Nazneen Ferdous

A comparison of the maximum simulated likelihood and composite marginal likelihood estimation approaches in the context of the multivariate ordered-response model

This chapter compares the performance of the maximum simulated likelihood (MSL) approach with the composite marginal likelihood (CML) approach in multivariate ordered-response situations. The ability of the two approaches to recover model parameters in simulated data sets is examined, as is the efficiency of estimated parameters and computational cost. Overall, the simulation results demonstrate the ability of the CML approach to recover the parameters very well in a 5–6 dimensional ordered-response choice model context. In addition, the CML recovers parameters as well as the MSL estimation approach in the simulation contexts used in this study, while also doing so at a substantially reduced computational cost. Further, any reduction in the efficiency of the CML approach relative to the MSL approach is in the range of nonexistent to small. When taken together with its conceptual and implementation simplicity, the CML approach appears to be a promising approach for the estimation of not only the multivariate ordered-response model considered here, but also for other analytically intractable econometric models.

A spatial panel ordered-response model with application to the analysis of urban land-use development intensity patterns

This paper proposes and estimates a spatial panel ordered-response probit model with temporal autoregressive error terms to analyze changes in urban land development intensity levels over time. Such a model structure maintains a close linkage between the land owner’s decision (unobserved to the analyst) and the land development intensity level (observed by the analyst) and accommodates spatial interactions between land owners that lead to spatial spillover effects. In addition, the model structure incorporates spatial heterogeneity as well as spatial heteroscedasticity. The resulting model is estimated using a composite marginal likelihood (CML) approach that does not require any simulation machinery and that can be applied to data sets of any size. A simulation exercise indicates that the CML approach recovers the model parameters very well, even in the presence of high spatial and temporal dependence. In addition, the simulation results demonstrate that ignoring spatial dependency and spatial heterogeneity when both are actually present will lead to bias in parameter estimation. A demonstration exercise applies the proposed model to examine urban land development intensity levels using parcel-level data from Austin, Texas.

Modeling the Influence of Family, Social Context, and Spatial Proximity on Use of Nonmotorized Transport Mode

This paper presents a joint model of the duration of walking and bicycling activity with the use of a hazard-based specification that recognizes the interval nature of time reported in activity–travel surveys. The model structure takes the form of a multilevel hazard-based model system that accounts for the range of interactions and spatial effects that might affect walking and bicycling mode use. In addition to the individual-specific factors, family (household-specific) interactions, social group (peer) influences, and spatial clustering effects are considered potential factors that contribute to heterogeneity in nonmotorized transport mode use behavior. The model system presented is capable of accommodating grouped duration responses often encountered in activity–travel surveys. A composite marginal likelihood estimation approach is adopted to estimate parameters in a computationally tractable manner. The model system is applied to a survey sample drawn from the recent 2009 National Household Travel Survey in the United States. Model results show that significant unobserved family-level, social group, and spatial proximity effects contribute to heterogeneity in walking and bicycling activity duration. The unobserved effects were also found to have a differential impact on bicycling activity duration, thus suggesting the need to treat and model walking and bicycling separately in transportation modeling systems.

Characteristics of Premium Transit Services that Affect Choice of Mode

Traditionally, travel models use travel time and cost to assess the usefulness of each mode of transportation to make a particular trip. Other factors that affect the selection of mode are accounted for using a single constant term that represents other attributes. In many cases, these attributes represent conditions that may not be the same for all trips. Travel forecasting models would benefit by incorporating an expanded list of non-traditional attributes so that the probability of using transit to make a trip is more specifically related to the characteristics of a potential transit journey. Potential non-traditional transit characteristics include on-board and station amenities, reliability, span of service, and service visibility/ branding. These characteristics are not typically directly considered in travel forecasting models. This research sought to improve the understanding of the full range of determinants for transit travel behavior and to offer practical solutions to practitioners seeking to represent and distinguish transit characteristics in travel forecasting models. The key findings of this research include the value of non-traditional transit service attributes on travelers’ choice of mode, in particular the influence of awareness and consideration of transit service on modal alternatives, and the importance of traveler attitudes toward both awareness and consideration of transit and on the choice of transit or auto in mode choice. The appendices present detailed research results including a state-of-the-practice literature review, survey instruments, models estimated by the research team, model testing, and model implementation and calibration results. The models demonstrate an approach for including non-traditional transit service attributes in the representation of both transit supply (networks) and demand (mode choice models), reducing the magnitude of the modal specific constant term while maintaining the ability of the model to forecast ridership on specific transit services. The testing conducted in this project included replacing transit access and service modes, such as drive to light rail or walk to local bus, as alternatives in the mode choice model with transit alternatives defined by the elements of the path, such as a short walk to transit path, a no-transfer path, or a premium service path.

Integration of National Long-Distance Passenger Travel Demand Model with Tennessee Statewide Model and Calibration to Big Data

The Tennessee Department of Transportation replaced the quick-response-based long-distance component in its statewide model by integrating the new national long-distance passenger travel demand model in a new statewide model and calibrating it to long-distance trips observed in cell phone origin–destination data. The national long-distance model is a tour-based simulation model developed from FHWA research on long-distance travel behavior and patterns. The tool allows the evaluation of many policy scenarios, including fare or service changes for various modes, such as commercial air, intercity bus, Amtrak rail, and highway travel. The availability of this tool presents an opportunity for state departments of transportation in developing statewide models. Commercial big data from cell phones for long-distance trips also pre-sents an opportunity and a new data source for long-distance travel patterns, which previously have been the subject of limited data collection, in the form of surveys. This project is the first to seize on both of these opportunities by integrating the national long-distance model with the new Tennessee statewide model and by processing big data for use as a calibration target for long-distance travel in a statewide model. The paper demonstrates the feasibility of integrating the national model with statewide models, the ability of the national model to be calibrated to new data sources, the ability to combine multiple big data sources, and the value of big data on long-distance travel, as well as important lessons on its expansion.

Multimodal Trip Generation Model to Assess Travel Impacts of Urban Developments in the District of Columbia

The research effort described in this paper aims to develop a state-of-the-practice methodology for estimating urban trip generation from mixed-use developments. The District Department of Transportation’s initiative focused on (a) developing and testing a data collection methodology, (b) collecting local data to complement the ITE’s national data in trip rate estimation, and (c) developing a model–tool that incorporates contextual factors identified as affecting overall trip rate as well as trip rate by mode. The final model accurately predicts total person trips and mode choice. The full set of models achieves better statistical performance in relation to average model error and goodness of fit than either ITE rates alone or other existing research. The model includes sensitivity to local environment and on-site components. The model advances site-level trip generation research in two major ways: first, it calculates total person trips independent of mode choice; second, it calculates mode choice with sensitivity to the amount of parking provided on site—a major finding in the connection between parking provision and travel behavior at a local-site level. The methodology allows agencies to improve their assessment of expected trips from proposed buildings and therefore the level of impact a planned building may have on the transportation system.

Implementation of a Practical Model System to Predict Long-Distance Travel for the Entire U.S. Population

This paper describes a model system designed and implemented to simulate long-distance travel for all U.S. households. The model system was created in the final phase of FHWA research project Foundational Knowledge to Support a Long-Distance Passenger Travel Demand Modeling Framework. It is a tour-based system simulating individual tours for individual households from a synthetic population. The models are disaggregate models of auto ownership, tour generation, tour duration, travel party size, tour destination choice, and tour mode choice. The model system runs in 1 to 2 h on a standard PC, simulating a full year’s long-distance tours for the entire U.S. population. The paper describes the model structure, input data, and software implementation and provides some results from the initial model calibration, validation, highway assignment, and sensitivity tests.

Transferability of Activity-Based Model Parameters

The Strategic Highway Research Program 2 (SHRP 2) program developed proof-of-concept Dynamic Integrated Models in partnership with planning organizations in Sacramento, California, and Jacksonville, Florida. “Dynamic Integrated Model” refers to an activity-based travel demand model linked with a feedback loop to a Dynamic Traffic Assignment (simulation) model. The goal of that research was to improve urban-scale modeling and network procedures to address operations or spot improvements that affect travel-time choice, route choice, mode choice, reliability, or emissions. Building a new activity-based model set for transportation planning is an expensive and time-consuming commitment. The objective of this research was to determine if activity-based model parameters can be successfully transferred from one community to another. If transfer of parameters could be shown to produce reasonable results, it could save development time and money. DaySim, an activity-based travel demand model originally developed in Sacramento, California, was applied to Jacksonville, Florida, with Sacramento parameters and then calibrated to the Jacksonville environment. DaySim was also applied to Tampa, Florida, with Sacramento parameters and then calibrated with local data. A statistical analysis was performed to identify significant differences between transferred parameters and parameters developed from local data. Variations in model performance on validation tests were also evaluated. The analyses identified specific model components that would be better transferred than re-estimated and others for which it would be better to re-estimate. A model with borrowed parameters must still be calibrated against local conditions. A significant finding of the research was that there must be a good match between the complexity of the source model to be transferred and the depth and coverage of data available for calibrating at the destination site. A second finding was that urban areas must be similar in key demographics such as household size, age, income, auto ownership, and trip purposes. This report will be of particular interest to planning organizations considering development of an activity-based travel demand model and, in general, to professionals who use travel demand models as part of the transportation planning process.

Comparison of Four-Step Versus Tour-Based Models for Prediction of Travel Behavior Before and After Transportation System Changes

The main objective of this study was to examine the performance of the Mid-Ohio Regional Planning Commission trip-based and tour-based frameworks in the context of three specific projects started and completed within the past 20 years in the Columbus, Ohio, metropolitan area. Region- and project-level comparisons of the performance of the trip- and tour-based models were made for three scenario years: 1990, 2000, and 2005. The region-level analysis was undertaken in the context of four travel dimensions on the basis of data availability and observed data to model output compatibility. These four dimensions were vehicle ownership, work flow distribution, work flow distribution by time of day, and average work trip travel time. The tour-based model performed better overall than the trip-based model for all four dimensions. The project-level comparative assessment of the predicted link volumes from the trip-based and the tour-based models was undertaken according to the observed link counts and roadway functional class. The results did not show any clear trends according to the performance of the models by functional class or year.