.5009 0.5031
vs. market share 0.3298 0.3320 0.3355 0.3382
Value of time (per hour)
In-vehicle time C$ 17 C$ 21 C$ 14 C$ 11
Out-of-vehicle time C$ 57 C$ 69 C$ 55 C$ 38
Signi®cance test rejecting CNL 20.6, 1, ) 23.6, 2, <0.001 )
models (v2 , DF, Sig.) <0.0001
 C.-H. Wen, F.S. Koppelman / Transportation Research Part B 35 (2001) 627±641 639

structure. This implies that the cross-elasticities between alternatives in a common nest are
reduced while those in common nests are increased, as expected. The relative value of these
parameters, as represented by the values of time, are reasonably stable over all the models
estimated.

4. Estimation and use of complex structural models

The development of multiple forms of GEV models with potentially large numbers of estimable
parameters raises important questions of model selection and use in analysis in both transpor-
tation and non-transportation contexts. Models with increased ¯exibility add to the estimation
complexity, the importance of analyst judgement, computational demands and the time required
searching for and selecting a preferred model structure. This task is interrelated with the task of
searching for and selecting a preferred utility function speci®cation. Horowitz (1991) raised the
concern that the increased ¯exibility of error structure speci®cation of the multinomial probit
model might lead to a proliferation of random e€ects parameters and thereby reduce the incentive
for modelers to develop enhanced utility function speci®cations. The same concern can be applied
to the search for and selection among alternative GEV models and the structural parameters that
de®ne each model type. Therefore, an important issue for additional research is the analysis and
understanding of interrelationships between model structure and parameters and utility function
speci®cation. The development of useful rules to guide the search among complex alternative
structures would provide the option of guiding the analyst and reducing both the search and
computational time associated with obtaining a preferred model.
A further issue is the usefulness of developing more complex GEV models when suitably
speci®ed multinomial probit and mixed logit models (Brownstone and Train, 1998; McFadden
and Train, 1997) can approximate all such models. Our perspective is that there is a place in the
set of analytic tools for models with di€erent levels of complexity in structure, estimation, in-
terpretation and application. Advanced research is likely to employ models with high degrees of
complexity. Professional practice, however, may be best served by the use of models, the com-
plexity of which is closely matched to the problem at hand; that is, use the minimally complex
model to capture and represent the behavior under study. We believe that the development of
models of varying degrees of complexity serves this purpose.

5. Conclusions

The GNL model adds useful ¯exibility to the family of GEV models by providing a more
¯exible structure for estimating di€erential cross-elasticities among pairs of alternatives. It also
provides a unifying structure for previously reported GEV models, with the exception of the NL
model, and provides a framework for understanding the properties of these models. This paper
demonstrates that the GNL model can be feasibly estimated and is useful in applied work.
An additional advantage is that the GNL model provides a structural framework for exploring
alternative cross-elasticity structures without necessarily estimating a large number of distinct
models as required in the estimation of the NL model.
640 C.-H. Wen, F.S. Koppelman / Transportation Research Part B 35 (2001) 627±641

Acknowledgements

This research was supported, in part, by NSF Grant DM-9313013 to the National Institute of
Statistical Sciences and, in part, by a dissertation year fellowship to the ®rst author from The
Transportation Center, Northwestern University. Insightful suggestions and comments by Vaneet
Sethi, John Gliebe and anonymous reviewers have contributed to the quality and clarity of this
paper. Further, Vaneet Sethi provided extensive support in validating derivations and estimation
results.

References

Aptech Systems, 1995. Gauss Applications: Constrained Maximum Likelihood. Aptech Systems Inc., Maple Valley, WA.
Bhat, C.R., 1995. A heteroscedastic extreme value model of intercity mode choice. Transportation Research B 29 (6),
471±483.
Bhat, C.R., 1997a. An endogenous segmentation mode choice model with an application to intercity travel.
Transportation Science 31, 34±48.
Bhat, C.R., 1997b. Covariance heterogeneity in nested logit models: econometric structure and application to intercity
travel. Transportation Research B 31, 11±21.
Bresnahan, T.F., Stern, S., Trajtenberg, M., 1997. Market segmentation and the sources of rents from innovation:
personal computers in the late 1980s. RAND Journal of Economics 28 (0), 17±44.
Brownstone, D., Train, K., 1998. Forecasting new product penetration with ¯exible substitution patterns. Journal of
Econometrics 89 (102), 109±129.
Chu, C.A., 1989. Paired combinatorial logit model for travel demand analysis. in: Proceedings of the Fifth World
Conference on Transportation Research, vol. 4. Ventura, CA, pp. 295±309.
Forinash, C.V., Koppelman, F.S., 1993. Application and interpretation of nested logit models of intercity mode choice.
In: Transportation Research Record 1413. TRB, National Research Council, Washington, DC, pp. 98±106.
Horowitz, J.H., 1983. Statistical comparison of non-nested probabilistic discrete choice models. Transportation Science
17 (3), 319±350.
Horowitz, J.L., 1991. Reconsidering the multinomial probit model. Transportation Research B 25, 433±438.
Koppelman, F.S., Wen, C., 1998. Alternative nested logit models: structure, properties and estimation. Transportation
Research B 32, 289±298.
Koppelman, F.S., Wen, C., 1999. Di€erent nested logit models: which are you using? In: Transportation Research
Record 1645. TRB, National Research Council, Washington, DC, pp. 1±7.
Koppelman, F.S., Wen, C.-H., 2000. The paired combinatorial logit model: properties, estimation and application.
Transportation Research B 34 (2), 75±89.
KPMG Peat Marwick, Koppelman, F.S., 1990. Analysis of the market demand for high speed rail in the Quebec±Ontario
corridor. Report Produced for Ontario/Quebec Rapid Train Task Force. KPMG Peat Marwick, Vienna, VA.
McFadden, D., 1973. Conditional logit analysis of qualitative choice behavior. In: Zaremmbka, P. (Ed.), Frontiers in
Econometrics. Academic Press, New York.
McFadden, D., 1978. Modeling the choice of residential location. In: Transportation Research Record 672. TRB,
National Research Council, Washington, DC, pp. 72±77.
McFadden, D., Train, K., 1997. Mixed MNL model for discrete response. Working paper, Department of Economics,
University of California, Berkeley.
Small, K., 1987. A discrete choice model for ordered alternatives. Econometrica 55 (2), 409±424.
Swait, J., 2000. Choice set generation within the generalized extreme value family of discrete choice models.
Transportation Research, in press.
Vovsha, P., 1997. The cross-nested logit model: application to mode choice in the Tel-Aviv metropolitan area. In:
Transportation Research Record 1607. TRB, National Research Council, Washington, DC, pp. 6±15.
 C.-H. Wen, F.S. Koppelman / Transportation Research Part B 35 (2001) 627±641 641

Voshva, P., 1999. E-mail to the lead author describing the FNL model, October.
Williams, H.C.W.L., 1977. On the formulation of travel demand models and economic evaluation measures of user
bene®t. Environment and Planning A 9 (3), 285±344.