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.

18 C H A P T E R 3 This research is about innovation that could be used to implement or guide how truck-activity data might be obtained. The research team relied on our experience and expertise as well as project panel input to identify possible strategies for doing so. This chapter introduces these strategies and presents initial feasibility reviews of them. In identifying possible new approaches, the researchers focused on improving existing approaches rather than creat- ing new ones. Given current institutional and political cli- mates, this seemed to be a reasonable and realistic approach. These are approaches from which the freight data community could see implementation and results within 5 to 7 years. The team addressed innovation as following a process of evolu- tion rather than a revolution in methods for obtaining truck- activity data. Innovation is risky, especially as it applies to the public sector. Every new undertaking involves some risk, but not all risk is created equal. Knight (1921) identified two types of risk: “risk proper” and uncertainty. Risk proper is real; it cannot be eliminated but, because its probability can be accounted for with reasonable accuracy, it can be managed and hedged against. Uncertainty occurs when the probability of risk proper is unknown; its costs and benefits also may be unknown or change over time depending on the actions of others. To inform the costs and benefits of hedging against the unknown, the research team conducted feasibility reviews of 10 proposed innovative strategies for obtaining truck-activity data. The team defined a “feasibility review” as an initial assess- ment of whether the data gathering strategy can be executed for the intended purpose. The team’s feasibility assessments of each strategy fall into the following five categories: 1. Technical: Capability to meet the technical requirements of implementing proposed strategy. This includes an assess- ment of industry and data coverage issues. 2. Institutional: Laws and formal provisions that define roles and responsibilities of all the organizations involved in implementing the strategy. 3. Operational: If the strategy is developed, will it be used? Operational issues include internal issues, such as labor objections, manager resistance, organizational conflicts and policies. External issues include social acceptability, legal aspects, and government regulations. 4. Geographic scalability: Level of geographies for which the strategy can be implemented. 5. Financial: Rough estimates of cost to see if they match general expectations or would have an acceptable return on investment. For quick reference, the researchers used a star rating system (1 star is low and 5 stars is very high). These reviews and rat- ings are based on our teams’ knowledge, expertise, and experi- ence, literature reviews, and expert interviews. The researchers gave star ratings to technical feasibility, institutional feasibil- ity, operational feasibility, geographic scalability, and financial Feasibility Review of Innovative Strategies CHAPTER 3 KEY TAKEAWAYS • The researchers assessed the feasibility of 10 proposed strategies for obtaining truck activity data with minimum uncertainty in their outcomes. • Rather than proposing disruptive strategies, the team evaluated the feasibility of evolutionary improvements in obtaining truck activity data. • The team identified three strategies for further consideration: a new VIUS-like survey, a national freight GPS framework, and an agent-based modeling approach.

19 feasibility of each alternative considered. The star ratings were summed to calculate a total feasibility score for each strategy. Hence, the total star scores could range from 5 to 25 stars. Total scores are provided after each strategy along with the research team’s general recommendations regard- ing feasibility and next steps. Each feasibility review begins with a general description of the strategy. Also identified are the reasons that the strategy would be useful for address- ing coverage of the trucking industry universe and in filling gaps in data coverage. The team classified strategies as being based on surveys (census-based and other surveys), operations data (tradi- tional and new sources), or administrative records or model- ing techniques. The strategies for which the team conducted feasibility reviews are as follows: • Census-Based Surveys 1. Expansion of the CFS 2. Expansion of the Trucking and Warehousing (T&W) Survey • Other Surveys 3. A New VIUS-Like Survey 4. A New Industry-Based Supply-Chain Survey • Traditional Operations-Type Data 5. Operations Data Analysis Platform • New Sources of Operations Data 6. National Freight GPS Framework 7. Investigate License Plate Tracking • Administrative Data 8. Analysis of Federal MCMIS Records 9. Analysis of R. L. Polk Data • Modeling Approaches 10. Agent-Based Modeling Identify Truck Movements on Network 3.1 Expanded Commodity Flow Survey The expansion of the Commodity Flow Survey entails multiple strategies. The CFS warrants these because it pro- vides the most comprehensive primary data on truck activ- ity. Most agree that it is the basis for understanding freight in America. The CFS provides detailed O/D data by detailed commodity for freight movements by all modes. The researchers explored the following five options for expanding the CFS: 1. Continuous survey approach to reduce costs and acceler- ate reporting, 2. Additional observations to deepen the geographic detail of the survey, 3. Expansions of industries sampled to broaden reporting representativeness, 4. Extending transportation questions to economic census, and 5. Reestablished import/export survey addendum. Of these, the team did not consider the following for feasibility review: • Continuous survey approach, because experience to date with the continuous American Community Survey (ACS) indicates this strategy would have very limited reward regarding costs or reporting. • Additional observations, because at current sample size levels the CFS is manageable and provides extensive coverage. Were greater observations to be considered, the issues of disclosure and cost would be critical. It may be worth considering in future research but holds limited potential in the short term. Of particular value could be a sampling plan that expanded sampling for establishments with very large sets of shipment destinations (a one-to-many versus a one-to-few production/ distribution structure), thus responding to potential dis- closure constraints. Three strategic paths may hold promise for expanding industry coverage. These are • A joint research effort at shared cost with the U.S. Depart- ment of Agriculture (USDA) on developing O/D data for farm to assembly points that would expand industries rep- resented in the CFS. • A research assessment of the industries covered by the economic census (and not covered in the CFS) in which introduction of limited questions (e.g., odometer readings, cost data, or expanded vehicle characteristics) can provide high-value transportation information. It is recognized that most of the industries not covered in the CFS gen- erate little in the way of major transportation shipments, either in volume, value, or distances shipped. The ultimate rationale would be completeness. • Assessment with Customs and Border Protection (CPB) of a prospective joint import/export survey, which, for a shared cost, would expand coverage to shipments originat- ing outside the United States. Table 3-1 summarizes what data an expanded CFS could provide by geography and trucking activity type. Similar tables later summarize potential for industry and data cover- age of the other innovations examined by the research team. 3.1.1 Technical Feasibility  Though the three strategies discussed below do not have any inherent design complexity, they would still all be complex undertakings because of the multiple stakeholders involved.

20 3.1.1.1 Developing O/D Data for Farm to Assembly Points Farm shipments to assembly points are not included in the CFS at present. Stockyards and grain storage facilities are included. An opportunity exists through the Census of Agri- culture to establish a survey procedure to fill this “farm to first assembler” gap. It would require design and resource consid- erations as well as a benefit/cost assessment. Of the industries that are not included in the current CFS, agriculture is the only one the researchers would call a shipping industry; the others are more readily defined as receiver industries. As noted, expanding CFS to cover O/D data for farm to assembly points would require data from the Census of Agriculture. The Census of Agriculture is conducted every 5 years by the USDA’s National Agricultural Statistics Ser- vice (NASS). It provides the only source of agricultural data for every U.S. county. Current questionnaires are tailored to seven geographic growing regions. As a general condition, the questionnaire establishes the number of cars and trucks held on the farm as of December 31 of the survey year. A current question in some questionnaires asks percentage of sales to local, regional, national, and international markets. The USDA just completed its fieldwork for its most recent iteration, which, like the economic census, is conducted in years ending in “2” and “7.” Transportation-related ques- tions on this census could establish modes of transportation of farm products and destination types and ranges for farm products. 3.1.1.2 Expanded Transportation Questions for Economic Census The economic census provides a detailed portrait of busi- ness activity for a calendar years ending in “2” and “7” from the national to the local level. For 2007, census forms were mailed to more than 4.7 million companies with one or more paid employees. All large- and medium-sized businesses receive a census form, but only a sample of small-employer businesses receives one. For 2007, there were more than 500 versions of the census form, each customized to partic- ular industries. By collecting separate information for each establishment, the economic census can include detailed data for each industry and area. Expanding transportation questions to the economic cen- sus would be a technically feasible adjunct to the CFS. This approach would require a comprehensive assessment of the elements of the economic census to establish which industries would be most important to access and what questions would yield the highest payoff. Questions that may be considered include own-account transportation for certain industries or fleet size and expenditures to yield important linkages in the national accounts and input/output (I/O) structure. 3.1.1.3 Import/Export Survey A weakness of the CFS is its inability to capture imports and exports. The CFS does not report imports and has only a limited sample of exports. Nevertheless, the Department of Transportation surveyed inland movements of goods in for- eign trade several decades ago. It drew a sample of import and export declarations and used them to trace shipments to the shipper/receiver and to establish the sources or destinations of the flows and modes employed. It was a very straight- forward and relatively low-cost undertaking with many ben- efits, which would be even more valuable now. The technical challenge for a new survey would be sampling electronic, rather than paper, records. This undertaking may be poten- tially complex. There are multiple federal agencies involved in promoting and managing exports. Imports are under the control of the County Business Patterns (CBP). 3.1.2 Institutional Feasibility  All of the three prospective activities are elements of exist- ing federal operations that have long histories of continuing operations. All have legal, mandated reporting requirements. The Bureau of Transportation Statistics (BTS) may have the legal authority to conduct such operations. The inter- actions with other federal statistical agencies, the necessity for Trucking Activity Types Geographic Detail State and Multi-County Regions County and Sub-County For-hire carrier O/D flows, ton-miles, transportation costs O/D flows Owner-operator O/D flows, ton-miles, transportation costs --- Shipper-owned trucking O/D flows, ton-miles, transportation costs --- Table 3-1. Expanded CFS: potential for industry and data coverage.

21 competent statistical design and logistical handling of these multifold efforts, and the multimodal character of the under- takings would make BTS the only feasible entity for these pro- spective programs. The strategies would require negotiating institutional responsibilities and cost coverage, which could require considerable time and resources and lower the insti- tutional feasibility rating. 3.1.3 Operational Feasibility  Operationally, there are some challenges. First, in general, response rates are declining for surveys and the proposed strategies would involve increased response burden. Although respondent burden can be eased by electronic reporting, the lack of standard formats for shipping records across all indus- tries could create an enormous data processing burden on Census with financial costs. Second, given the quinquennial structure of two of the surveys (Census of Agriculture and Economic Census), timing for testing and preparation may be an issue. These two surveys are complex given their num- ber of specialized sub-questionnaires involved. Selecting the right starting place for each CFS expansion activity would be key to success. An inland origins and destinations survey would have to account for the long-term plans of the customs service to estab- lish a complete tracing capability, particularly for imports. An O/D would need to support, and not conflict with, a CBP reporting system. That said, duplication of some elements may be necessary, particularly those CBP is precluded from divulging or unwilling to divulge. 3.1.4 Geographic Scalability  As all of the strategies are essentially expanding the indus- try coverage and not the geographic specificity, there would be no perceived gains in geographic coverage such as county- to-county flows. 3.1.5 Financial Feasibility  These three efforts were specifically selected for their rela- tively low cost. There is design and institutional complexity in all of them so that internal staff resources might become an issue at some stage, though likely not in the preliminary stages of development in which scale and scope are defined. For the linkages to the Census of Agriculture and Economic Census, incremental costs of additional questions appended to existing surveys are typically not an expensive under- taking. Nevertheless, given survey expenses, competition for space on forms may be great, and the impact of additional items on overall response rates would be of concern, as might potential costs. The follow-on survey of imports and exports would be a new survey (although using existing administra- tive records as the sample frame) and would entail greater financial risks than the other two activities. The greatest finan- cial burden would be in preparing shipment information to generate estimates. To retain a similar level of data quality with an expanded sample of shipments would be quite an undertaking. Tripling the number of records might require a substantial increase in the budget. With an expanded sample, respondent attrition becomes a more serious issue as well. Total Feasibility Score: 9. These three options were considered because they are relatively low-cost expansions to the CFS. Addressing the quality issues associated with expanded CFS sample might dilute the potential low-cost expansion. The potential for improved industry and data coverage is not great. Next steps might include discussions with USDA, Economic Census, and CBP to gauge their receptivity to an add-on activity. Such discussions would need to be prefaced by a review of the industries covered and current questions asked for two of the surveys; the third dis- cussion with CBP would be more general. Given the uncer- tainty of the receptivity of these federal agencies, the team did not proceed with a detailed implementation scenario. 3.2 Expansion of the Trucking and Warehousing Survey The Trucking and Warehousing Survey, for industries in Truck Transportation (NAICS 484), is a part of the Services Annual Survey (SAS). The existing dataset goes back at least to 1998. The NAICS 484 coverage includes all carriers, both employer establishments and non-employers (owner- operators). The survey covers all for-hire (TL, LTL), heavy and tractor-trailer, light or delivery services identified by NAICS code; private carriage is not included. Thus, not only does it cover a significant portion of the trucking industry universe, but also it has significant potential, as shown in Table 3-2, to relate to other modes of freight transport with the same collec- tion methods and definitions. The Trucking and Warehousing Survey represents the most comprehensive set of transporta- tion industries available, with reporting on an annual and quarterly basis. Its weakness is that it does not encompass shipper-owned vehicle fleets engaged in own-account haulage. The survey at present covers several data elements of interest. • Motor Carrier Revenue by commodity classes, • End-of-year fleet size by type, • Fuel expenditures, • Payroll, and • Purchased freight transportation.

22 The main focus of the survey is revenues; thus, it has the potential to gather important information on value- miles by commodity and vehicle type. It is also a potential source of derived information on VMT and ton-miles (see Table 3-3). Relating the revenue items collected to discrete transportation values such as VMT and tonnage would provide valuable current information, conceivably even quarterly, to the extent not already employed in FAF and other estimates. The research team proposes the following three potential approaches for obtaining data: • Derivation: The data can be tested and analyzed as a poten- tial source of modeled information on broad VMT, ton- nage, and value for carriers and owner-operators. Other opportunities include fuel costs linkages to VMT. To the extent feasible, where data were derived from revenue data, quarterly estimates would be possible. Present quarterly reporting provides only broad total revenue values for employer and non-employer firms. • Add-ons: There should be opportunities for question add- ons that do not impede response rates. There is precedent for such supplemental questions in the SAS series. Con- ceivable add-ons such as fleet VMT, tonnages moved, or broad commodity classification could be tested. • Survey sample expansion: A third area of opportunity would be expanding the survey samples to assure state-level coverage. There would be considerable cost implications of doing so. This strategy should not be considered in the short term. 3.2.1 Technical Feasibility  The surveys are ongoing activities of the Census Bureau. These data are now on a substantially sound footing for the future and represent a valuable potential resource about the trucking industry and its activities. Absent extreme fiscal conditions, they appear to be steady over time. The model- ing activities to derive the additional data elements (VMT, ton-miles, value-miles) are technically feasible if time and NAICS Code NAICS Category No. of Sub- Categories No. Freight- Related No. Truck- Related 481 Air Transportation 4 2 n/a 482 Not Used 0 0 n/a 483 Water Transportation 6 3 n/a 484 Truck Transportation 6 6 6 485 Transit 11 0 n/a 486 Pipeline Transportation 4 4 n/a 487 Scenic and Sightseeing 3 0 n/a 488 Towing and Other Support 12 4 4 49 Courier and Express Delivery 6 6 6 All 52 25 16 Note: Excludes NAICS 482 Rail and 491 Postal Service. Table 3-2. NAICS industries included in the Services Annual Survey. Trucking Activity Types Geographic Detail State and Multi- County Regions State and Multi- County Regions For-hire carrier VMT, Ton-Miles, Value-Miles -- Owner-operator VMT, Ton-Miles, Value-Miles --- Shipper-owned trucking --- --- Table 3-3. Expanded Trucking and Warehouse Survey: potential for industry and data coverage.

23 resource intensive. Technically, changes can be made in the survey design. For example, in 2009 the SAS was expanded to cover a greater share of gross domestic product (GDP). Add- ing questions to gather additional data items would not incur substantial added expense, and would appear to be feasible, to the extent that such items do not reduce response rates. 3.2.2 Institutional Feasibility  There would be no legal constraints on deriving modeled information. Some constraints may arise should the need to work with individual records surface. In this case, sworn sta- tus for researchers or other procedures, such as anonymizing the data, could be necessary. Another possible legal issue would revolve around asking supplemental questions (which likely would not have the force of required response) in a survey that is required by law. None of these present insurmountable legal questions. The natural entity for such an undertaking would be BTS. It would have the skills and perhaps the human resources for undertaking such an effort. 3.2.3 Operational Feasibility  Operationally, there are likely few challenges because the surveys are ongoing activities of the Census Bureau. Given the annual structure of the surveys, timing for testing and preparation would not be as onerous as in a quinquennial survey. Data collection might also take place in alternating years or over another period. 3.2.4 Geographic Scalability  This survey currently has a national scale, but cost would be the only limit in expanding it to the state level. Establish- ments in the list have addresses identified and therefore, conceivably, could be analyzed by corridor or region. 3.2.5 Financial Feasibility  The modeling or derivation activities would require a focus of human resources that may pose some opportunity costs for DOT. This activity might be better housed at a university transportation research center. Beyond that, supplemental questions should require few additional financial resources. An early strategy would be to start with a limited question set for a small number of industries. This might include, for example, one question for the long-distance truck freight component (NAICS 484121) or long-distance specialized freight (484230). Although adding questions might not pose a financial burden, expanding the sample to provide state or regional detail could entail substantial expense. 3.3 New VIUS-Like Survey The loss of VIUS was a problem for road-based freight sta- tistics. It provided crucial vehicle characteristics and differ- entiation of vehicle activity by range, products carried, miles traveled, industry, and type of operator at the state level. The survey was cancelled due to financial problems, hence, rein- stating or expanding it would likely raise financial issues. A new program modeled on VIUS could again become the baseline for truck freight activity, differentiating business from personal use of light trucks, providing coverage of all categories of trucking operations, and establishing the uni- verse characteristics of the truck fleet: for hire, public, private, owner-operator, off-road, rental, and others. It would likely be straightforward to incorporate all vehi- cles into the survey and obtain a national framework picture of the entire motor vehicle fleet’s characteristics and activ- ity. This was at one time a valued endeavor, but expansion of TIUS to VIUS only got as far as the name change in 1997. The strategy would entail possible expansion in three areas. • Appling the new Canadian Vehicle Use Survey (CVUS) and developing a parallel U.S. prototype. • Working with R. L. Polk to obtain commercial vehicle- registration or MCMIS records to obtain motor carrier information for sampling purposes. • Expanding the sample frame to include private vehicles, government, and motor carriers of passengers to obtain a comprehensive picture of over-the-road vehicle movements. All of these are feasible either as simultaneous or separate activities. VIUS represents a critical area of development of road transportation statistics. It recognizes and uses vehicle regis- trations unique to road transportation rather than establish- ments as its universe for trucking activity statistics. In addition, VIUS can obtain VMT for all industries for which vehicles are sampled and ton-miles for industries not covered by the CFS. VIUS obtained VMT for the vehicle types covered irrespective of industry, largely getting around CFS coverage issues. Ton-miles could be estimated from Total Feasibility Score: 13. None of the three activities were pursued in detailed implementation scenarios. The proposed strategies might have great use in deriving data but they could not provide timely sub-annual data. The team’s final report states that university-based or other researchers may consider the challenge of modeling or derivation work. The notion of expanding such work to state or regional detail may be considered a long-term research opportunity.

24 payloads and VMT, and truck VMT could be estimated from VIUS payloads times FAF tonnages (see Table 3-4). 3.3.1 Technical Feasibility  VIUS was conducted by the Census Bureau within the last 10 years and can be readily reinstated in its previous form without any delay caused by technical concerns. However, applying the CVUS design to the U.S. context could provide opportunities for potential improvements in survey effi- ciency and data quality. The CVUS now underway requires that a sample of vehicles monitor total daily trip making with onboard equipment requiring some driver input (e.g., trip purpose, occupancy). The technical design elements of this comprehensive vehicle survey (i.e., passenger and commercial vehicles) have been worked out for the Canadian application. This would make the technical feasibility higher for U.S. implementation. FHWA is considering moving forward with a pilot dem- onstration of the CVUS technology in the United States. A potential future activity might be a harmonized, cross- border survey in which the questions would be the same so the data could be made useful for either country. How- ever, each country would be responsible for in-country deployment. The technical feasibility of this strategy cannot ignore the challenges associated with data collection through an in- vehicle device rather than a traditional survey questionnaire. There is precedent for the use of such technology to collect travel information from the use of GPS devices in the con- duct of regional household travel surveys. Much research has investigated issues related to non-response and response bias. Privacy issues have not been given as much attention in the household travel survey literature. One of the core challenges is an agreed framework for data management. Data manage- ment, in this sense, entails policies and “best” practices that relate to data collection, storage, and analysis. It addresses data governance, transparency, value distribution, data own- ership, and data privacy. The previous VIUS used the vehicle registrations sup- plied by the Polk Company as the sampling frame. This same sampling frame could be used in a CVUS-like applica- tion. Polk has maintained those records for decades. As an alternative, the MCMIS could be employed as a sampling frame. If Polk were used, a sampling frame could be pur- chased in the format desired. If MCMIS were used, signifi- cant data processing would be needed to obtain a usable frame. Polk allows for multiple types of vehicle databases, covering (1) automobiles and light-duty trucks and (2) heavy- duty trucks and other vehicles, which would be useful for expanding the survey population. Regardless of the frame itself, the CVUS approach requires a sample be drawn throughout the year to reflect seasonal variation in driving, whereas the old VIUS asked for year-end totals in a single sample. The rolling sample would allow multiple uses of CVUS hardware (i.e., send it back, download data and clean, use in next sample wave). 3.3.2 Institutional Feasibility  BTS would likely lead any new VIUS-like efforts. There do not appear to be any legal barriers to BTS observing the CVUS or participating in the ongoing North American Transporta- tion Statistics interchange. It is not necessary for the Census Bureau to have a role in VIUS-like activity. If a direct Polk relationship is considered, then publication of some findings might be an issue regarding proprietary materials. More specialized and detailed access would require negotiation with Polk regarding its interests and protection of its arrangements with states and attendant costs. A relationship with FMCSA for use of MCMIS may help accomplish a new VIUS without proprietary or other cost concerns. A purely internal DOT effort built around the MCMIS also could be evaluated. 3.3.3 Operational Feasibility  Operational feasibility would be influenced by the same issues relating to survey participation, response bias, and data management (e.g., privacy, governance, etc.)—that were raised under technical feasibility. These issues would need to Trucking Activity Types Geographic Detail State and Multi- County Regions State and Multi- County Regions For-hire carrier VMT, Ton, Ton-Miles, VMT, Ton, Ton-Miles, Owner-operator VMT, Ton, Ton-Miles VMT, Ton, Ton-Miles Shipper-owned trucking VMT, Ton, Ton-Miles VMT, Ton, Ton-Miles Table 3-4. New VIUS-like survey: potential for industry and data coverage.

25 be addressed through future research or in conjunction with a pilot study. Other operational issues include the following: • There may need to be operational changes at the Census Bureau were the survey conducted under the their auspices. • Minor operational changes might be required regarding sampling and supporting any prospective data mining operation. A similar observation would apply to the Cana- dian relationship. 3.3.4 Geographic Scalability  Like the Trucking and Warehousing Survey, the fundamen- tal structure of VIUS is national, but the sample could expand to yield state estimates. The VIUS sample was drawn from the state vehicle-registration files managed by R. L. Polk. This made the resulting national survey similar to 51 state surveys done with the same questionnaire at the same time. 3.3.5 Financial Feasibility  Designing and implementing a VIUS-like full vehicle sur- vey would be a significant undertaking. An analysis conducted for FHWA’s Office of Freight Management and Operations in 2010 estimated program costs for the expanded survey at about $12 million per survey cycle (ORNL, 2010). This was a maximum; costs, depending on design, could be as low as $9 million. In addition, the application of a CVUS approach would entail further costs in the short run, but perhaps lower- ing costs in the long run. For example, current estimates of the CVUS hardware costs are approximately $800 per unit, although costs are expected to decrease with technological evolution and a bulk purchase (as has been the case with using GPS equipment in household travel surveys). Financial resources will be central to this effort, which led the research team to giving financial feasibility a 1-star rating. 3.4 New Industry-Based Supply-Chain Survey As TRB Circular E-C169 highlighted, supply-chain data are critical for transportation planning and policy making. TRB Special Report 304: How We Travel: A Sustainable National Program for Travel Data also identified supply-chain data as a major gap in freight transportation data. There is currently no data program responsible for capturing, analyzing, and delivering supply-chain information. A new survey would capture intercity data on freight shipments from origin to intermediate handling and warehousing locations, and to final destination. This new survey would provide informa- tion on intermodal freight (i.e., movements of freight from origin to destination using two more end-on modes) that is not currently captured by the CFS. It also could provide more detailed O/D information than the CFS. CFS survey respondents typically do not know how their products reach their final destinations; they can only report the mode by which products leave their establishment. This new survey would provide information necessary to understand where goods originate, where they are ultimately consumed, and the modes of travel between the two. Focusing on supply chains as the sampling unit would expand coverage of the trucking industry to activities not currently included in the CFS, such as agriculture, retail, construction, and even imports. Truck- ing industry coverage would be extended to all for-hire and private fleets as well as parcel deliveries. Such a survey could conceivably help fill data gaps on VMT, tons/ton-miles, value/ value-miles, O/D flow, and truck transportation costs for spe- cific supply chains (see Table 3-5). 3.4.1 Technical Feasibility  Supply-chain surveys require a design that collects infor- mation from shippers, receivers, and intermediaries. This is such a technically challenging endeavor that the practicality of such a survey is questionable. A 2004 French shipper sur- vey remains the only attempt to conduct such a survey on a national scale (Rizet, 2008). The French experience identified a number of problems with achieving good response rates, and with higher costs per successful response than with tra- ditional shipper surveys, such as the CFS. In particular, the probability of getting a complete description of the complex supply chain and its elements for sampling proved to be low. One reason for this is the diversity of supply chains (which we address in terms of the need for developing a typology of supply chains below). The supply chain for a local restaurant that buys ingredients from the local farmers market serviced by local farmers is less complicated than the supply chain for a chain of car dealers that obtain cars from several manufac- turers who supply cars from multiple factories that obtain Total Feasibility Score: 13. Based on the outcomes of the feasibility analysis, the research team produced a detailed implementation scenario (see Chapter 4) to evaluate: (1) establishing a relationship with Statistics Canada to apply CVUS methodology and implementation, and (2) holding preliminary discussions about a public- private approach permitting development and use/sale of data, in which BTS could develop the data and the con- tracted company would act as the marketing entity, or, if that is not feasible, researching other funding sources to establish a new VIUS-like survey.

26 components from many suppliers who use materials from many sources who, in turn, get raw materials from multiple origins. The remainder of this section discusses some of the contingent issues. Freight transport across industry sectors is commonly mea- sured and described (loosely) using two terms: commodity flows (CFs), and vehicle flows (VFs) (Blanquart et al., 2012). CFs are represented by an O/D matrix focused on the type and quantity of goods moved through VFs represented by traffic flows in different modes, where the focus is on the vehicle and its operation. VFs are the result of logistics decisions made by carriers. A supply-chain view connects these, focusing on the commodity, the value, the volume, the origin, and the ultimate destination, and on intermediate routing of prod- ucts from production to consumption to recycling. This view focuses on senders, receivers, and intermediaries as well as the operational (VF) activities of transporters. The total sup- ply chain and range of processes that need to be included are numerous and varied, making them difficult to sample and challenging to analyze. In addition, collecting the required data from companies can be a very involved and complicated task. Researchers would need to consider the willingness of companies to provide information, data confidentiality issues, and data terminology issues (which can vary between sectors and also between companies in the same sector). The complexity of a supply-chain survey design could be simplified by developing typologies of supply chains. How- ever, the researcher team’s literature search uncovered no use- ful typologies that could be used for survey sampling. The typologies in existence relate to supply-chain management, not supply-chain logistics. Examples of supply-chain man- agement typologies include (1) procurement, production, distribution, and sales or (2) number of stages and forms of integration. These typologies are not useful for a survey try- ing to design efficient and explanatory sampling approaches or generating a unit of analysis from which generalizations can be made to a larger population. The development of typology that would be useful for sampling supply chains would entail a large R&D activity. Researchers from North Dakota State University are research- ing the use of FAF-CFS databases to determine a typology or systematic classification of supply chains and their relation- ships to commodity flows. One could simplify the technical requirements of a supply- chain survey by focusing on supply chains for a few or single commodities. Many prior survey efforts have done just this. Studies have worked backward from a commodity (such as strawberry yogurt) to the origins of its components (e.g., strawberries, yogurt culture, and sugar beets) (Browne and Allen, 2004). Alternatively, one could take a category of prod- ucts like organic vegetables and trace their delivery; chains also could be prioritized by the volume in a corridor, region, or nation. The benefit-cost ratio of such an exercise, however, may be low, given its low inferential power. Supply chains also are dynamic. Decisions are made in real time and are typically short term in nature. Hence, policy- makers need to consider the long-term value of tracking one or a few commodities through their chain. Perhaps, a more fruitful approach would be to conduct more qualita- tive research on how shippers decide about modes and routes. This would be very useful basic research but much less imme- diately applicable to the objectives of NCFRP Project 39. Finally, although NCFRP Report 29 focuses on trucking activity, a supply chain is multimodal and intermodal. Hence, while a supply-chain survey would yield much rich data, it would not focus wholly on truck movements. 3.4.2 Institutional Feasibility  Much of the supply-chain data is privately held and ship- pers and carriers consider their supply chains to be a strategic advantage. While supply-chain managers do share information on challenges and risks in annual surveys by organizations such as PriceWaterhouseCoopers (PwC), McKinsey, CSC, UPS, and Trucking Activity Types Geographic Detail State and Multi- County Regions State and Multi- County Regions For-hire carrier VMT, Ton/Ton-Miles, Value/Value-Miles, O/D, Transportation Costs VMT, Ton/Ton-Miles, Value/Value-Miles, O/D, Transportation Costs Owner-operator VMT, Ton/Ton-Miles, Value/Value-Miles, O/D, Transportation Costs VMT, Ton/Ton-Miles, Value/Value-Miles, O/D, Transportation Costs Shipper-owned trucking VMT, Ton/Ton-Miles, Value/Value-Miles, O/D, Transportation Costs VMT, Ton/Ton-Miles, Value/Value-Miles, O/D, Transportation Costs Table 3-5. New industry-based supply-chain survey: potential for industry and data coverage.

27 Deloitte, such data are highly aggregate and are more attitudinal than behavioral. The sponsor of a supply chain survey obtain- ing relevant information for transportation policy and plan- ning would need to be a government agency (e.g., sub-unit of the Department of Transportation, Department of Agriculture, Department of Commerce, or Census Bureau). Collecting pri- vate sector supply-chain data for public purposes while protect- ing private interests could be challenging. This would be even more difficult if focusing on one supply chain or one corridor. 3.4.3 Operational Feasibility  The complexity and variability of supply chains require more sophisticated approaches to data collection, integration, and fusion that would capture end-to-end flows, modes, interchange points, and routes. Even if one were to focus on administrative or sensor data (i.e., RFID), the complexity of multi-level data collection points is substantial. Establishing these approaches would require a substantial research and development effort. 3.4.4 Geographic Scalability  Boundary issues are a challenge. Supply chains cross juris- dictions, regions, states, and even countries. For example, most processed food supply chains involve numerous trans- formations of an input being transformed into something else in order to supply consumers with a final product. Exam- ples include making flour out of grain or transforming olives into oil. Depending on the complexity of the product being studied, this can result in the study of the entire system, and involve enormous numbers of other products and the pro- cesses involved in their production and supply. 3.4.5 Financial Feasibility  The lack of precedence for comprehensive supply-chain surveys indicates that the development of a new supply-chain survey would require much research and development. The complexity of such a survey in sampling, levels of analysis, and post-processing of data indicates that this survey is likely to be double or triple the current cost of CFS. 3.5 Operations Data Analysis Platform Operations data is one of the richest sources of timely and detailed data about trucks traveling on public infra- structure. These data represent private, for-hire (TL, LTL), heavy and tractor-trailer, and light or delivery services trucks. They represent many industries not currently covered by the CFS. Although extensively collected by states and reported to FHWA, the currently generated reports do not include VMT and do not use weigh-in-motion (WIM) data to look at freight flows. This program would create tools to estimate truck VMT based on operations data as well as freight flows based on Classification Counts and WIM data. This strategy acknowledges that the current sample size for WIM equip- ment is too small to measure truck weights on individual routes or on functional classes of routes in each state. That is why the strategy focuses on developing a platform for access- ing and visualizing the data in conjunction with data from other sources. 3.5.1 Technical Feasibility  Roadway operations data are collected by sensors installed by state departments of transportation in America’s roadway network. There are more than 6,000 installed in America’s roads, continually collecting data. Although a vast majority of these sensors only collect volume data, some are able to dif- ferentiate the types of vehicles that are on the roads, allowing for analysis of fleet mix and a comprehensive understanding of truck volumes. About 800 devices collect WIM data, which captures the total weight, axle weights, and even axle spacing of trucks passing through these stations. Because operations data are collected continually, they can be used to analyze the underlying traffic patterns. This same data is used in conjunction with HPMS data to project gen- eral traffic VMT and could be used with additional analysis to calculate truck VMT and to understand regional truck flows and ton-miles (see Table 3-6). 3.5.2 Institutional Feasibility  Currently, all states must report these data to FHWA, which collects the data in the Traffic Monitoring Analysis System (TMAS) (Tang, 2010). This system could provide an excel- lent institutional platform for continued analysis of opera- tions data to fill gaps in available truck data. FHWA, however, focuses on national-level data collection and may not have incentives to calculate local truck data. At the same time, state- level traffic monitoring and count programs are constantly striving to derive regional and local values from the count programs. Both the national and state-level institutional approaches are valid, and would provide valuable inputs, with different but overlapping interests. Total Feasibility Score: 7. The research team did not examine a detailed implementation scenario. The cost- benefit ratio makes it a risky investment of public funds in this resource-constrained environment. The team suggests two research activities that could lead to needed insights for supply-chain research: (1) exploratory research on develop- ing a typology of supply chains for the purpose of designing an efficient and effective sampling approach, and (2) quali- tative research on how shippers choose among modes and routes.

28 3.5.3 Operational Feasibility  Analysis tools could be constructed easily to produce accurate truck VMT counts using operations data available from publically available HPMS reports. Such data, used in conjunction with operations data from a states-count pro- gram or the TMAS database, could help in understanding the amount of freight that flows in monitored corridors by truck. Since TMAS is already a funded program and every state has a mandated count program managed by the state DOT, there are few extra responsibilities or expenses incurred in cre- ating these new datasets. This project also could be designed as a TMAS module with fixed inputs and outputs that would not require additional funding beyond the cost of creation. 3.5.4 Geographic Scalability  Operations data by its very nature creates excellent national- and state-level aggregate data but becomes less reliable below the state level. As more WIM and count sites come online, however, more reliable detail is available in local areas. This program allows states to decide how to use funding for their count programs. 3.5.5 Financial Feasibility  This strategy focuses on development of a module to extract truck VMT, ton/ton-miles, and vehicle speed from existing data. It does not consider expansion of the current WIM or sensor systems. Costs would cover creating the data processing and analysis tools based on currently available operations data. The project could be accomplished and made available to FHWA with a budget around $500,000. Trucking Activity Types Geographic Detail State and Multi- County Regions State and Multi- County Regions For-hire carrier VMT, Ton-Miles --- Owner-operator VMT, Ton-Miles --- Shipper-owned trucking VMT, Ton-Miles --- Table 3-6. Operations data analysis platform: potential for industry and data coverage. Trucking Activity Types Geographic Detail State and Multi- County Regions State and Multi- County Regions For-hire carrier VMT, Ton-Miles, Value-Miles VMT, Ton-Miles, Value-Miles Owner-operator VMT, Ton-Miles, Value-Miles VMT, Ton-Miles, Value-Miles Shipper-owned trucking VMT, Ton-Miles, Value-Miles VMT, Ton-Miles, Value-Miles Table 3-7. National freight GPS framework: potential for industry and data coverage. Total Feasibility Score: 20. The development of a mod- ule to extract truck VMT, ton/ton-miles and vehicle speed from existing data would be helpful. 3.6 National Freight GPS Framework The rapid uptake of GPS technology in the last 10 years is the most notable example of recent technological innovation in transportation. GPS data has the potential to fill important gaps in truck-activity data. Requiring all trucks, or a statisti- cally valid sample of trucks, to report GPS data for all trips would create a near perfect dataset for understanding truck movements in the United States. Such a dataset could pro- duce VMT with unprecedented accuracy. Adding commod- ity and cargo weight records to the data requirements would create leading economic indicators with a high level of spatial and temporal resolution (see Table 3-7). 3.6.1 Technical Feasibility  This strategy relies heavily on the recent rapid prolifera- tion of inexpensive and reliable GPS devices. The trucking

29 industry has already largely adopted the technology to pro- vide data for fleet management, greatly reducing the barriers for adoption of GPS devices for this strategy. Nevertheless, relying on this strategy would require resolv- ing several other technological issues. One primary challenge is incentivizing truck operator willingness to share truck trace data from in-vehicle tracking devices. Chapter 4 presents a strategy (My Trip Matters, or MTM) that would rely on information and services exchange to overcome this challenge. Other challenges have to do with the sheer size of the data being collected. For example, the amount of data generated by this project would require significant amounts of infrastructure for storage. New tools would need to be developed to ana- lyze the data, both in real-time and off line. Recent research into the development and implementation of algorithms to compress and query GPS trace data, such as SQUISH, offer near-term solutions for both storage and analyses (Muckell et al., 2009; 2013). At the same time, FHWA has been “harvesting” GPS data from trucks and developing travel time information (see http://ops.fhwa.dot.gov/freight/time.htm). Thus, the data collection and analysis requirements of this project are well within the scope of currently available technology. Creating a valid sampling frame from which to general- ize observations to the broader population of trucks would be the greatest technical challenge for this strategy. CVUS provides an excellent model for how this might be imple- mented. Although there are several possible sources that could be used to develop a sampling frame, such as state vehicle registration from R. L. Polk and MCMIS, they may not suffice for a valid sampling frame and would require further investigation. 3.6.2 Institutional Feasibility  Collecting and analyzing GPS data from trucks has his- torically been difficult for transportation planners and researchers. Although several studies have collected truck GPS data from private sources, these have recognized road- blocks to broad-based GPS collection. Many fleets currently produce GPS data, but their prime use for these data is to facilitate “real-time” tracking and fleet management. There are limited examples of using these data for histories analy- ses, such as applications to facilitate backhauls (Muckell et al., 2009). As a result, there are only very limited systems in place for data extraction and manipulation. It has been easier in some cases for researchers to buy the data from the cellular providers who facilitate the collection of data for truck fleet management. As previously mentioned, FHWA sponsored a project with ATRI to assemble a large, broad-based collection of truck GPS data from a number of fleets across America. Privacy and proprietary information concerns, however, limited the number of variables collected. Although some form of mandated reporting from selected trucks would be ideal for this strategy, its implementation would be a tremendous institutional undertaking that would likely be met by strong resistance from the transportation industry. Mandated reporting is a worthy long-term goal, so other options may be considered to demonstrate the viability of collecting GPS data from trucks as a source of analytical data. For example, a pilot program could leverage existing GPS data from ATRI data, foster the collection of new GPS data on a voluntary basis, or use a data-sharing agreement for an existing GPS truck data program. 3.6.3 Operational Feasibility  A national freight GPS framework could give an unprec- edented level of access to data on freight moving by truck. It has the potential to generate high-quality data for truck VMT, truck speed, and freight performance measure data. In addi- tion, it would be feasible to append information to the GPS trace data. For example, adding waybill data to the GPS for- mat would allow for the collection of commodity flow data at better spatial and temporal resolution than is currently avail- able in CFS. The first approach to this project is to conduct a pilot study with a small number of trucks providing GPS trace data and to build the project infrastructure while looking for a long- term solution for valid national, regional, and local sampling frames. By restricting the initial study to volunteer organiza- tions, purchased data, or to a state geography where the state- level government is interested in collecting such detailed data, this project can immediately begin operations. 3.6.4 Geographic Scalability  The nature of this project allows it to scale to all geographic levels, or even to be deployed in single instances, given geo- graphic restrictions. The long-term goal of the project would be to cover the entire country with sufficient samples to sup- port analysis at the state, regional, and local levels. Examples of state-level existing programs include Oregon’s weight-mile data collection using smartphone GPS (Bell and Figliozzi, 2013) and Washington State’s GPS-based truck freight per- formance measure platform (McCormack et al., 2010). An example of a multi-geographic platform for truck analysis is being deployed through the Ministry of Transpor- tation Ontario. Their iCorridor project is supported through the “harvest” and visualization of GPS data from trucks in Canada and the United States and provides information to many different geographic users (see http://www.gbnrtc.org/ files/9813/2769/5483/Tardif.pdf).

30 3.6.5 Financial Feasibility  The largest financial barrier to implementing a national freight GPS framework would be the collection of data. There are a number of options for collecting GPS data from the nation’s private trucking fleets that would reduce cost barriers. The first would provide devices for each vehicle to be tracked. According to truckinfo.net, there are 15.5 million trucks in the United States, of which 2 million are tractor-trailers. Mar- ket research shows a number of GPS tracking devices priced around $50 a unit. To cover all trucks would cost $775 mil- lion or $100 million to cover only tractor-trailers, while a 1 percent sample would cost $7.5 million for all trucks and $1 million for tractor-trailers. Because a large number of fleets already use GPS data in their operations, it may be possible to offer incentives for these companies to provide their existing data. This would be a much more cost-effective option than providing specific physical devices, allowing the market and fleets to decide which devices serve their needs best, while producing the valuable GPS data for analysis. Also, the FHWA system dis- cussed above could be leveraged for financial benefits. The last option to consider would be to purchase the data from third parties that are currently supplying data services to trucking fleets. In addition to the data collection costs, this project would need to account for data storage and processing costs as well as the development of the software to analyze the data and create reports or visualizations. The cost of hosting and band- width may be as high as $50,000 per year. The project would need to be maintained by a team of three to five workers, whose salary and overhead costs may be between $300,000 and $500,000 per year. Another strategy for storage would be to use a secure data storage center, such as the Secure Transportation Data Center (STDC) (see http://www.nrel.gov/vehiclesandfuels/ secure_transportation_data.html). STDC provides secure storage for household travel survey GPS data and allows access to these data for research and analysis through pro- cedures similar to those used with the Census Bureau for access to microdata. Data users are not required to be on- 3.7 Data from License Plate Recognition (LPR) Systems LPR is an image-processing technology used to identify vehicles by their license plates. LPR systems use cameras, computer hardware, and software to capture an image of a license plate, recognize its characters by converting them into readable text, and check the license plate against designated databases for identification. Use of LPR among law enforce- ment agencies across the country has been increasing rapidly in recent years. In addition, LPR is used for toll collection and traffic management on major arteries. A subset of the data from such systems—specifically, large volumes of observa- tions of different commercial vehicles at different locations and points in time—could be analyzed to develop informa- tion about truck travel patterns. This capability could lead to the development of more detailed O/D flow data particularly in urban areas (see Table 3-8). 3.7.1 Technical Feasibility  The use of LPR to capture data for characterizing goods movement activity is conceptually attractive; it could be largely automated following initial integration efforts, and the deployment of LPR cameras—both fixed and mobile—for law enforcement or other purposes (such as tolling) is rapidly increasing across the country (Lum et al., 2010). There are, however, a number of difficult technical challenges that would need to be overcome in order for this strategy to succeed. Total Feasibility Score: 20. Based on the outcomes of the feasibility analysis, the research team produced a detailed implementation scenario (see Chapter 4) that investigated the requirements for starting a limited pilot program allowing for immediate returns in freight data availability and capable of growing into an essential data program. site at the center to use the GPS data but can access the trace data with secure logins. Trucking Activity Types Geographic Detail State and Multi- County Regions State and Multi- County Regions For-hire carrier --- O/D Owner-operator --- O/D Shipper-owned trucking --- O/D Table 3-8. LPR systems: potential for industry and data coverage.

31 The first challenge relates to accessing data on a national basis. LPR systems are typically acquired and operated by individual jurisdictions, or perhaps by several neighboring jurisdictions. Some of these systems maintain databases of historical LPR observations for a period of time that might range from a few days to a few years. Such databases could, in theory, provide access to the large volumes of historical observations that would be needed to analyze travel patterns of commercial vehicles. Currently, mainly because of privacy considerations, LPR data are used for real-time purposes and are not typically archived—regardless of application. Accessing archived data, however, would require a signifi- cant amount of systems integration work. Because different systems are used by different jurisdictions, the systems inte- gration work would need to be repeated for many different jurisdictions. More work has been done in terms of data sharing across systems in the law enforcement sector than in the transporta- tion sector. For example, Nlets, the International Justice and Public Safety Network (www.nlets.org), is currently develop- ing a national portal for sharing LPR data across jurisdic- tions. Such a portal could collect LPR data for commercial vehicles at a national scale. The design of the Nlets portal, however, is intended to facilitate checks on specific vehicles. For example, the portal might be used to determine whether a vehicle owned by a fugitive suspect in a robbery in one city has been observed by LPR in any other cities across the country. The system is not designed for automated collection of large volumes of LPR records from multiple jurisdictions across the country. That is, it is not set up to allow for the type of data mining that would be necessary to evaluate commercial vehicle flows and trips. For the foreseeable future, collection of data at a national scale would require system integration on a jurisdiction-by-jurisdiction basis. The second issue is that the location of cameras—either at fixed locations or mounted on police cruisers—is deter- mined by agency needs, which may not overlap particularly well with the most helpful points for observing goods move- ment flows. For example, police cruisers are more likely to circulate through high-crime areas than around trucking hubs (assuming the two do not overlap). Open-road tolling operations are only on select facilities in a region. Also, for the foreseeable future, much of the forecast market for addi- tional LPR technology acquisition will be in law enforcement, even though other end uses include access control, travel time measurement, parking time management, and tollways (IHS IMS Research, 2012). A third challenge relates to the difficulty of inferring ori- gins and destinations assuming that the network of LPR readers is relatively sparse. Imagine that a commercial vehicle is observed at Point A and then again, after some time has passed, at Point B. It would be impossible to discern whether, during the period between the two observations, the vehicle stopped to make a pickup or a drop-off, or perhaps simply stopped for a lunch break. Unless cameras were spread ubiq- uitously throughout a region—unlikely in the near term at least—this challenge would be difficult to overcome. Addi- tionally, it would be impossible to determine whether a vehicle observed by LPR is carrying a load or deadheading, further complicating the difficulty of inferring O-D informa- tion from a series of LPR records. A final challenge relates to license plates for commer- cial vehicles. First, depending on ownership and logistics arrangements, the license plate on the cab of a tractor-trailer configuration may differ from the plate on the trailer. Second, some vehicles may have multiple license plates for the differ- ent states in which they travel, and many LPR systems are not currently capable of reading multiple license plates for a given vehicle. Both of these issues may compound the challenges of accurately identifying certain commercial vehicles. 3.7.2 Institutional Feasibility  Beyond the technical obstacles discussed, there are three institutional barriers to using LPR systems for goods move- ment data. First, many law enforcement and other agencies are contending with privacy-related concerns associated with their use of LPR technology (Lum et al., 2010). To avoid com- pounding those concerns, many agencies and jurisdictions may be unwilling to share their historical LPR data for any purposes other than legitimate law enforcement activities. Second, there are legal questions yet to be resolved regard- ing appropriate constraints on use and storage of LPR data. Lum et al. (2010) consider a spectrum of possible uses that ranges from quickly identifying vehicles of interest (e.g., a stolen vehicle, vehicle that should pay a toll) in real time to archiving the data over much longer periods to facilitate more complex historical analyses (e.g., to perhaps construct a suspect’s travel patterns in the days leading up to a crime in some future investigation or to analyze the travel patterns of toll road users). Although past legal precedent suggests that use of LPR at the simpler end of the spectrum appears to be acceptable, there is much greater uncertainty regard- ing the collection and long-term storage and analytic uses of LPR data (Lum et al., 2010). Until this issue is resolved in the courts, it remains uncertain whether long-term storage of historical LPR data will be viewed as constitutional. This could have a significant effect on the viability of collecting such data for goods movement analysis. Third, vehicles travel freely among different jurisdictions, so it is important that LPR systems can share data among agencies in order to assure cov- erage. Because LPR systems are, for the most part, owned and operated by individual agencies, sharing data requires added effort beyond normal operation. As with any technology,

32 different vendors use varying data formats and transfer pro- tocols that make sharing difficult. As technologies mature, markets typically converge toward standardized formats, easing interoperability. For now, it is possible to share data only with agencies using the same brand of LPR system. 3.7.3 Operational Feasibility  Assuming the necessary systems integration work and the development of suitable algorithms for evaluating LPR were completed, this approach to collecting goods movement data would be largely automated. Ongoing operational require- ments are likely to be modest, and there should be no burden on users or respondents. 3.7.4 Geographic Scalability  Much of the geographic coverage for LPR systems is designed to serve specific agency needs and is likely to be concentrated at certain spots on the network. LPR data may therefore be most useful for examining origins and destinations for goods movement at the metropolitan scale. If augmented with additional observations—for example, from weigh- in-motion stations—along major highways, the data might support analysis of origins and designations for regional or interstate travel as well. 3.7.5 Financial Feasibility  Assuming the need to integrate with many disparate LPR systems across the country as discussed earlier, the initial expense for this strategy is likely to be cost prohibitive. If such an investment were made, however, ongoing operating costs could be quite modest. A detailed analysis of these data with the objective of assessing their suitability as a sampling frame could enable an alter- native establishment-based sampling frame outside of the Census Bureau. The frame could be used for a new VIUS-like survey or to build the truck GPS framework. 3.8.1 Technical Feasibility  It would be technically feasible to obtain and use the MCMIS file as a sampling frame. There are many different reports available from MCMIS, but the Census File Extract is the file of interest here. This file contains a record for every company or entity that operates a vehicle subject to the Federal Motor Carrier Safety Regulations (FMCSR) or Hazardous Materials Regulations (HMR). That includes any company, from owner-operators to large companies that operate commercial trucks in the United States. The cen- sus file for this research was from July 7, 2012, and contains records for 1,627,638 entities registered with FMCSA. There are more than 130 data elements for each entry including physical address information, the number and type of trucks owned and leased, the types of commodities hauled by the entity, whether the commodity was hauled less or more than 100 miles, and whether it was hauled across or within states. The data also includes total VMT for the year for the entire company. Through the analysis of these data, particularly VMT and commodity information, an efficient, stratified sampling approach for VIUS (or other surveys) could be designed. The team’s review indicated that most of the data is complete and current, although stringent quality controls would be needed to guard against misreporting of informa- tion by respondents. Because MCMIS is being reviewed for its appropriateness as a sampling frame, a table indicating coverage in terms of data elements and trucking activity types is not presented. Such a table is not relevant. 3.8.2 Institutional Feasibility  These data belong to FMCSA, a DOT administration. Thus, it is unlikely that institutional issues would arise if an internal DOT effort is built upon MCMIS. 3.8.3 Operational Feasibility  There are no significant operational conflicts to evaluating MCMIS data for use as a sampling frame. The largest data source for the census file, Form MCS-150, must be updated biennially. The long-term availability of the data is secure. Metadata and documentation are free and publically avail- able on the MCMIS website, making identification of the required data items relatively easy. Total Feasibility Score: 12. This strategy, while intrigu- ing, faces a daunting array of technical, institutional, and legal challenges or concerns. Further pursuit of this strat- egy in the near term does not appear promising, although it would be worthwhile to monitor the deployment and integration of LPR systems for ongoing developments that could make it more feasible in future years. 3.8 Analysis of the MCMIS for Use as Sampling Frame MCMIS is an establishment-based source of truck-activity data. It contains more than 1 million trucking company records with information about size, how they operate, what com- modities they carry, and how many miles their trucks travel.

33 3.8.4 Geographic Scalability  This could be applied at any geographic level. One has the exact address of the firms and would be able to geocode the stratified sample desired and determine its geographic reach. 3.8.5 Financial Feasibility  The cost to obtain the data is inexpensive. Any financial outlay would be to cover the work required to process and output data upon request. files of truck registrations identified as being active as of July 1, 2002. Information in the files can then be used to stratify the frame by geography and truck characteristics. The 50 states and the District of Columbia comprise the 51 geographic strata. Body type and gross vehicle weight (GVW) deter- mined the following five truck strata: 1) Pickups; 2) Minivans, other light vans, and sport utilities; 3) Light single-unit trucks (GVW < 26,000 lb); 4) Heavy single-unit trucks (GVW ≥ 26,000 lb); and 5) Truck-tractors. Polk-based percentage data were then applied to state sup- plied total registered vehicle data to obtain final counts of each of the five vehicle segments. For trend purposes, a simi- lar sampling approach could be envisioned for future, rein- stated VIUS-like surveys. Polk handles all processing of registration data and little information is available about the accuracy of the data. One purchases the processed files directly from Polk. As with the MCMIS data, the research team presents a table indicating data element and trucking activity type coverage. 3.9.2 Institutional Feasibility  If Polk data were used it would have to be purchased by the agency that owns and operates a new VIUS-like survey. In previous VIUS iterations, this purchaser would have been the Bureau of the Census. This does not preclude DOT from holding the institutional responsibility. There is precedent for FHWA purchasing Polk data, which is currently used to calibrate truck information reported in the HPMS. This has been done through a sole-source contract with Polk. Also the National Highway Traffic Safety Administration (NHTSA) already makes extensive use of the files. 3.9.3 Operational Feasibility  Polk is a privately held consumer marketing and infor- mation company, which started motor vehicle statistics operations in 1922. Since then, Polk has maintained com- prehensive vehicle databases for sale. Given its long-standing legacy as a data and information supplier, there could be future operational issues that would prevent obtaining the required commercial database, and this may be an issue. Polk has been acquired by IHS, which may affect future business operations. 3.9.4 Geographic Scalability Geographic scalability does not apply. Total Feasibility Score: 16. In Chapter 4, the research team evaluates MCMIS (relative to R. L. Polk) as a poten- tial sampling frame for a VIUS-like survey or truck GPS framework. 3.9 Analysis of the R. L. Polk Data R. L. Polk and Company provides the opportunity for a vehicle-based sampling frame for a new VIUS-like survey, truck GPS framework, or for data mining of truck-related characteristics. Polk compiles Department of Motor Vehicle registration information from the 50 states and the District of Columbia. In 2010, Polk launched the National Vehicle Population Profile (NVPP) that provides a near-census of Class 1-8 vehicles in the United States and Canada. In 2013, Polk released an enhanced version of its data that is VIN-based and updated in near real time. Without purchasing the data, it is not possible to access its file structures. 3.9.1 Technical Feasibility  VIUS (formerly Truck Inventory and Use Survey), as well as other surveys, have used a Polk database as the sampling frame. Using the vehicle-based frame allows for stratification that could achieve greater coverage of the trucking industry than an establishment-based survey. A shortcoming is that registration data contains no descriptive information beyond what Polk derives from the VIN. The data would include truck-activity data; registration information (i.e., the data that Polk could provide) would include registration date, vehicle identification number, make, model year, fuel type, body type, wheelbase, and gross vehicle weight. One can use these data to determine the vehicle-based truck universe is (i.e., what is on the roads), but not what these trucks are doing. Prior experience with VIUS indicates that it would be tech- nically feasible to use the Polk database as a sampling frame. As in the past, it would be necessary to select a single date for which to obtain the files of truck registrations. For instance, the sampling frame for the 2002 VIUS was constructed from

34 3.9.5 Financial Feasibility  The cost of Polk data varies by the amount and type of infor- mation purchased. There do not appear to be published price schedules, and cost proposals are produced on a customer-by- customer basis. The government, however, may be able to negotiate the price. sub-regional geography and off-the-shelf network assign- ment methods are used to fit the flows to the network using time and other impedance factors, adjusting the flows so that payloads per year divided by 365 do not exceed aver- age annual daily traffic (AADT) from HPMS. This method requires several assumptions, including that (1) truck O/D and truck tonnage O/D geography are the same, (2) errors in disaggregation of regional flows to the county level balance out, and (3) there is no way of assigning intra-zonal traffic. Since the FAF is largely based on CFS, it is subject to the same limitations of CFS; namely, limits in commodities covered, timeliness of data, and local coverage. This strategy evaluates whether ABM could produce improved network assignment models. 3.10.1 Technical Feasibility  The ABM approach is technically feasible. An agent-based model represents individual decisionmakers whose interactions form a system. Experimenting on this system by changing agent parameters or performing optimizations can provide insight into the study objectives. Agent-based simulations have been used to analyze patterns of pedestrian behavior, urban traffic, and emergency evacuations. Basic components of an agent-based model are agents and an environment in which these agents interact. The model has to define behavior rules and attributes for agents so that their interaction with each other and the environment can be simulated. Developing a model for truck movements and exploring model results by changing parameters such as the number of trucks, speeds, times of departure, and route selec- tion rules can yield insights into truck traffic patterns. The first step to develop an agent-based model is to define agent. An agent-based approach to truck traffic can take the perspective of individual drivers and their decisions on travel times, route selection, and speed to observe emerging patterns of traffic flow in the region of interest. Truck drivers follow instructions from their companies on destination and route selection, although they may have more initiative when they select local roads depending on the traffic situation. Another option is to define carrier companies as agents. Carriers Total Feasibility Score: 13. In Chapter 4, the research team further evaluates R. L. Polk data as a potential sam- pling frame for a VIUS-like survey and as a source of other data to support such a survey activity, but it received a lower feasibility score than did MCMIS. 3.10 Agent-Based Modeling to Identify Truck Movements on Network This strategy attempts to improve upon the current meth- ods through the use of more powerful simulation-based mod- eling techniques. Such techniques are the primary analysis tool for designing or understanding complex systems, such as truck movements on a network. When linked with optimi- zation techniques, agent-based modeling (ABM) techniques have been effectively used for representing such systems. With a sufficiently complex simulation model, the strategy has the potential to represent a large part of the trucking industry universe, both interregional as well as intraregional truck movements. There is also the possibility of filling gaps in VMT and O/D flow data. The researchers tested the feasibil- ity of using agent-based modeling (see Table 3-9). The current best model of applying powerful modeling techniques to produce freight information is the FAF. The FAF has two basic products: an O/D matrix of tons by mode and a highway network loaded with truck payloads as a sur- rogate for trucks. The approach used to identify actual truck movements is an approximation. Current methods rely on transforming tons carried by truck-only into truck payloads based on payload-per-vehicle estimates from the VIUS and weigh-in-motion data. O/D flows are then disaggregated to Trucking Activity Types Geographic Detail State and Multi- County Regions State and Multi- County Regions For-hire carrier VMT, O/D VMT, O/D Owner-operator VMT, O/D VMT, O/D Shipper-owned trucking VMT, O/D VMT, O/D Table 3-9. Agent-based modeling: potential for industry and data coverage.

35 schedule trips and routes for their trucks, based on an inter- nal decision rule and orders from shippers. Since the question is about the number of trucks and their route assignment, a higher resolution model that includes truck driver deci- sions might not bring additional benefits and unnecessar- ily increase the modeling burden. So, a carrier-based model would be preferred. The second step is to define a decision rule for the agents, which is the primary challenge in an ABM effort. The diffi- culty in modeling carrier-company behavior is to estimate how detailed decision rules are. Some carriers might use optimiza- tion algorithms and others might follow simple rules, with a primary and secondary choice for each of their destinations. The supply-chain academic, consultancy, and third-party logistics communities already have a good handle on the private-sector decision calculus and “micro” drivers of mode shifts that can be used to estimate the necessary decision rules. The third component of the model is the network on which carrier companies operate. The network in the model can represent the highway system with nodes and links between them. The capacity and the speed on different parts of these links must be included in the model, because it influences how carrier companies operate. Assuming an adequate representation of the carriers is com- plete, the agent-based model must go through a verification and validation process. Verification ensures that the model is working as intended and that its calculations yield proper results. Validation determines if the agent is representing the reality accurately enough for the purposes of the model. Note that an agent-based model should not try to emulate reality with 100 percent accuracy. After the simulation is set up, the number of carriers and their trucks that enter the road net- work from the origin locations can be adjusted to real-world levels (much like the FAF does now), and the ensuing traffic flows can be observed. The validation stage can include com- parison of simulated traffic flow data with truck counts from different locations to calibrate model parameters. Although not plentiful, there are examples of ongoing research using agent-based models. One is “A Conceptual Framework for Agent-Based Modeling of Logistics Services” by the University of Toronto (Roorda et al., 2010). The effort is at a conceptual stage, with data requirements currently under investigation. Another example is work being performed for the Chicago Metropolitan Agency for Planning (CMAP). More examples of current research are presented in Chapter 4. 3.10.2 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. 3.10.3 Operational Feasibility  The operational feasibility of this strategy is questionable due to data inadequacies. GPS data from trucks could be one component of developing a carrier-company decision- making model. GPS data on trucks potentially includes truck identification number, coordinates, speed, travel heading at a given time, and stops. Data on change in coordinates and time stamps can determine if a truck had stopped or just slowed down in traffic. Stops may be classified as intended (e.g., loading or unloading) or unintended (e.g., congestion, traffic lights, or intersections). These data do not give information on activity at a stop location; it is still unknown if the truck unloaded goods or received new cargo before continuing to a new zone or if it perhaps simply stopped for a lunch break. Combining land- use data with GPS data on stop locations can only be a limited improvement, as there are many types of establishments in the same type of land-use area. Moreover, GPS data do not necessarily include truck type, owner, cargo, or depots. Another difficulty with GPS data is that carriers’ decision rules might not be observable from truck locations. Carriers’ routing choices might depend on availability of their personnel, their own priorities for different clients, maximum amount of hours a driver is allowed to drive before resting, etc. GPS data must be combined with other sources of data to derive a carrier decision-making model. Even after a decision-making model for carriers is complete, simulating this behavior to obtain truck count data between origin and destination locations will rest on the assumption that the model is accurately cap- turing many different decision rules of various carriers. This assumption is not easy to justify. The main difficulty in using an agent-based model is that it is unclear what types of data sources are necessary to model carrier behavior. Identifying the relevant quantitative and qualitative data sources, obtaining them, and validating the model require a significant amount of effort. Truck GPS data alone is not enough to build such a model. 3.10.4 Geographic Scalability  Assuming that the necessary 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, and regional. 3.10.5 Financial Feasibility  Without an existing, necessary 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.

36 Total feasibility scores indicated the levels of risk (low score) and the chances of success (high scores) for the 10 data gathering strategies. The researchers developed detailed implementation plans for three strategies with high scores: a new VIUS-like survey, national freight GPS framework, and ABM approach. A fourth strategy, operations data analysis platform, scored high on feasibility. Total Feasibility Score: 15. In Chapter 4, the research team investigates the state of the practice in freight-related ABM activities and research further to determine whether this could be a potential source of missing information in the near future. In addition, NCFRP Report 26 presents the findings of a study on developing commodity flow data at the state and local levels. An ABM approach may be useful for pulling together the different types of data into the FAF for a richer dataset.