Data Fusion

Developing a Big Data fusion tool to understand travel demand trends and measure transport project impacts.

This research, initially funded by the San Francisco County Transportation Authority, aimed to develop software tools to support the fusion and analysis of large, passively collected data sources for the purpose of measuring and monitoring transit system performance. This study used San Francisco as a case study, taking advantage of the automated vehicle location (AVL) and automated passenger count (APC) data available on the city transit system. These data are expanded General Transit Feed Specification (GTFS) data, which were used as a measure of the full set of scheduled transit service. Reporting and visualization tools were developed to explore transit performance changes over time.

Subsequent research funded by University College London, built upon these tools to incorporate longitudinal data on changes in highway conditions and in the drivers of transportation demand, such as households, employment and fuel price. The highway data component uses GPS traces from a fleet of 500 taxis to impute roadway speeds on streets in San Francisco. The resulting data visualizations quickly highlight the choke points in the network and are used to monitor the changes that occur through time.

The combined system was applied to quantify the factors contributing to the divergent ridership trends on BART rapid transit versus MUNI bus in recent years. It suggests that emerging shared mobility and active transport modes, combined with demographic changes may serve as a greater drag on bus ridership than on rail.

Close Project

Bicycle Route Choice

Using smartphone location data to estimate the air quality benefits of bicycle infrastructure.

The Association for Monterey Bay Area Governments (AMBAG) sought to develop a model to estimate the emissions reduction benefits of building new bicycle facilities. The model needed to be implemented as a stand-alone tool that could be freely distributed to its member agencies. This challenge was met by developing a model that combined bicycle route choice with an incremental logit bicycle mode choice model and an emissions calculator. GPS traces from bicycle users, collected via the CycleTracks app, were used to estimate the route choice preferences for different types of bicycle facilities. The California Household Travel Survey was used to estimate a scaling coefficient on the bicycle utility for use in mode choice. The stand-alone model was implemented in an Adobe ActionScript graphical user interface.

Close Project

Activity-Based Modeling

Updating an activity-based model to study congestion pricing in San Francisco.

The San Francisco County Transportation Authoirty (SFCTA) sought to use its existing activity-based travel model, SF-CHAMP, to study the impact of area and cordon pricing policies in downtown San Francisco. To accomplish this, a set of model improvement were undertaken to improve the models treatment of pricing. These including expanding the geographical bounds of SF-CHAMP to cover the 9-county Bay Area, estimating new time-of-day models that are properly responsive to peak period pricing, and leveraging the microsimulation nature of the models to draw values of time from a continuous distribution rather than using averages. This value-of-time improvement was the first of its kind in an activity-based model and allowed the model reflect the full diversity of the sample population with respect to their willingness to pay tolls.

Close Project

Statewide Modeling

Assessing the utility and costs of statewide travel demand models.

This research aims to assess the expected cost and utility that can be expected to accrue from developing or upgrading a statewide travel demand model.

The core audience for this report is technical or planning staff at state transportation agencies who must make a recommendation on whether and how to engage in statewide modeling at their agency. To make a recommendation, those individuals need to specify a budget request both in terms of external costs (consultants and materials) and in terms of staff time. They also need to be able to, at a minimum, articulate, and preferably quantify, the value they expect to achieve by pursuing the proposed approach.

This ongoing research will quantify the costs of statewide models, and both identify and quantify the benefits of statewide models in a range of situations. It will accomplish this using a novel approach that combines data on the revealed outcomes of existing statewide models, with the collective professional judgment of statewide modelers. This will be done by offering statewide modelers a scenario-based survey in which they are asked to estimate the costs of a potential modeling project, as well as the utility of the model for different types of applications. The project involves extensive outreach to all 50 state DOTs to administer the survey as a telephone-based interview.

This is the only study we are aware of to provide a rigorous measure of the utility of different types of models for different applications, and provides a template from which to start the utility assessment for forecasts.

Close Project

Zephyr Foundation

Launching a non-profit foundation for improving travel analysis methods.

The Zephyr Foundation’s mission is to advance rigorous transportation and land use decision-making for the public good by advocating for, facilitating, and supporting improved travel analysis and forecasting methods. It was born from a series of workshops aimed at finding ways to make travel modeling and travel forecasting more scientifically rigorous. By scientifically rigorous, we mean that it is transferrable, transparent and subject to empirical testing of what works. The goal is systematic improvement, avoiding a random walk between a variety of novel methods, without the evaluation necessary to choose the most promising.

Greg Erhardt is one of a small group of instigators who have been working to advance this effort. It is taking the form of a not-for-profit foundation as a vehicle for pushing the industry forward. This structure provides a home for the effort independent of individual commercial interests, and allows for a greater degree of flexibility than government agencies can achieve.

Close Project

Transit Smart Card Data

Evaluating transit smart card data with privacy restrictions and limited penetration rates.

This research analyzes data from the Clipper Card system in the San Francisco Bay Area, and provides evidence for other agencies seeking to understand the value and limitations of their own data. It considers how the data can be both useful to the transportation planning process, while a the same time respectful of the privacy of the users. The evaluation goes on to compare this new data set to onboard transit survey data for the same transit systems. It finds that the smart card data under-represent minority and low-income travelers, relative to the onboard surveys, potentially creating equity issues if the data are used in planning without accounting for these biases. Broadly, this is an exploration of how to take advantage of both Big Data sources and more traditional transportation surveys.

Close Project

Optimal, Responsive and Efficient Transit

Transit-Serving Communities Optimally, Responsively and Efficiently (T-SCORE), a Tier 1 University Transportation Center.

The T-SCORE Center aims to define a set strategic visions that will guide public transportation into a sustainable and resilient future, and to equip local planners with the tools needed to translate their chosen vision into their own community. Those strategic visions might include a focus on serving those riders who need transit the most, consolidation into high-volume capacity-constrained corridors, integrated on-demand multi-modal transit, or multi-modal pricing and incentive strategies. The visions must necessarily include a COVID-19 recovery strategy.

A core component of this research is the Multi-Modal Optimization and Simulation Track, which builds upon our previous NSF-funded research to optimize the allocation of fixed-route and on-demand transit vehicles. We are integrating this optimization with a multi-agent simulation (MATSim) that considers competition from ride-hailing, which we found to be a key contributor to declining transit ridership. MATSim provides a dynamic traffic simulation that works for large regions, and allows agents to choose their route, time-of-day and mode within the framework. However, it requires daily activity patterns as input, which we are reading from ActivitySim. This combined framework will provide transit agencies with a starting point corresponding to each strategic direction, enabling them to be nimbler in adapting to changing conditions. We are developing a prototype for San Francisco and Salt Lake City, with the expectation that it can be deployed more broadly in the future.

The T-SCORE Center is a USDOT-funded University Transportation Center involving a partnership between Georgia Tech, University of Kentucky, University of Tennessee and Brigham Young University. Workforce development through enhanced educational activities, stakeholder engagement, and technology transfer are key elements of T-SCORE at all stages of its work.

Close Project

Causes of Transit Ridership Decline

Analyzing the causes of the recent transit ridership decline and evaluating the effectiveness of potential responses.

In 2019, transit ridership in the United States declined for the fifth consecutive year. Buses were the most affected with the lowest ridership levels since the 1970’s. Even rail declined the last few years following an upward trend since 2009. As ridership declines, agencies lose fare revenue and often reduce service to meet budgets, resulting in further ridership losses.

The recent decline in transit ridership is particularly worrisome because traditional factors of ridership do not seem to be involved. Although US transit agencies experienced drastic service cuts following the recession, overall vehicle revenue miles rebounded to their 2010-level by 2015 and have kept growing ever since. Meanwhile, urban population and employment rates, which are both typically associated with high ridership have risen substantially in the same period.

Transit Cooperative Research Program (TCRP) A-43: Recent Decline in Public Transportation Ridership: Analysis, Causes, Responses has been the most comprehensive effort to understand transit ridership change. The project employs a two-phase research approach that considers changes at the system, route and stop level, then runs simulation models to test the effectiveness of potential responses.

Close Project

Traffic Forecast Accuracy

Measuring the historic accuracy of project-level traffic forecasts.

The evidence on forecast accuracy remains limited, with only a small set of empirical studies examining non-toll traffic forecast accuracy in the United States. A major barrier to such research has been the lack of data, largely because it is cumbersome to compile information on forecasts made years earlier. This research aims to fill that gap, focusing specifically on project-level traffic forecasts of public roads in the US. It assembles the largest known database of traffic forecast accuracy, composed of information about traffic forecasts and about measured outcomes after the projects open. It reports on the accuracy of these forecasts and factors related to accuracy. It goes on to consider a series of case studies aimed at providing better understanding of the sources of forecast inaccuracy. Together, these provide empirical evidence about the accuracy of past traffic forecasts. Recognizing that no forecasts will be perfectly accurate, it is prudent to quantify the expected inaccuracy around traffic forecasts and consider that uncertainty in making decisions. This research provides a means of estimating the range of uncertainty around a forecast using a technique called quantile regression.

An output of this research is a guidance document providing recommendations for state departments of transportation, metropolitan planning organizations, or other agencies that produce project-level traffic forecasts. In addition to describing the recommended method for quantifying the uncertainty around a traffic forecast, the guidance documents describes a process for the continual improvement of forecast accuracy. That process involves compiling the relevant data, periodically reporting forecast accuracy, and using past accuracy assessments to improve traffic forecasting methods.

Close Project

Ride-Hailing in San Francisco

Measuring the effect of ride-hailing on congestion, transit ridership and traffic safety in San Francisco.

This research involves assess the effects of Transportation Network Companies (TNCs), such as Uber and Lyft, on the transportation system in San Francisco. It focuses on three related effects: congestion, transit ridership, and traffic safety.

The advent and assimilation of Transportation Network Companies (TNC’s) into the urban traffic ecosystem affects urban transportation in a different manner than the extant transport components in terms of lane occupancy, curb occupancy, parking requirements, trip generation stimuli, etc. Their effect in these three dimensions is not immediately clear, and depends on the modes that TNC users otherwise would have used, as well as other changes in travel behavior and the operational characteristics of the roadways. This assessment is conducted empirically, for the period from 2010 to 2016, corresponding to the emergence of TNCs. It does so using a unique set collected by researchers at Northeastern University from the Application Programming Interfaces (APIs) of Uber and Lyft to identify the location of TNC vehicles.

The results of this research have been covered in media publications worldwide, including in The Boston Globe, The Wall Street Journal, The Ken NPR's Science Friday.

Close Project

A Linked-Longitudinal American Community Survey

Linking restricted-use Census data to measure changes in communting behavior, teleworking and gentrification.

Past research on commuting behavior typically observes people at only one point in time. That approach lends valuable insight, but also imposes severe limitations on the kinds of inferences that can be made about commuting behavior. This new project overcomes those limitations by drawing on data collected in the American Community Survey (ACS). The ACS is an annual survey of about 2 million households conducted by the US Census Bureau since 2005. Among other data, the ACS records where people live, where they work, what they earn, and how they get to work. Utilizing a secure data center to protect privacy, this project identifies and links the responses of individuals who happen to be surveyed in more than one year, creating a longitudinal data set. By observing the change in commuting behavior over a period long enough to observe changes in transportation infrastructure, the resulting data set overcomes the limitations of surveys that observe people at only one point in time.

These new data are used to address four questions: 1) By how much can changes to the residential built environment reduce car commuting? 2) How does teleworking affect commuting distance? 3) How does teleworking affect wage growth? And 4) Who is leaving transit-rich areas, who is replacing them, and how do their commutes change? The data and analyses are made widely available through the network of 33 Federal Statistical Research Data Centers, enabling new directions for future research on a broad range of transportation, economic and social outcomes.

Close Project

VMT Reduction through Mode Shift

Using a smartphone-enabled travel survey measure mode shift potential in the Twin Cities.

In this study, we consider the potential to reduce vehicle miles traveled (VMT) in the Twin Cities region that might be achieved by incentivizing people to switch from traveling in cars to walking, biking or riding transit. We do so by estimating the maximum feasible shift of car trips to alternative modes given existing land-use and travel patterns, as observed in 380,000 real-world trips reported in the 2019 and 2021 Travel Behavior Inventory (TBI) surveys. For each observed car trip, we calculate the best available walk path, bike path, and transit path, considering the quality of the available infrastructure and transit service. We consider an option to be feasible if we observe a substantial number (5% or more) of people using that mode under similar circumstances, and if the traveler has sufficient time to complete the trip within their travel schedule. Further, we consider a feasible option to be competitive if its travel time is within 15 minutes of the car trip it would replace.

Close Project

Induced Vehicle Travel

Measuring changes resulting from real-world projects to quantify the "fundamental law of road congestion".

Induced vehicle travel is the increase in car and truck travel attributable to an increase in road capacity. The link between highway capacity expansion and increased VMT is both consistent with economic theory and supported by strong empirical evidence. However, a disconnect remains between this empirical evidence of induced travel and the travel demand models used to forecast traffic volumes and speeds when evaluating proposed road projects. This research will empirically measure induced travel for real-world roadway projects and compare those estimates to the results produced by travel demand forecasts.

Close Project