Impacts of Policy-Induced Freight Modal Shifts (2019)

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.

71 This chapter presents the methodological approach selected to estimate mode choice models in this research. Selecting the methodological approach required careful consideration of alter- native modeling methodologies, data requirements, and data availability, and the ability of these to provide sound estimates of the impacts of market changes and public policy on freight mode choice. The key features of the most important modeling approaches are summarized in Table 17. As shown, freight mode/vehicle choice models can be divided into two major families: econo- metric and supply-chain-based models. Econometric models use random utility theory and advanced econometric techniques to capture the decision dynamics at the core of the freight mode/vehicle choice process, while supply-chain-based models replicate a chain of logistical decisions made with the aim of minimizing logistical costs (transport plus inventory costs) by finding the mode/vehicle that provides the optimal combination of shipment size and frequency of delivery. Econometric models are subdivided into two sub-families: shipment-level models and market-share models. It is worth mentioning that, although the ITIC model is used (FRA 2005) in the United States, the freight mode/vehicle choice models in use in all other countries— developed and developing—are econometric shipment-level models. Overall, econometric models provide the best trade-off between the ability to capture the decision dynamics that drive the choice of freight mode/vehicle and practicality, so these models provide the best modeling foundation to predict the effects of alternative mode choice policy scenarios in real-life settings. These conclusions are in agreement with the chief recommen- dations of the truck size and weight research road map (Special Report 328 2018). Additional information about the rationale for these conclusions follows. Best Choice: Shipment-Level Models Shipment-level models are typically estimated as discrete choice models. Discrete choice models try to capture the decision-making behavior of the agent in charge of making deci- sions on mode/vehicle choice, using random utility theory (Manski 1973, Manski 1977). Dis- crete choice models express the probability of choosing a mode/vehicle as a function of such variables as shipment size, cost and time for each mode/vehicle alternative, commodity type, and the attributes of the shipper. These utility functions, because of their empirical nature, capture the effects of the key independent variables without having to collect highly detailed data about each and every variable that could influence freight mode/vehicle choice. The data needed to estimate and use these modes are disaggregate, typically at the ship- ment level. The power of discrete choice models resides in their ability to consider the effect of spatial variables at a fine level of detail, e.g., the distance between an establishment and the C H A P T E R 5 Modeling Approach Selected

72 Impacts of Policy-Induced Freight Modal Shifts closest rail terminal or the effects of the frequency of rail service on waiting times. Discrete choice models provide a better platform to capture the interdependence between shipment size and the mode choice (endogeneity) and multiple ways to consider the interactions among vari- ous vehicle types for a given mode. The methodological alternatives include nested logit models with choices of mode/vehicles and shipment sizes (Chiang et al. 1981), advanced models that explicitly treat shipment size as a continuous model that is embedded in the mode/vehicle choice model (McFadden et al. 1986b), and techniques that use instrumental variables (Holguín-Veras 2002). The latter approach was the one used to estimate the freight mode choice models in this project. Second Best Choice: Aggregate (Market-Share) Models Aggregate market-share models express the market share of a mode/vehicle as a function of variables (such as commodity type) and the average characteristics of the competing mode/ vehicles (e.g., time and cost). These models were the first econometric models used to model mode choice in the 1960s and 1970s, when large-scale freight origin-destination surveys were routinely conducted. However, because of their aggregate nature, these models cannot consider the role played by attributes such as shipment size and other shipment-level or establishment- level factors that influence mode/vehicle choice. Thus, these models are less able to capture the key dynamics that influence mode/vehicle choice and are much less sensitive to policy variables. Also, collecting the data needed to calibrate these models is not a trivial endeavor, as doing so typically requires large-scale freight surveys. Without access to large-scale data, such as the CFS, the estimation of market-share models is bound to be expensive because of the required data collection effort. Shipment Level Models Market Share Models Recommendation Best Choice Second Best Choice Not Recommended Methodology Random Utility Theory Regression (OLS) Models Optimization Techniques Data needed for calibration Disaggregate data with choices, and attributes of modes/vehicle, shipments, and shippers Market shares by commodity type, attributes of vehicle/modes and shipments No calibration needed Data needed for application Travel time and cost by modes, commodity types, sample of shipments Travel time and cost by modes and commodity type Disaggregate data about all the components of transport and inventory costs, travel and cost by vehicle modes, disaggregate demand to be transported from shippers to receivers Output produced Probability that an establishment selects a vehicle/mode Market shares among modes/vehicles by commodity type Choice of mode vehicle and shipment size, for an establishment and shipment, from shipper to receiver Ease of use, practicality Once estimated, relatively easy to use and practical Once estimated, relatively easy to use and practical Easy to use Quality of output High Low-medium As high as the quality of the input data Considers vehicle choice? Yes Yes Yes Advantages Able to consider the behavior of freight agents Fast and easy to model, if sufficient data are available. Replicates the real-life decision process Disadvantages Requires shipment level data Needs large amounts of calibration data, forecast- ing ability is in question Unavailability of input data Approach Supply Chain Methods (Inventory Theory) Econometric Approach Table 17. Summary of features of alternative modeling approaches.

Modeling Approach Selected 73 Issues with Supply-Chain-Based Models Supply-chain-based models attempt to replicate the process followed by private-sector com- panies when making mode/vehicle decisions. These models consider in great detail all the dif- ferent cost components that influence logistics costs (the summation of transportation and inventory costs), as well as detailed demand data such as the amount of cargo that a shipper needs to deliver to each customer. Although at first sight, the notion of replicating the decisions of individual companies seems like a good idea, the data requirements are so difficult to meet that these models cannot reliably be used in practice. The most widely used model in this family is the ITIC (FRA 2005, FHWA 2011). The fundamental weakness of these models is the tremendous challenge of collecting the data needed. The onerous cost of collecting the data needed by ITIC is a consequence of the confi- dential nature of private-sector data, particularly cost data, and the profound heterogeneity of business conditions, where different industry sectors and companies have different transpor- tation and inventory costs. It is worth mentioning that ITIC’s developers were undoubtedly aware of the challenge associated with securing the input data as revealed in the “Data Over- view” section (p. 17) of “Intermodal Transportation and Inventory Cost Model: Highway to Rail Intermodal” (FRA 2005): While these are the data needs of the model, the problem is that publicly available sources of disaggre- gate data are difficult to find. This is true in spite of the fact that hundreds of thousands of shipments are made every day by manufacturing companies and product distributors throughout this country as well as overseas. Each of these shipments is fully documented, the movement is billed for, and the transportation charges paid to trucking companies, railroads, airlines, barge lines, pipeline companies, and other freight carriers. The data collection problem is caused by the fact that it is against the law for carriers to reveal the names of shippers and receivers, the product amounts that are shipped and the origins and destinations of individual movements without that shipper’s individual approval. Only in very rare circumstances, such as the Southern California Association of County Governments Port and Mode Diversion Model (Leachman et al. 2005, Leachman 2008), can supply chain models be used to meaningfully analyze policy alternatives. In this case, the data needed—detailed records of import containers—were available. However, for studies of mode/ vehicle choice in domestic cargo, such records are not readily available and are extremely hard to assemble. The challenge of collecting the data needed to run the ITIC has been recognized by ITIC users, who sometimes have elected to use it with aggregate estimates of demand (e.g., from the FAF) and with default values of the input data cost items. Since the ITIC was not designed to use aggregate data, using aggregate estimates that combine thousands of individual shipments leads to an incorrect calculation of logistics costs and freight mode/vehicle choices. More- over, assuming that all of the data items required by the ITIC are equal for all industry sectors neglects to consider the heterogeneous nature of business activity. Thus, it is not advisable to use the ITIC in such a way. In summary, econometric models provide the best combination of conceptual validity and practicality. These models are based on empirical data that reflect the real-life patterns of freight mode choice in the United States, require a fraction of the data needed to use supply chain models, and are backed by solid theory. The decision concerning which type of econo- metric model to use, shipment-level or market share, should take into account that shipment- level models are better able to accurately estimate the impacts of policies and/or changes in market conditions. The use of market-share models should be confined to long-term planning exercises where the objective is to obtain a general idea about future market shares.