Estimating the Value of Truck Travel Time Reliability (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.

4 This chapter reviews the literature to build an understanding of how roadway unreliability affects shippers and motor carriers, focusing on previous attempts to estimate the VOR. The chapter focuses on freight, but several studies on the reliability of passenger transportation are also reviewed. This review benefited greatly from the research on reliability done under TRB’s second Strategic Highway Research Program (SHRP 2), which included 25 research projects on a variety of topics relating to analyzing, modeling, and mitigating roadway unreliability. 2.1 Travel Time Unreliability 2.1.1 Definition and Causes Reliability has been defined many different ways in the literature. This study adopted the definition used by SHRP 2, which defined reliability as “the lack of variability in travel times” (Mahmassani et al. 2014, p. 10). The most important aspect of this definition is that it takes the perspective of system users and defines reliability relative to expectations. That is, reliability is defined as the travel time uncertainty faced by roadway users. In its application to the study of truck performance, this definition • Does not include uncertainty in the loading and unloading of cargo; • Does not include unreliability resulting from vehicle breakdowns, accidents, or driver issues; • Does not consider uncertainty in costs (variable tolls or volatile fuel prices might cause ship- ping uncertainty, however these costs are ignored because they do not affect travel time); • Assumes that hours-of-service regulations do not cause unreliability because trucking companies consider these regulations in their schedules; and • Assumes that truck drivers and schedulers have the ability to anticipate travel times for different routes at specific times of the day. This last point is important, because it stresses that defining reliability requires an understanding of how truck operators predict travel times and set delivery schedules. Some of the factors likely considered are time of day (peak versus off-peak conditions), day of the week, holidays, season (if winter weather is an issue), long-standing construction work zones, recurring sporting events, and any other known factor that is expected to have an impact on travel time. This assumption is reasonable, as most truck operators are sophisticated in their reliance on travel planning applica- tions and first-hand experience of routes to set delivery schedules that are realistic. Focusing on travel times was useful in limiting the scope of this study; however, future research could define reliability more broadly. Hirschman et al. (2016) found through interviews and surveys that shippers and motor carriers often face significant uncertainty in finding the right vehicles and drivers, loading and unloading cargo, and meeting hours-of-service regulations. C H A P T E R 2 Background

Background 5 Unreliability in travel time can be caused by demand factors that affect vehicle volumes or supply factors that affect system throughput (Mahmassani et al. 2014). These include • Demand factors: – Issues with alternative modes, – Special events, and – Random fluctuations in demand. • Supply factors: – Crashes, – Weather, – Work zones, – Malfunctioning of traffic control devices, and – Random individual behavior (limiting capacity and disrupting operations). Unreliability can also be caused by major exceptional events such as unexpected natural dis- asters or the failure of infrastructure. These “system killers” are typically not included in reli- ability studies because they have fundamentally different impacts and responses (Landergren et al. 2015). 2.1.2 Measurement The increasing availability of GPS data has created many opportunities for measuring road- way performance. In the United States, these data have been packaged by the Federal Highway Administration into the National Performance Management Research Data Set, which is avail- able to transportation agencies for planning analyses. Private companies have developed similar data products, many of which providing greater spatial and temporal detail than the NPMRDS. (More information on these data is provided in Section 5.2.1.) The most appropriate way to measure reliability depends on how the measurement will be used. The measure should address how reliability is being defined and should be compatible with its intended use (e.g., performance monitoring, demand modeling, benefit–cost analysis). The measures used in previous studies can be categorized as • Standard deviation of travel time, • Proportion of shipments delayed, • Interquartile range, • Average delay if delayed, and • Probability density functions of travel time. The standard deviation of travel time has been popular in transportation modeling studies because it is easy to calculate and typically leads to well-fitting models (de Jong and Bliemer 2015). However, it is sensitive to outliers and does not capture the tail of the distribution well. It is also hard to communicate to survey takers and nontechnical readers. In response to these limitations, several studies have modeled reliability as the proportion of shipments delayed (Bolis and Maggi 2003; Beuthe and Bouffioux 2008; Puckett and Rasciute 2010; Zamparini et al. 2011), because this measure is easier to conceptualize and communicate. However, the proportion of shipments delayed has the key weakness of ignoring the magnitude of delays. The average delay if delayed measure addresses this issue (Miao et al. 2011; Kawasaki et al. 2014); however, it is more common in the passenger transportation literature, where it was first formulated (Noland and Small 1995). The main drawback of this measure, which is often called the “scheduling approach,” is that it requires information about desired arrival times that is difficult to obtain or simulate from available data. Measures of the interquartile range have been used primarily in the United States to monitor the reliability of the roadway network. These include the Travel Time Index (95th percentile

6 Estimating the Value of Truck Travel Time Reliability travel time over 50th percentile travel time), the buffer index (95th percentile travel time over average travel time), buffer time (95th percentile travel time minus average travel time), and the planning time index (95th percentile travel time over free-flow travel time), among others (Cambridge Systematics, Inc., and Texas Transportation Institute 2005). The U.S. Department of Transportation recently required that all states calculate the truck travel time reliability index (TTTR), which is a function of the 95th percentile travel time and 50th percentile travel time, on Interstates for several periods of the day. This requirement solidified the use of interquartile range measures in freight planning nationwide. To date, this type of measure has not been used for modeling travel choice or in benefit–cost analyses. Interquartile range measures are not common outside the United States. Researchers and planners in Europe, New Zealand and Australia have a clear preference for the standard devia- tion measure (de Jong and Bliemer 2015). Figure 2-1 shows the results of a survey of freight reliability experts (mostly in academia) that asked them how reliability should be measured for benefit–cost analyses. The majority preferred the standard deviation measure for the reasons described above. In Sweden, this measure is being used by the government to track the perfor- mance of different freight modes (Kruger et al. 2013). Measuring reliability through probability density functions offers many advantages. This approach was implemented by List et al. (2014) to capture greater detail in identifying the causes of unreliability, particularly when the distributions are multimodal (have multiple peaks), as tends to happen on arterials. However, this approach has not been used since because of its lack of tractability. Regardless of the measure used, enough data are needed to ensure that estimates are statisti- cally stable, especially given that reliability is a phenomenon that is observed from the tail of the distribution. Figliozzi et al. (2011) provided guidance for how to aggregate roadway segments and analysis time periods to ensure that reliability estimates are based on enough data. 2.1.3 Impacts and Costs Travel time unreliability negatively affects shippers and motor carriers in many ways. Landergren et al. (2015) described how unreliability can raise costs, decrease service quality, and reduce competitiveness. The survey and interviews conducted by Hirschman et al. (2016) fr eq ue nc y 10 9 8 7 6 5 4 3 2 1 0 Source: de Jong and Bliemer (2015). Figure 2-1. Expert preference for reliability measurement in BCA.

Background 7 showed how unreliability can affect firms in many ways, leading them to adopt various short- and long-term strategies to cope with the uncertainty. A delayed shipment can generate the following costs: • Transportation costs: These costs, which result from trucks taking longer to make a delivery, include driver wages, fuel, maintenance, and so forth. Perhaps more significantly, late deliv- eries can also tie up drivers and vehicles, thereby preventing them from completing other deliveries. If contingencies are not in place, this may lead to cascading delays that cripple operations. Delays may also affect customer satisfaction and even lead to lost customers. • Shipper and receiver costs: Unreliability affects firms differently. For some, a delay may merely inconvenience staff. For others, a delay of a critical input can affect production and even lead to downstream impacts throughout the supply chain. Delays may also lead to missed trans- shipments at airports, seaports, or intermodal terminals, which can result in even greater delays if service is infrequent. Unreliability increases the costs of all shipments, not just those that are delayed. The slack that is built into the system to deal with this uncertainty represents additional costs in lost pro- ductivity. VOR estimates should consider both the costs of shipments being delayed and the costs of the mitigation strategies implemented by motor carriers and shippers to avoid these delays (sometimes unsuccessfully). The costs of unreliability ultimately affect business productivity by increasing input costs and decreasing the market area that can be served competitively, thereby depressing demand and reducing economies of scale (Weisbrod et al. 2001). Both of these are detrimental to busi- ness productivity. In the long run, areas with persistently high levels of unreliability could lose businesses and economic activity to areas with better traffic operations. 2.1.4 Nonlinearity Unreliability has long been understood in the literature to have nonlinear costs. Many of the costs are triggered only if delays are higher than a threshold. Figure 2-2 represents some of these relationships found in the literature. This diagram incorporates several concepts from Fowkes and Whiteing (2006), who explored a model of schedule delay with a slack period in Shipper/ Receiver Costs Expected Arrival Time Transportation Costs Early Arrival Delivery Window Contingencies Activated Late Delivery Travel Time Travel Time Veh. Operation & Driver Wages Other Shipments Affected Figure 2-2. Conceptual costs of late arrivals.

8 Estimating the Value of Truck Travel Time Reliability which no cost was accrued, a point at which delay costs increased rapidly, and a point at which contingencies had been activated and further delays were not costly. Early arrivals can also be problematic if, for example, there is not enough space to accommodate incoming trucks. 2.1.5 Contingencies and Responses The survey conducted by Hirschman et al. (2016) obtained valuable information about how shippers and motor carriers respond to unreliability (readers are encouraged to read this study for the complete results). The main findings were that shippers • Most frequently add a buffer time to a shipment’s departure schedule; • Seek improvements in route planning, including the development of alternative routes and use of better technology to track cargo in real time; • Increase the number of trucks and drivers on routes where the risks and costs of delay are high; • Rarely increase inventories to manage unreliability; • Include penalties for late deliveries (10 percent of respondents); • Require higher on-time performance (defined as 95 percent or better) for shipments that are expedited, involve intermodal transfers, have high value, or contain perishable products; • Consider delays as little as 1 hour to be unacceptable (14 percent of respondents); and • Consider a response to be necessary if a shipment does not arrive within 4 hours (74 percent of respondents). Motor carriers were also surveyed as part of this study. The main findings were that they • Almost always have delivery windows specified in their contracts; • Face penalties for late deliveries (50 percent of respondents); • Triggered operational responses for delays of 4 hours (74 percent of respondents); • Mentioned that delays of 2 hours caused operational issues (50 percent of respondents); • Responded to unreliability in ways similar to shippers: by improving route planning, adding new trucks, improving real-time tracking, and even outsourcing certain shipments. Many other studies have investigated how freight users respond to unreliability. Landergren et al. (2015) described how the key performance indicators used by businesses shape how they perceive and respond to unreliability, stressing that different industries operate differ- ently. Kuipers and Rozemeijer (2006) emphasized the time horizon of the responses (shown in Table 2-1), distinguishing between operational responses that can be implemented on a case- by-case basis and strategic responses that have a long-run effect on supply chains and distribu- tion centers (DCs). McKinnon et al. (2008) categorized responses on the basis of the magnitude of the unreliability and used sector data and interviews to assess the sensitivity of different indus- tries to unreliability. Even though most shippers are sophisticated and carefully consider unreliability in their operations, research in behavioral economics suggests that individuals and organizations are often biased in how they make decisions, especially when uncertainty and risks are involved (Kahneman and Tversky 1979). Some of these biases might include • Imprecise characterizations of risks, particularly with rare high-consequence events; • Risk aversion or risk proneness; and • Incomplete information or unsophistication in interpreting or responding to uncertainty. All these factors can lead to responses that do not reflect the actual costs of unreliability. In the passenger literature, several researchers have incorporated behavioral considerations in study- ing transportation choices. Polak (1987) considered the risk aversion and risk proneness of travelers, and other studies have focused on the importance of rare but intensely negative or

Background 9 positive events in defining preferences. Carrel et al. (2013) found that passengers reacted more negatively to unreliability that appeared to be the fault of transit agencies. A similar situation might be occurring in freight, where shippers behave differently according to the source of unreliability, particularly if it is perceived to be the fault of motor carriers. Previous work has not considered these behavioral factors in studying shipment unreliability. 2.1.6 Heterogeneity The most prevalent conclusion in the literature is that the costs of unreliability—the VOR— can vary widely depending on many factors. Following are some of the more important ones: • Commodity and industry. The commodities shipped and industries involved have been found to have a large impact on unreliability costs. Most VOR studies have considered aggre- gated commodities and industries such as finished products, perishable goods, dry bulk, and containerized cargo. • Shipment size. The larger the shipment, the greater the impact if the shipment is delayed. Some studies have controlled for this by estimating the VOR for different shipment sizes, while others have reported VOR per ton shipped. No study, however, has directly controlled for the value of the shipment. It is possible that shipment value correlates more strongly with unreliability costs than tonnage. • Type of transportation service. Most studies have considered the type of transporta- tion service either directly or indirectly. This includes whether a shipment is truckload or less than truckload and whether the truck is owned and operated by the shipper or con- tracted out to a motor carrier. Some studies have even considered third-party logistics (3PL) providers. Service type is important because it affects who sees what costs and their ability to respond. Company Type Operational Measures Tactical Measures Strategic Measures Road transport companies Earlier departure of trucks (and later return) Delivery at an earlier time Use of more trucks Use of backup trucks Making of better agreements with shippers on delivery times Broadening of planning horizon Use of night distribution Use of planning software Use of mobile telephone Consolidation of transport networks with other transport companies Strategic cooperation with other transport companies Use of consolidation centers Increase in the number of DCs Movement of DCs toward important customer locations Design of new and innovative logistics concepts Shippers Relaxation of transport planning Longer opening hours of facilities Assignment of longer time windows per truck More use of information communication technology control tools Adapting the level of stocks Narrowing of the planning horizon Allowance of night distribution Increase in the size of DCs to increase the level of flexibility in stock- keeping practices Increase in the number of DCs Design of new and innovative logistics concepts Source: Kuipers and Rozemeijer (2006). Table 2-1. Measures taken in response to travel time unreliability.

10 Estimating the Value of Truck Travel Time Reliability • Firm sophistication. Sophisticated firms are likely to respond more effectively to unreliability because they likely have better technology, greater familiarity with routes (and knowledge of traffic conditions), and more assets at their disposal to redeploy. Some studies have used firm size as a proxy for sophistication. • Supply chain structure. If a shipment is part of a pull or just-in-time supply chain, delays will be costlier because there is less inventory, if any, at the destination. Also, shipments heading to intermodal terminals might be more sensitive to delays because they have a fixed schedule to meet. Bolis and Maggi (2003) found that many of these logistical factors had a substantial impact on unreliability costs. Hirschman et al. (2016) found that shippers’ sensitivity to reli- ability depended on whether delivery schedules were static (regular supply points to regular delivery points) or dynamic (as-needed replenishments). • Shipment frequency. Delays can potentially be costlier if shipments are infrequent; however, this has not been found conclusively in previous studies. • Geography. A few studies have considered the effect of geography on unreliability costs. Results show that VOR does not vary significantly within countries in Europe (where the studies were conducted); however, this might not hold in the United States because of its greater size and economic diversity. • Intercity versus urban. Intercity shipments are very different from urban shipments, from the types of vehicles used to the sources of unreliability involved. • Shipment distance. Another way of considering the distinction between intercity and urban shipments is to consider shipment distances. Longer shipments could potentially have higher unreliability costs because there are fewer alternatives available than for urban shipments. • Contractual agreements. Some researchers have considered different types of agreements between carriers and shippers, including on-time performance, shipment scheduling, and size of delivery windows. List et al. (2014) conducted interviews of shippers to ascertain their sensitivity to reliability and found considerable heterogeneity in their responses. The study divided respondents into eight categories on the basis of their level of flexibility, level of operational adaptability, and costs of unreliability. Their study found that reliability was not an important factor for a subset of respondents, while for others, it was paramount to their operations (particularly involving time-sensitive, high-value goods). 2.2 Estimating Freight VOR Despite reliability being paramount in freight transportation, there is no consensus in the literature about the VOR for this type of transportation. Since 2000, there have been 13 studies that estimated the VOR of freight transportation by truck; however, they have taken place in 11 different countries, defined reliability in three different ways, and adopted many different estimation assumptions. In fact, many of these studies are obsolete, given the contemporary understanding of surveying and modeling. Nonetheless, each of them was useful in advancing some aspect of the literature, and, therefore, this project benefited in one way or another from all of them. This section focuses on describing the major lessons learned and delineating the strengths and weaknesses of the different approaches used. 2.2.1 Different Approaches The idea that the time spent in an activity can be monetized was first introduced by Becker (1965). Since then, researchers have used many approaches to estimate the value of time (VOT) for a variety of activities. In transportation, this type of analysis has been extended to estimate the value of other travel attributes, such as comfort, waiting times, and reliability. Estimating

Background 11 VOT and VOR is challenging because travel time and reliability are goods that are not traded in markets and therefore are not priced by the supply–demand equilibrium. There are essentially two ways that nonmarket goods such as these can be valued: They can either be (1) modeled through the optimization of approximate factor costs or (2) inferred from the implied trade-offs of behavior (see Figure 2-3). Approach 1 develops a ground-level understanding of how unreliability affects the produc- tion of goods and represents this understanding in an optimization model. Previous studies have developed this understanding by interviewing freight users, conducting case studies, and analyzing aggregate data. Several previous studies have taken this approach, including Noland and Small (1995), who formulated a model that described how the cost of being early or late affects reliability valuation and departure times. Kittelson & Associates (2013) modeled reliability valuation as an investment problem, in which travelers selected insurance policies to protect them against the risk of shipments being delayed. Sadabadi et al. (2015) continued this line of research by relaxing several statistical assumptions and providing a simpler way for calculating optimal insurance premiums. Hirschman et al. (2016) also followed this approach by using insights from surveys and interviews to develop a tool that approximates the importance of reliability on the basis of buffer times and their costs. Although Approach 1 provides a useful approximation, in practice, it is difficult to obtain accurate estimates with this approach because the information required is typically confidential and rarely shared with public agencies or researchers. Moreover, analysts are likely to have dif- ficulties reconstructing the factors and constraints that influence logistic decisions, particularly for different types of firms, supply chains, and all of the other factors that influence reliability valuation (see Section 2.1.6). Because of these difficulties, most researchers have resorted to Approach 2. This approach observes the behavior of decision-makers and models how they make trade-offs. As shown in Figure 2-3, behavior can be observed from aggregate data or disaggregate data. Aggregate data consists of observations at a regional or sectoral level, while disaggregate data consists of obser- vations for each decision-maker, and even each decision. In theory, it is possible to estimate the VOR from aggregate data; however, this has not been done in the literature because there is no good method of controlling for all other factors that influence transportation decisions. Virtually all the previous studies have relied on disaggregate data to estimate the VOR. Most of these data have been collected through stated-preference surveys, although a couple of older studies have used revealed-preference data. The next section describes these efforts. Estimation of VOR Optimization of Approximate Factor Costs Infer from Behavior Aggregate Data Disaggregate Data Revealed Preference Data Stated Preference Survey Data Aggregate Data Interviews Case Studies Figure 2-3. Research approaches to estimating the VOR.

12 Estimating the Value of Truck Travel Time Reliability 2.2.2 Stated-Preference Studies Interest in estimating the VOR of trucking has picked up in the past couple of decades. Reli- ability had always been known to be important for freight, but only as surveying capabilities improved and models became more flexible has estimating VOR become possible. Table 2-2 lists the main attempts found in the literature. Although this list focuses on freight, the pas- senger transportation literature was also reviewed for ideas and best practices (see Noland and Polak 2002 for a review). For a comprehensive overview of the freight VOR literature, see Shams et al. (2017). A review of the literature on estimating the VOT in freight, which is similar in several ways to estimating the VOR, can be found in Feo-Valero et al. (2011) and de Jong (2014). The first researcher to attempt to model how reliability affects the decisions of shippers was Winston (1981). That study used revealed-preference data from before deregulation to model freight transportation demand as a function of various factors, including reliability. After dereg- ulation in the late 1970s, this type of disaggregated shipper data became very difficult to obtain because of confidentiality, and revealed-preference studies of this kind became virtually impos- sible to conduct. Instead, researchers have resorted to stated-preference surveys that ask participants to choose from hypothetical alternatives. The alternatives are carefully designed to expose how respon- dents make trade-offs between different attributes. For estimating VOR, these had to include at least costs and reliability. Small et al. (1999) used one of these surveys to estimate the VOR for both passengers and freight, although the sample collected for freight was insufficient to yield statistically significant results. The study also noted difficulties in conducting the survey by phone, which has not been attempted by other studies since. The literature after Small et al. (1999) branched out into (1) estimating the willingness of trucks to pay tolls to achieve faster and more reliable travel times and (2) estimating the impor- tance that shippers and truck operators place on reliability. The toll studies took place in the United States (Kawamura 1999; Miao et al. 2011), and had the objective of assessing different levels of tolling. Kawamura (1999) looked at tolls on SR-91 in California, and Miao et al. (2011) interviewed truck drivers in Texas and Wisconsin about different tolling scenarios. The VOR estimates in these studies are not relevant because they only considered trade-offs in routing and assumed shipment characteristics were fixed. The second branch of the literature has focused on quantifying the value of reliability rela- tive to other attributes, such as travel time and cost. Even though several studies have been conducted in this area since 2000, their results vary widely because the methods, assumptions, and contexts have been very different (Shams et al. 2017). It is difficult to compare the results of these studies, which is why this review focuses on methods and assumptions instead of on specific estimates. This review skips the fundamentals of stated-preference surveys and model- ing; however, Appendix A provides some best practices in the estimation of VOR. 2.2.2.1 Key Innovations In stated-preference surveys, the hypothetical choices presented should be familiar to the respondents. The quality of the results depends on whether respondents can answer from exper- tise and experience, which is more likely when the choices resemble their day-to-day opera- tions. Some studies have sought to accomplish this by pivoting the choice questions around the characteristics of a shipment previously arranged by the respondent (Wigan et al. 2000; Maier et al. 2002; Halse et al. 2010; de Jong et al. 2014; Jin and Shams 2016). Other studies have used adaptive stated-preference surveys in which the choice questions were based on the responses to previous choice questions (Bolis and Maggi 2003; Fowkes and Whiteing 2006). However, these

Year Author(s) Place Reliability Measure Modes Heterogeneity Survey Sample Size 1981 Winston United States Travel time coefficient variation Road and rail Commodities and modes RP data ~2,500 shipments 1999 Small et al. United States Schedule delay Road 4 commodity groups SP survey 20 carriers 1999 Kawamura United States: California Schedule delay Road Business type, shipment weight SP survey—pay toll to avoid congestion 70 motor carriers 2000 Wigan et al. Australia Proportion of shipments late Road Intercity, urban, urban multiple stops Contextual SP survey of shippers—within mode 43 shippers 2000 Kurri et al. Finland Schedule delay Road and rail Commodity type SP survey shippers—within modes 236 road shipments, 162 rail shipments 2002 Maier et al. Austria Proportion of shipments on time during year Road and rail Geography, commodity group, transport decision-maker Contextual SP survey of shippers 148 experiments with 74 shippers 2003 Bolis and Maggi Switzerland and Italy Proportion of shipments late Road Shipment weight, geography, firm JIT, distance, intermodality Adaptive SP survey 22 shippers 2006 Fowkes and Whiteing Great Britain Schedule delay Road and rail 9 commodities, shipment type, time of day Adaptive SP survey 49 shippers 2008 Beuthe and Bouffioux Belgium Proportion of shipments late Road, rail, water and multimodal Shipping distance, commodity value, type and weight SP survey 113 shippers 2010 Halse et al. Norway Travel time standard deviation and schedule delay Road and rail Shipper versus carrier SP survey 640 shippers, 117 carriers 2010 Puckett and Rasciute Australia Probability of being late Truck Shipper, carrier SP survey 129 shippers, 136 motor carriers 2011 Miao et al. United States: Texas and Wisconsin Schedule delay Road Geography, transport provider SP survey of drivers and fleet dispatchers—pay toll to avoid congestion 111 truck drivers 2011 Zamparini et al. Tanzania Proportion of shipments late Road Transport provider, value density of goods SP survey—rank and weight attributes 24 shippers 2014 Kawasaki et al. Southeast Asia Schedule delay Road None SP survey shippers 48 shippers 2014 de Jong et al. Netherlands Travel time standard deviation Road, rail, air, and water Containerization, truck size, mode SP survey of shippers and carriers 249 road shippers, 166 motor carriers, 397 other modes 2016 Jin and Shams United States: Florida Travel time standard deviation, in hours and ton-hours Road Carrier type, 4 commodity groups, perishability SP survey 35 shippers, 108 carriers, 7 3PL providers Note: RP = revealed preference; SP = stated preference; JIT = just-in-time. Table 2-2. Valuation of freight reliability literature.

14 Estimating the Value of Truck Travel Time Reliability adaptive surveys have undesirable statistical properties that make it difficult to estimate the VOR (discussed further in Appendix A, Section A.2). All stated-preference studies on truck VOR have described hypothetical alternatives at least in terms of costs, time, and reliability. A few have considered other attributes such as mode, flex- ibility, frequency, and in-transit damage (see Table A-3 in Appendix A). Bolis and Maggi (2003) were innovative in considering logistic variables such as type of supply chain and intended receiver (intermediate or final consumption), finding that these variables have a large effect on VOR. Fowkes and Whiteing (2006) were innovative in using a more realistic nonlinear repre- sentation of delay, in which costs are not incurred for small delays during a slack period or large delays that exceed a threshold. Kurri et al. (2000) calculated the VOT and VOR per ton shipped, an approach later repeated by Jin and Shams (2016). Most studies have asked respondents to choose from hypothetical alternatives; however, a few have tried other approaches. Beuthe and Bouffioux (2008) asked respondents to rank 25 different transport alternatives, where each alternative was defined by six attributes (frequency, time, reli- ability, flexibility, damage, and cost). Zamparini et al. (2011) asked respondents to rank and weight attributes directly. Adaptive stated-preference surveys have asked respondents to rate alternatives from 0 to 100 to assess how strongly respondents preferred their selection (Bolis and Maggi 2003; Fowkes and Whiteing 2006). At the time, these studies argued that asking respondents to rank or rate alternatives provided additional information that could be useful in modeling. However, in the past decade a consensus has formed that these types of surveys are inferior because responses are less stable and more open to interpretation (Johnston et al. 2017). Stated-preference surveys with binary choices produce results that are more consistent and easier to model. The representativeness of VOR estimates has not been a major concern in the literature, even though it is critical for the applicability of these results in planning analyses. The focus has been on achieving enough responses to generate statistically significant estimates rather than on how well the responses represent freight activity in a region. The only exception found was Halse et al. (2010), who attempted to improve representativeness by surveying firms at random from a representative sample of shippers and motor carriers (by number of firms and number of ship- ments); however response rates were ultimately higher for certain subgroups, and no corrections were implemented in the model. Halse et al. (2010) compared the characteristics of firms that responded with the characteristics of those that did not and found that larger firms were over- represented in the sample. Compensating for these biases is critical to developing VOR estimates that can be used in planning analyses. One approach, which has not been adopted in the literature, involves re-weighting the sample according to population shares (Johnston et al. 2017). 2.2.2.2 Major Recent Studies The largest study in the literature was conducted by de Jong et al. (2014). This study received 800 responses from shippers and motor carriers in the Netherlands across multiple freight modes (road, rail, air, inland waterways, and sea). However, despite this large sample, the study was not able to estimate some of the effects of interest and concluded that a larger sample was needed, especially to capture heterogeneity between respondents. The only recent study conducted in the United States was that of Jin and Shams (2016). This study achieved a moderate sample of shippers and motor carriers in Florida (responses of 97 respondents used in modeling), and estimated truck VOR for different types of cargo and respondents. It also used several of the innovations introduced previously, such as presenting VOR on a tonnage basis, pivoting the choice questions as a function of typical shipments, con- sidering logistical variables, and simplifying the survey to reduce cognitive burden. All recent major studies in this area have relied on stated-preference surveys.

Background 15 2.2.2.3 Limitations Stated-preference surveys can be affected by several well-known biases (see Ortúzar and Willumsen 2001 for a review). Some respondents could be influenced subconsciously by the motivation of the study to select responses that place a greater importance on reliability, or they could even purposefully game their choices because they feel they might benefit if the study finds higher VORs. Self-selection could also bias the results if the people that think reliability is important have a higher propensity to participate, or if employees of more-sophisticated firms are more likely to take the survey. Weisbrod et al. (2001) concluded that stated-preference surveys can potentially under- report the true costs of truck unreliability to the economy. Companies adjust their businesses in response to unreliability, which might include purchasing a more expensive warehouse closer to clients or having a larger fleet of trucks in anticipation of some being stuck in con- gestion. However, when taking a stated-preference survey, the employees of these companies were unlikely to think about these medium- or long-run costs, even if they were prompted to do so in the survey. Therefore, the valuation arising from this survey might represent the marginal VOR, not the total economic value.