The main goal of NCFRP 44 was “…to develop a handbook for public practitioners that describes the factors shippers and carriers consider when choosing freight modes and provides an analytical methodology for public practitioners to quantify the probability and outcomes of policy-induced modal shifts…”. To achieve this goal, the team undertook a major effort to secure access to the confidential Commodity Flow Survey (CFS) microdata—the most comprehensive freight dataset in the US—complement the CFS with confidential shipper data and modal data, and use state of the art econometric modeling techniques. This significant research effort overcame some of the most significant challenges to the study of freight mode choice in the US.
Working with the CFS microdata file allowed the team to go well beyond previous similar research. This unique dataset provided the cornerstone of a behavior modeling effort to gain insight into the roles played by important variables such as door-to-door travel times (referred to in this report as “transit times”), freight rates, and commodity types. These analyses were complemented with involved in-depth-analyses of the historical patterns of freight mode shares, and a technical identification of the influencing factors at the market (macro) level, and the shipper (micro) level that shape freight mode choices. The team conducted an in-depth technical review of the potential modeling methodologies, both econometric and supply-chain based that could be used to develop freight mode choice models. Based on this review, the team conducted a critical evaluation of the advantages and disadvantages of the various methodologies and selected the most appropriate ones to be pursued in NCFRP 44.
This project presented substantial challenges to the research team. The research effort gathered the 4.5 million records in the confidential CFS microdata file, merged them with the even larger Longitudinal Business Database (LBD), prepared custom made data sets with the modal attributes—transit times, freight rates, and three different versions of generalized costs—and merged these data to prepare the master dataset for estimation of freight mode choice models. Using this unique data set, the team estimated four sets of market-share models and twelve sets of shipment-level models; and analyzed the more than thirteen hundred models estimated to identify the ones that meet the conditions of being conceptually valid and statistically significant.
In addition to the statistical modeling effort, the team also collected information from numerous market participants to inform the final model structures and policy analysis. The team conducted in-depth-interviews with ten market participants comprising four shippers, four receivers and two carriers to gain insights about their mode choice decisions. In addition, the team conducted six case studies of freight mode policy efforts in the US providing a review of the policy objectives, structure and implementation. Based on data related to two of these case studies, the Crescent and Heartland Corridors, the team conducted numerical experiments using hypothetical examples. These experiments were conducted to gain insight on the ability of the estimated models to produce sound estimates of the impacts of hypothetical policies.
Taken together, this effort is likely to be the most comprehensive research effort on freight mode choice in the US.
The most unique aspect of the work was the use of several confidential datasets. The use of the CFS and other confidential data, pioneered by the team during the research on freight generation conducted as part of NCFRP Report 37 “Using Commodity Flow Survey Microdata to Estimate the Generation of Freight, Freight Trip Generation, and Service Trips”, provided the team with access to vast amounts of high quality data that enabled to team to gain insight into freight mode choice behavior otherwise unreachable. These data provided an extremely solid foundation for the estimation of freight mode choice models.
At the same time, the size of the resulting datasets and the protocols to ensure security and privacy of the confidential data had the unintended effect of creating working conditions far from typical in transportation research project. Briefly explaining these aspects is essential to understand the decisions made by the team. Because of the confidential nature of the data, they could only be accessed at one of the secured Census Bureau’s Research Data Centers (RDCs), following strict protocols of confidentiality and privacy, which required team members to secure Special Sworn Status as Census Bureau’s agents. More than one hundred person-trips were conducted by team members to the RDC in New York City to access the data and perform the econometric analyses. A complicating factor was that, due to the tremendous size of the joint dataset, the computation time required to estimate some of the models took days, and in some cases weeks. Frequently, to ensure that there was enough computing power for the rest of the users of the Census Bureau servers, the server administrators had to cancel the computer jobs submitted by the team. Thus, the team had to re-submit the jobs the next time they traveled to the RDC, which is when they were able to check the status of the submitted jobs. The net result was a very protracted modeling effort. However, these inconveniences were a small price to pay for the access to the quality data available at the Census Bureau.
In conducting the case studies, the team encountered a different set of challenges, most notably, the lack of publicly available data about public sector experiences in freight mode policy. To overcome this situation, the team critically examined public information, conducted interviews, and analyzed the sparse data available to gain insight into the effects of public sector efforts to foster changes in mode shares. The high-level perspective provided by the case studies was complemented with the bottom-up view provided by a distinguished group of private sector leaders that participated in the in-depth-interviews (IDIs) with the team. The participants represent a wide range of industry sectors, company sizes, with operations that cover the entire geography of the US. During the IDIs, they provided their views about how to enhance the performance of the major freight modes in the US.
The chief findings produced by these efforts are summarized next.
Development of Freight Choice Models
The team conducted a rigorous review of the modeling techniques that could fulfill the objective of providing “…an analytical methodology for public practitioners to quantify the probability and outcomes of policy-induced modal shifts” The review, which included econometric and supply chain based models, concluded that econometric techniques represent the best combination of conceptual validity, ability to capture the effects of the various key variables, and practicality. The review also concluded that supply-chain based approaches, such as the Intermodal Transportation and Inventory Cost (ITIC), are not viable alternatives for freight mode choice analyses on account of their onerous input data requirements, which are very difficult to meet in practical applications. These conclusions are in agreement with current international practices.
Two different types of econometric models were estimated. The first family of models, referred to as market-share models, attempt to directly estimate the market share between rail and truck for a given commodity type as a function of the average values of transit times, freight rates, and generalized costs. The second family, referred to as shipment-level models, estimate the probability that a given shipment be sent by either rail or truck. A uniquely important feature of shipment-level models is that they are able to consider in minute level of detail the impacts of policy efforts that affect transit times and rates. In contrast, market share modes can only consider aggregate average variables. The econometric estimation of these models required assembling a data set containing the shipment data from the CFS, the shipper data from the Longitudinal Business Database, and the modal data (i.e., transit times, freight rates, and generalized costs) which had to be prepared by the team.
Notwithstanding the team’s access to data not typically available to researchers, the data assembled imposed notable constraints to the number of modes that could be considered, and the number of variables that could be incorporated in the models as explanatory variables of freight mode choice. The number of modes considered was constrained by the CFS data because the vast majority of the data are for truck and the rail modes (rail and intermodal). The analyses conducted by the team concluded that the shipment data for the other modes was too small to try to estimate models that include these modes. Another consideration was the substantial effort required to prepare the modal data for the modes with smaller number of observations. As a result, the freight mode choice only consider the rail and truck alternatives.
Ideally, the modal data should include all the relevant variables—such as transit times, freight rates, reliability, quality of service—that influence freight mode choice, for the same time period the CFS was conducted. The obstacle is that data of this nature were not found and probably do not exist (with the exception the 2012 Confidential Waybill Sample, which was used by the team). As a result, the team had to impute transit times and freight rates using sensible assumptions about travel speeds, transfer times, unit costs, among others. Unfortunately, variables such as reliability cannot be imputed a posteriori on the basis of the data available. As a result, such variables could not be included in the models.
Ultimately, a total of 1,346 models were estimated. The shipment size models amounted to 318 models. The market-share mode choice models totaled 266; while, 762 shipment-level mode choice models were produced. To ensure the models could support a wide range of potential applications, the team applied three different weighting schemes to ensure the models replicated the market shares in the CFS microdata, the domestic market shares embedded in the Freight Analysis Framework (FAF) version 4; and the total cargo handled in the country, both domestic plus imports and exports by commodity at the level of 2-digits North America Industrial Classification System (NAICS) (Freight Analysis Framework 2018).
The number of unweighted shipment-level model estimated are 254 (6 pooled models; 249 commodity-wise models). Weighted shipment-level models totaled 508; 250 of which are weighted using domestic cargo only (3 pooled models and 247 commodity-wise models) and 258 weighted using total cargo which include in addition to the domestic cargo, imports and exports (7 pooled; 251 commodity-wise models).
The modeling results indicate that the freight mode choice can be explained better using the generalized cost (specifically with 5% opportunity costs) as the independent variable. The team recommends the use of the models based on generalized costs because they consider the role of transit times, albeit indirectly. Only three models with transit times and freight rates as separate independent were found. The type of commodity at 2-digit Standard Classification of Transported Goods (SCTG) was also found to influence the mode choice. All commodities display inclination towards selecting trucks when all other modal attributes for truck and rail are same, though the strength of the inclination varies across commodity types.
Opportunity for Mode Shifts / Case Studies
The team conducted six case studies of freight mode policy efforts in the US, providing a review of the policy objectives, structure and implementation. The case studies, selected with input and approval of the study panel, spanned a variety of policy types, modes, and geographical areas. The chief lessons from the case studies are summarized in this section.
The Palouse River & Coulee City (PCC) Short-Line Freight System
The state purchased a 300-mile short-line freight system located in eastern Washington State, providing grain shippers with an alternative to truck to barge. This public investment has spurred private-sector commitment, maintained transportation options, and provided shippers with alternatives that allowed them to minimize transportation costs, fostering regional economic development.
The Crescent Corridor
The Corridor is an over $2.5 billion rail infrastructure improvement project operated by the Norfolk Southern Railway (NS). The corridor, under development since 2008, consists of a 2,500-mile network of existing rail lines that extends from New Jersey to Memphis and on to New Orleans. Corridor projects include straightening curves, adding signals, and building new track and rail terminals. NS also partnered with five states to improve the system and develop regional intermodal freight distribution centers. While the project has appeared to have shifted a significant amount of freight from trucks to rail intermodal, it appears to be only a fraction of initial forecasts, illustrating the difficulty in both shifting freight among modes and in forecasting those shifts.
The Heartland Corridor
The result of a $397 million public-private partnership completed in 2010, the corridor connects the Port of Virginia to major destinations in the Midwest including Chicago, Detroit, Columbus and Cincinnati. The partnership included NS, the Federal Highway Administration (FHWA), and the States of Virginia, West Virginia and Ohio. It raised clearances in 28 tunnels and 24 other overhead obstructions to allow the transport of double-stacked intermodal trains. It included intermodal capacity improvements made at the Rickenbacker Airport in Ohio and new intermodal terminals in Roanoke, VA and Prichard, WV. Given at least a doubling of intermodal traffic on the corridor, there is a lesson learned in that policymakers should explore investments that greatly improve efficiency at a reasonable cost and implement them where they appear feasible and efficient. However, there is also a cautionary tale in the case of the $30 million Pritchard terminal, which in a recent quarter was operating at only five percent of breakeven.
Chicago Region Environmental and Transportation Efficiency Program (CREATE)
A collection of rail and roadway improvement projects, this $3.8 billion public-private partnership consists of over 70 different projects. Members of the partnership include USDOT, Illinois DOT, six major freight rail carriers, and two passenger train systems. CREATE program projects have achieved a significant reduction in delays experienced by freight and passenger trains, as well as truck freight and passenger automobiles. CREATE represents the first time state and local governments have collaborated with the railroad industry to solve the problem of auto and rail congestion on such a large scale and a lesson learned is that these partnerships can work. However, critics contend that the rail industry has not shared in the costs to the extent they have enjoyed the benefits, so there is a lesson learned there as well.
Albany Express Barge
The Port Authority of New York State and New Jersey (PANYNJ) continually searches for alternative routes to move containerized cargo to and from ports to avoid the increased amount of road, bridge and tunnel congestion in the region. The Albany Express Barge Service started in 2003, but was suspended in 2006, with reasons cited included a lack of the funding, a lack of interest from shippers, and higher than anticipated transportation costs. Recent plans to restart the service in 2014 have not succeeded. The major lesson learned is mode shift from truck to container-on-barge is difficult to establish, with a major challenge being to reach the minimum volume of cargo that ensures financially viable operations. However, factors that promote container-on-barge programs are not going away including New York City congestion, environmental issues, and truck driver shortages.
Truck Route Management and Community Impact Reduction Study
Less than 10 percent of freight tonnage in the New York region is carried by modes other than truck. The relatively low rail mode share can be attributed in part to limited freight rail connections, especially to geographic Long Island, and in part to historical reliance on rail-to-barge car floats that by the middle of the 20th century were no longer competitive. The City’s DOT and other related agencies have focused on freight and have undertaken several large-scale plans and initiatives including the 2015 Urban Freight Initiatives, the Smart Truck Management Plan, NYC DOT Strategic Plan 2016, and Freight NYC. New York City’s size, geography, and limited transportation infrastructure all contribute to a difficult environment for freight. Truck size and weight policy alternatives are restricted by the policies and politics of multiple jurisdictions, aging and inadequate bridges, safety concerns on local streets, and the lack of alternative freight rail and freight barge infrastructure. The rapid expected growth of freight movements presents a difficult challenge to overcome. However, given the lack of funding for large scale transformational projects, policymakers and planners must continue to use all of the tools at their command.
Overall, the study team, based on the results of the case studies, recommends that policymakers and planners, when considering mode shift policies and projects:
- Understand the economics of mode choice;
- Consider vehicle types within modes;
- Expect resistance to change;
- Recognize the need for time and longevity;
- Design policies or projects to change the economic decision;
- Develop partners and encourage private investment;
- Examine the potential for economic development; and,
- Analyze the benefits and costs to each stakeholder
One last important lesson learned is that attempting to start new intermodal services or terminals to foster changes in freight mode choices from truck to rail or water from scratch is difficult. Evaluate risk thoroughly, forecast carefully and conservatively, line up business in advance, seek partners, and minimize up-front investment.
The in-depth-interviews revealed that the top four factors influencing mode choice are:
- Freight Rates: The in-depth interviews (IDIs) identified two factors most frequently mentioned as influencing mode choice. One is the rate associated with the mode though it is never the only determinant of mode choice; the choice is usually associated with other factors, as the company needs to maintain a certain level of quality in its operations. So, if the cheapest mode option does not provide them the minimum level of service required, typically that mode will not be selected.
- Quality of service: The second of the two factors most frequently mentioned as influencing mode choice was quality of service, which encompasses “reliability” and “level of service”. The need for a high level of service leads to a preference for trucks, due to their service and flexibility. A reoccurring theme in the IDIs was the need to balance both cost and reliability in selecting the mode. However, on-time deliveries is a major factor that drives the decision because if time is a factor, most companies will opt for more expensive options to get the shipment to the destination on time.
- Product type: The type of product being shipped is another significant factor influencing mode choice. One of the receivers specified that high-valued products are sent by truck due to timeliness of the demand, while standard products that are low-valued are shipped by rail. Mode choice is also based on whether the product is perishable or non-perishable.
- Seasonal changes: Seasonal changes include the effects of season-related weather on modes, as well as varying sales periods. Inland waterways in some locations are closed during the winter due to freezing, and for rail, winter weather may also result in delays. This results in the decreased use of rail, intermodal, and barge, and an increased use of trucks when weather is a concern.
The top three suggestions for improving factors affecting mode choice from the IDIs are:
- More consistency in rail delivery times: Most of the receivers and shippers are willing to use rail if the service is up to their standards. A shipper pointed out that it is not the longer transit time that affects the use of rail, but rather the inconsistency of those transit times, pointing out that even an early arrival of goods can lead to storage problems and extra costs. One of the shippers expressed willingness to use rail if it could offer competitive pricing, and more consistent service.
- Dredging and preserving the land for waterways: An inland waterway carrier is convinced that it could increase its market share, with a few improvements. The first is the dredging of local canals that have not been dredged for years. The second is to preserve the land along rivers for loading/storage, which are now being converted into either residential or recreational center.
- Increase the allowable weight limits on trucks: A shipper and large receiver said they would encourage an increase in the allowable weight limits for trucks. The shipper found that trucks would often reach their weight limits before getting filled. So, increasing those limits would allow for more being loaded on a single truck, which reduces the transportation costs and lessens externalities by reducing the number of truck trips required for a given amount of cargo.
This team conducted numerical experiments to illustrate some of the possible applications of the mode choice models estimated as a part of this project. To give the numerical experiments a real-life flavor, the team decided to use the Heartland and Crescent Corridors as inspiration for the numerical experiments. It is worth noting that these applications do not purport to be an evaluation of the real-life impacts of the selected projects, as they are nothing more than “real-life inspired scenarios.”
To construct these scenarios, basic details of the projects, publicly available data, and reasonable assumptions are used. The analyses demonstrate the application of both types of mode choice models: a) market-share mode choice models, and b) shipment-level mode choice models. The application of the market-share models to the scenario inspired by the Heartland Corridor estimated the impact of a shorter route on mode split for selected commodity types. The analysis for Crescent Corridor focused on the impact of travel distance and transit times on mode split for selected commodity types. The application of shipment-level models for the Heartland and Crescent Corridors focused on estimating the mode split along each corridor, for selected commodity groups. In all cases, the numerical experiments yielded sensible results that conform the expectations.