Making Trucks Count: Innovative Strategies for Obtaining Comprehensive Truck Activity Data (2014)

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.

37 C H A P T E R 4 This chapter provides detailed implementation plans for the three strategies that fared best in the feasibility reviews illustrated in the previous chapter. These plans can be consid- ered guides for further developing the program. 4.1 Using GPS Traces to Understand Trucking Activities Trucks equipped with GPS, or drivers carrying GPS- enabled devices (e.g., smartphones), create traces of the movement of each truck. Amassing these data using cloud technologies creates an innovative dataset that can provide origins and destinations by time of day. Appended with multiple attributes (e.g., commodity, ownership), many additional dimensions of truck activities can be created. 4.1.1 Current State of the Opportunity 4.1.1.1 Identification of Present Conditions The recent use of GPS trace data for truck-activity data patterns (e.g., Lioa, 2009; Greaves and Figliozzi, 2008; Zhao et al., 2011; Tardif and Li, 2011; Sharman and Roorda, 2010; Holguín-Veras et al., 2011a), the use of smartphones for GPS truck traces (Bell and Figliozzi, 2012), and GPS truck traces for backhaul analysis (Muckell et al., 2009), have provided ample evidence that this method of collecting data is via- ble. McCormack et al. (2011) demonstrated that GPS truck traces could be used to identify bottlenecks. In addition, GPS truck trace data can be used for performance measures (McCormack et al., 2010). The Freight Performance Measures Program funded by FHWA used third-party ownership of GPS truck trace data through an agreement with ATRI. The data were used pri- marily to assign average speeds to the national road network. Because of the third-party ownership arrangement, a small number of exploratory studies were conducted with these data. In addition, neither the characteristics of the trucks or firms were made available. The data feeds were opportunistic rather than generated from a randomly sampled set of trucks, making generalization impossible. FHWA is beginning a new program for data it will collect and own. As part of this new effort, there are plans to collect GPS truck traces from trucks that have been randomly sampled. To date, the FHWA approach has not been used to produce aggregate statistics other than average speed. A Transport Canada (2013) pilot study, however, demonstrated the capability of capturing enhanced GPS trace data suitable for creating aggregate statistics. 4.1.1.2 Data Sources and/or Analytical Methods With access to the original raw GPS truck trace data, it is possible to identify the origins and destinations by geocoding Detailed Implementation Strategies CHAPTER 4 KEY TAKEAWAYS • Of three implementation scenarios for a na- tional freight GPS framework, one pertaining to a new app, My Trip Matters, could produce detailed O/D and VMT data at multiple levels of geography within 5 years. • A new VIUS-like survey would likely follow the model of the Canadian Vehicle Use Survey (CVUS) as a wholly internal U.S. DOT program. • Agent-based modeling is largely an academic exercise at this time, and continued FHWA and TRB support would continue on a long-term development path.

38 the latitude and longitude variables. Merging these data with geographic information systems (GIS), or Internet mapping services (e.g., Google Maps, Open Street Maps), makes it pos- sible to identify the parcel where the trace began or ended, and to infer the type of load hauled from the attributes in the GIS data. A recent executive order by the Obama Administration describes details for this conceptual model (White House Press Release, 2013a). Figure 4-1 illustrates how cloud-based technologies can accomplish the goals of the executive order. The same strategy could be used to amass GPS truck trace data. The raw data would be assembled digitally in the Infor- mation Layer, made “mobile” through the Platform Layer, and then visualized in the Presentation Layer (White House Press Release, 2013b). General Transit Feed Specifications (GTFS and GTFS-R) is a good example of this concept where the Information Layer contains the transit scheduling and real-time transit GPS trace data assembled in a standardized format, moved to Google Maps for the Platform Layer, and visualized as bus routes in the Presentation Layer. The execu- tive order addresses the need for protecting the privacy, con- fidentially, and security of the data throughout each of layers. Any GPS truck trace data used in this strategy would receive review and full protection from disclosure, while still being made useful for statistics and other planning uses. For exam- ple, assembling the raw traces and storing them in a secure data center protects the privacy of the data, while only dis- playing aggregate de-identified visualizations mitigates any risk of the disclosure of protected information. As indicated in Table 4-1, GPS truck traces generate ori- gins and destinations by route, time of day, and speed. At the corridor level, direct metrics can be extracted from the traces. At larger geographies, cloud-based technologies can be used to “amass” the traces, yielding metro-area, multi-state, and national metrics. As mentioned, the data currently available have been gathered from cooperating, rather than randomly sampled, firms. As a result, it is not possible to generalize from the data using traditional statistical methods. The amassing process is not a statistical method and only provides the char- acteristics of the traces contained in the total volume of data. The raw GPS traces also provide VMT by route, time of day, and speed for each vehicle instrumented (see Table 4-2). A GPS trace accompanied by information on the type of vehi- cle, commodity being hauled, and whether truck is loaded or empty, as gathered by traditional surveys or electronically, can result in a multi-attributed trace with a large amount of useful data. Table 4-3 displays how origin and destination for all of parameters (except calculated averages) can be obtained directly. The direct metrics can be amassed for larger geogra- phies. Table 4-4 illustrates the gain in information for VMT by vehicle type, commodity type, load type, route used, time of day, and speed that is possible should multi-attributed GPS data be available. 4.1.1.3 Key Challenges Although some firms have been willing to provide GPS truck trace data through a third-party provider, others have been willing to share their data directly using a non-disclosure Source: Digital Government (White House press release, 2013b) Security & Privacy Information Layer Platform Layer Presentation Layer CUSTOMERS Figure 4-1. Layers of digital services. Geography Type of GPS Data Vehicle Type Commodity Type Load Type Average Load Route Used Time of Day Speed Average Speed Truck Stop NA I NA NA D D D C Corridor NA I NA NA D D D E Metro Area NA I(A) NA NA D(A) E(A) E(A) E(A) State NA I(A) NA NA D(A) E(A) E(A) E(A) Multi-State Region NA I(A) NA NA D(A) E(A) E(A) E(A) National NA I(A) NA NA D(A) E(A) E(A) E(A) Legend: D = Direct, E = Estimated, C = Calculated, I = Inferred, NA = Not Available, A = Amassed Table 4-1. Current GPS data for O/D.

39 Geography Type of GPS Data Vehicle Type Commodity Type Load Type Average Load Route Used Time of Day Speed Average Speed Truck Stop NA I NA NA D D D C Corridor NA I NA NA D D D E Metro Area NA I(A) NA NA D(A) E(A) E(A) E(A) State NA I(A) NA NA D(A) E(A) E(A) E(A) Multi-State Region NA I(A) NA NA D(A) E(A) E(A) E(A) National NA I(A) NA NA D(A) E(A) E(A) E(A) Legend: D = Direct, E = Estimated, C = Calculated, I = Inferred, NA = Not Available, A = Amassed Table 4-2. Current GPS data for VMT. Geography Type of GPS Data Vehicle Type Commodity Type Load Type Average Load Route Used Time of Day Speed Average Speed Truck Stop D D D C D D D C Corridor D D D E D D D C Metro Area D(A) D(A) D(A) C(A) D(A) D(A) D(A) C(A) State D(A) D(A) D(A) C(A) D(A) D(A) D(A) C(A) Multi-State Region D(A) D(A) D(A) C(A) D(A) D(A) D(A) C(A) National D(A) D(A) D(A) C(A) D(A) D(A) D(A) C(A) Legend: D = Direct, E = Estimated, C = Calculated, I = Inferred, NA = Not Available, A = Amassed Table 4-3. Multi-attributed GPS data for O/D. Geography Type of GPS Data Vehicle Type Commodity Type Load Type Average Load Route Used Time of Day Speed Average Speed Truck Stop D D D C D D D C Corridor D D D E D D D C Metro Area D(A) D(A) D(A) C(A) D(A) D(A) E(A) C(A) State D(A) D(A) D(A) C(A) D(A) D(A) E(A) C(A) Multi-State Region D(A) D(A) D(A) C(A) D(A) D(A) E(A) C(A) National D(A) D(A) D(A) C(A) D(A) D(A) E(A) C(A) Legend: D = Direct, E = Estimated, C = Calculated, I = Inferred, NA = Not Available, A = Amassed Table 4-4. Multi-attributed GPS data for VMT. agreement (NDA). For those not willing to share their data, more education may be needed on the value to their business of good data for freight planning. Lawson et al. (2002) were able to achieve a 60 percent response rate from private truck- ing firms when asking them about problems they faced while using the road network in Oregon. This traditional phone survey collected rich descriptive information on infrastruc- ture and operations-related problems (e.g., congestion, dan- gerous curves) at specific locations. Having a mechanism for capturing and reporting problems facing the trucking indus- try and making this information available for planning pur- poses has benefits for the freight community. These same infrastructure and operations issues can be identified using GPS and GIS. The raw GPS trace data can be kept confidential through the use of a secure data center (e.g., the Transportation Secure Data Center; see http://www.nrel. gov/vehiclesandfuels/secure_transportation_data.html). In a secure data center, algorithms capable of compressing and querying the GPS trace data can be used to identify infra- structure and operations problems. The dual use of the GPS

40 trace data to provide VMT and O/D metrics increases the value of the traces. The trucking industry would be contrib- uting to the betterment of the road network they rely upon through sharing privacy-protected data. 4.1.1.4 Key Opportunities The concept of collecting GPS trace data for trucks has been demonstrated in the Canadian Vehicle Use Study (see http://www.tc.gc.ca/eng/policy/aca-cvus-menu-2294.htm for documentation on methods, instrumentation, and ques- tions). The CVUS equips vehicles with GPS data loggers and an interface that allows for the capture of trip purpose and location, configuration and body style, cargo weight on board, and commodity description. It used a nested strati- fied sample survey to target owners of multiple vehicles. The pilot test used a random sample of 500 heavy vehicles, with one-third participating in the test. The study concluded that recruitment was difficult and driver’s cooperation with enter- ing data by hand was challenging. One of the recognized opportunities for GPS trace data is the very large generation of location data. To achieve the highest quality of data, collection intervals need to be very small to provide the greatest granularity (e.g., every second, every 5 seconds). The resulting massive dataset (BIG DATA) can be managed using new algorithms designed specifically for use with GPS. Amassing techniques include the use of single trace compression algorithms (Muckell et al., 2013) or multiple trace compression (Birnbaum et al., 2013). Although it is possible to grow the sample by deploying and redeploying data loggers, Bell and Figliozzi (2012) dem- onstrated the ease of using smartphones in the Truck Road Use Electronics (TRUE) data collection system. The State of Oregon currently collects a weight-mile tax (WMT) from all commercial trucks operating within the state. The TRUE data has advantages over previous GPS truck trace data due to its level of disaggregation and ability to report vehicle and commodity types. Smartphones have internal GPS and can be transformed into data collection instruments through the use of apps. The GPS data that is created can be incorporated into an Information Layer, managed in a Platform Layer, and visualized in a Presentation Layer. More than 92 percent of members in the Owner-Operator Independent Drivers Association have cell phones, with 33 per- cent owning a smartphone/iPhone/Blackberry, 8 percent owning a tablet, and 3 percent owning an e-reader. Although the number using these technologies has been growing, more than half of those who could download apps are reluctant to do so—perhaps because the average age of the membership is 55. Although cross-tabulations of mobile device ownership by age of members is not available, perhaps more will use smart devices as younger drivers enter the trucking industry. Drivers also may use such devices more as their price drops. At the same time, as the business advantages of using a smart device increase, firms will have a greater incentive to outfit their fleets with appropriate technologies to capture these benefits. 4.1.1.5 Expected Results Two key issues with respect to the use of GPS trace data are determining what the most cost-effective device is and getting cooperation from a fleet manager for GPS-enabled onboard equipment or driver cooperation with mobile devices. The cost of data loggers has decreased over the years and the size of the equipment has diminished. Most importantly, in the last 2 years, the capability of smart devices to provide high- quality GPS trace data means a dedicated GPS data logger is no longer required. Because GPS trace data can be generated from any number of devices and is formatted the same way, there is no need to purchase or install dedicated equipment on a truck. If the truck already has GPS-enabled devices on board, they can be used, as can any GPS mobile device the driver car- ries. The resulting data will be the same. The use of a smart device app makes the data transmission from the device easy. The cost of using the app depends on the mobile plan used with the device. Other means are more expensive (e.g., a dedi- cated transmission system from a GPS device). Alternatively, the data can just be stored in a data logger, although only storing data will not allow for real-time transmissions. One advantage of using GPS on smart devices is the capacity of sending less costly periodic transmissions. The second issue is getting cooperation from a fleet man- ager for GPS-enabled onboard equipment or driver coopera- tion with mobile devices. If the fleet manager or third-party provider can charge for the GPS trace data feeds, they will see it as a profit center. A driver needs to receive some compensation or benefits for providing data. This concept of “service value data harvesting” uses the original methodology of a “passive harvest” (e.g., traditional GPS data collection) and returns valuable information to the data provider in real time (e.g., apps that provide users instantaneous context data based on their current location or routes chosen). Drivers would receive information regarding events or circumstances related to their location (e.g., weather conditions, traffic conditions, alerts). These locational services already exist in the private sector. Many apps ask the mobile device owner for permission to extract location information from the device on a regular basis. Resistance to providing location information is reduced when the mobile device user receives benefits in return. Truck- ing companies perceiving data benefits would be willing to participate in an innovative data collection program, privacy protected and secured, capable of generating necessary truck data metrics.

41 4.1.2 Implementation Scenarios Three potential implementation scenarios are identified in this section. These can best be distinguished as a voluntary program, a mandatory program, and a voluntary program that would be differentiated by the provision of informa- tion in exchange for the capture of GPS data. The research team suggests that the third scenario be researched further by FHWA, perhaps as some sort of public-private partnership. 4.1.2.1 Voluntary GPS Collection Trucking companies use GPS equipment of their choice (e.g., fleet management system) and archive and transmit the GPS traces from each of their trucks using their U.S.DOT vehicle number. The administrative data associated with the U.S.DOT number would be appended to the GPS trace and post-processed into the standardized format used for truck metrics. The raw data could be transmitted to a private, secure center and post-processed or transmitted to FHWA for post- processing. In all cases, the data would be privacy protected and kept secure while in the Platform Layer. It would only be made available to other users in the Presentation Layer, where there is no risk of disclosure. Data processing services could be offered by secure third-party providers, trucking asso- ciations, and ATRI, or similar entities. Using a standardized post-processing methodology makes data from any source part of the collection and analysis effort. 4.1.2.2 Required GPS Collection for Hazmat Trucks Trucks hauling hazardous material (hazmat) loads would be required to transmit GPS trace data for safety and security of their load. The GPS traces would be associated with truck- specific and load information and post-processed for truck metrics. Trucks involved in any accidents or incidents would be subject to very high fines if not current in their transmit- tals. Trucks hauling any type of load could receive a subsidy on their insurance costs if they participate in the GPS trace transmission data program. Additional incentives could be introduced for streamlining licensing and permitting and ensuring the availability of truck parking services. 4.1.2.3 GPS Collection in Exchange for New Information Service: “My Trip Matters” An app (e.g., My Trip Matters [MTM]) would be devel- oped to provide operators with valued services and informa- tion in exchange for information on their truck movements. This business model is one that has been successful in the app world, relating data emanating from location-based sys- tems (Herrera et al., 2009; Hann et al., 2002; Cruickshanks and Waterson, 2012). In this proposed strategy, the app would have three service levels that could be offered to truck operators, each with progressively more detailed information exchange requirements and capabilities. The first level (e.g., basic) would prompt the user with the statement My Trip Matters would like to use your current location. Don’t allow/OK. If the user indicates OK, then the GPS data from the device will be transmitted to a secure data center to be amassed in an Information Layer that compresses the trace data. The Plat- form Layer would query the GPS compressed traces for route, time of day, and speed, using a standardized format (e.g., similar to GTFS) with industry acceptable buffers regarding exact locations (e.g., block, census tract, county) for the origin and the destination of the trace. The trace data also would be processed using a series of queries to produce performance measures and to detect problems experienced on the infra- structure or in operation. The Presentation Layer would dis- play a variety of visualizations regarding all processed metrics, using any browser on a base map desired by the end user. The Presentation Layer could be accessed through password- protected portals, or be viewed in a stylized manner that pro- tects against disclosure or unintended locational information at levels deemed unacceptable by the data provider. The second service level (e.g., premium) also would prompt the user with the statement My Trip Matters would like to use your current location. Don’t allow/OK. When indicating OK, the user would complete a one-time short demographic survey screen asking if the driver is a for-hire carrier or an owner-operator and whether the truck being driven is a ship- per-owned truck or a service truck (construction, utility, or other services). These demographic details would be trans- mitted with the GPS trace data and compressed in a similar manner as the Basic MTM. At the Presentation Layer, this service level would also illustrate metrics for route, time of day, speed, O/D, VMT, and problems on the infrastructure by type of driver and type of truck. The third level (e.g., platinum), would include the func- tionality of the second level, but would have an additional pop-up to query the driver for information on the commodity hauled, the registered and actual weight of the truck, and the load status (empty or loaded). This pop-up could be motion and time sensitive and could appear after a specific length of time or lack of motion to ask the driver if the information is still correct. The driver also could use voice-recognition technologies to complete the device’s survey. When these platinum-level data are transmitted, the entire set of variables would accompany the GPS trace and be compressed. The Pre- sentation Layer would be able to display any combination of the available variables for analysis. There would be metrics for route, time of day, speed, O/D, VMT, and problems on the infrastructure by type of driver and type of truck, commod- ity, weight of the truck, actual weight, and load status.

42 An available national sampling frame (e.g., MCMIS filtered for a stratified sampling of firms) could be used to recruit truck operators into the app-enabled data collection system. E-mail addresses are available in the set of attributes in the MCMIS census file. A traditional survey recruitment strategy could be used to contact firms and offer a random sample the use of the MTM app. To encourage participation in the use of MTM apps, the apps could have users compete for points, score points for a lottery, or “keep score” on their own statistics for best perfor- mance, longest trip, etc. Users could also receive feedback on the benefits their data have made to the planning process, par- ticularly with respect to future planning for extreme weather events. Another potential incentive that the app offers is a bet- ter safety score as a result of submitting data. The resulting data would be weighted at a future date in the Platform Layer with Travel Monitoring Analysis System (TMAS) continuous count data and weigh-in-motion (WIM) data. The weighted data would be visualized in the Presenta- tion Layer and illustrated in a manner to alert the user that the visualizations are estimations based on a random sampling strategy rather than on original trace data. An additional data stream could be appended using Commercial Vehicle Infor- mation Systems and Networks (CVISN) (see www.fmcsa.dot. gove/facts-research/cvisn/index.htm) information and could provide validation for some of the desired data fields. 4.1.3 Potential 5-Year Progress for MTM The entire MTM data program could be up and running in 5 years. Freight planners and researchers would be able to generate statistics and identify and monitor problems on the infrastructure system. Any additional third-party GPS or FHWA Performance Measures GPS trace data could be appended in the Platform Layer and visualized in the Presen- tation Layer, in addition to any other GIS shapefile features or attributes. With sufficient volumes of data, regions would be able to visualize more disaggregated data. This would be similar to strategies used in the American Community Survey (ACS) by which 5-year aggregates increase the level of detail provided without violating disclosure rules. An added feature that could be included within a five-year period would be integrating 511 system feeds. Currently, states produce 511-information and transmit alerts and warnings in a variety of ways (e.g., Washington State DOT sends emails for incidents on its freight system). Within 5 years, MTM could consolidate all transmissions from each state into the Plat- form Layer and provide MTM users with customized feeds (e.g., user-defined information) in their Presentation Layer in a timeframe chosen by the user. This location service would provide the MTM user with all relevant information across states and on any route, particularly current weather condi- tions and any forecast window the user chooses. Members of the International Reciprocity Program (IRP) who travel in more than one state would benefit the most from a harmo- nized 511 system. All of the spatially-relevant information could be assem- bled and appended to the GPS traces, regardless of the user display choices, and made available for safety and behav- ioral analyses conducted at the secure data center and only released in aggregate or visualized in the Presentation Layer with appropriate disclosure rules (e.g., minimum number of traces to prevent exposure). The Platform Layer would archive all weather-related data for the traveled portions of the network, making it available for asset management and safety analysis. This aspect of the MTM system illustrates the “capture data once, use it many times” strategy. New sets of metrics could be generated on all the previously described factors by characteristics (e.g., VMT by vehicle type) and with all the newly appended conditions and situations data (e.g., VMT by vehicle type by extreme weather condition). The MTM user can find information on the most effi- cient routing. (Fleet management and GPS services already have this capability, but not in a consolidated manner with the ability for users to give feedback or confirmations of the information.) This feature would allow MTM users to query the best route from their origin to their next destination and receive information that is most relevant to freight commu- nity members. This service could include “off-hour” delivery information in large urban areas. 4.1.4 Rationale for Recommendation of MTM 4.1.4.1 Technical Feasibility The technology required to develop a location service app is well developed. The GPS processing techniques will require some experimental research to refine existing algorithms or develop new algorithms to report the VMT, O/D, and speed parameters. These techniques are already known and can be made into Open Source processes. The more advanced met- rics would require algorithm development. The compression techniques for single and multiple GPS traces are available in Open Source, but further research will be needed to compress appended attributes to the trace. The data could remain in its original form, but would require extension storage capacity— available as cloud storage. Although the use of these tech- niques is new to the freight community, they are already in use by private industry and app developers. 4.1.4.2 Institutional Feasibility FHWA has been successful in obtaining GPS trace data from a third-party vendor, although the use of the data was

43 constrained. The current contract that FHWA is complet- ing will allow it ownership of the GPS data and the ability to share the data (assuming non-disclosure agreements) with state DOTs and MPOs. Using an app to capture and trans- mit the GPS trace data to a secure location is supported by a presidential executive order, which also provides a strategy to manage Open Data and provides protections for privacy and security. The Transportation Secure Data Center (TSDC) has been safeguarding GPS trace data that derive from travel sur- veys for several years and has a mature program to make the data available for research purposes without disclosing it. The provisions of the Open Data Policy provide for the protection of privacy and confidentially. The ability to garner cooperation in providing GPS data remains a challenge at the federal level. Precedence for man- dating the collection of freight in federal legislation can be found in SAFETEA-LU. 4.1.4.3 Operational Feasibility The current abilities of data management systems and platform processing with cloud-based services are now well established. There are very few risks involved with retriev- ing GPS trace data from a smart device, processing it on a Platform Layer, and visualizing it on a Presentation Layer. Although these services are limited in freight transportation, they have been demonstrated to provide services similar to the ones required. With a standardized data format and Open Data post-processing program, any number of options could be used to transmit data. An alternative to a central repository would be the web- accessible archiving process being used for the transit industry—the GTFS data exchange that allows the Platform Layer to act as a cloud aggregator (see http://www.gtfs- data-exchange.com/). The organization of data sources could remain flexible and adaptable regarding the data collector’s preference and still be completely accessible for a Presen- tation Layer. This greatly reduces resource commitments. Alternatively, all post-processed data could be housed in an FHWA cloud environment. 4.1.4.4 Geographic Scalability There are no known constraints with respect to geographic scalability with the transmission and processing of data, given the recent availability of cloud-based technologies. There could be variations in participation by regions of the United States (e.g., truck drivers on the West Coast being more tech- ready than those in the Midwest). Using a random sampling frame would make it possible to generate the spatial resolu- tion desired for varying purposes (e.g., over-sampling at the local level for small geographies). If the data program is federally mandated, then concerns regarding sufficient coverage would be overcome with nearly complete participation by the freight community. 4.1.4.5 Financial Feasibility An Open Data system that relies on existing mobile devices, inexpensive cloud-based flexible data storage contracts, and cost-effective development of algorithms to extract the vari- ous metrics would have no large operational costs. Although there are no large operational costs anticipated when using an app-based data collection, storage, and visualization strat- egy for creating innovative truck-activity data, there would be costs associated with the development of the Web-based infrastructure. For example, a decision would need to be made regarding which devices would be used for data col- lection. Costs for the development of “native” apps capable of generating GPS data could cost approximately $50,000 per operating system type. The data generated by the MTM app would need to be stored, and although storage is much cheaper with cloud-based third-party facilities, a general over- all cost of $60,000 per year for the data program (including storage, maintenance, and preliminary data processing) would be sufficient for very large volumes of data. To easily view, ana- lyze, and use the generated data for reporting out statistics and planning purposes, the development of a visualization app may be of benefit. The cost of developing a visualization tool would be approximately $150,000. Thus, the MTM data program, with two data collection operating systems, a stor- age program, and a visualization/data analytics app would cost, in total, $310,000 in the first year, and $60,000 (or less) to operate in future years. Another important cost consideration is the need for computer services and personnel to maintain the platform services and keep current with browser capacities for pre- sentation services. Staff and third-party providers would be needed to maintain the code and move seamlessly across plat- forms and through upgrades in the various “working parts” of the system. These costs are present in all cloud-based tech- nology deployments and not specifically due to the needs of the freight community. Further, technology-oriented solu- tions have the added advantage of fostering innovation and attracting private-sector participation in the development and implementation process. 4.2 A New VIUS-Like Survey The TIUS/VIUS series, which had been initiated in 1963 and conducted as part of the economic census, ended in 2002 due to financial constraints. This survey was central to provid- ing national- and state-level measures of the scope and char- acter of the trucking industry. A new survey would establish a

44 system with consideration of opportunities and requirements for modifications and expansions of the approach and content of the old VIUS. 4.2.1 Current State of the Opportunity 4.2.1.1 Identification of Present Conditions Loss of the VIUS affected road-based freight statistics. VIUS provided vehicle characteristics and differentiation of vehicle activity by range, products carried, miles traveled, industry, and type of operator, all at the state level. The end of the VIUS program has created a gap in national and state understanding of fundamental trucking activities and fleet attributes. One gap is the ability to differentiate personal and business uses of pickups and sport-utility vehicles. No new statistical approach has arisen to replace the VIUS capabilities. 4.2.1.2 Data Sources and Analytical Methods The new survey would follow the same analytical methods as the historic VIUS. That survey was conducted as part of the economic census every 5 years, with a sample purchased by the Bureau of the Census from the R.L. Polk Company, a provider of state vehicle-registration information. The survey was mailed to owners of more than 100,000 selected vehicles in each year of the survey. It was managed by Census as a standard procedure mail-out/mail-back survey of the economic census. The sample was stratified by geography and truck characteris- tics. The geographic strata were the 50 states and the District of Columbia. The five truck characteristics strata were • Pickups; • Minivans, other light vans and sport utilities; • Light single-unit trucks (GVW 26,000 lbs. or less); • Heavy single-unit trucks (GVW 26,00 lbs. or more); and • Truck-tractors. Individual state summary reports, as well as a national summary report, were produced. 4.2.1.3 Key Challenges The VIUS survey was cancelled due to financial issues. Therefore, establishing a comparable or expanded survey would be a financial issue. Such a survey could again become the baseline for truck freight activity, establishing the uni- verse characteristics of the truck fleet. It also could serve as an effective basic guide to design and survey planning for other truck surveys and statistical analyses. The researchers are unaware of challenges in the survey design or process that would threaten a new survey program. In each quinquennial undertaking, additions and deletions were made to keep the survey current and relevant to both private and public users. Like other similar surveys, the VIUS suffered from decreas- ing response rates, from 90 percent in 1992 to less than 80 per- cent in 2002, even with mandatory reporting (ORNL, 2010). 4.2.1.4 Key Opportunities One opportunity would be to reinstate the survey as it would have been done in 2009 when it was defunded. While this has the benefit of clarity of purpose and execution, other options to the traditional approach are possible. Justifying reinstatement requires not only demonstrating the value of the data but also evaluating prospective methods in light of past experience and present opportunities. Several opportunities are drawn from the previous analy- ses undertaken in this research effort. These opportunities can be stratified into three major groups. 1. Consideration of a new, and perhaps expanded, sample frame employing the traditional Polk procedure or the MCMIS as the source of the frame. Such consideration could include private vehicles, government, and motor carriers of passengers employing the state registration files via Polk. Such consideration would have to recognize that the MCMIS could not address personal passenger vehicles. 2. Analysis of state vehicle-registration records working with R.L. Polk and of federal records maintained by MCMIS for their ability to provide supplemental data supporting any vehicle-based data collection effort. 3. Monitoring CVUS and consideration of a parallel U.S. prototype. At the time of this publication, FHWA is pro- gressing with a pilot demonstration of the CVUS technol- ogy in the United States. These are shown in Figure 4-2 and described further below. 4.2.1.5 Sample Frame Opportunities A VIUS-like survey could assist road transportation statistics. It would recognize and use a third statistical universe unique to road transportation, vehicle-registration files, in addition to the traditional household and establishment universes. One possible weakness of the past VIUS approach is that its sampling of registrations were not fully used. Particularly, the VIUS excluded vehicles owned by governments, ambulances, buses, motor homes, and automobiles. Such consideration is beyond the scope of this work; however such an expansion could be a potential information resource in areas where no effective information sources now exist. Another sampling consideration is that, although drawn from publicly owned state registration files, the Polk resources

45 are the product of a private vendor. This would require gov- ernment purchases of private products, with the expense and complexities regarding uses and disclosure that entails. Proper survey design would require investigation of alternatives. One alternative would be for U.S.DOT to contract directly with Polk rather than through the intermediation of the Census Bureau. Thus, the U.S.DOT would own, fund, and manage the survey directly as a central part of its statistical programs. One potential conflict would be the use of legally supported manda- tory reporting. The Census Bureau employs such reporting in this survey under Title 13. Conceivably, U.S.DOT could employ it under Title 49, subject to legal review. Another potential option would be to use MCMIS. This system provides extensive detail on establishments that own motor carrier vehicles of passengers or freight. Evaluation of the basic file establishes that there is extensive detail about the owning establishments, such as the nature of their industry, whether their transport structure is interstate or intrastate in nature, general characteristics of approximately 30 com- modity types carried, a general vehicle classification, and a count of vehicles and drivers by range of operations. Safety characteristics and hazmat characteristics are also a compo- nent of the dataset. The MCMIS data may be a potential in-house data resource for U.S.DOT. One possible constraint of the data is a sam- ple frame based on establishments and their characteristics without detailed identification and characteristics of indi- vidual vehicles. Therefore, the VIUS approach would have to RENEWED VIUS SUPPORT DATA SAMPLE FRAME POLK POLKMCMIS MCMIS MONITOR CANADIAN CVUS POSSIBLE CANADIAN US COMPARABILITY Figure 4-2. Key approaches to establish a new VIUS-like survey.

46 sample establishments and then, in a second phase, have those establishments sample their own vehicles, given instructions from the surveying agency. A caveat is having establishments draw their owns samples may introduce bias and create edit- ing work for those doing the processing. This would be much like the sampling strategy of the CFS but would need to be considered and pretested with necessary control mechanisms. It may only be able to be used as a sample frame for highly targeted approaches, but might be able to offer advantages of cost and control in using MCMIS. 4.2.1.6 Supplemental Data Support to the VIUS Approach Both R.L. Polk and MCMIS prospectively provide oppor- tunities to develop data about the truck fleet and its activities. R.L. Polk maintains a Web-based dataset as a public service and as a guide and advertisement for its more tailored prod- ucts. This dataset is derived from a process of monitoring all state vehicle-registration systems for changes in registrations with considerable vehicle detail. Its Commercial Vehicle Mar- ket Intelligence Report summarizes “Market Performance,” i.e., registrations of GVW3 through 8 by make, month, engine type, and trailer registrations by type. Registrations include new and used vehicles. It may be helpful to examine further opportunities to employ these data for independent analysis or to supplement VIUS efforts. Similarly, the potential of MCMIS as a statistical data source may supplement the VIUS undertaking. MCMIS has universal coverage of all truck fleet owners with summary descriptions of their vehicle fleets. However, the files are con- tinuously updated with no single “snapshot” of what exists at a given time. FMCSA staff said that those listed are not obliged to notify FMCSA when they cease business, so it can be difficult to find those responsible for updating the data or providing further information. As noted previously, the establishment records in MCMIS provide information at a summary level on truck fleets owned, including the following: • Establishment characteristics carrier/shipper interstate or intrastate, • Cargo transported (30 freight categories), • Equipment owned/leased (5 freight categories), • Drivers employed/leased by interstate/intrastate within or outside 100 miles, and • Highly detailed hazardous materials carried. In addition to R.L. Polk and MCMIS, the Services Annual Survey and other sources such as the Cass Freight Index could be examined further for their effectiveness in supporting the VIUS design and ultimate dataset. SAS reports trucks owned by type, revenues received by 11 types of products, operat- ing expenses such as personnel, fuels, and purchased freight transportation. These are produced for various truck operators, general freight carriers, TL/LTL, and specialized carriers for local and long distance. The SAS only covers for-hire motor carriers rather than all establishments with trucks. 4.2.1.7 Monitoring CVUS CVUS may produce as much data as previous VIUS-like approaches and do so more quickly and cheaply. CVUS draws its sample from provincial registration files so that distances, x-y coordinate locations, and fuel consumption can be recorded, conducted on a quarterly basis through use of an instrument connected to the engine port. In the present design, CVUS includes all light vehicles (less than 4.5 tons), medium trucks (between 4.5 and 15 tons), and heavy trucks (15 tons or more). It excludes such vehicles as motorcycles, buses, motorized equipment, and off-road vehicles. The present plan also permits the recording device to obtain answers to queries posed to the vehicle operator. These have included trip purpose information and a count of persons on board. Questions can be tailored to truck operators as desired. The passenger version of the survey has been under testing since late 2011 with more than 8,000 vehicles tested generating a dataset with more than 500 million observations. The truck version has had a single test so far with interested carriers in the summer of 2012. Successful data were developed from 72 vehicles with 440,000 observations per truck. These data are accumulated from a data-logging device, the Ottoview autonomous electronic data logger, connected to the engine, which monitors all aspects of the vehicle’s function. A touch screen device also is being tested for the driver. This device obtains reason for stop, trip purpose, land use at location of stop, vehicle configuration, vehicle body type, cargo unit measure and weight, and broad cargo description (8 categories). Each time the driver picks up or delivers goods, the driver is instructed to record a new trip. Trips defined as vocational (no goods delivered/contractors, etc.) are given a different set of ques- tions. Some stops, such as those for fueling, changing drivers, food/rest, or equipment checks, are not considered trip stops. A more recent comprehensive full-scale test in Ontario achieved a 41 percent acceptance rate, but data were only captured among a third of vehicles. In addition to recruit- ment issues, this test had to deal with driver lack of skill in responding, data port issues, and survey fatigue among large fleets. In short, there are both technical and logistical threats to potential success. Under the existing approach, the provinces agree to have their registration files accessed for sampling purposes, and truck operators (and private users in the passenger survey)

47 must be recruited to have instruments placed on their vehi- cles. CVUS offers various inducements to gain acceptance. Such a survey approach can provide VIUS-like data products in even greater specificity, precision, and speed. It is, in effect, a full-scale vehicle-based O/D survey. 4.2.1.8 Expected Results All of the lines of inquiry on a VIUS-like dataset can be con- ducted in parallel and a design for the reinstatement of the VIUS completed over a 2-year cycle. Given sufficient lead times, the financial structure for the survey also can be put in place. 4.2.2 Implementation Scenarios The multi-path process envisioned generates several sce- narios, as follow: Scenario A: Direct Reinstatement assumes that the past pro- cess is the best model for the present design. Scenario B: MCMIS Alternative assumes that a MCMIS sample frame is viable and a wholly internal DOT program is established. Scenario C: CVUS Model Implemented assumes that moni- toring of CVUS activity in Canada demonstrates that this approach can be the basis for a new VIUS structure and replaces the existing options in the long term. 4.2.3 Potential 5-Year Progress It is envisioned that the multi-path process could be com- pleted and a survey system put in place within 5 years. Given the path selected, design would require 6 months to a year, and pre- testing would require another year. The historical VIUS process was a quinquennial undertaking conducted in parallel with the economic census in years ending in 2 or 7. This would create a mismatch in schedules and could force a delay until 2022 if the 2017 window were missed. There are no reasons to tie the survey to the economic census, although that was its genesis. If DOT were to take responsibility, it could conduct the survey when- ever scheduling permitted, such as an annual process. When it is determined whether CVUS can be implemented in the United States, such a survey could be implemented in 5 years. 4.2.4 Rationale for Recommendation that a VIUS-Like Survey be Harmonious with CVUS 4.2.4.1 Technical Feasibility The survey was conducted by the Census Bureau within the last 10 years, so the documentation is likely available to reestablish a new VIUS-like survey. Mining state vehicle-registration records would be straight- forward. The R.L. Polk Company has maintained those records for decades. It may be possible to negotiate with each state jurisdiction separately, but this may be complex and costly. Polk provides access to summary material from the states at no cost on its website. There would likely be no tech- nical barriers to overcome, other than the issue of declining response rates common to other survey approaches. Expansion of TIUS to VIUS involved only the name change in 1997. Incorporating all vehicles into the survey and obtaining a national framework of the entire motor vehicle fleet’s characteristics and activity represents a third universe, beyond households and establishments, that would be avail- able to the transportation community. There may be technical challenges with the MCMIS option in sample and survey design. Respondents are expected to update their records every 2 years and there are lags and gaps in reporting. CVUS is underway, providing instruments to a sample of vehicles to monitor total daily trip making with onboard equipment requiring some driver input (e.g., trip purpose, occupancy). The United States could follow this development and perhaps a more formalized relationship. Early results indicate that caution may be required in pursuing such a sur- vey on a large scale. 4.2.4.2 Institutional Feasibility BTS would fulfill the administrative roles for all these pro- spective activities. The relationship with Canada is directly BTS’s role and responsibility. It is also possible that VIUS reinstatement and expansion, and the data mining activi- ties of registration files, could be undertaken by FHWA or FMCSA. The Census Bureau may not need to have a role in VIUS activity—a direct Polk relationship with DOT might prove more effective in engaging ultimate users closer to the product. An internal DOT effort built around MCMIS might involve no more than inter-administration coordination. Publication of findings might be an issue in a Polk arrange- ment, particularly with proprietary materials. More special- ized and detailed access might require negotiation with Polk regarding protection of their arrangements with states and attendant costs. A similar relationship with FMCSA would not have the proprietary concerns but would have public disclosure constraints. There may not be legal barriers to an observer status regarding CVUS for the U.S. BTS and the North American Transportation Statistics interchange. 4.2.4.3 Operational Feasibility There are minimal operational conflicts to reinstating the survey. There probably would be small operational changes

48 at the Census Bureau were the survey to be expanded to a full VIUS. One operational option is an annual approach that culminates in a detailed survey over 3 to 5 years. This would likely add to purchase costs with Polk. Minor opera- tional changes also might be required of Polk for sampling and supporting any prospective data mining operation. It would be beneficial to consider the MCMIS approach in parallel with the Polk effort. FMCSA may provide copies of files to BTS for research and analysis. There do not appear to be any operational constraints to a prospective Canadian relationship. 4.2.4.4 Geographic Scalability Given the nature of the registration files, the survey prob- ably would be a state-level survey, with roughly equal state observations, aggregated for a national summary picture. The 2010 Oak Ridge National Laboratory (ORNL) study proposed a series of sampling options to improve national statistical quality at lower cost but apparently at the expense of quality. If the intent were to establish national truck VMT and ton-miles, then the sampling design could be modified considerably. 4.2.4.5 Financial Feasibility Financial resources will be central to this effort. There likely would be minor but not substantial fees involved with Polk. Much of the materials of interest are available free from Polk and MCMIS. Data processing costs can be more with the MCMIS approach. Were the survey approach to become an annual effort, the costs of selecting the Polk sample could be more than mining the registration files. Similarly, monitoring CVUS would probably entail no more than a minor cost. Of course, instituting such a sur- vey in the United States would be a long-term and expen- sive undertaking, with a possible trade-off between the costs of an annual survey and that of using onboard equip- ment. The logic of the CVUS approach might well favor a MCMIS approach. The reinstatement of VIUS and expansion to a full vehicle survey would probably be a significant undertaking. An analysis conducted for the FHWA Office of Freight Manage- ment and Operations by ORNL in 2010 estimated program costs for the expanded survey at about $12 million per sur- vey cycle. This was a maximum, and costs—depending on design—might be as low as $9 million. Such costs would be subject to negotiations with R.L. Polk for developing the sample structure. The MCMIS approach might be slightly less expensive; where file operations are substituted for pur- chase costs as in a Polk arrangement, this approach may have greater survey expenses. 4.3 Agent-Based Models for Freight Transportation Agent-based modeling (ABM) is a new modeling approach in the field of freight transportation modeling. A main char- acteristic of these frameworks is that firms are modeled as the decision-making units, they interact with each other directly within the modeling system, and their future behavior may be altered by past interactions. This approach seeks to improve state-of-practice models that are typically four-stage mod- els, through better representation of firm behavior and their logistics and supply chains decisions (Hensher and Figliozzi, 2007; Roorda et al., 2010). This research addresses innova- tive strategies for obtaining truck-activity data. Such strate- gies typically include data “collection” or data “capture.” This strategy deals with data “production,” that is, data that are the outcome of model simulations. References for use of modeling-based approaches for the creation of new data (e.g., a new data source) include NCFRP Report 26: Guidebook for Developing Subnational Commodity Flow Data (http://www.trb.org/Publications/Blurbs/169330. aspx); NCHRP Report 738/NCFRP Report 19: Freight Trip Generation and Land Use (http://onlinepubs.trb.org/online- pubs/nchrp/nchrp_rpt_739.pdf); and NCFRP Project 25-1, Estimating Freight Generation Using Commodity Flow Sur- vey Microdata (http://apps.trb.org/cmsfeed/TRBNetProject Display.asp?ProjectID=3492). 4.3.1 Current State of the Opportunity 4.3.1.1 Identification of Present Conditions Agent-based modeling is “a computational method that enables a researcher to create, analyze, and experiment with models composed of agents that interact within an environ- ment” (Gilert and Terna, 2000). ABMs are often also character- ized by agent learning and complex behavior that is emergent from models with simple rules. In the transportation literature, agent-based models are sometimes referred to as “microsimu- lation models.” Strictly speaking, agent-based modeling can be considered an extension to microsimulation, in which agents interact, learn, and adapt, often leading to emergent behaviors. Several frameworks have been proposed recently (Roorda et al., 2010; Samimi et al., 2010; and Liedtke, 2009). These proposed frameworks vary in their details but typically they model the interaction between firms (strictly speaking, firms can include one or more business establishments, although some published papers used the terms firms and business establishments interchangeably) that act as shippers, carriers, and receivers, and include the following key processes: • How firms start, select their locations, and make other long-term strategic decisions (in most cases this is treated as exogenous to the framework);

49 • How firms select other firms to supply goods and services (a supply-chain choice), resulting in the exchange of a certain fixed amount of commodity over a defined period of time; • How firms, as shippers, determine shipment size and fre- quency (although sometimes the receiving firm is also involved in this decision-making process); • How firms, as shippers, make logistics decisions (choice of mode—trains vs. trucks, whether to consolidate— bundling of shipments or use warehouses or distribution centers, vehicle types and vehicle scheduling) or make out- sourcing decisions (choice of carriers); and • How firms, as carriers, make route choice (this often includes tour-based modeling). The main focus of this strategy, truck movements, is pro- duced at the end of all these processes. Some of the individual processes, such as tour-based modeling and logistics choices, are not ABMs per se, since agents only make decisions but do not interact with one another directly in making the deci- sions. Nevertheless, it is important that all of the processes are discussed here, as they are integral parts of the ABM framework. Many of the proposed frameworks are similar in prin- ciple, but different research groups make different technical choices to operationalize the ABM framework. For example, to model how firms select other firms to supply goods and services, Wisetjindawat et al. (2007) used multinomial logit choice models to determine the supply chain based on buyer- supplier pair characteristics and characteristics of the indus- try; Liedtke (2009) proposed the use of a guided Monte Carlo algorithm that considers product availability, transportation costs, and supply-chain vulnerability (sensitivity to distance); Roorda et al. (2010) proposed a method that examines in great details how firms interact in markets, form alliances, and agree to contracts; and Outwater et al. (2012) plans to use game theory to understand firms’ procurement strategies and thus commodity flow origins and destinations (from notes taken at “An Agent-Based Economic Extension to CMAP’s Mesoscale Freight Model,” Chicago Metropolitan Agency for Planning (CMAP), September 19, 2013). To date, research in ABM for forecasting freight demand has no operational prototype for policy analysis. Although Outwater and colleagues are developing a demonstration model for CMAP, it is a proof of concept model, which is not calibrated with real data. If implemented, ABM could be used to generate informa- tion on supply chains (i.e., the interactions among shippers, receivers, and intermediaries). Currently, there is no data pro- gram that is responsible for capturing, analyzing, and deliv- ering supply-chain information. One of the key challenges for obtaining this information is the complexity of a supply chain. The ABM models specifically simulate the simulta- neous operations and interactions of multiple agents, in an attempt to re-create and predict the appearance of complex phenomena. The process is one of emergence from the lower (micro) level of systems to a higher (macro) level. As such, a key notion is that simple behavioral rules generate complex behavior. These principles could, and are, being used (e.g., the model for CMAP) to re-create supply chains. 4.3.1.2 Data Sources and/or Analytical Methods The development of ABM requires highly disaggregate data, and this is the biggest challenge in implementing this approach. Early on, the main difficulty was an uncertainty about what types of data sources would be necessary to model carrier behavior. Identifying the relevant quantitative and qualitative data sources, obtaining them, and validating the model would seem to require a significant amount of effort. Without an existing data platform, the initial expense for this strategy would likely be cost prohibitive. If such an investment were made, the model development costs could be quite modest. Samimi et al. (2010) has outlined the four types of data that are needed for the development of an ABM for freight transportation that provide focused direction on the contents of a necessary data platform: (1) business establishments, (2) shipments and supply chains, (3) the freight transporta- tion networks, and (4) zone-to-zone freight flows. Business establishments—To provide the necessary base population of business establishments, information about the establishments by geographic zone is needed. In the ideal case, for each establishment there is information on employee size, annual turnover, square footage, number of franchises, types and amounts of the commodities coming into and shipped out of the establishment by tonnage, and dollar value. Samimi et al. (2010) commented that obtain- ing accurate information on the commodities is challenging, since most firms work with a wide variety of goods. Samimi recommended the best available data for a model in Chicago was County Business Patterns (CBP). It is an annually updated dataset that provides the number of estab- lishments, number of employees, and payroll data by industry classification by geographical zone. However, the usefulness of this dataset is limited due to data suppression for business confidentiality reasons. There are many missing values in the disaggregate dataset, and key information such as the number of employees and payroll data are only provided in aggre- gated form. Urban et al. (2012) subsequently documented the use of this data for the development of the Chicago model for CMAP. To get information about commodities coming into and shipped out of an establishment, Samimi et al. (2010) rec- ommended the use of “input-output accounts” that provide

50 general information on the type of industries producing or using a specific type of commodity. Input-output accounts also provide information on the values of the required com- modities to produce a unit output of an industry. This dataset is also used in Urban et al. (2012). Shipments and supply chains—Samimi et al. (2010) dis- cussed supply-chain data and shipment data under a single heading, as it is conceivable that one can construct the supply chain for a given firm based on the detailed information about the shipments. Ideally, for each acting agent involved in the shipping process (shippers, receivers, and other intermedi- aries), information regarding their primary activity, number of employees, annual turnover, establishment square foot- age, and number of branches would be useful. Additionally, the shipment characteristics (e.g., weight, value, dimensions, time sensitivity, commodity type, origin, and destination of the commodity) and shipping specifications (haul time, cost, mode, and damage risk of the shipping process) also would be important. The CFS provides a foundation for these data elements. Samimi et al. (2010) developed a tailored survey (known as FAME) to fill this data gap. Similarly, in Canada, to under- stand supply chains, Cavalcante and Roorda (2013), have conducted their own survey (a stated preference survey) to understand how contracts are formed between shippers and carriers. Shipment and supply-chain data are more available in Europe. Advances in the understanding of shipment size and mode choice were made possible by the Swedish Com- modity Flow Survey (CFS) (e.g., De Jong and Johnson, 2009; Habibi, 2010; and Windisch et al., 2010), and the French ship- ment database known as ECHO (e.g., Arunotayanun and Polak, 2009; Combes, 2010). Freight transportation networks—As with all transpor- tation models, a representation of the highway and road net- work would be needed. Samimi et al. (2010) recommended the North American Transportation Atlas Data (NORTAD) be made available to the public by BTS. Zone-to-zone freight movements—The commodity flow between each zone pair for each specific commodity type is required for the validation of such a framework. If available, the dollar values and tonnage figures are necessary, and ship- ping distances would be useful. The categorization of com- modities such as distinguishing functional products that would be efficiency driven (e.g., meat and dairy) from inno- vative products (e.g., electronics) that would be price driven, is an important model design decision that needs to be made at an early stage of model development. Outwater and col- leagues have made some progress in this area as part of the model development for CMAP. Samimi et al. (2010) recommended the FAF as the most comprehensive survey-based data that is available publicly. It provides information on the commodity value and tonnage, by origin and destination zones, mode of transportation, and type of the commodity. Other data—Furthermore, for model validation, Samimi et al. (2010) recommended that shipment volumes at termi- nals and mode share for different final products may be used to assess the accuracy of the supply chains and logistics modules, and average daily truck traffic on links and roadside intercept survey of trucks may be used for validating the final truck flows. Additionally, not discussed in Samimi et al. (2010), or most other freight transportation research articles reviewed, is the potential use of panel surveys (i.e., undertaking interviews/ surveys with the same businesses multiple times at regular intervals). Although panel surveys are relatively difficult to conduct compared with cross-sectional data, they are par- ticularly suitable for measuring processes and transitions (Hassan et al., 2010). This kind of data will be particularly useful for understanding and validating emergent behaviors observed in ABMs. 4.3.1.3 Key Challenges There are challenges in both conceptual development as well as data collection, as follows: • Agents in businesses are highly heterogeneous and diverse (compare to passenger transportation models). For exam- ple, firms may differ in size by order of magnitude (whereas household sizes do not vary as much). Furthermore, there are huge variations in how firms operate (e.g., some want to minimize uncertainty, some want to minimize regrets, and some focus on maximizing efficiency). • Data availability is a barrier to progress. Data related to supply chains and logistics are often proprietary; firms tend to be unwilling to participate in surveys because the value of time is high in the business context. Within a firm, it is often not clear who makes logistics decisions. The data availability for small firms can be problematic. • ABM for freight transportation is a new kind of model. Its functionality and limitations may be unfamiliar to the transportation planning and policy-making community. For this new tool to become mainstream in policy analy- sis, dialogues with the DOT, MPOs, and other stakeholders about ABM would be needed. 4.3.1.4 Key Opportunities Draw on existing behavioral research. Although many of the ABM frameworks for freight transportation were devel- oped recently, many behavioral studies of the decision-making processes within ABM took place before the existence of these frameworks. A summary of relevant behavioral freight models that can be potentially adapted is presented in Table 4-5. The

51 development of new ABM models can be accelerated by draw- ing heavily on these previous and ongoing behavioral studies. Build on private-sector supply-chain and logistics research. Extensive research in logistics planning and supply-chain management has been done by the private sector. Research on ABM of freight demand modeling can build on this. Exam- ples of research that bridge this private-sector and public- sector gap include Maurer (2008) who integrated a commer- cial software package for logistics decision making (known as CAST) with a national freight model and Friedrich (2010) who demonstrated that the optimization techniques used in logistics planning of the German food sector can be used for wider freight transportation forecasting. 4.3.1.5 Expected Results Results that can be expected from these models are freight shipments by mode and path for origins and destinations. These results are expected to be more behaviorally realistic and accurate than those from conventional models. However, ABM modeling for freight transportation is still in its infancy, so the probability that it can be used as a suc- cessful data production strategy for truck-activity data in the short term (5 years) is low. 4.3.2 Implementation Scenarios Two implementation scenarios are discussed: a rapid devel- opment path and a long-term development path. The rapid development path would rely on the adaptation of existing freight behavior models and minimum data collection to cal- ibrate existing models, whereas the long-term development path would entail new behavioral research, data collection, and concept development. The rapid development path would result in a model for demonstration purposes only. It would be just a proof of concept and could not be used for forecasting purposes. The ABM model being developed for CMAP is in this category. An enhanced supplier choice model that builds on concepts from game theory is expected to be ready by June 2014. The long-term development path would generate new behavioral research through new data collection and concept development. There is consensus in the academic literature that better understanding of supply chains and logistics will be beneficial (De Jong, 2013). The understanding of mar- ket interactions and contract formation is a topic that is less understood but important to freight demand forecasting, as it recognizes that “freight transport is the output of an economic market which converts commodity flow into vehicle flows” (Cavalcante and Roorda, 2013). Previous studies sug- gested that confidentiality issues can be a problem. One way to circumvent such issues is the use of stated preference sur- veys as demonstrated in Cavalcante and Roorda (2013). This research team suggests that research on the use of ABM in freight modeling continue. The team believes that the long-term development path is the most likely. At the same time, furthering the development of such models is not high risk, as the unique benefits of applying ABM methods will Relevant Behavioral Freight Models Li ed tk e (20 06 ) H ol gu ín -V er as e t a l. (20 07 ) D e Jo ng a nd B en - A ki va (2 00 7) Co m be s an d Le ur en t (20 07 ) M au re r ( 20 08 ) Fr ie sz e t a l. (20 08 ) W an g an d Ho lg uí n- Ve ra s (20 09 ) D e Jo ng a nd Jo hn so n (20 09 ) A ru no ta ya nu n an d Po la k (20 09 ) H ol gu ín -V er as e t a l. (20 11 b) Firm synthesis X -- -- -- -- -- -- -- -- -- Supplier choice/shipper- carrier interactions/carrier- receiver interactions X X -- X -- X -- -- X -- Shipment size and frequency choice X -- X -- X -- -- X -- X Lo gi st ic s de ci si on s Mode choice -- -- -- -- X -- -- -- -- X Consolidation/Use of warehouse -- -- -- -- X -- -- -- -- -- Vehicle type choice -- -- -- -- X -- -- -- -- -- Vehicle scheduling -- X -- -- -- -- -- -- -- -- Route planning and tour formation X -- -- -- X -- X -- -- -- Source: De Jong et al. (2013) Table 4-5. Summary of relevant behavioral freight models since 2004.

52 become documented during this development time, which in itself will provoke increasing interest in its application. The benefits can best be characterized as follows: • ABM gives insights into causes of emergent phenomena— Emergent phenomena result from the interactions of individual entities, which are difficult to understand and predict. Truck activity results from the behavior of, and interactions among, many different players including indi- vidual vehicle drivers. ABM may be the only way to model the resulting traffic behaviors. • ABM uses a natural description of a network—When one is attempting to describe truck movements, ABM makes the model seem closer to reality. ABM has the potential to describe what firms (suppliers and receivers) and individu- als (truck operators) actually do in terms of the movement of freight. 4.3.3 Potential 5-Year Progress Developing agent-based models for freight transporta- tion will be a long-term effort. It is highly unlikely that a fully operational prototype would be available in the next 5 years. Development is expected to be slow. At the time of the writing of this report (late 2013), to the best of the authors’ knowl- edge there was one transportation consultancy in the United States (Resource Systems Group) that was actively working on this topic. However, it is a very active area in academia with research teams from Germany, Japan, Canada, and the United States making significant contributions to the field. Another U.S. consultancy, Cambridge Systematics, worked on the CMAP ABM in 2010 to 2011. In the next 5 years, we will likely see a gradual development process, with elements of the ABM framework being introduced to existing practical freight models. In particular, the understanding of some decision pro- cesses can be considered more advanced, namely shipment size and frequency and tour-based modeling. 4.3.4 Rationale for Recommendation that ABM Implementation Would Be a Long-Term Activity 4.3.4.1 Technical Feasibility As noted in this chapter, the number of agent-based appli- cations in freight modeling has been growing in recent years. ABM appears well suited for obtaining a picture of truck activity, in which large numbers of firms and individuals interact in complex ways. The technical feasibility of this approach is increasing and there may be further applications of ABM as a companion to microsimulation. 4.3.4.2 Institutional Feasibility The institutional feasibility of this strategy is high, given that the FAF and augments to it have been administered and financed by FHWA. FHWA is sponsoring current develop- mental work on ABM. 4.3.4.3 Operational Feasibility The challenge with implementing ABM in this context is the acquisition or availability of data. As noted in this chap- ter, specific data are needed to support ABM model develop- ment. Data need to be documented, their sources identified, and their utility validated. 4.3.4.4 Geographic Scalability Assuming that the data were available and that an agent- based simulation model could be developed, the information obtained would be at all levels of geography—national, state, regional. 4.3.4.5 Financial Feasibility The financial resources for model development over the long-term would be minor. Building a necessary data infra- structure to support ABM at the national level would involve more extensive costs. If a GPS framework were to be imple- mented, it would reduce the cost requirements. However, the transportation community would still be left with finding a source of commodity-specific flows at the requisite level of geography. One future source might come with the advent of the Internet of Things. The Internet of Things is a scenario in which objects, animals, or people are provided with unique identifiers and the ability to automatically transfer data over a network without requiring human-to-human or human- to-computer interaction. Increasingly, objects are embedded with Radio Frequency Identification (RFID) and sensors to track how products move through supply chains, and that generates masses of (real-time or logged) location data (see, for example, McKinsey, 2013). If these data can be made available to researchers, they may provide a rich source of information to support modeling of the freight transporta- tion system.