National Academies Press: OpenBook

Cell Phone Location Data for Travel Behavior Analysis (2018)

Chapter: Chapter 3 - A Planner s View of Cell Phone Data

« Previous: Chapter 2 - Travel Behavior from Cell Phone Data
Page 15
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 15
Page 16
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 16
Page 17
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 17
Page 18
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 18
Page 19
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 19
Page 20
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 20
Page 21
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 21
Page 22
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 22
Page 23
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 23
Page 24
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 24
Page 25
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 25
Page 26
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 26
Page 27
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 27
Page 28
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 28
Page 29
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 29
Page 30
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 30
Page 31
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 31
Page 32
Suggested Citation:"Chapter 3 - A Planner s View of Cell Phone Data." National Academies of Sciences, Engineering, and Medicine. 2018. Cell Phone Location Data for Travel Behavior Analysis. Washington, DC: The National Academies Press. doi: 10.17226/25189.
×
Page 32

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

15 Transportation agencies across the country have either purchased or are evaluating the purchase of anonymized and aggregate cell phone data to supplement, enhance, or even replace traditional data sources in support of planning and modeling projects. Agency staff and practitioners need to be clear about the terminology used to describe cell phone data derived from call detail records (CDRs). It is even more important for them to understand the methods used to translate locational data into estimates of travel flows and generate the final data sets used to support transportation planning. This chapter has four major sections: • Section 3.1 focuses on big data and cell phone CDR data, especially as they pertain to data commonly available to transportation planners. • Section 3.2 focuses on the policy needs of planning agencies and on how cell phone data may be used to support the analyses required by these policy needs. Interviews of agency staff are summarized so readers can gain from their perspectives. This section includes a checklist for planners to use when thinking about the uses of CDR data, the strengths and weaknesses of these data, and efforts to open up vendor products that may appear to be a black box (see Box 1-1). • Section 3.3 focuses on the utility of cell phone data as (a) a source of travel information, (b) an input in estimating travel demand matrices, and (c) a substitute for different modeling com- ponents. The chapter discusses how traditional and CDR sources of data can address ques- tions related to sampling and expansion, lists how key elements of daily travel are recorded under each option, and presents how aggregate analyses and model components can be built with each data source. • Section 3.4 describes the research approach used to shed light into the black box and to make multiway comparisons with similar measures of travel from traditional surveys and model results to highlight the strengths and weaknesses of each source of data. 3.1 Cell Phone Data in Transportation Planning This section focuses on key aspects of cell phone data commercially available to transportation planning agencies and presents commonly used terms, standardized location analytics proce- dures, and information on sample sizes. 3.1.1 What Is Big Data? Commercial cell phone data sold to transportation agencies are categorized as big data. Although this term conveys that these data sets are large, other elements are relevant within C H A P T E R 3 A Planner’s View of Cell Phone Data

16 Cell Phone Location Data for Travel Behavior Analysis this description. Data streams that are termed “big data” meet one or more of the following “V” criteria:1 • Volume, the size of the amount of data. In the case of cell phone data, information is obtained from millions of devices nationally, making these databases massive in size. • Velocity, the speed at which data are generated and processed. Most cell phone data vendors obtain live streams of data and handle, process, and store them in near real time. • Variety in the type of transmitted data, which makes processing challenging. In the case of cell phone data used in location analytics, the most relevant information includes geographic ele- ments, alphanumeric device IDs, time stamps, and dates. These may sometimes be combined with additional sources of information (e.g., land use or topography) to support waypoint analyses.2 • Valence, the measure of the degree of connectedness of the data. This attribute is relevant in cell phone data analytics where device locations are imputed using transportation and land use data and where heuristics are used to understand the difference between traffic stops and actual trips ends. In addition, device IDs are monitored over a longer period and machine learning algorithms are used to impute home and primary daytime locations. • Veracity, a measure of the accuracy and usefulness of the data. Veracity is an important con- sideration when data elements collected passively are being evaluated. Algorithms described in Chapters 4 through 8 use principles grounded in transportation planning to improve the veracity of passive cell phone data. • Value, that which is gained from processing the data. Value is the center of the new wave of collecting, analyzing, and interpreting data for transportation planning and modeling purposes. 3.1.2 What Are CDRs? A CDR is a record produced by telecommunications equipment that documents the details of an incoming or outgoing call, an incoming or outgoing text message, or a connection to an app, web browser, or e-mail database. The communication passes through the device, which creates a record of the transactions made but not the content of the communication. CDR records contain only metadata, not the actual data packets being transferred. This prop- erty makes CDRs ideal for use in anonymized data analytics. Key dimensions of financial and operational CDR data are as follows (see also Figure 3-1): • CDR data capture all metadata that trigger a contact with the telecommunications towers, including calls, text messages, and Internet data access. • Both active and passive signals trigger the capture of CDR data from a single cell phone: – Active signals include cases when users make calls, send text messages, or use the Internet actively. – Passive signals are recorded when users receive calls or text messages. – Passive signals are also received from apps accessing the phone either continuously or periodically (e.g., e-mail, music, news, social media, and sports and fitness apps). 1The earliest reference to “three Vs” was made by Doug Laney (2001), who discussed three dimensions of increasing volume, velocity, and variety in an unpublished research note. Examples of references in the literature include a discussion of four dimensions (IBM 2014) and a discussion of five dimensions (Scott 2015). 2Waypoints are intermediate observations between the inferred trip ends while the cell phone device is moving. Trip ends can provide information about the locations of activities supported by travel. Waypoints can provide information about the paths used while the device is traveling between those inferred trip ends.

A Planner’s View of Cell Phone Data 17 • The frequency and amount of CDR data generation depend on user usage patterns. Frequent users, through repeated use of their devices, provide a lot of information about their daily travel as compared with occasional or infrequent users. • The amount of CDR data also depends on the number of apps installed by users on their cell phones and the frequency with which each installed app interacts with telecommunications towers. • Most CDR data (e.g., start and end time of transactions and device ID) are accurate, but the location of a device is less accurate, given that device locations are estimated with meth- ods, such as triangulation, that often produce geographic errors. Vendors have indicated that errors in CDR data range from tens to hundreds of meters and depend on geography and the density of cell towers. • The CDR data available from cell phone service providers include financial information maintained by the cell phone service providers to support billing to their subscribers. Finan- cial records are the data most likely to be maintained and available for additional processing. • Providers also collect operational data processed to deliver services efficiently to subscribers. 3.1.3 How Are Locations Determined? Using trace data from CDRs to identify the locations of devices is a critical step in supporting transportation planning and model development, application, and validation. Several analytical procedures have been implemented to convert trace data into locational and stop data: • Devices may be traced through analysis of CDR data or use of GPS tracking on devices. • Cell phone traces may be obtained by evaluating CDR transactions that include information about device location and the start and end times of transactions. Depending on the tech- nology being used by the device and the number of transactions with telecommunications towers, locational traces of a device may either be few and far between or may be fast and furious. The frequency with which these traces are collected does not affect the accuracy of the mapping. However, the accuracy of CDR trace data ranges from tens to hundreds of meters. • Cell phone traces may also be obtained by tracking GPS outputs from cell phones. These are obtained when GPS options on mobile phones are turned on. The GPS traces may be obtained (a) through apps when their location settings are turned on and (b) through apps that track location continuously in the background or by using a native GPS system. Although GPS traces are more accurate than CDR data, not all devices have GPS traces turned on all the time. The locational information from GPS-enabled cell phones is much more accurate than locations inferred from triangulated CDR observations. Financial Data Operational Data CDR CDR plus Calls, text, and web access Network pinging the phone Active: outgoing control by user Constant communication with various apps accessing the phone Passive: incoming control by phone Location information is derived from cell tower locations only Handset locations are triangulated with more sightings, more signals, and more accurate locations Signal frequency is low Signal frequency depends on settings and use Location accuracy is within 1 mile Positional accuracy is within hundreds of feet Figure 3-1. Types of call detail record data.

18 Cell Phone Location Data for Travel Behavior Analysis • Cell phone traces obtained through CDR or GPS data are synthesized through sophisticated algorithms to provide estimates of locations, activities, and travel patterns. To do this, the analyst needs to address the following issues: – Qualifying criteria are needed to determine which devices have sufficient trace information to support further algorithmic investigation. These qualifying criteria are defined both in terms of the number of days with trace activity and in terms of the amount of activity itself. – A minimum threshold must be defined so as to recognize a true stop as opposed to a traffic stop. With regard to the nature of the stop, an algorithm must be developed to identify the location of the stop by exploring several nearby activity stops and “averaging” them to arrive at a location. – Regular nighttime locations (corresponding to home) and regular daytime locations (reflecting work or a school or university) need to be synthesized by using multiday data and a machine learning algorithm. – Locations that are not regularly visited and that do not qualify as home or work locations are much more difficult to identify. Several geolocation algorithms can be applied to iden- tify these types of activities and their locations. – Algorithms are used to analyze consecutive stops and assign travel between them. The time of travel is determined through the time stamps of waypoint locations or, if there are no waypoints, by using heuristic algorithms based on other observed data. – In cases where sufficient waypoints exist, the analyst may be able to use map-matching algo- rithms to map the route taken by the device. In other cases, standard assignment algorithms can be run to evaluate network loading.3 3.1.4 What Are the Challenges of Using CDR Data? As with every analytical method where inferences about travel need to be drawn on the basis of a data sample, measuring travel and activity patterns by relying on cell phone location data presents challenges. Given that location data are collected on the basis of users’ active and passive use of cell phones, end users of these data sources need to consider the following: • Not all movements may be recorded. Phones may be turned off; devices may not interact with telecommunications systems; individuals may not travel with their cell phones; or multiple users may share the same phone. • The same movements may be recorded more than once, as individuals may carry more than one cell phone at any given time. • Devices may travel through urban canyons in dense urban areas with multiple cell towers or through areas with poor cell tower coverage. In such cases, location analytics become less reliable. • In cases of neighboring cell towers, cell phone signals may jump from one tower to another, so that it may appear that the device traveled between two different points, resulting in a spurious trip. Similarly, short “pseudotrips” may be inferred from changes in triangulation patterns that result in slightly distinct location inferences. • Movements within a large office campus or hospital need to be identified, given that they do not affect transportation system performance. • Although anchor locations such as home and work can be accurately located by repeated obser- vations over multiple days and machine learning algorithms, the algorithms are less successful in identifying locations that reflect nonregular activity points outside home and work. • Current research suggests that cell phone location data cannot readily measure key modeling metrics such as – Mode of travel, – Party size, 3Although this report does not focus on the waypoints between inferred trip ends, such information could be valuable.

A Planner’s View of Cell Phone Data 19 – Activity stops at which the activity duration is shorter than the threshold used in the algorithm, and – Activity purposes for locations other than home and work. • Cell phone carrier restrictions aimed at preserving intellectual property and user privacy do not allow the end user to know how exactly the data are collected. 3.2 Transportation Planner Needs In the context of population growth in cities and the restructuring of urban economies and societies, the fundamental task of transportation planners, modelers, and engineers—to effec- tively move people and goods—has become increasingly challenging. Meanwhile, transporta- tion services and infrastructure greatly affect economic growth and quality of life. Today’s planning agencies focus on various issues when evaluating policy options: mobility, access to jobs, safety, environmental justice, land use and zoning, congestion, air quality and transportation emissions, transit utilization and fare policies, bike–pedestrian improvements, roadway construction, asset management, tolling and managed lanes, and socio economic zonal data. These issues need to be studied not only under current-day conditions, but also under future scenarios. Agencies use various data to support such analyses, including Census planning tools, custom survey data, and sensor data (from tolling, transit, and road- way agencies). Regional planning relies on behavioral statistical models that link these data sources and pro- vide a framework that explains today’s travel behavior (Ben-Akiva and Lerman, 1985). Regional models provide snapshots of system performance by mode and time of day for different facilities across a region and are validated with observed measures of flows and levels of service. Regional models are also applied under future estimates of population and employment using different transportation supply and policy scenarios to estimate the effect on travel flows and system performance. The underlying principles of the behavioral approach to modeling are as follows: • The observed behavior of a traveler as a rational decision maker is key to understanding his or her observed choices in a spatial and temporal context. • The analyst observes travel decisions and makes inferences about the aspects of traveler behavior that are not observed. • Individual travelers are rational decision makers who use an “expected utility” approach to make trade-offs and evaluate and choose among alternatives. • Individuals are assumed to have perfect information about the alternatives available to them and the attributes of each alternative. • The individual behavioral approach is broad and applies to total trip making, the destinations and stops of individual trips, the timing of trips within a given day, and the mode choice deci- sions that individuals make. Within this framework, planning agencies are now tasked with assessing how and where advanced sensor-driven analytics tools (such as trip tables based on cell phone data) fit in. 3.2.1 Perspectives of Transportation Planners The objective of the interviews with agency staff was to discuss the cell phone data avail- able, how agencies intended to use that data, and how modeling practitioners actually use that data. The research team interviewed modeling practitioners about how cell phone data could be used to develop and enhance transportation models. The interviewees—regional

20 Cell Phone Location Data for Travel Behavior Analysis planning staff, academics, and modeling practitioners—understand the data needed for traditional transportation models and the more extensive data requirements for detailed activity-based models. Most of the participants have conducted or used traditional household, onboard, and inter- cept surveys and have considered using nontraditional sources of data to develop, augment, or validate regional and corridor-level models. Participants reflected a mix of backgrounds in Federal research, department of transportation (DOT) and metropolitan planning organiza- tion (MPO) experience, and academic research. The questions listed in Table 3-1 were used as a rough interview guide to frame the discussion. This section summarizes the key findings from the discussions under each group of questions. Key Question Attributes of CDR Data What are the greatest agency needs that cell phone data can address? • Passenger versus freight flows • Regional versus corridor-level flows • Insights into internal versus external flows • Representation of all regional travel versus focus on specific individual markets • Understanding of “underreported travel” (e.g., short and nonmotorized trips) How do agencies perceive the potential value and role of cell phone data? • Replacement versus augmentation of traditional data • Cell phone data as a source to use in model estimation versus validation • Option to update existing data periodically without a 10- year regional survey • Targeted use of cell data for geographic markets or traveler segments What was the agency’s view of the cell data it acquired or examined? • Were the cell phone data usable and useful? • What types of cell phone data did the agency use or consider? • What was the greatest value of these data? • What were the weaknesses of the data that might require revisiting? What was the agency’s experience working with these data? • What was the format of the cell data, and what issues were encountered regarding the ease of use, processing, and storage of the data? • What were the strengths of the cell data? The critical weaknesses? The ways to overcome the weaknesses? • What assumptions was the agency willing to accept when using these data? • What additional analyses did the agency conduct to “validate” these data? Does the agency believe that the modeling paradigm will change? • Should cell phone data be fit in current model formulations? • Should model formulations change in response to these data? How? For which purposes does the agency intend to use these data going forward? • Willingness to invest in these data? • Interest in additional testing before making a commitment to use these data? • Wait for the products to mature? • Plans for using cell data as part of a new data collection cycle? Source: Cambridge Systematics, Inc. Table 3-1. Interview outline with CDR data users.

A Planner’s View of Cell Phone Data 21 3.2.1.1 Agency Needs That Can Be Addressed by Cell Phone Data Agency staff recognized the need for increasingly complex travel demand models to address today’s more nuanced and policy-sensitive questions. Issues included congestion by time of day, the mix of passenger and freight flows, introduction of technology and new modes, optimization of existing facilities, and experimentation with new tools (e.g., congestion pricing and technology) to better manage travel demand. Participants also recognized the increasing costs of traditional data collection and the time and resources required to collect, analyze, and interpret such data within an advanced modeling framework. They were intrigued by the potential uses of new sources of data that can be harvested faster, more often, and at a lower cost than traditional data sources. In addition, participants realized that cell phone–derived data cannot address all of an agency’s planning needs, especially the nuanced and policy-sensitive questions that activity- based models address. Participants mentioned concerns about sampling bias and their desire to model both passenger and freight flows in their regions. Also, participants recognized the value of new sources of data as reflected in the data’s scale and cost. The staying power of such data will be greater if the data can provide answers regard- ing individual model components or markets currently not addressed by regional models (e.g., external models, long-distance travel, corridor studies, and travel to special events). 3.2.1.2 Potential Value of Cell Phone Data Participants assessed the potential value of cell phone data as compared with the pros and cons of traditional data and methods and were skeptical about some of the cell phone–based data products available in the marketplace, given that some of these products represent a black box to agency staff as users of the data. The paradigm that has served the transportation community well over the years has evolved along with increasing transparency in the methods used by academia and the industry. The state of the art has progressed with healthy debates about the value of different data sources, analysis methods, validation approaches, and forecasting mechanisms. Transportation practitioners are eager to dive into the details of new data to better recognize and appreciate their value. Trans- portation practitioners are also accustomed to specifying their data needs and have a high level of control that is less amenable to ready-to-use data products and analysis solutions. In this light, agency staff, practitioners, and academics alike were intrigued by the promise of new data sources. They recognized that improvements in data and in the methods used to ana- lyze them could improve the products available in the marketplace. They also were interested in gaining a better understanding of the underlying assumptions, so as to better evaluate for them- selves the strengths and weaknesses of new data product options, and were interested in helping shape the design of these data products to better reflect their own priorities. 3.2.1.3 Staying Power of the Behavioral Modeling Paradigm There was general consensus about the value of the behavioral paradigm in guiding the devel- opment of traditional four-step models and more sophisticated activity-based travel demand models. Industry and academia believe that the study of individuals’ travel decisions and their activities are key to the analysis, understanding, and forecasting of travel patterns. The skepti- cism sometimes expressed about prepackaged data products reflects, at least in part, analysts’ preference for specifying their own detailed data requirements and developing customized travel demand models to fit their region’s modeling needs for a range of analytical purposes. Although the discussion did not focus on model evolution, the behavioral paradigm itself has changed over time from crude aggregate models to individual-level disaggregate estimation to sophisticated

22 Cell Phone Location Data for Travel Behavior Analysis activity-based models. The behavioral paradigm may evolve again in response to emerging pol- icy and analysis needs and may benefit from current ongoing research on locational data in transportation and related fields. The participants acknowledged that it is likely that research in refining inferences based on locational data and machine learning methods may spur development of new analytical meth- ods. In this case, it can be expected that the behavioral principles and methods of travel behavior analysis will evolve to best take advantage of cell phone data sources, accompanying land use data, and new flexible forms of personal surveys. 3.2.1.4 Experience with and Future Uses of Cell Phone Data The participants’ experiences with data products derived from cell phone data varied. Most of the uses focused on passenger travel at a regional or corridor level, with one application of cell phone data for travel at a national level. In addition to conducting the interviews, the research team reviewed an in-depth presentation by Ron Milam at the Washington Council of Governments. Participants agreed that cell phone–derived data cannot address all of an agency’s planning needs, especially with regard to nuanced and policy-sensitive questions. Concern was also expressed about the gap between the resolution provided by cell phone data and that provided by today’s sophisticated activity-based models with regard to travel for different purposes, at dif- ferent times of day, and by different members of a household. The following topics were raised: • The use of cell phone data for estimation of an activity model was mentioned as problematic, given the aggregate nature of the cell phone–derived data. • Concern about sample bias reflected in differences in ownership and use of cell phones across different market segments was discussed. • The power of cell phone data in providing origin–destination (O-D) flows was recognized but needs to be coupled with checks against existing data. • The accuracy of O-D data at different levels of geographic detail is a major consideration in the practitioner community. • The inability to differentiate between passenger and freight traffic is another limitation present in current cell phone data products. Participants recognized the value of quick access to travel data outside the long cycle of house- hold surveys. They also indicated that cell phone–derived travel data could provide input to individual model components, as follows: • External models, a promising area in which cell phone data could replace current methods or allow for more frequent model updates; • Long-distance travel and corridor studies, which require a lower level of geographic detail, could benefit from cell phone O-D flows; and • Travel to special events and traffic to special generators were mentioned as examples of promising applications of cell phone–derived travel data. 3.2.2 A Planner’s List of Do’s and Don’ts for Cell Data Findings from practitioner interviews can be synthesized into three key areas that agency staff must evaluate before making the decision to purchase and use cell phone data for plan- ning and modeling purposes. Positioning cell phone data within an agency’s overall data and modeling program, recognizing the strengths and weaknesses of this data stream, and identifying the questions to ask the data vendor are critical to extracting the most value from these data.

A Planner’s View of Cell Phone Data 23 3.2.2.1 Data Collection Program First and foremost, planners should position cell phone data requests within the bigger pic- ture of the agency’s data collection program that responds to the different analysis requests and modeling requirements. In particular, questions that agency staff and practitioners need to ask themselves and discuss in detail with data vendors include the following: • Where do cell phone data fit within the bigger picture of the agency’s overall planning func- tions? Can cell phone data as packaged by the vendors be used to – Support existing behavioral analysis tools and models? – Provide a snapshot of travel today and a forecast of future travel? • What are the agency’s data collection and analysis strategies? How can the agency benefit from integrating cell phone data into its data collection program? Can cell phone data be used to – Augment or replace the collection of regional household surveys? – Replace special purpose surveys at airports and other special generators? – Support visitor surveys or external travel data collection efforts? • More critically, agencies must evaluate whether and how cell phone data may support or aug- ment travel demand models. Specifically, can cell phone data be used to – Estimate regional travel demand model systems? – Develop models at a level of resolution similar to that of today’s tools? – Replace individual model components instead of an entire model system? – Support freestanding analyses such as time-of-day travel? – Validate existing travel behavior models on a regular basis? – Refresh model inputs more regularly than every 10 years, which is itself an ambitious stan- dard not met by most planning agencies? 3.2.2.2 Evaluation of CDR Data Products Second, planning agency staff need to recognize the strengths and the limitations of commer- cial cell phone data across three dimensions—the data themselves, the procedures vendors use, and the applicability of the data for different planning and modeling uses. • Data content, technology, and the structure of cell phone–derived data sources are different from traditional survey data sources. • Vendors use their own proprietary analytical procedures to convert cell phone data into usable data products. – Agencies that regularly collect survey data to develop and update sophisticated models to their own exact specifications need to understand the vendors’ underlying methods. – Smaller agencies that collect data less often and maintain simpler travel demand models may recognize the benefit of these forms of data and also need to appreciate the underlying methods. – In both cases, agency staff may be willing to accept that they will rely more on a black box approach to gain the benefit of frequent data and model updates that use cell phone data. • The utility and value of cell phone data vary according to the intended purposes of different agencies and departments within an agency. – Cell phone–derived data products are not likely to be appropriate for model estimation, given that less detail is available regarding travel purposes, sample weighting and expansion, and sensitivity by market segment. – Cell phone–derived data may be appropriate to enhance analyses of travel by time of day, special events, and external travel. – Cell phone data can also be used to help validate individual model components or to refresh selected model components.

24 Cell Phone Location Data for Travel Behavior Analysis 3.2.2.3 Questions to Ask CDR Data Vendors As a third and final step, agency staff evaluating cell phone data should ask vendors a range of specific questions and hold discussions with their colleagues to help determine the value of cell phone data packages. • What is the technology supported in the data product (e.g., 4G versus 3G)? How does the technology affect the frequency of signals and their location accuracy? • What is the geographic coverage of devices in the study region? What is the density of service or cell towers? What effect does the density have on the spatial accuracy of the data? • Who are the providers of cell phone data in the region and what is their subscriber base and market share? Are the CDR data from a representative sample of the region’s cell phone users that reflects the regional population? – What is the profile of devices and users in the marketplace? This issue is especially important in cases where the subscribers’ socioeconomic and usage profiles differ across vendors. – What are the types of analysis that would be most appropriate given the local sample of cell phone data collected? Vendors can be asked to provide details on why they believe cell phone data are appropriate for different analyses. 3.2.2.4 Understanding the Attributes of the Black Box • In summary, a discussion that clarifies aspects of a final product that is now a black box will help agency staff and practitioners better understand the underlying assumptions for the product. Such a dialogue will help users of the data evaluate key aspects of the data set and will provide input to influence the design of future products. Agency staff can be part of a two- way educational effort that may inform software vendors in developing and refining future cell phone data products in response to the planning and modeling needs of transportation agencies. • Increased collaboration between industry, planning agencies, and academia can accomplish this task even if business realities prevent the sharing of proprietary methods by vendors. At a minimum, increased transparency of the black box through input provided by users of the data will increase the usefulness and value of the tool to planning agencies. Recognition of the strengths and weaknesses of CDR data and the underlying analysis assumptions will increase confidence in cell phone data products. • Agency staff need to formulate their own understanding of the utility of CDR data from a practitioner’s perspective. In particular, it is important to identify the agency objectives that can be supported by CDR data and those for which other data sources are more appropriate. Thinking about how CDR data will be implemented and analyzed within an agency’s data col- lection and planning support cycle will help crystallize the agency’s approach to purchasing and utilizing these data. 3.3 Utility of Cell Phone Data Cell phone data can provide metrics with which to understand snapshots of today’s travel demand and can provide data that can be analyzed in a modeling framework to forecast future travel. This chapter discusses the utility of cell phone data in providing a bird’s-eye view of travel demand patterns. Cell phone data are compared with traditional travel surveys in Section 3.3.1. Typical model outputs from each data source are compared in Section 3.3.2. Finally, the ways in which different model elements can be captured by traditional surveys and cell phone data are addressed in Section 3.3.3.

A Planner’s View of Cell Phone Data 25 3.3.1 CDR Data Utility as a Source of Travel Data Because of the large sample size of cell phone data matrices and their lower unit costs as compared with traditional household travel surveys, there are often discussions about replacing household travel surveys with cell phone data. Table 3-2 contributes to this discussion by show- ing the strengths and weaknesses of cell phone data as compared with survey data: • The lower unit cost and lower total cost of cell phone data allow the collection of a much larger data set that often spans multiple days. which offers a key advantage over household surveys Basis of Travel Data Variable of Interest Traditional Survey Cell Phone Use Sample Size Sampling rate (%) 0.5–2 15–35 Sample size for a region with 1,000,000 households 3,000,000 population 5,000–20,000 households 15,000–60,000 individuals 150,000–350,000 households 450,000–1,050,000 individuals Sampling Strata Unit of analysis Individual and household Cell phone Sampling unit Household Cell phone Sampling by geography As fine-grained as block group level Feasible at aggregate level Sampling by market segment Yes N/A Socioeconomic Information Respondent attributes Age Gender Worker status Student status Occupation Work hours Ability to telecommute Ethnicity Nighttime location of cell phone Daytime location of cell phone Household attributes Size Number of vehicles Number of workers Income Life cycle Residential location Number of children N/A Survey Expansion Sampling rate and size Small sample size requires careful sampling and expansion Large data set—relatively robust sample sizes for expansion Household attributes Used in all household travel survey expansion No data available for detailed weighting on basis of personal or household attributes Respondent attributes Increasingly used in weighting of surveys for activity-based models Geography Attributes Often carried out at county level; smaller geographic levels possible, depending on sample sizes Some geography-based expansion feasible, but not at individual or household levels Source: Cambridge Systematics, Inc. Note: N/A = not available. Table 3-2. Sampling and expansion by data source.

26 Cell Phone Location Data for Travel Behavior Analysis that use a smaller sample focus mostly on travel during a single day, and are administered every 10 years or so. • The units of the analysis are different. Surveys focus on households and individual travelers within a household, while cell phone data rely on devices. Data products built on cell phone data are weaker for the following reasons: – Individuals may own multiple devices, members of a household may share devices, and individuals may carry different devices at different times of day. These patterns are not accounted for and may affect data quality. – The use of the device as the sampling unit does not allow for differential sampling by mar- ket segment or by detailed geography and prevents the analyst from focusing on specific segments of greater interest to a region. • Cell phone data do not include any household or individual socioeconomic information linked to each device because of privacy considerations. The absence of this information lim- its the resolution of cell phone data for model estimation. – In traditional surveys, detailed household and individual data are collected and offer critical input to the model estimation. – Traveler and household socioeconomics allow the segmentation of the market and the evaluation of travel responses to different pricing and level of service scenarios. – Recent research has focused on inferring socioeconomic characteristics at the home end and linking cell phone records to those characteristics. • Survey expansion methods also differ significantly and provide an advantage to traditional survey methods. – Cell phone data most likely need to be expanded by using the device owner’s residential address, which is in turn inferred by the nighttime location of a device. There are also poten- tial biases in ownership that are not possible to identify and correct during sample expansion. – In contrast, household survey expansion and weighting are much more detailed and take into account socioeconomic and geographic criteria that provide a richer and more bal- anced sample across different criteria. 3.3.2 CDR Data Utility as a Source of Travel Demand Metrics The lack of socioeconomic data in trip tables and travel estimates derived from cell phone data limits the use of CDR data as a full-fledged replacement for travel models. Today’s activity models provide detailed estimates about the type of traveler who uses different modes and facili- ties of the transportation system to reach various destinations at different times of day. Table 3-3 lists the elements that traditional models provide and discusses how traditional and cell phone data sources can address each model component. • Metrics. Cell phone data can provide metrics of aggregate measures of residential travel, visi- tor travel, and travel at external stations. However, they cannot support fine-grained analyses by purpose and market segment. • Total travel. Daily travel is underreported in surveys because respondents may not report short trips that are considered less important (Bricka and Bhat 2006, NuStats 2002, and Zmud and Wolf 2003). Monitoring of cell phone traces over multiple days offers an objective way to observe travel. The drawbacks of CDR data include – The device as the unit of the analysis instead of the individual; – The need to infer stops, activities, and purposes; – Potential gaps in cell phone signals in lower-density areas; and – The possibility that travelers do not carry their cell phone at all times. • Home, work, and nonwork activities. Repeated observations of a device during night hours and during a typical workday provide a robust definition of home-based work and school

A Planner’s View of Cell Phone Data 27 or university activities. However, cell phone–derived data do not provide detail for activities other than home and work. • Spatial resolution. Locations other than home and work are also subject to reporting errors by respondents in traditional surveys. In the case of cell phone data sources, locations are inferred by analyzing cell phone traces and using assumptions about an activity introducing the potential for spatial error. • Purposes and joint travel. Surveys have a clear advantage, given that they provide detailed information about all types of purposes and joint travel with other members of the household. • Temporal resolution. Cell phone data record traces with accuracy as long as they are linked to a call, message, or Internet data access. Under these conditions, they may be preferable to surveys where recording of time is approximate and less detailed. Variable of Interest Travel Data from Traditional Surveys Travel Data Based on Cell Phone Use Total daily travel Self-reported in survey diaries. Travel may be underreported. Prompted recall offers an improvement. Passive cell signals over days may offer more robust metrics than surveys. Unit is device-trips rather than person-trips. Quality depends on CDR data density. Time of travel Self-reported in survey diaries. Times may be inaccurate and incomplete. Accurate time stamps. Need to infer activity and link it to the time stamp versus en route travel. Stops versus activities Self-reported in survey diaries. Detailed log of stops and activities. Good detail on all travel purposes. Need to infer stops, activities, segments. Nonwork purposes are difficult to infer. Location of activities Self-reported in survey diaries. Smart geocoding needed to match. Prompted recall offers an improvement. Difficult to infer the location of activities. A challenge in mixed land use areas. Travel purpose Self-reported in survey diaries. Prompted recall offers an improvement. Home and work locations are inferred. Poor inference on nonhome and nonwork. Joint travel Self-reported in survey diaries. Risk of underreporting. Prompted recall offers an improvement. Not feasible to record or capture. Mode of travel Self-reported in survey diaries. Good detail by tour and segment. Walk and bike trips may be underreported. Not readily inferred. Route assignment Not usually captured in surveys. Depends on trace data and algorithm. Tour generation Self-reported in detail in a survey. Analysis by using heuristics and rules. Data products do not include chains. Only aggregate trips are sold. Source: Cambridge Systematics, Inc. Table 3-3. Recording of travel elements.

28 Cell Phone Location Data for Travel Behavior Analysis • Travel modes. Mode choice can be imputed in cell phone data tables by using a combination of travel speeds and perhaps some transit routing information. However, this method is not yet reliable, especially in large urban areas that experience congestion and where the speed difference across modes may be smaller. • Tour metrics. Commercially available CDR data provide aggregate trips between zones and do not include travel at the tour level in contrast to the high level of individual travel detail offered by activity-based models. • Traffic assignment. Traditional surveys do not capture detailed route information, whereas cell phone–derived data do. Cell phone–derived data are advantageous in this regard, as long as the transactions made produce enough signals to reflect the entire route. 3.3.3 CDR Data Utility for Model Components The labor-intensive efforts and considerable data resources allocated to the development of a regional model do not support frequent updates or reestimation of regional models. In most cases, model updates are limited to reflecting the availability of new sources of control data such as the Census Transportation Planning Products, journey-to-work data, American Community Survey (ACS) data, and up-to-date traffic or ridership counts. Models are also updated to account for the effect of major transportation investments such as new highways, the introduction of tolls, or the introduction of new transit services. It is also possible that major changes in residential or commercial land use, such as urban revitalization, development of a new major employment cluster, or the introduction of new sports or conference facilities, may motivate the update of the model to better quantify their likely effects on travel. It is often necessary to develop freestanding modules to examine special event travel; changes in travel that originates outside the region or traverses the region; effect of new facilities on visi- tor travel; or needs of visitors who travel within a region. Table 3-4 provides a list of model-related tasks where CDR data can provide input instead of, or in addition to, traditional survey methods. For each module, the key attributes of each data source are outlined to highlight the corresponding strengths and weaknesses of traditional survey data and CDR data. 3.3.3.1 Seasonality of Travel The granularity of policy questions changes over time, and it may be desirable to address ques- tions related to the seasonality of travel, especially for regions with major differences in travel patterns by season. Rolling samples of surveys similar in concept to the ACS or multiple snapshots of travel by season based on CDR data can achieve this objective and should be considered. The calculation of the absolute or percentage change from season to season using seasonal data provides an additional off-model estimate of relative change over a typical day model approach. 3.3.3.2 Visitor Travel Patterns This element may be related to the seasonality of regional travel but it may also affect metro- politan regions where tourism is a key component and driver of the local economy. Visitor and establishment surveys can be supplemented through the use of CDR data. 3.3.3.3 Special Generators Travel patterns to and from nodes of intense recurring or nonrecurring activity, such as airports, shopping malls, and sports or cultural events, can have a major effect on travel infrastructure

A Planner’s View of Cell Phone Data 29 and display significant peaking. A better understanding of these events can benefit from off- model components that can be readily updated by using periodic surveys or snapshots that use CDR data. Processed CDR data may be used to assess site-specific event studies such as airports, concerts, or sporting events. Limitations in inferring, aggregating, and expanding CDR data and the difficulties in providing path traces make these locations unsuitable for evacuation, emer- gency response, or other route-based studies. 3.3.3.4 Year-to-Year Variation A weakness of existing regional models is that the time between model updates may easily exceed 10 years and may therefore miss subtle or more important trends in population and employment growth or stagnation. As a result, planners may under- or overestimate corre- sponding increases or slowdowns in travel. Surveys or CDR data that are collected periodically can provide more frequent updates of the factors that affect travel. The identification of trends in data and model results can be valuable tools to account for growth in intermediate years until a new regional survey and regional model are completed. As with seasonality effects, an analyst can focus on calculating absolute or percentage changes from year to year to provide an estimate of upward or downward trends in travel compared with the base-year model. These estimates can also prove valuable in updating forecast year estimates in cases where observed trends greatly exceed or significantly lag projected patterns. 3.3.3.5 External Travel Traditionally, external stations or zones are used to supplement a regional model, especially in metropolitan areas that interact heavily with cities, counties, and states outside the model Variable of Interest Travel Data from Traditional Surveys Travel Data Based on Cell Phone Use Seasonal variation A well-thought-out sampling plan. Continuous data collection by season. CDR data by month of the year. Differences in seasonal travel. Visitor travel patterns Targeted detailed visitor surveys. Airports, train stations, highway rest areas, hotels, and popular visitor sites. Home as nighttime device location. Differentiate visitor from residential devices. Special generators Specialized surveys at airports, malls, or special event sites, Supplement to regional surveys. Data on mode, time of day, origin of trip, and socioeconomic detail. CDR data for “event days”. Capture of time of day and trip origin. Mode inference is weak. Socioeconomic data not available. Year-to-year variation Longitudinal/panel or rolling sample data. Measurement of change over time. CDR data sets from different years. Measurement of change in patterns. External travel License plate capture at cordon line. Follow-up survey of auto owners. Bluetooth data as an option. Definition of external cordon line. Number of devices crossing cordon. Home origin to measure external travel. Source: Cambridge Systematics, Inc. Table 3-4. Travel elements for aggregate analysis and metrics.

30 Cell Phone Location Data for Travel Behavior Analysis area boundaries. Traditional license plate number recording and follow-up survey methods are labor intensive and can be supplemented or replaced by CDR data to provide snapshots of total external–internal or through travel. Key assumptions related to the home location of the device need to be accepted as part of this method. 3.4 Research Framework Raw cell phone data, which include an identifier of the cell device, are exceedingly hard to obtain because of the confidential information that could be inferred from those CDR data. Although this project used raw CDR data for the Boston, Massachusetts, area that had been obtained for research purposes, the research team recognized that it is not likely that such a set of raw CDR data will be made available again. At the present time, only aggregated results of processed CDR data can be obtained commer- cially. The methods used by vendors to process this data are proprietary and are not disclosed. From a practitioner’s standpoint, it is a challenge to ascertain the quality of the end product without an understanding of the procedures that drive the end product. To bridge this intel- lectual gap, the research team used the following three-step process: 1. The team analyzed the raw CDR data and described the processing methods used to infer trip ends and activities from the raw data. The open and transparent procedures allow practitio- ners to get a bird’s-eye view of the techniques used in the field of cell phone data processing. 2. The team compared the results of its processing of the raw CDR data with commercially pro- cessed CDR data for the same geography. In cases of similar results, it was concluded that the methods used to process the raw CDR data discussed in this report were broadly comparable to those used in preparing the commercial data. In cases of differences, the team documented the differences and discussed how they might be reconciled. 3. The team compared results generated from the commercially available CDR products and the custom analysis of the raw CDR data with two independent transportation sources: regional household travel surveys and the regional travel demand model in Boston. These compari- sons were necessary to identify the strengths and weaknesses of the CDR data and to develop a roadmap to enable practitioners to use CDR data effectively in the development of transport modeling and analysis. This analysis will allow practitioners to assess the value of this new CDR data stream as com- pared with that of traditional surveys and models. On the one hand, CDR data offer a much larger volume of data on travel observed over a long period. However, despite their sample size and the advantage of repeated observations, CDR data are less detailed and require inferences to be made regarding locations, activities, purpose of travel, and the time of day of travel. CDR data also do not provide information on users’ socioeconomic characteristics, which are a key part of traditional and activity-based models. On the other hand, household surveys are well tested, their strengths and weaknesses are well understood, and they are currently evolving through the use of technology. Surveys are generally more expensive, have a much smaller sample size, and are collected infrequently as compared with CDR data. However, they offer the great advantage of providing household and person- level travel data, including accurate information on activities and purpose. They also provide detailed socioeconomic characteristics for each respondent, which allows for the development of nuanced models of daily travel at a disaggregate level. If the outcomes of the two data sets are similar and the processes used to infer trip ends and activities are understood and considered acceptable, processed CDR data may offer a suitable and acceptable supplement to household surveys. Under such a scenario, CDR data can be used

A Planner’s View of Cell Phone Data 31 for a range of purposes, from estimating travel demand models or model components to provid- ing selected model outputs for estimation or validation to serving as interim data sets between consecutive travel behavior survey efforts. 3.4.1 Research Method The overarching goal of this research is to present a method that extracts activity locations (stay points), labels activity types (“home,” “work,” and “other”), estimates O-D trip matrices, and assigns traffic in the road network by analyzing raw cell phone data (Jiang et al. 2013, Alexander et al. 2015). The next sections present a flexible, modular, and computationally effi- cient software platform built to implement these methods, which are analogous to procedures of traditional travel demand models. This system enables researchers to import raw cell phone data to produce trip matrices and road usage patterns in any city (Toole et al. 2015). It also visualizes these outputs to communi- cate mobility patterns effectively to planners, stakeholders, and decision makers. The platform is an alternative to proprietary transportation software packages and has been built specifically to handle massive mobile phone data sets and additional open-source data. The research team used the Boston metropolitan area as a case study for analyzing cell phone records. The gamut of travel demand estimation using big data is presented through a discus- sion of methods, validation, implementation, and applications. Given the scope of this analysis, cases from other continents, such as Latin America and Europe, are not included, although these were also tested (Toole et al. 2015) to confirm the flexibility and applicability of the modeling framework. 3.4.2 Multiway Comparisons: A Case Study To mirror the thinking and approaches used by transportation modelers and planners, the case study compared and contrasted travel demand estimation results from CDR data, traditional sur- vey data, Census data, and the Boston regional model. These comparisons allowed the research team to get an understanding of the strengths and weaknesses of cell phone data and how these data could be incorporated into different aspects of travel demand modeling. The sources compared included the following: • Travel purposes, including home-based work trips (HBW), home-based other trips (HBO), and non-home-based trips (NHB); • Time-of-day patterns, including a.m. peak (6 to 9 a.m.), midday (9 a.m. to 3 p.m.), p.m. peak (3 to 7 p.m.), and early evening/night (7 p.m. to 6 a.m.); and • Geographic aggregation, including Census tracts and towns. Model comparisons relied on the following data sets and models: • Travel demand estimates based on raw CDR data. The data set included 2 million cell phone subscribers in the Boston metropolitan area for 2 months in 2010. Two different methods were used to extract travel patterns, which were compared with the other data sources for validation and evaluation purposes. • O-D matrices by a CDR data provider. These proprietary results by a third party use 3 months of 2015 CDR data that are adjusted by using the 2010 Census for the Boston region. The estimation methods and procedures are proprietary and not known to the project team. • 2010 Boston Travel Demand Model. The Central Transportation Planning Staff model results were used as the baseline for comparisons with the demand estimates from the raw CDR data and the commercially available third-party CDR data.

32 Cell Phone Location Data for Travel Behavior Analysis • Census Transportation Planning Products. These data were used to obtain journey-to-work travel flows for 2010 (Federal Highway Administration 2013) and to validate home and work inferences and the commuting flows estimated with the raw CDR data. • 2009 National Household Travel Survey (Federal Highway Administration 2009). This sur- vey provides information on the departure time distribution used in analyzing raw cell phone data. The survey data were also compared with the raw CDR data estimation results in terms of trip purpose distribution. • 2011 Massachusetts Travel Survey (Massachusetts Department of Transportation 2012). This survey, completed in 2011, was used as another independent source for evaluating the travel demand results obtained with the CDR data. The survey data were compared with the raw CDR estimates on trip purpose and travel departure time. 3.5 Summary This chapter takes the perspective of transportation agency staff and practitioners who are evaluating the purchase of aggregate cell phone data to supplement, enhance, or complement traditional data sources to support planning and modeling projects. The properties of big data and how CDR data fit in this picture are discussed in Section 3.1, which outlines the way CDR data are used to determine locations and the challenges of relying on the CDR data commonly available to transportation planners. In Section 3.2, the chapter shifts gears to consider the perspective of transportation planners and academics and discusses inter- views with practitioners working in MPOs, DOTs, and federal agencies. This section includes a checklist that planners can use when thinking about the uses of CDR data, the strengths and weaknesses of these data, and efforts to open up the black box. Section 3.3 describes the utility of CDR data as a source that may replace or augment tra- ditional surveys. The option of using CDR data to develop travel demand modeling metrics is discussed, along with the value of CDR data in developing individual model components. Summary tables are used to compare traditional surveys and CDR data with regard to sampling and expansion; how key elements of daily travel are recorded in surveys and CDR data; and how travel elements for aggregate analysis and model components can be captured by each data source. Section 3.4 concludes the discussion by presenting the research framework and case study approach used in Chapters 4 through 8. These sections describe how data and models from the Boston region were analyzed to compare and contrast measures derived from traditional sources to evaluate the strengths and weaknesses of CDR data.

Next: Chapter 4 - Description of Raw Data »
Cell Phone Location Data for Travel Behavior Analysis Get This Book
×
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

TRB's National Cooperative Highway Research Program (NCHRP) Research Report 868: Cell Phone Location Data for Travel Behavior Analysis presents guidelines for transportation planners and travel modelers on how to evaluate the extent to which cell phone location data and associated products accurately depict travel. The report identifies whether and how these extensive data resources can be used to improve understanding of travel characteristics and the ability to model travel patterns and behavior more effectively. It also supports the evaluation of the strengths and weaknesses of anonymized call detail record locations from cell phone data. The report includes guidelines for transportation practitioners and agency staff with a vested interest in developing and applying new methods of capturing travel data from cell phones to enhance travel models.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!