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.
59 This chapter summarizes the analysis conducted by the NCFRP Project 44 research team on methodologies that could explain and forecast freight mode choice. The literature review under- taken as part of this research can be found in full in the latter part of this section. The literature review focused on mode choice behavior; factors influencing mode decisions; the relationship between shippers and carriers; and the tools available for mode choice analysis, particularly the ITIC models (FRA 2005). The review provided key insights on the advantages and disadvan- tages of the various methodologies. In general terms, the methodologies can be classified into econometric and supply chain models. Econometric modeling is the most widely used; out of over 40 references analyzed by the research team, 64 percent used econometric modeling and the remaining 36 percent used supply chain models. Econometric Models Econometrics uses statistical and probability theory to obtain empirical models that represent a real-life economic process. In most cases, econometric modeling entails using a functional form with parameters estimated using real-life data. In doing so, it is essential to ensure that (1) the functional form provides a meaningful representation of the phenomenon being mod- eled, in terms of the actual equation used and the variables, dependent(s) and independent; (2) the data contain sufficient variability to capture the wide range of behaviors exhibited in real-life; and (3) appropriate estimation procedures are used to calibrate the model parameters. These considerations play an important role in influencing the selection of econometric tech- niques to estimate freight mode choice. Freight mode choice could be studied at the level of markets or shipments. Market-share models seek to explain the collective decisions of the hundreds, or thousands, of individual decision makers that are implicitly captured in the market share of a given freight mode. In contrast, shipment-level freight mode choice models seek to represent the decision-making process used when a shipper decides on the freight mode to be used to transport a specific shipment. It is widely acknowledged that shipment-level models are the best methodological alterna- tive because they (1) establish a more direct connection between the choice being modeled and the independent variables, (2) use the data in the most efficient way, and (3) enable a seamless incorporation of policy variables. The branches of econometrics used to estimate these models (market-share and shipment-level) are (1) continuous dependent variable models and (2) dis- crete outcome models. These models are described next. Table 13 shows the main types of econometric models used. C H A P T E R 4 Overview of Available Methodologies
60 Impacts of Policy-Induced Freight Modal Shifts Continuous Dependent Variable Models These models estimate mode choice as a continuous dependent variable, typically the market share of a freight mode. Frequently used independent variables are (1) average values of modal attributes, such as costs, transit times, distances and (2) average values of shipment attributes, such as shipment size, commodity type, origin, or destination. Different mathematical func- tional forms are available. One of the most appealing forms is the logistic form shown in Equa- tion (1) because it is able to replicate the range of values of market shares that are bound between zero and one. where Ui = Î²0i + Î²1X1i + Î²2X2i + . . . , Uj = the utility of choosing mode j, Î²0i = mode i specific parameters, Î²1, Î²2, . . . = attribute specific parameters, X1i, X2i, . . . = mode i attributes, and Pi = market share of mode i. As shown, Equation (1) expresses the market share of a given mode i as a function of its attri- butes and of the other alternatives. The higher the performance of a given mode, the higher the market share estimated by Equation (1). In the case of two modes, i, j, Equation (1) reduces to Equation (2) could be linearized by taking natural logarithms and obtaining the inverse, then expressing Ui as a function of corresponding explanatory variables, as shown in Equation (3). Thus, the parameters of the models can be obtained by Ordinary Least Squares (OLS) techniques. In the case of multiple modes, the estimation becomes a set of N-1 linear equations, as shown in Equation (4), where N is the number of mode choice options â = (1)P e e i U U j i j = + (2)P e e e i U U U i i j ln 1 1 ln . . . (3)0 0 1 1 1 2 2 2 P P P X X X X i j i j i j i j i( ) ( ) ( )âï£«ï£ï£¬ ï£¶ ï£¸ï£· = ï£« ï£ï£¬ ï£¶ ï£¸ï£· = Î² â Î² + Î² â + Î² â + P P X X X X k i k i k i k i k iln . . . (4)0 0 1 1 1 2 2 2( ) ( ) ( ) ï£« ï£ ï£¬ ï£¶ ï£¸ ï£· = Î² â Î² + Î² â + Î² â + â â Market Share (Aggregate) Models Shipment-Level (Disaggregate) Models Market share of a mode Probability of a shipment selecting a mode Modal attributes (average costs, travel times, and distances), shipment characteristics (average shipment size and commodity type) Modal attributes (costs, travel times, and distances), shipment characteristics (shipment size and commodity type) Ordinary Least Squares (OLS) regression, and other forms of continuous dependent variable models Discrete choice models (probit, logit, nested), Discrete-continuous Attribute/Method Dependent Variable(s) Independent Variable(s) Econometric Techniques Table 13. Various econometric models used in the literature.
Overview of Available Methodologies 61 However, since OLS cannot solve the system of linear equations, estimating market-share models for multiple modes requires advanced techniques such as maximum likelihood estimation. Discrete Choice Models These models estimate freight mode choice by computing the probability that the corre- sponding decision-maker selects mode i to transport shipment n. Hence, these models are disag- gregate in nature, where the dependent variable is discrete. Discrete choice models, also known as discrete outcome models, are based on the random utility theory, which states that the user selects the choice that maximizes the utility derived from it. However, since it is not possible to have perfect information about the alternatives considered in the decision process, the choice problem is expressed in probabilistic terms. Thus, the utility of choice i for shipment n (Uin) is given by Ben-Akiva and Lerman (2010): where Î²i = vector of estimable parameters, Xin = vector of observable characteristics, and ein = error terms. Different models can be derived using different assumptions for the error terms. The most widely used approaches are binary logit, in cases where there are only two choices, or multi- nomial logit (MNL), in cases where there are more than two modal alternatives. The logit model assumes that the error terms are Extreme Value Type 1 (Gumbel) distributed. This assumption leads to the estimation for probability of an individual n selecting an alternative i as given below: Like OLS, the MNL model is based on a few assumptions. The MNL model assumes homo- geneity across individual variables in the data, independence of irrelevant alternatives (IIA), independently and identically distributed error terms, and common variance of the error terms (homoscedasticity). However, quite frequently, the alternatives being considered share com- mon attributes. In these cases, the assumption of independence embedded in the MNL breaks down, and the MNL produces erroneous results. To overcome this issue, more advanced forms of discrete choice models have been developed. To this effect, Jiang et al. (1999) adopted a nested logit model, and Kim (2002) introduced a random heterogeneity logit model. Norojono and Young (2003) adopted a heteroscedastic extreme value method; Arunotaya- nun and Polak (2007b), Patterson et al. (2007), and Abate and de Jong (2014) use mixed MNL models, where the Î²i in Equation (6) is assumed to follow a specified distribution. Other models include an MNL-MNL Archimedean class of copula model, like the one used by Pourabdollahi et al. (2013). Discrete-Continuous Models An important feature of freight mode choice models is that the decision makers make two choices at the same time. The first is the choice of the freight mode, and the second is the choice of the corresponding shipment size. These two choices, hopefully, minimize total logistics cost (the summation of transportation and inventory cost). Thus, the econometric interactions between the choice of mode and shipment size must be econometrically considered. It turns out = Î² + Îµ (5)U Xin i in in â ( ) = Î² Î² (6)p i e e n X X i in i in
62 Impacts of Policy-Induced Freight Modal Shifts that the decision concerning shipment size has a tremendous influence on freight mode choice. It has been found that the closer the shipment size is to the capacity of a freight vehicle or mode, the more likely it is that vehicle or mode will be selected to transport the shipment (HolguÃn- Veras 2002). In the words of Samuelson (1977), the decision on shipment size is âmode deter- mining.â Capturing this combination of decisions requires the use of discrete-continuous models. The mathematical form for discrete-continuous models is similar to Equation (5), but includes endogenous variable(s), Zn, as shown in Equation (7). There are multiple approaches to addressing the issues created by the correlation between the discrete and continuous choices. A partial list of techniques include (1) the control func- tion or instrumental variable approach (Heckman 1976); (2) the Berry, Levinsohn, and Pakes approach (Louviere et al. 2005); (3) the dual approach (Matzkin 2004); (4) the latent variable approach (Ben-Akiva and Boccara 1995); and (5) the special regressor approach (Lewbel 1998). The technique used in this research is the instrumental variable approach, which was selected for practicality reasons. In this technique, the values of Zn are replaced with the estimated values from an instrumental variable, as shown in Equation (8) (HolguÃn-Veras 2002): where Yn = vector of instrumental variables (exogenous) and dn = error terms. The majority of the publications on freight mode choice modeling use the instrumental vari- able approach with various functional forms and exogenous variables as in Equation (8). For example, Abdelwahab and Sargious (1991), Abdelwahab and Sargious (1992), and Abdelwahab (1998) adopt a maximum likelihood binary probit model with continuous shipment sizes for various modes, estimated as a function of other exogenous variables. HolguÃn-Veras (2002) uses an instrumental variable approach MNL with continuous shipment size. Dewey et al. (2002) and Lloret-Batlle and Combes (2013) use OLS to estimate the demand function, where the shipment size is calculated using the inventory theory approach (discussed later in this section, under Supply Chain Models of Freight Mode Choice). Supply Chain Models of Freight Mode Choice Supply chain models attempt to mathematically replicate the processes followed by business managers in decisions pertaining to replenishment and distribution, such as order frequency, shipment size, and the like. Multiple considerations come into play when managers decide on the combination of order frequency and shipment sizes to be used for a given time period. In relatively steady conditions, where the demand for a given product is reasonably constant, man- agers could select the combination of order frequency and shipment size that minimizes the total logistic costs, i.e., the sum of transportation and inventory costs. In this context, transportation costs that are high relative to inventory costs lead to infrequent orders of large shipments while the converse situation leads to frequent orders of small shipments. Thus, in the first case the use of large freight vehicles or modes is favored, whereas in the second case the use of small freight vehicles or modes is favored. A different situation arises if the demand for a product is highly variable. In these cases, the need to maintain a safety stock may influence order patterns and ultimately freight vehicle and mode choice. Inventory management models are used to make these decisions. = Î² + Î± + Îµ (7)U X Zn n n n = Î³ + Î´ (8)Z Yn n n
Overview of Available Methodologies 63 The Economic Order Quantity Model Inventory management models calculate the shipment size needed to minimize the total logis- tics cost, which includes inventory as well as transportation costs. These principles are embed- ded in the economic order quantity (EOQ) model (Harris 1915), which is widely used in supply chain modeling (Sampson 2014). The inventory required depends on the choice of mode, fre- quency, reliability, transit time, freight rate, and capacity of the inflow. The inventory quantity also depends on the cost of storage, the quantity required in a time of crisis to run the business or maintain customer loyalty, and the difference between inflow and outflow rates. The outflow rate is affected by demand, which in turn depends on factors such as season, price of substitutes and complements, the economy, and others, as shown in Figure 32. The EOQ model assumes that demand is constant, and does not change over time. The opti- mal shipment size is that which minimizes the total logistics cost to the firm, including the inventory and transportation costs, which can be calculated with the EOQ. Equations (9) and (10) show the EOQ calculations for optimal shipment size and duration of time between orders. where D = demand (units/year), S = ordering cost ($/order), H = storage cost ($/unit/year), and Q = shipment size. The inverse of the time between orders (T) gives the frequency of the deliveries (F) in number of deliveries per year, is as shown in Equation (11) The minimum total logistics cost (TC) is given in Equation (12) â = 2 (9)Q DS H â = 2 (10)T S HD = =1 2 (11)F T HD S 2 (12)TC SD Q HQ= + Inventory/ Stock Inflow Outflow Mode Frequency Reliability Capacity Demand Transit time Freight rate Storage costs Safety stock required Life span Damages Figure 32. Factors influencing inflow, inventory, and outflow.
64 Impacts of Policy-Induced Freight Modal Shifts ITIC Model The FRA (2005) developed the ITIC model to estimate the impacts of alternative policies on freight mode choice. The ITIC model was preceded by the Translog Shipper Cost Model developed by Roberts (1981) and the Intermodal Competitive Model developed by Association of American Railroads. Two different versions of the ITIC were developed, the state tool, ITIC-ST, and the intermodal tool, ITIC-IM. ITIC-IM helps users estimate the shift from truck to rail intermodal service. ITIC-ST estimates potential shifts from highway freight traffic to different truck configura- tions or rail intermodal service. The ITIC models are disaggregate models that minimize the total logistics costs. The ITIC models consider various attributes of a shipment, a shipper/receiver, a commodity, and modes under consideration to select the mode that minimizes the total logistics costs. The costs include ordering, capital carrying in transit, capital carrying in inventory, ware- housing, loading and unloading, safety stock carrying, and loss and damage claims. The ITIC also considers reliability, described as the variability in the ordering lead time, using a Gamma distribution. Table 14 summarizes the inputs, parameters, and models used by ITIC tools. As shown in Table 14, the ITICâs data input requirements are onerous. The data ITIC requires, such as storage and handling and line-haul charges, vary widely depending on the type of com- modity, and using average values will lead to erroneous results. ITIC also requires commercially Datasets Parameters ITIC-ST (1) Freight Analysis Framework (2) Rail carload and intermodal data (3) Truckload movement data General: (1) Commodity type (2) Annual use (lb/year) (3) Shipment size (lb) (4) Shipment density (5) Unit value ($) (6) Origin state (7) Destination state For each mode and truck type: (8) Cost per mile (9) Distance (miles) (10) Maximum payload (11) Railcar type (rail carload, TOFC, COFC) (12) Junction frequency (13) Rail revenue per pound of shipment (14) Rail variable cost (15) Distance by rail (miles) (16) Drayage in miles (first and last leg) (1) Operating days per year (2) Variability in demand (3) Order processing cost (4) Inventory costs: capital, tax, insurance, and depreciation (5) Shipment method (pallets) (6) Size, weight of pallet (7) Loading/unloading costs and time (8) Storage, handling costs (9) Reliability in transit time ITIC-IM Not applicable (1) Commodity type (2) Annual use (lb/year) (3) Shipment size (lb) (4) Unit value ($) (5) Origin state (6) Destination state (7) Cost by truck per mile (3-axle tractor with 2-axle semitrailer) (8) Distance by truck (miles) (9) Maximum payload of truck (10) Railcar type (rail carload, TOFC, COFC) (11) Junction frequency (12) Rail revenue per pound of shipment (13) Rail variable cost (14) Distance by rail (miles) (15) Drayage in miles (first and last leg) (1) Reliability in transit time (does not account for variability in the demand) (2) Safety stock (3) Carrying cost includes: capital tied up in the inventory, pilferage, transfers, handling, and storage for safety stock (4) Loading/unloading costs and time (5) Cost of drayage (6) Overhead costs (per mile of truck) (7) Rail junction interchange delays Inputs Required Parameters AssumedModel Source: (FRA 2005, FHWA 2006) Table 14. ITIC inputs, parameters, and models.
Overview of Available Methodologies 65 sensitive data, which are practically impossible to obtain, such as the annual volume of cargo sent by the shipper to each individual customer (after all, the ITIC is a shipment-level model). Essentially, the ITIC requires data that are either extremely expensive or, in some cases, virtually impossible to collect, particularly for planning and policy studies, which also involve forecasts of future conditions. As a result, notwithstanding its practicality and ease of use, the ITIC would probably not provide a reliable approach to study the impacts of public-sector policies on freight mode choice. Summary of Relevant Literature This section summarizes the literature on freight mode choice. Publications are divided into econometric modeling and supply chain modeling. Of these two approaches, econometric modeling is widely used, and is also recommended for analyzing freight mode choice policies. Econometric modeling is also recommended by various publications. For example, these two approaches were analyzed by Gray (1982), who recommends the econometric approach because of the physical restrictions and obligations from the optimization techniques, and the impor- tance of loyalty between the shipper/carrier and receiver on mode choice that can be captured by econometric models. Winston (1983) compared the model significance produced by econo- metric methods (aggregate and disaggregate) and optimization models. He concluded that the disaggregate data would produce a better model as it could capture the finite-level details affecting the modal choice. The sections below provide a brief review of the relevant literature. Econometric Modeling McGinnis et al. (1981) analyzed freight mode choice using a one-way analysis of variance and MNL on a group of shippers. The authors found that the shippers that use for-hire truckload shippers do not transport high-value products but assign a high importance to special services (diversion, stop-off privileges, and processing en route) and avoiding damage of goods. Shippers using LTL tend to ship smaller quantities, give more importance to speed and reliability, and are also concerned with loss and damage of goods. Rail-dominant users attach a high importance to freight rates and availability of special services; the products they ship are usually not fragile and are shipped in larger quantities. Parcel shippers tend to utilize small package services with high product values, so they give lower importance to freight rates than to loss and damage compared to the users of other modes. Finally, private carriage shippers, who tend to have fragile products that are difficult to handle and a high proportion of intra-company ship- ments, place a great deal of importance on speed and reliability. Winston (1981) used a binary probit model to study mode choice and elasticities and concluded that shippers of perishable goods and their supporting goods, such as packing materials and metals, give high priority to the quality of a modeâs services. As the transit time of the mode increases, the firmâs inventory cost also increases. The traffic is more elastic to price than service quality for goods with large transportation costs, whereas agricultural products and chemical petroleum products would display a huge modal change to rail with an increase in the quality of service. Young et al. (1982) analyzed the predominant variables affecting mode choice using an elimination-by-aspects model for shippers including manufacturers and non-manufacturers. For a typical shipper, the attributes that influence a choice of rail most significantly are conve- nience, risk of damage, reliability, capacity and freight rates. Young et al. (1982) concluded that reliability is the most significant attribute for manufacturers, whereas capacity and freight rate are the most significant attributes for non-manufacturers. Winston (1983) provided a detailed review of freight mode choice modeling techniques, aggregate and disaggregate models. He concluded that the disaggregate models provide more accurate analysis and an estimate of user satisfaction for a given mode, depending on the availability of data.
66 Impacts of Policy-Induced Freight Modal Shifts McFadden et al. (1986a) were the first to analyze shipment size and mode choice simultane- ously using a discrete-continuous model with produce transportation data. McFadden et al. (1986a) found that shipment size is dependent on the attributes of the modes, such as freight rate, transit time, frequency, and reliability. In policy-induced mode shifts, the total logistics pattern will be affected, so the shipment size will need to be adjusted to be compatible with the new mode. The authors (McFadden et al. 1986a) conclude that in markets dominated by motor carriers, diversion to rail can be achieved with sizeable improvements in transit time, along with fixing the rates in response to market demand. Wilson et al. (1986) used linear logit models to examine the factors that influence the mode choice decisions for the alternatives of hired truck, private truck, and rail. Mode choice is affected by transit time, reliability in transit time, and length of hauls. Rail is more likely to be selected for long hauls and truck for short hauls; this is consistent with expected shipper behavior. As frequency increases, there is a move away from hired truck to private truck. The provision of pickup services and shipment tracking promotes the use of rail. Intercept terms of private truck and rail had the highest explanatory power, implying that factors not considered in the model are influencing choices. Abdelwahab and Sargious (1990) explained the use of instrumental variable approaches to the discrete-continuous choice model for the choices of mode and shipment size, and Abdelwahab and Sargious (1991) provided two alternative estimation methods that can be modeled with data similar to that of a conventional mode choice model. The two alternative estimation methods are (1) the simultaneous equation model, which uses two equations to predict ship- ment size by each mode and a third equation to predict the choice of mode and (2) a two- stage least squares (TSLS) estimation, which uses maximum likelihood probit to estimate the mode choice and OLS to estimate shipment size. Similarly, Abdelwahab and Sargious (1992) provided the three-equation model, i.e., two equations from TSLS for shipment size in trucks and rail and one equation from a binary probit model to identify the mode by which the ship- ment was transported. The results of cross-elasticities indicated the existence of some degree of competition between the two modes, which seemed to intensify as the direct elasticities of demand for either mode increased. Nam (1997) found that the disaggregate models by com- modity type produce better estimates for freight mode choice. Jiang et al. (1999) used a nested logit model to analyze mode choice and found that with an increase in transportation distance and shipment frequency, the probability of choosing rail and combined modes increases, and an increase in the shipment size leads to an increase in the probability of choosing rail. HolguÃn-Veras (2002) used a discrete-continuous model with shipment size as the continu- ous variable, distance as the instrument, an MNL model for the choice of vehicle (truck type), and elasticities calculated using both MNL and heteroscedastic extreme value methods (Bhat 1995). The results suggest that carriers chose pickups or trucks more often than semi-trailers. Any change in the cost or shipment size in trucks affects the market share of other modes nega- tively. Imposing permissible axle load limits causes more congestion due to the increased num- ber of pickups. HolguÃn-Veras (2002) emphasizes the limitations of commodity-based models, as they do not account for empty trips, which make up 15 to 40 percent of total trips (HolguÃn- Veras 1984). Kim (2002) used inherent random heterogeneity logit models to assess the effect of cost, arrival time, and reliability on selecting the preferred mode among ferry, new ferry, shuttle, and through rail. Norojono and Young (2003) analyzed the hierarchical preference data using a nested logit model and heteroscedastic extreme value model to analyze the final results. The authors found that users give more emphasis to delivery time and safety (quality). Time of departure and distance to rail terminal are not significant. The rail mode is more elastic to cost and time, which may explain the greater decrease in the rail mode compared to trucks. Train and Wilson (2006) improved the existing aggregate models used by the U.S. Army Corps of Engineers to estimate shippersâ choice between rail and barge, by replacing them with dis- aggregate MNL models that include spatial effects. The authors found that rate, transit times, and
Overview of Available Methodologies 67 reliability have statistically significant effects on choices, and a number of unobserved character- istics have important effects. Shippers located close to a river are more likely to use the river than to use rail, and as the distance from the river increases, shippers switch to rail. Arunotayanun and Polak (2007a) used a mixed MNL model to analyze the data for random taste heterogeneity and panel effects for three modes: small truck, large truck, and rail. Results indicated that electronic commodities are not sensitive to either cost or time; leather commodities are not sensitive to cost and only moderately sensitive to time. Departure time is a significant attribute for electronics, whereas train formation is significant in the case of food. Food and electronic commodities are sensitive to frequency of service in terms of small trucks, and textiles are sensitive to frequency of service in terms of rail. The authors revealed the existence of significant amounts of taste hetero- geneity among shippers in relation to service attributes, quality, and flexibility. Patterson et al. (2007) analyzed freight mode choice accounting for random effects using a mixed logit model and found that increases in cost, damage risk, and security risk decrease the probability that a carrier will be chosen, whereas an increase in on-time reliability increases that probability. The authors concluded that shippers mistrust rail and are very cautious about using rail to move their consignments; therefore, increasing rail market share may be tremendously challenging. Cavalcante and Roorda (2010) used a discrete-continuous model with shipment size as the continuous variable and vehicle-type as the discrete variable. The results indicated that small vehicles are more likely to be chosen for higher-value, time-sensitive shipments and services (as opposed to goods shipments) and larger vehicles for long-distance shipments. De Jong (2009) concluded that discrete-continuous is the preferred model, a discrete-discrete model might be preferable when there are no data available, and the model that assumes independence of mode and shipment is not preferred. Samimi et al. (2011) examined the competition between truck and rail for a given commodity movement, using binary choice (probit and logit) models. This research found that rail is preferred for longer distances and large shipments. Rail is found to be more sensitive to cost, whereas truck shippers are found to be sensitive to shipping time. Ship- pers who chose truck in the 2 years previous to the research were less likely to choose rail over truck, and the conclusion is drawn that modal decisions with respect to fuel cost are inelastic. Pourabdollahi et al. (2013) used an MNL-MNL copula expression to derive the probabilities of choosing a mode and shipment size. The results indicate that shipping cost is the most sig- nificant variable. Lloret-Batlle and Combes (2013) showed that the Box-Cox transformation to density times the air distance, which covers an extremely large range of values, improved the likelihood function substantially. Abate and de Jong (2014) investigated the allocation of differ- ent truck sizes and the factors affecting the allocation. A discrete choice continuous model with shipment size as the continuous variable is used using the MNL model. The choice of vehicle includes the five truck types of increasing capacities. As an extension to HolguÃn-Veras (2002), this paper introduces a cross-alternative correlation parameter assuming normally distributed disturbances, to overcome the independence of irrelevant alternatives assumption in MNL. The results were verified by a nested logit with rigid trucks grouped in one nest. The probability of choosing heavy vehicles decreases with an increase in the fixed costs or vehicle age and will increase with an increase in the distance or fleet. Whether the carrier is an owner or hired is also significant in vehicle choice, as hired vehicles can aggregate the demand. Supply Chain Modeling In addition to the ITIC model, there are a number of publications that use supply chain or inventory theory methods. For example, Baumol and Vinoud (1970) combined standard inven- tory theory with the abstract mode technique to estimate freight demand for various modes. The method used by Baumol and Vinoud (1970) estimated the change in mode, quantity, and fre- quency for the respective change in relevant attributes such as shipping time or shipping cost by minimizing the total logistics costs. The logistics cost function minimized constitutes total direct
68 Impacts of Policy-Induced Freight Modal Shifts shipping cost, in-transit carrying cost, ordering cost, and recipientâs inventory carrying cost, which is derived to obtain the optimal frequency of reordering. Samuelson (1977) modeled freight tar- iffs to yield a better understanding of rate structure and explained mode choice using mileage, shipment size, weight, density and value of the commodity, special handling requirements, and geographic region. The significant variables are mileage, weight, and value of the commodity in terms of rail carload models. Railroads favor value-of-service pricing, emphasizing the value of the commodity in rate determination, while trucks favor cost-of-service pricing, emphasizing density of the commodity in rate determination. Cunningham (1982) provided a theoretical model for the supply chain method and concluded that model choice is a function of costs incurred by the competing modes of carriage, the shipperâs predispositions toward the various competing modes and carriers, and the total transportation and non-transportation cost to the shipper. Hall (1985) explained the interdependence of shipment size and mode choice by analyzing three transport alternatives: FTL contract carriers (focused on large shipments) and LTL carriers and United Parcel Service (for small shipments). Furthermore, Hall included capacity constraints of the various modes in the model that minimize the sum of transportation and inventory costs to estimate the optimal shipment size. Tyworth (1992) integrated inventory elements such as order- ing replenishments and holding (cycle, safety, and in transit) stock with transportation compo- nents such as freight rate, speed, and the consistency of delivery to define each mode or carrier, in order to obtain optimal costs. Casavant et al. (1993) used the least cost spatial equilibrium model using the linear programming that minimizes the total cost, made up of the assembly cost, elevation cost, and shipment and handling cost, subject to several constraints such as produc- tion, elevator, and storage capacities. Leachman et al. (2005) evaluated the economic viability of additional port user fees and the potential for using the revenue from the fees to improve the infrastructure, using a supply chain model that minimizes the total logistics costs of the importers using the port. Two scenarios were considered: As-Is (without improvements in shipment lead times) and Congestion Relief (including the impact of major improvements in lead time distribu- tions). An aggregate demand curve of port demand versus fee value was constructed; the slope of this curve is the elasticity of imports with respect to user fees. Leachman et al. (2005) included in the model different strategies for the importation process, varying from direct shipping to consolidation-deconsolidation practices for import cargo at San Pedro Bay ports. Blauwens et al. (2006) used an inventory-theoretic framework to estimate implications from different mode shift policies that aim to decrease congestion. Results indicate that an increase in the transportation cost of roads by 20 percent would increase the rail mode share from 16 to 27 percent. Leachman (2008) developed a supply chain model and used it to determine the most cost-effective way for each importer to transport from Asia to regional distribution centers in the United States. Leachman (2008) considered alternative ways to reach regional distribution centers including direct shipping, exclusive shipment for a single commodity, or consolidation- deconsolidation of various goods. Brogan et al. (2013) used supply chain methods to analyze the possibility of shifting from less fuel-efficient modes to more fuel-efficient and greener options with hypothetical changes in policies. Freight Modes, Scope, and Variables from the Literature This section discusses the modes, scope, and variables considered in the freight mode choice literature. This discussion will provide a bigger picture of the advantages and limitations of various methodologies, datasets, and estimation procedures. The types of mode, scope, and geo- graphical level included in each publication on freight mode choice are shown in Table 15. The majority of the recent literature deals with different types of trucks (Arunotayanun and Polak 2007a, Cavalcante and Roorda 2010, Lloret-Batlle and Combes 2013, Abate and de Jong 2014). There are just two references that consider all four mode choices. Only three references consider
Overview of Available Methodologies 69 the air mode and waterways. Table 15 also shows that some references (e.g., Cunningham 1982 and Gray 1982) did not consider any specific mode, as these studies are theoretical. No study on supply chain models considers the air mode. Truck and rail are the predominant mode shares studied in research on freight mode choice. Many references on econometric models concen- trate on practical applications, while the majority of studies on supply chain methods are theo- retical. Mode choice studies predominantly focus on national-level freight movements, while there are only a few publications on regional-level freight flows. However, a deeper look into the modeling, data, and geography considered in the literature reveal that none of the publica- tions has a solid nationwide model estimated using a reasonable sample size to provide better estimates for modal splits. NCFRP Project 44 is the first to fill this gap by using 2012 confidential CFS data with nearly two million observations at the national level. T ru ck R ai l P or ts / W at er w ay s A ir T he or et ic al P ra ct ic al E m pi ri ca l N at io na l R eg io na l McGinnis et al.(1981) Winston (1981) Young et al. (1982) Winston (1983) McFadden et al. (1986b) Wilson et al. (1986) Wilson et al.(1988) Abdelwahab and Sargious (1990) Abdelwahab and Sargious (1991) Abdelwahab and Sargious (1992) Nam (1997) Abdelwahab (1998) Jiang et al. (1999) Dewey et al. (2002) HolguÃn-Veras (2002) Kim (2002) Norojono and Young (2003) Train and Wilson (2006) Arunotayanun and Polak (2007a) Patterson et al. (2007) De Jong (2009) Cavalcante and Roorda (2010) Samimi et al. (2011) Pourabdollahi et al. (2013) Lloret-Batlle and Combes (2013) Abate and de Jong (2014) Baumol and Vinoud (1970) Samuelson (1977) Cunningham (1982) Gray (1982) Hall (1985) McGinnis (1989) Casavant et al. (1993) FRA (2005) Leachman et al. (2005) FHWA (2006) Blauwens et al. (2006) Leachman (2008) Stewart et al. (2008) Combes (2010) Brogan et al. (2013) Total 36 31 7 3 9 27 4 20 10 References Modes Considered Scope Geography NA NA NA NA NA Econometric Models Supply Chain Models NA Note: Represents that different truck types are considered. Table 15. Modes, scope, and geography considered in freight mode choice literature.
70 Impacts of Policy-Induced Freight Modal Shifts Table 16 shows the variables considered in publications on freight mode choice, with freight rate, transit time, shipment size, and reliability being some of the most important variables. Table 16 shows that econometric models are able to provide solid depictions of freight mode choice with a smaller number of independent variables than supply chain models, as the maxi- mum number of variables required for econometric models is around seven while that of supply chain models is twelve. In Table 14, the ITIC models (FRA 2005, FHWA 2006) consider 20 to 25 variables each and require a huge amount of data to estimate the mode choice. Data on reli- ability, capacity, and inventory costs are especially difficult to obtain for all commodities in the United States, and are needed to utilize these models to estimate nationwide freight modal shifts. F re ig ht R at e T ra ns it T im e R el ia bi lit y C os t of D am ag e L en gt h of H au l F re qu en cy C ap ac it y Sh ip m en t Si ze C om m od it y T yp e Sh ip m en t V al ue In du st ry S ec to r Sh ip m en t D en si ty In ve nt or y C os ts In te ra ct io n of a a nd b A cc es s to T er m in al s McGinnis et al.(1981) Winston (1981) Young et al. (1982) Winston (1983) McFadden et al. (1986b) Wilson et al. (1986) Wilson et al.(1988) Abdelwahab and Sargious (1990) Abdelwahab and Sargious (1991) Abdelwahab and Sargious (1992) Nam (1997) Abdelwahab (1998) Jiang et al. (1999) Dewey et al. (2002) HolguÃn-Veras (2002) Kim (2002) Norojono and Young (2003) Train and Wilson (2006) Arunotayanun and Polak (2007a) Patterson et al. (2007) De Jong (2009) Cavalcante and Roorda (2010) Samimi et al. (2011) Pourabdollahi et al. (2013) Lloret-Batlle and Combes (2013) Abate and de Jong (2014) Baumol and Vinoud (1970) Samuelson (1977) Cunningham (1982) Gray (1982) Hall (1985) McGinnis (1989) Casavant et al. (1993) Leachman et al. (2005) Blauwens et al. (2006) Leachman (2008) Stewart et al. (2008) Combes (2010) Brogan et al. (2013) Total 36 23 18 16 14 8 4 23 17 11 8 6 8 3 5 Supply Chain Models References a) Modal Attributes b) Commodity Attributes c) Other Econometric Models Note: ITIC models are presented in Table 14. Table 16. Significant variables considered in freight mode choice literature.