U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
202-366-4000


Skip to content
Facebook iconYouTube iconTwitter iconFlickr iconLinkedInInstagram

Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations

 
REPORT
This report is an archived publication and may contain dated technical, contact, and link information
Back to Publication List        
Publication Number:  FHWA-HRT-13-022    Date:  August 2013
Publication Number: FHWA-HRT-13-022
Date: August 2013

 

Synthesis of Traveler Choice Research: Improving Modeling Accuracy for Better Transportation Decisionmaking

Travel Behavior Models Review

Activity-Based Models

The advantages of activity-based models are diverse but can be organized in to the following four categories, which are later discussed in more detail:

·        They can identify the influence of time, destination, and mode on trip attributes.

·        They capture long-term behavior, overcoming the limitations of tour-based models by including activity patterns outside the daily schedule in addition to time dependency, destination, and mode.

·        They capture certain characteristics of individual-based decisions beyond aggregate traffic analysis zones.

·        They capture short-term decision shifts that may have substantial impacts at the network level. Linkage of interpersonal decisions that are crucial to policies such HOV lanes can be taken into account.

These are also the main reasons to incorporate an activity-based framework rather than a trip- or tour-based framework.

Within activity-based travel demand models, there are two primary modeling paradigms: utility-based econometric models and rule-based computational process models. These models are not exclusive but have different philosophical approaches and are therefore fundamental in how the travel generation process is understood. Combinations of the two models can be found within agent-based simulation models, which are discussed later in this report.

Utility-Based Econometric Models

Utility-based econometric models have their roots in economic consumer choice theory, which says that individuals maximize their utility from the choices they make. These models consist of a number of different choice-based models for the individual’s activity-travel decisions. These models can be enriched by other utility-based models, such as hazard models for time durations. The set of economic equations builds the structure to model the relationships among the traveler characteristics, the network characteristics that allow the individual to travel, and the environment characteristics that describe the place to perform activities and further restrictions on the traveler’s behavior.

The following are existing utility-based econometric activity-based model systems:

·        Greater Portland METRO.(88)

·        San Francisco County Transportation Authority.(89)

·        New York Metropolitan Transit Council.(90)

·        Mid-Ohio Regional Planning Commission (Columbus, OH).(91)

·        Sacramento Area Council of Governments.(92)

·        Atlanta Regional Commission.(93)

·        Comprehensive Econometric Micro-Simulator for Daily Activity-Travel Patterns.(94)

·        Florida Activity Mobility Simulator.(95)

All of the listed models can be categorized as full individual day pattern models or linked full individual day pattern models. The full individual day pattern models follow the concept of an overarching daily activity-travel pattern proposed by Bowman and Ben-Akiva.(96) These models are based on an underlying system of multinomial logit or nested logit models in a particular hierarchy. The linked full individual day pattern models enhance the models previously described by allowing intrahousehold interactions in activity-travel engagement models. The Columbus, OH, and Atlanta, GA, models are examples of such models.

Types and Advances

To meet expectations and improve analyses, one of the natural responses from transportation modelers is to develop more sophisticated modeling forms, ranging from simple logit and probit to nested and mixed logit to Bayesian procedures and so on. The travel behavior model evolved from simple regressions based on aggregate, revealed preference (RP) data to more sophisticated mathematical functions based on disaggregate, SP data.(97) For example, the logit family of models progressed from the binary logit model to the multinomial logit model or the conditional logit model and then to the nested logit model and the mixed logit model. As early as 1972, the binary logit model was utilized in intercity travel mode choice. The multinomial logit model was primarily used to model multiple choices. Ben-Akiva derived the nested logit model that is designed to capture correlations among alternatives.(98) At the time of this report, mixed logit is considered the most promising discrete choice model that is intuitive, practical, and powerful. It combines the flexibility of probit with the tractability of logit.

At the same time, a large number of transportation professionals have devoted effort to identifying and evaluating major utility factors besides time and cost to improve the predictability of utility functions. Hensher used an early example to incorporate comfort and convenience in a travel mode choice model.(99) Algers et al. included comfort and convenience in a study on the value of travel time.(100) Later, Liu et al. proposed a conceptual framework that includes travel time, monetary cost, comfort/convenience, and safety/security in the travel choice models.(101) In a recent choice model, Ben-Akiva et al. integrated latent variables to model attitudes and perceptions and their influence on choices.(102)

The underlying concept to understand behavior changes from operational interventions has to account for the problem that behavior switching is not the same as observation of a certain behavior. Behavior switching is less rational, as travelers tend to stick with what they are used to, which can be described as inertia. The decision rule is no longer based only on maximizing the utility but includes a switch if the utility of switching exceeds a threshold. The generalized indifference band framework is shown in figure 2.

 

Pr open parenthesis X equals 1 close parenthesis if and only if Pr open parenthesis U is greater than or equal to 0 close parenthesis.

Figure 2. Equation. Generalized Indifference Band Framework.

 

The random utility formulation has to account for time and location at different days and can be formulated as shown in figure 3.

 

U subscript ijt equals V subscript ijt plus epsilon subscript ijt.

Figure 3. Equation. Random Utility Formulation.

Where:

Vijt = f (Zi, Nijt, NNijt, Vijt).

Zi = Traveler characteristics.

Nijt = Time-dependent network attributes by decisionmaker i at decision location j at day t.

NNijt = Time-dependent non-network characteristics by decisionmaker i at decision location j at day t.

Vijt = Time-dependent vehicle characteristics by decisionmaker i at decision location j at day t.

εijt  = Error terms correlated over different times of day, locations, and days.

To incorporate interpersonal interactions, a utility maximization approach can be included. Each individual’s utility is calculated jointly and singly, and a joint decision is made if the difference of the best alternatives exceeds a certain threshold, as shown in figure 4 and figure 5.

 

Delta U subscript si equals the sum of U subscript ijt subscript single minus the sum of U subscript ijt subscript joint.

Figure 4. Equation. Joint Versus Single Utility Threshold.

 

Delta U subscript T equals the sum of open parenthesis alpha subscript i times U subscript si plus ellipsis plus alpha subscript k times U subscript sk close parenthesis.

Figure 5. Equation. Weighted Sum of Joint Versus Single Utility Threshold.

Where:

α = Each individual’s weight of influence.

Rule-Based Computational Process Models

Rule-based computational process models are developed on the premise that individuals do not always act rationally and so do not maximize their utility. Instead, individuals rely on a process that contains complex if-then rules to solve a task, similar to a production system model. These models have problems describing the statistical significance of the factors that affect the rules and are therefore not always best for understanding behavior changes based on experiments or for predicting future changes. The following are existing rule-based computational process activity-based model systems:

·        CARLA: Combinatorial Algorithm for Rescheduling Lists of Activities.(103)

·        STARCHILD: Simulation of Travel/Activity Responses to Complex Household Interactive Logistic Decisions.(104)

·        SCHEDULER.(105)

·        AMOS: Activity-Mobility Simulator.(106)

·        SMASH: Simulation Model of Activity Scheduling Heuristics.(107)

·        ALBATROSS: A Learning-Based Transportation Oriented Simulation System.(108)

·        TASHA: Travel Activity Scheduler for Household Agents.(109)

Agent-Based Modeling Systems

The Transportation Analysis and Simulation System (TRANSIMS) and the Multi-Agent Transport Simulation Toolkit (MATSim) are agent-based activity-based modeling systems that were originally designed to account for the full disaggregate representation of individual travel behavior.(110,111) They were mainly developed to capture characteristics of individual-based decisions beyond aggregate traffic analysis zones and short-term decision shifts that may have substantial impact at the network level in conjunction with dynamic traffic assignment. Thus, the feedback is based on aggregated system information, not on specific agent information. If one is not interested in the Nash equilibrium, traditional dynamic and non-dynamic traffic assignment approaches fail because there is no access to human behavior.(112) Since these agent-based systems do not find a clear equilibrium, it is unclear what solution they produce. Further, by decoupling the demand side (activity-based models) from the supply side (either assignment or simulation models), the activity models typically compute probabilities for a large number of alternatives, which demands an explicit choice set. Accounting for such alternative sets in real-size networks would result in very long computation times.(113)

 

Federal Highway Administration | 1200 New Jersey Avenue, SE | Washington, DC 20590 | 202-366-4000
Turner-Fairbank Highway Research Center | 6300 Georgetown Pike | McLean, VA | 22101