This paper employs extensions to the MNL model to relax the known characteristic of MNL elasticities being a function of the size of variables which can cause under-sensitivity at low values and over-sensitivity at higher values and demonstrates the advantages of adopting the Generalized Nested Logit formulation in place of the more restrictive MNL and NL Generalized Extreme Value models. The resultant model, which includes distance scaling of distance related variables, the flexible covariance structure of the Generalized Nested Logit model and differential variances for first and second choice stated preference data. The analysis is undertaken in the context of intercity travel with multiple modes and multiple rail service classes. The resultant model strongly rejects more basic MNL and NL models and provides a highly intuitive interpretation of intercity choice behavior.
This study models airline passengers' no-show and standby behaviour using passenger and outbound and inbound itinerary information. It describes factors that influence no-show and standby behaviour in continental US markets
using multinomial and nested logit models. The model accounts for early standby opportunities on other itineraries flown by the carrier of interest, i.e., it incorporates information about booking levels on the carrier's itineraries departing on the same day. Empirical and validation analyses highlight the importance of distinguishing between outbound and inbound itineraries. It concludes by describing a framework carriers can use to implement the no-show model.
This is the first study of airline travelers' no-show and standby behavior based on passenger and directional outbound/inbound itinerary data. The paper describes passengers' behavior based on estimation of a multinomial logit model for domestic U.S. itineraries departing in March 2001 or March 2002. This enables us to explore behavioral differences based on passenger and itinerary characteristics as well as identify differences in rescheduling behavior occurring after September 11, 2001. Benefits of using passenger data to improve forecasting accuracy and support a broad range of managerial decisions are described.
Intra-household interactions constitute an important aspect in modeling activity and travel-related decisions. Recognition of this importance has recently produced a growing body of research on various aspects of modeling intra-household interactions and group decision making mechanisms as well as first attempts to incorporate intra-household interactions in regional travel demand models. This paper presents an attempt to build a general framework for incorporation of intra-household interactions in the regional travel demand model. The approach distinguishes between three principal levels of intra-household interactions: 1) Coordinated principal daily pattern types, 2) Episodic joint activity and travel, 3) Intra-household allocation of maintenance activities. The adopted models are discrete choice constructs of the Generalized Extreme Value class. These models together create an analytical framework for integrative modeling of the daily activity and travel of multiple household members, taking into account their interactions.
The assumption of independently and identically distributed (IID) error terms in the Multinomial Logit (MNL) model leads to its infamous IIA property. Relaxation of the IID assumption has been undertaken along a number of isolated dimensions leading to the development of a rich set of discrete choice models that are more flexible than the MNL model. In some cases, these more general models lose the mathematically convenient closed-form structure of the MNL. In this paper, we combine the Generalized Nested Logit model that allows for non-independent errors, the Heteroscedastic MNL which allows non-constant errors across observations, and the Covariance Heterogeneous NL model which allows for non-constant correlation structure across observations. The resulting model, called the heterogeneous GNL model extends our ability to represent the complex behavioral processes involved in choice decision-making. The value and need for the additional modeling complexity of the HGNL model is tested in the empirical context of mode and rail service class choice behavior for long distance intercity travel.
The daily activity-travel patterns of individuals often include interactions with other household members, which we observe in the form of joint activity participation and shared rides. Explicit representation of joint activity patterns is a widespread deficiency in extant travel forecasting models and remains a relatively under-developed area of travel behavior research. In this paper, we identify several spatially defined tour patterns found in weekday household survey data that describe this form of inter-agent decision making. Using pairs of household decision makers as our subjects, we develop a structural discrete choice model that predicts the separate, parallel choices of full-day tour patterns by both persons, subject to the higher level constraint imposed by their joint selection of one of several spatial interaction patterns, one of which may be no interaction. We apply this model to the household survey data, drawing inferences from the household and person attributes that prove to be significant predictors of pattern choices, such as commitment to work schedules, auto availability, commuting distance and the presence of children in the household. Parameterization of an importance function in the models shows that in making joint activity-travel decisions significantly greater emphasis is placed on the individual utilities of workers relative to non-workers and on the utilities of women in households with very young children. The model and methods are prototypes for tour-based travel forecasting systems to represent the complex interaction between household members in an integrated model structure.
Choice-based samples are often used in transportation and other fields to obtain better information about alternatives that are infrequently chosen in the population. However, standard maximum likelihood estimation (MLE) techniques do not obtain consistent estimates of the underlying model parameters. The most widely used approach to obtaining consistent, but not efficient, estimators is to use weighted exogenous sampling maximum likelihood (WESML) estimators. Efficient estimators of all parameters can be obtained using choice-based samples when using the multinomial logit (MNL) model with a full set of alternative specific constants with an adjustment to account for differences between sample and population choice frequencies. This paper extends this result to the estimation of nested logit (NL). The ability to recover efficient parameter estimates for NL models using ESML estimators provides practitioners and researchers with a practical method to obtain estimators that are both consistent and asymptotically efficient for NL models using choice-based data.
This study reports the results of aggregate air-travel itinerary share models estimated using data from all East West markets in the United States and Canada. These models predict airline ridership at the itinerary level and aid carriers in long and intermediate term decision-making. Official and comprehensive schedule and bookings data is used to estimate generalized extreme value models capturing the inter-itinerary competition dynamic along three dimensions: time of day, carrier and level-of-service (nonstop, direct, single-connect, double-connect). Models incorporate one, two or three of these dimensions simultaneously. Model structures considered include multinomial logit and variations of the nested logit model (two-level nested logit, two-level weighted nested logit, three-level nested logit, three level weighted nested logit and nested weighted nested logit). Independent variables for the models measure various itinerary service characteristics such as level-of-service, connection quality, carrier attributes, aircraft type, and departure time. The results are intuitive, and the advanced models outperform the more basic specifications with regard to statistical tests and behavioral interpretations, giving insight into the competitive dynamic of air-carrier itineraries.
This paper investigates the role of history dependency in explaining activity-travel
patterns of commuters. Specifically, the paper examines the extent to which
one day's activity engagement affects activity frequencies and activity durations
of the next day. The analysis utilizes two-day activity survey data collected
in 1996 in the San Francisco Bay Area. Models of daily activity engagement
and time allocation are estimated as a function of the previous day's activity
pattern to understand day-to-day dependency in activity engagement. Results
from the model estimation effort are used to draw conclusions regarding the
extent to which history dependency exists (within a two-day time frame) in
modeling different activity types. The results suggest that there is a strong
positive history dependency in activity engagement between days within a 48
hour time frame.
Traditionally, mode choice models distinguish between drive-alone and shared ride modes, leaving the network assignment models to predict the assignment of vehicles to toll and HOV facilities. If the shortest generalized cost path in the user-equilibrium assignment is a toll or HOV path, then the trip becomes a toll or HOV trip. This research develops mode choice models that include the use of general-purpose highways; toll roads and HOV lanes simultaneously with choice of the drive-alone and shared ride modes. Multinomial logit and nested logit models are estimated for this full set of alternatives. The models are estimated from a sample of data enriched by special surveys of toll, HOV and transit users in the Houston region. These data provide an empirical basis for studying the behavior of toll and HOV users that is not normally available.
The results indicate that time saved by using these facilities has a higher
utility weight than time differences between other modes. Furthermore, there
is an additional benefit for each mile traveled on a toll or HOV facility,
which is partially offset by any excess in total travel distance necessary
to use the toll or HOV facility. The additional preference for toll and HOV
facilities can be explained by a perception of lower travel time, less driving
stress and/or higher reliability on these facilities. These results suggest
that selecting a least-cost path in trip assignment is not sufficient for
modeling the use of toll and HOV facilities.
This study reports the results of aggregate air-travel itinerary share models
estimated at the city?pair level for all city-pairs in the United States.
These models determine the factors that influence airline ridership at the
itinerary level and support carrier decision-making. The models are estimated
using aggregate multinomial logit methodology and use comprehensive data.
Independent variables for the models measure various itinerary service characteristics:
level-of-service, connection quality, carrier, carrier market presence, fares,
aircraft size and type, and time of day. The results are intuitive, and validation
tests indicate that the models outperform existing methods. Finally, the impacts
of changing various itinerary service attributes on carrier market share are
discussed.
This paper presents research into the development of a utility-based model
of household decision making regarding joint activity participation. Joint
participation represents an important portion of maintenance and leisure activities
and an important element of daily/weekly household travel-activity decision-making.
Joint activities arise from the same sources as independent activities, individual
and household needs, but are motivated by different considerations, including
efficiency, companionship and/or altruism. The importance of each motivating
factor is a function of the type of activity, household roles, employment
status, the number and age distribution of children and automobile ownership.
The types of activities most likely to be pursued jointly are maintenance
and leisure activities, while subsistence activities are least likely to be
pursued jointly. Accordingly, we limit our consideration to joint maintenance
and leisure activities.
We assume that decisions regarding joint activity participation are made in
the context of an entire day or week's schedule of activities for all parties
concerned. Joint activities are thus interdependent with other decisions,
such as time of day, location choice and travel mode choice. The complexity
of this process is likely to create tighter constraints on joint activity
than on individual activity participation due to the need to consider more
than one person's schedule. We expect this complexity to reduce the elasticity
of joint activity decisions, compared with individual activity choices. Scheduling
decisions for joint and independent activities are modeled simultaneously
to represent tradeoffs between activities, joint participation and household
members.
Abstract: Transportation researchers and urban geographers have long recognized that women in two-worker households commute for less time and for shorter distances than their male counterparts. This research develops multinomial logit models of residential location choice for two-worker households in the San Francisco Bay Metropolitan Area to identify household characteristics that account for differences between female and male commuting behavior. The results indicate that the shorter commutes of female workers relative to male workers are most pronounced when children are present. This result is consistent with hypotheses in the literature that argue that the female worker's greater participation in household maintenance and, when children are present, child-rearing responsibilities leads to their shorter commutes. In addition, the occupation of the male worker and the relative order of the last residential change and the last change in the female worker's workplace were found to be important determinants of female and male commuting behavior.
This paper employs extensions to the MNL moel to relax the known characteristic of MNL elasticities being a function of the size of variables which can cause under-sensitivity at low values and over-sensitivity at higher values and demonstrates the advantages of adopting the Generalized Nested Logit formulation in place of the more restrictive MNL and NL Generalized Extreme Value models. The resultant model, which includes distance scaling of distance related variables, the flexible covariance structure of the Generalized Nested Logit model and differential variances for first and second choice stated preference data. The analysis is undertaken in the context of intercity travel with multiple modes and multiple rail service classes. The resultant model strongly rejects more basic MNL and NL models and provides a hgihly intuitive interpretation of intercity choice behavior.
The generalized nested logit model is a new member of the generalized extreme value family of models. The GNL provides a higher degree of flexibility in the estimation of substitution or cross-elasticity between pairs of alternatives than previously developed generalized extreme value models. The generalized nested logit model includes the paired combinatorial logit and cross-nested logit models as special cases. It also includes the product differentiation 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 generalized nested logit model includes the two-level nested logit model as a special case and can approximate closely multi-level nested logit models. The generalized nested logit model accommodates differential 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 confirms the statistical superiority of the generalized nested logit model to the paired combinatorial logit, cross-nested logit and nested logit models and indicates important differences in cross-elasticity relationships across pairs of alternatives.
Key words: Discrete choice; Random utility models; Travel demand; Logit; Intercity Travel
Abstract: This paper develops a conceptual framework for the generation of activity and travel patterns in the context of more general structures and presents an integrated model system as a step toward development of an improved travel demand forecasting model system. We propose a two-stage structure to model activity and travel behavior. The first stage, the stop generation and stop/auto allocation models, consists of the choices for the number of household maintenance stops and the allocation of stops and autos to household members. The second stage, the tour formation model, includes the choices for the number of tours and the assignment of stops to tours for each individual, conditional on the choices in the first stage. Empirical results demonstrate that individual and household socio-demographics are important factors affecting the first stage choices, the generation of maintenance stops and the allocation of stops and autos among household members, and the second stage choices, the number of tours and the assignment of stops to tours.
Random utility maximization discrete choice models are widely used in transportation and other fields to represent the choice of one among a set of mutually exclusive alternatives. The decision maker, in each case, is assumed to choose the alternative with the highest utility to him/her. The utility to the decision maker of each alternative is not completely known by the modeler; thus, the modeler represents the utility by a deterministic portion which is a function of the attributes of the alternative and the characteristics of the decision-maker and an additive random component which represents unknown and/or unobservable components of the decision maker’s utility function.
Early development of choice models was based on the assumption that the error terms were multivariate normal or independently and identically Type I extreme value (gumbel) distributed (Johnson and Kotz, 1970). The multivariate normal assumption leads to the multinomial probit (MNP) model (Daganzo, 1979); the independent and identical gumbel assumption leads to the multinomial logit (MNL) model (McFadden, 1973). The probit model allows complete flexibility in the variance-covariance structure of the error terms but it’s use requires numerical integration of a multi-dimensional normal distribution. The multinomial logit probabilities can be evaluated directly but the assumption that the error terms are independently and identically distributed across alternatives and cases (individuals, households or choice repetitions) places important limitations on the competitive relationships among the alternatives. Developments in the structure of discrete choice models have been directed at either reducing the computational burden associated with the multinomial probit model (McFadden, 1989; Hajivassiliou and McFadden, 1990; Börsch-Supan and Hajivassiliou, 1992; Keane, 1994) or increasing the flexibility of extreme value models.
Two approaches have been taken to enhance the flexibility of the MNL model. One approach, the development of open form discrete choice models is discussed by Bhat in another chapter of this handbook. This chapter describes the development of closed form models which relax the assumption of independent and identically distributed random error terms in the multinomial logit model to provide a more realistic representation of choice probabilities.
(please click here to open abstract in doc format)
The Independence of Irrelevant Alternatives (IIA) property of the multinomial logit (MNL) model imposes the restriction of zero covariance between the utilities of pairs of alternatives. This restriction is inappropriate for many choice situations; those in which some pairs or sets of alternatives share the same unobserved attributes. The nested logit (NL) model relaxes the zero covariance restriction of the MNL model but imposes the restriction of equal covariance among all alternatives in a common nest and zero covariance otherwise. The paired combinatorial logit (PCL) model relaxes these restrictions further by allowing different covariances for each pair of alternatives. This relaxation enables the estimation of differential competitive relationships between each pair of alternatives. The closed form of the PCL model retains the computational advantages of other logit models while the more flexible error correlation structure, compared to the MNL model and NL models, enables better representation of many choice situations. This paper describes the derivation, structure, properties and estimation of the PCL model. The empirical results demonstrate that the PCL model is statistically superior to the MNL and NL models and may lead to importantly different travel forecasts and policy decisions.
Abstract : The adoption of disaggregate analysis in transportation and other fields has led to widespread use of choice models to describe the influence of the characteristics of decision makers and the attributes of alternatives on choices. The multinomial logit model (MNL) is most used due to its relative simplicity, potential to add new alternatives, its ease of estimation and the wide availability of estimation software. However, concerns about the restrictive assumptions of the MNL (independent and identical distribution of error terms) and its properties has led to a search for more robust model structures. The nested logit (NL) model has become widely used in a variety of contexts due to its ability to represent similarities (covariance of error terms) among groups of alternatives. It is not widely recognized that there are two different forms of the nested logit model. These different models, which we describe as the utility maximization nested logit (UMNL) and the non-normalized nested logit (NNNL), have very different properties which produce different estimation results, behavioral interpretations and forecasts. The use of nested logit estimation requires a thoughtful choice between these different model structures. This paper describes and compares the model structure and properties of the UMNL and NNNL models and illustrates the differences between them analytically and empirically. While the selection of one or the other structure is a matter of judgement, we prefer the UMNL model because it is based on utility maximization and it has intuitively reasonable elasticity relationships and interpretation of utility parameters across alternatives.
The central basis of the activity-based approach to travel demand modeling is that individuals' activity-travel patterns are a result of their time-use decisions within a continuous time domain. This paper reviews theoretical and empirical research in the time-use area, emphasizing the need to examine activities in the context or setting in which they occur. The review demonstrates that substantial progress has been made in the past five years and identifies some reasons for this the accelerated rate of progress in thus field. The paper concludes that the field of time-use and its relevance to activity-travel modeling has gone substantially past the "tip of the iceberg", though it certainly still has a good part of the "iceberg" to uncover. Important future areas of research are identified and discussed.
Abstract: In the development of residential location choice models, multi-collinearity among spatial socioeconomic status and family status attributes prohibits the inclusion of a large number of these attributes in a single specification. This paper tests alternative methods of representing these characteristics with a smaller number of summary or representative measures. Factor analysis is performed on the tract-level attributes to produce factor scores that are used in residential location choice models. The factor analysis results are also used to select one representative socioeconomic status variable and one representative family status variable that are used in choice models. These approaches are tested against a base model that uses a full set of eleven attributes. To incorporate systematic taste variation, the factors and representative attributes are interacted with household income and life cycle variables. The factor analysis approach provides a clearer interpretation of the influence of spatial variables than inclusion of multiple variables, which obtains many non-significant and counter-intuitive parameters, or use of a single representative variable, which is not readily interpreted with respect to impact of the variables it represents.
Since the beginning of civilization, the viability and economic success of communities have been, to a major extent, determined by the efficiency of the transportation infrastructure. To make informed transportation infrastructure planning decisions, planners and engineers have to be able to forecast the response of transportation demand to changes in the attributes of the transportation system and changes in the attributes of the people using the transportation system. Travel demand models are used for this purpose; specifically, travel demand models are used to predict travel characteristics and usage of transport services under alternative socio-economic scenarios, and for alternative transport service and land-use configurations.
The need for realistic representations of behavior in travel demand modeling
is well acknowledged in the literature. This need is particularly acute today
as emphasis shifts from evaluating long-term investment-based capital improvement
strategies to understanding travel behavior responses to shorter-term congestion
management policies such as alternate work schedules, telecommuting, and congestion-pricing.
The result has been an increasing realization in the field that the traditional
statistically-oriented trip-based modeling approach to travel demand analysis
needs to be replaced by a more behaviorally-oriented activity-based modeling
approach.
Two distinctly different nested logit (NL) models have been widely used in both research and applications. The differences, not widely recognized, between these models will substantial influence estimation results, behavioral interpretation and policy analysis. The McFadden NL is derived from random utility theory; the Daly or non-normalized NL 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 different NL models identifying important differences in both direct and cross-elasticities of the different model structures. An empirical application demonstrates that the different NL models produce dramatically different results with respect to nesting structure and the relative importance of utility components. Thus, the selection of one or another of the NL models has important consequences for model interpretation and prediction with consequent impacts on policy analysis. The McFadden model is preferred because of its basis in utility theory, intuitively reasonable elasticity relationships and a clear interpretation of utility function parameters across alternatives.