An empirical study of the deviation between actual and shortest travel time paths

Shortest distance and time route vs. actual route for one person on one day.
Shortest distance and time route vs. actual route for one person on one day.

Recent working paper:

Few empirical studies of revealed route characteristics have been reported in the literature. This study challenges the widely applied shortest-path assumption by evaluating routes followed by residents of the Minneapolis–St. Paul metropolitan area, as measured by the GPS Component of the 2010 Twin Cities Travel Behavior Inventory conducted by the Metropolitan Council. It finds that most travelers used paths longer than the shortest path. This is in part a function of trip distance, trip circuity, number of turns, and age of the driver. Some reasons for these findings are conjectured.

Do People Use the Shortest Path? An Empirical Test of Wardrop’s First Principle

Recently published

Fig 5. The number of speed observations on each link during the entire study period.
Fig 5. The number of speed observations on each link during the entire study period.


Most recent route choice models, following either the random utility maximization or rule-based paradigm, require explicit enumeration of feasible routes. The quality of model estimation and prediction is sensitive to the appropriateness of the consideration set. However, few empirical studies of revealed route characteristics have been reported in the literature. This study evaluates the widely applied shortest path assumption by evaluating routes followed by residents of the Minneapolis—St. Paul metropolitan area. Accurate Global Positioning System (GPS) and Geographic Information System (GIS) data were employed to reveal routes people used over an eight to thirteen week period. Most people did not choose the shortest path. Using three weeks of that data, we find that current route choice set generation algorithms do not reveal the majority of paths that individuals took. Findings from this study may guide future efforts in building better route choice models.

Analyzing Multiday Route Choice Behavior using GPS Data

Recent working paper

Three Layer Neural Network for Traveler Classification
Three Layer Neural Network for Traveler Classification

Understanding variability in daily behavior is one of the most important missions in travelbehavior modeling. In traditional method, in order to find the differences, respondents were asked to list the used multiday paths. The quality of results is sensitive to the accuracy of respondents’ memories. However, few empirical studies of revealed route characteristics, chosen by the travelers day-to-day, have been reported in the literature. In this study, accurate Global Position Systems (GPS) and Geographic Information System (GIS) data were employed to reveal multiday routes people used, to study multiday route choice behavior for the same origin-destination (OD) trips. Travelers are classified into three kinds based on their route types. A two-stage route choice  process is proposed. After analyzing the characteristics of different types of travelers, a neural  network was adopted to classify travelers and model route choice behavior. An empirical study  using GPS data collected in Minneapolis-St. Paul metropolitan area was carried out in the  following part. It finds that most travelers follow the same route during commute trips on successive days. The results indicate that neural network framework can classify travelers and model route choice well.

 Keywords: multiday travel behavior, day-to-day modeling, route choice behavior, GPS data, neural networks

Open Traffic

Open Traffic is a new initiative to make GPS traffic data open and available to the public and others, by linking it with OpenStreetMap. It is organized by Conveyal, MapBox, and MapZen with support from the World Bank. The Code is of course open source as well.

OpenTraffic is a free, global traffic speed data set linked to OpenStreetMap built with open source software.

Traffic speed data is a critical input to many transportation related applications. Fortunately many users who need speed data also produce the inputs necessary to create annonmyized traffic statistics.

OpenTraffic provides the space and tools to share traffic statistics from connected vehicles and mobile services. We support the development of analysis and routing tools that enable cities, businesses, and individuals to make use of this data.

How it Works

OpenTraffic connects anyone with real-time or archived GPS location data to processing technology, data storage, and routing and analysis applications.

Location data privacy is paramount. We allow contributors to share anonymized traffic speed statistcs from derived GPS data without disclosing individuals’ location information. In return, data contributors help build a global traffic speed data set that can be used in routing and analysis applications.

The OpenTraffic platform is comprised of several components to make it easy to share and use traffic data:

GPS Probes

GPS probe data can be generated from a variety of sources, including mobile applications or fixed GPS hardware. GPS data can be processed in real-time or archived and transmitted for batch analysis. The OpenTraffic platform has a variety of open source tools to help you load your GPS data from existing sources or connect to Amazon AWS Kinesis streams to manage real-time flows of any size.

Traffic Engine

GPS data is linked to the OpenStreetMap network via Traffic Engine. Data is converted from GPS locations to roadway speed observations and anonymized before being aggregated. As open source software, you control where Traffic Engine is deployed, allowing full control over GPS trace data. Simply install Traffic Engine and load your GPS data to start generating traffic data.

Data Pool

Once anonymized, traffic statistics are added to the global OpenTraffic data pool. By pooling data many different data sources are merged together to provide a seamless global data set, free for use by any application.

Get Involved

We are working with vehicle fleet operators, app developers, and governments to develop and operate the OpenTraffic platform. Learn how you can contribute and benefit: Contact Us

Multiday GPS Travel Behavior Data for Travel Analysis

I am pleased to report that a new FHWA-supported study: “Multiday GPS Travel Behavior Data for Travel Analysis” (PDF) is out. Our contribution is Chapter 4.0 An Empirical Study of the Deviation between Actual and Shortest-Travel-Time Paths. The report was coordinated by RSG, and the overview is below:

Introduction to Report

By Mark Bradley (RSG)

The use of GPS devices to collect trip-specific data as part of household travel surveys has increased steadily in recent years, and will likely become the main mode of travel survey data collection in the future as smartphone-based platforms for collecting travel data come into use. Compared to diary-based methods, the advantages of GPS data capture include the following:

  • The time and location of each trip end can be captured with more precision.
  • There is less potential for respondents to omit entire trips or activities from the survey.
  • The data can be used to trace the route traveled for any particular trip.
  • It becomes more cost-effective to capture multiple days of travel for each respondent.

These unique aspects of GPS data enable new types of behavioral analysis relative to those conducted with more traditional travel survey data. In particular, multiday data capture, in combination with more precise and complete travel data on each day, allows researchers to investigate day-to-day variability in travel behavior at the individual and household level. Such analyses can provide more insight into peoples’ travel patterns at a broader level, and guide future efforts in modeling and predicting travel behavior and designing transportation policies.

Large-sample, multiday GPS datasets from household travel surveys are still relatively limited in quantity, as is the expertise required to process point-by-point GPS trace data into trip-level data that can be used by most analysts. To address these issues, the US Department of Transportation and the National Renewable Energy Laboratory (NREL) have created the Transportation Secure Data Center (TSDC).[1] The TSDC allows researchers to access preprocessed data from almost one dozen different multiday GPS travel datasets from across the United States; it also allows researchers to analyze these data in a secure environment that ensures the protection of data privacy.

The two main objectives of this project are: 1) to provide new examples of the type of valuable research that can be done using multiday GPS travel survey data; and 2) to demonstrate that such research can be conducted in the TSDC research environment. Each of the following four chapters describes a research project that was funded and carried out as part of this project. The four research topics were originally specified by RSG, with input from FHWA, and then further refined by the authors during the course of their research.

In “The Effect of Day-to-Day Travel Time Variability on Auto Travel Choices,” Jennifer Dill, PhD, and Joseph Broach, PhD (candidate), of Portland State University address the important research topic of measuring the effect of auto network reliability on drivers’ choices. Using data from a 7-day vehicle-based GPS survey in the Atlanta region and a longer-duration vehicle-based GPS survey in the Seattle region, the authors identified several cases where respondents made multiple car trips between the same origin-destination (O-D) pairs during the survey period, and measured the actual experienced day-to-day travel time variation for those O-D pairs. The authors report several interesting analyses showing that such variability is related to trip and traveler characteristics, including trip purpose, distance, and household income.

In “Multiday Variation in Time Use and Destination Choice in the Bay Area Using the California Household Travel Survey,” Kate Deutsch-Burgner, PhD, of Data Perspectives Consulting, investigates day-to-day variation in the number, types, and level of dispersion (distance) of destinations visited during the specific days of a 3-day person-based GPS survey in the California Bay Area. Using the technique of latent class cluster analysis (LCCA), she is able to distinguish clearly different patterns of variability in terms of number of trips and type and dispersion of destinations. This analysis method shows promise for addressing the complexity of multiday travel data, and may become even more useful as future person-based (e.g., smartphone-based) GPS datasets include a greater number of travel days and a potentially wider variety of different patterns across the days.

In “Capturing Personal Modality Styles Using Multiday GPS Data-Findings from the San Francisco Bay Area,” Yanzhi “Ann” Xu, PhD, and Randall Guensler, PhD, of Trans/AQ, Inc., analyze the same multiday GPS dataset from the Bay Area that was used for the analysis described in the preceding chapter. In this analysis, however, the focus is on day-to-day variation in mode choice-research for which person-based, rather than vehicle-based, GPS data collection is clearly necessary. The authors were able to identify distinct groups of individuals in terms of whether they always used the same mode or used a variety of modes, and in terms of whether auto or alternative modes were used more often. They were also successful in relating these groupings to different person and household characteristics. The propensity to use multiple modes would benefit standard travel modeling methods, as someone who usually uses auto but also uses transit one or two days per week may be more likely to increase his or her transit use in response to service changes, as compared to someone who never uses transit at all.

Finally, in “An Empirical Study of the Deviation between Actual and Shortest-Travel-Time Paths,” Wenyun Tang, PhD (candidate), and David Levinson, PhD, of the University of Minnesota, use multiday person-based GPS data from the Minneapolis region Travel Behavior Inventory (TBI) to determine how often drivers use the shortest path for their home-to-work trip, and look at patterns in the deviation in travel time between the shortest path and the actual path. In terms of day-to-day variability, the authors were not able to identify many cases where respondents made the same direct home-to-work auto trip on multiple days. This outcome indicates that analyses that measure travel behavior across multiple days (rather than simply treating them as separate single days) will tend to require large sample sizes, particularly when the analysis focuses on a specific type of behavior (e.g., direct home-to-work auto trips).

The research presented in the following four chapters provides interesting findings in their own right, and insights into the types of research designs and methods that will be valuable in analyzing multiday GPS data as it becomes more ubiquitous and accessible in the future. The authors generally recognize that their methods could benefit from larger sample sizes, in terms of the number of respondents, and particularly in terms of the number of days per respondent. (For example, use of 7-day GPS data capture periods would allow analysis of patterns, including both weekdays and weekends.) The authors also note the critical importance of how the GPS trace data are processed into trip-level data, and the need for evolving practices and standards in GPS data processing. Finally, the authors describe the value of the TSDC in making these unique datasets available while providing a secure and productive research environment.

A survival analysis-based choice set formation approach for single-destination choice using GPS travel data

SDC figure

Recent working paper:

This research investigates how land use and road network structure influence home-based single-destination choice in the context of trip chains, using the in-vehicle GPS travel data in the Minneapolis-St. Paul Metropolitan area. We propose a new choice set formation approach which combines survival analysis and random selection. Our empirical findings reveal that: (1) Accessibility and diversity of services at the destination influences individuals’ destination choice. (2) Route-specific network measures such as turn index, speed discontinuity, and trip chains’ travel time saving ratio also display statistically significant effects on destination choice. Our approach contributes to methodologies in modeling destination choice. The results improve our understanding on travel behavior and have implications on transportation and land use planning.

This paper is part of Arthur Huang’s Dissertation.

Accessibility, network structure, and consumers’ destination choice: a GIS analysis of GPS travel data

Working paper:

Walking areas around a trip destination
Walking areas around a trip destination
  • Huang, Arthur and Levinson, David (2011) Accessibility, network structure, and consumers’ destination choice: a GIS analysis of GPS travel data.

    Anecdotal and empirical evidence has shown that road networks, destination accessibility, and travelers’ choice of destination are closely related. Nevertheless, there have not been systematic investigations linking individuals’ travel behavior and retail clusters at the microscopic level. Based on GPS travel data in the Twin Cities, this paper analyzes the impacts of travelers’ interactions with road network structure and clustering of services at the destination on travelers’ destination choice. A multinomial logit model is adopted. The results reveal that higher accessibility and diversity of services in adjacent zones of a destination are associated with greater attractiveness of a destination. Further, the diversity and accessibility of establishments in an area are often highly correlated. In terms of network structure, a destination with a more circuitous or discontinuous route dampens its appeal. Answering where and why people choose to patronize certain places, our planning, our findings shed light on the design of road networks and clusters from a travel behavior perspective.

    (working paper)

Google Killed Map Traffic Estimates Because It Just Didn’t Work

From Gizmodo: Google Killed Map Traffic Estimates Because It Just Didn’t Work

If you’re wondering how road traffic’s gonna slow you today, don’t turn to Google Maps anymore—the site’s killed its estimates. Not because it wasn’t popular. It turns out those road calculations didn’t exactly correlate to, you know, reality.
The Atlantic describes the discovery of perturbed Maps users, who complained to Google when they noticed the change. Its answer?

[W]e have decided that our information systems behind this feature were not as good as they could be. Therefore, we have taken this offline and are currently working to come up with a better, more accurate solution. We are always working to bring you the best Google Maps experience with updates like these!”

Translation: traffic didn’t work. And as the Atlantic’s Nicholas Jackson asks, how could Google be sucking down so much locational data from Android drivers and be botching it to the point that they pulled it down entirely? [The Atlantic]”

A big defeat for the biggest information provider. But using in-vehicle GPS on mobile phones as a probe is coming, and will eventually get it right (approximately, if lagged). The problem of course is that traffic is dynamic, and even a 5 minute lag will be quite off if there is an incident or something non-steady state. However as a signal of whether things are normal, it probably works.

Information provision is probably best for what an individual will not know from routine behavior—random incidents and unfamiliar territory. The qualitative conclusion that incidents and the unexpected are where the greatest gains from traveler information are to be found reinforces the results from our simulations. Those models show that a low level of probes can provide useful information by rapidly detecting incidents, whereas a much greater number is needed to provide any gains from recurring congestion.

New air guidance system threatened with delays

The Washington Post reports on NextGen:

New air guidance system threatened with delays : “Now the Obama administration has embarked on the single most ambitious and expensive
national transportation project since completion of the interstate highway system: a program called the Next Generation Air Transportation System (NextGen).
The NextGen concept sounds simple: Replace an air traffic system based on 60-year-old radar with a satellite-based Global Positioning System (GPS) network that would be far more versatile and efficient. In reality, it is an extraordinarily complex undertaking, threatened with delay by airline fears that the government will not deliver the system in time to justify their expenditures.
NextGen demands the largest investment ever made in civil aviation: between $29 billion and $42 billion for equipment, software and training by 2025. The cost would be shared by a federal government struggling with budget constraints and an airline industry that has been drained by years of recession and high fuel prices. Those tensions over funding threaten to slow the launch of NextGen, despite near-universal support for the program, and delays could prove costly.”