Methodology for Estimating the Value of Travel Time Reliability for Truck Freight System Users (2016)

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.

1 S u m m a r y Study Goals and Approach The goal of NCHRP Project 08-99 is to estimate the value of travel time reliability for truck freight system users and to develop and demonstrate a methodology for it to be applied. In 2013, the Transportation Research Board (TRB) initiated the process leading to NCHRP 08-99, when it issued a broad research statement “Truck Freight Benefit Methodology.” That state- ment described a two-step research process, comprised of: “ . . . (an) initial, qualitative data collection effort to inform the design of a survey of truck dependent industry sectors and ensure the survey captures the relevant information appropriately. This exploratory phase will be followed by a larger scale survey that will collect direct schedule delay and cost information, as well as responses to stated preference questions directly from the trucking community.” This study fulfills the first exploratory phase in the research sequence—e.g., the qualitative data collection effort and development of a “first order” valuation model. While traditional stated preference analysis can provide a rigorous analytic framework, the researchers believe—along with TRB—that it is first necessary to obtain as nuanced as possible an understanding of the complexity of supply chains in trucking before designing a truly well-constructed stated preference analysis. It was felt that a necessary condition for further statistical study would be to gain more understanding of the economic trade- offs that shippers and truckers make when faced with uncertainty, given the complexity of truck logistics arising from fragmentation of decision making, and the length of the value chain. Accordingly, the approach was to conduct a program of survey research within the indus- try (Chapter 3), to begin to obtain a more complete picture than now exists of how different trucking industry “actors” are affected by travel time uncertainty, how they respond to and mitigate its effects when then can, and what the impacts of uncertainty are to their supply chain costs. This understanding was used to develop a valuation model (Chapter 4). Following that, the model is demonstrated through two case study applications (Chapter 5). A proposed research plan for the follow-up study, which would build off the information and tools developed in NCHRP 08-99 follows (Chapter 8). More specifically, the basic elements of the research and modeling were comprised of the following major activities. 1. An in depth program of survey research was undertaken, including online surveys and industry interviews, to improve understanding of trucker and shipper behavior in response to travel time uncertainty. The survey research plan included an initial round of electronic surveys, followed by selected interviews of truck industry stakeholders. A Methodology for Estimating the Value of Travel Time Reliability for Truck Freight System Users

2follow-up electronic survey was conducted to narrow in on shipper and trucker behavior with respect to buffering truck delivery schedules. 2. An economic valuation framework was determined, and an Excel-based modeling tool, the Truck Freight Reliability Valuation Model, was developed. The modeling framework reflects what was learned in the survey and interview research. The model, as described in detail in Chapter 4, estimates the additional costs per trip for various trip parameters, given varying levels of travel time unreliability. The Buffer Index is used as the primary metric of uncertainty in the model. 3. Use of the model is demonstrated for two case studies. The case studies involved two real world planning situations—a major truck freight corridor in Georgia, and a U.S.–Mexico truck border crossing. The demonstrations applied corridor level data from existing studies to the per trip modeling framework to estimate the additional annual economic costs of unreliability, given available data on Buffer Index values prevailing at chokepoints, AADT truck volumes, and median trip times and speeds. 4. Finally, extending the results of this study, a more complete survey and modeling protocol is proposed for a next phase of economic valuation research beyond the completion of NCHRP 08-99. Survey and Interview Research The survey and interview research represents the central data gathering phase of the study effort. The purpose was to help provide a more complete “behavioral” understanding of how truck service providers and shippers view unreliability, what thresholds of unreliability are significant and require actions to mitigate, and how supply chain participants respond in terms of strategies and behaviors to mitigate unreliability. This phase comprised three different efforts: (1) an initial online survey of shippers and truck transportation service providers; (2) in depth interviews of shippers and transporta- tion service providers (interviewees are not specifically drawn from the online survey roster); and (3) a shorter but more focused follow-up online survey, narrowing in on the most common response to unreliable conditions—adding buffer time to truck schedules. Add- ing buffer time to truck schedules to minimize the probability of late delivery/arrival at the truck destination provided the conceptual basis for the valuation methodology, as described in Chapter 4. Online Surveys The survey instruments were developed by the research team and reviewed by the NCHRP Project 08-99 Panel, and the execution, distribution, and collection of the survey, and its results, were conducted by Tompkins International, which assembles and maintains the Supply Chain Consortium participant data listing. The Consortium is a cooperative of over 350 participating retail, manufacturing, and wholesale/distribution companies as members. In addition, Tompkins International maintains a comprehensive database of over 10,000 freight service providers, which provides much wider access for conducting surveys. The Consortium uses online analysis tools, topic forums, and peer networking information and makes it available for supply chain executives and practitioners. The survey was delivered in online electronic format. In total the surveys achieved about a 3% response rate, with the shippers’ survey receiving 93 responses and the transportation provider survey receiving 76. More specifically, the final responses rates were 2.4% for the shippers’ survey and 2.1% for the transportation providers’ survey. Tompkins regards these as average response rates for surveys of this type. Because the Tompkins data base of transportation providers and shippers is comprised of several thousand potential participants, the approximately 3% total response rate yielded a total of

3 169 responses. While still a small response rate, it is large enough to begin to yield distribu- tions that converge toward large sample size distributions. The number of responses however, limits the ability to make any generalizable statements about individual submarkets within the 169 surveys received. While the overall response rate was nearly 3% for the survey as measured by the number of responses compared to all people asked to respond to the survey, the effective response rate was higher. A better indicator of the effective response rate to the survey is found by comparing the number of responses versus the number companies in the survey, since multiple people from each company are asked to respond to the survey. The response rate as measured by companies participating in the survey is just under 10% as a little more than three people per company were invited to respond to the survey. Findings from the initial round of surveys suggested that buffer time is one key to under- standing and measuring the economic value of unreliability in trucking. However, it was felt that there was not a sufficiently clear picture of how and why trucking and supply chain service providers buffer schedules. This led to a fairly quick and focused second round of follow-up online surveys to explore this topic further. Major Survey Findings • Route planning is the most frequently employed strategy, both for shippers and truckers. Route planning—including adding buffer time, developing routing alternatives, and real time truck tracking—may be regarded as flexible responses to uncertainty in truck on-time performance. Route planning may be utilized where the costs and risks of unexpected delay are manageable. Other strategies could be significantly more costly especially in the short run. • As part of their route planning, the strategy most commonly employed by shippers and logistics providers to ensure on-time performance with a high degree of probability is to add buffer time to the truck departure schedule. While a range of other strategies, such as driver teaming, outsourcing, mode shifting, and adding more trucks and drivers were identified in our surveys and interviews, the basic first line of defense is to add buffer time. • Use of driver teams, or adding trucks to routes, may be viewed as incrementally more costly strategies and would be employed where the risks and the costs of delay and missed deliveries are high and the ability to manage those risks through route planning and truck tracking are more limited—for example, where detour routes are unavailable or entail very significant increases in travel times and costs. • Increased buffer inventory is infrequently cited as a strategy to respond to increased unreliability. This result, which runs counter to commonly held conceptions about supply chain practices, has important implication for the analysis of reliability in general, as the standard approach to the valuation of unreliability typically involves estimating the carrying cost of increased inventory. However, the results suggest that this approach may be at least partially mistaken, as shippers do not appear to want to hold more inventory as a counter measure against increased unreliability in truck transport. • When asked about shipment characteristics that cause shippers to shoot for very high on-time performance—95% or better—expedited shipments, cargo transfers to other modes, high value shipments, and perishable products were all identified as major factors in buffering aggressively against late delivery. Interviews A series of one-on-one interviews were conducted with trucking operators/owners and shippers to further articulate the causes of, impacts of, and responses to unreliable truck

4travel times. These interviews both validate and further articulate the information obtained from the online surveys. The companies interviewed represented a cross section of shipper and commodity types, as well as the range of trucking operations. The organizations interviewed use over-the-road long distance trucking, both within the United States and across the border into Canada and Mexico, as well as operations that are limited to multi-state regions or a specific region; truckload and less-than-truckload (LTL) movements; local pick-up and delivery (P&D); and drayage operations involving ports, airports, and rail yards. The trucking companies and shippers operate in a mix of settings, including urban, suburban, and rural. Truck movements may be from one origin to one destination or involve multiple stops. The trucking operations include companies with their own drivers and companies that use independent owner operators. The shippers interviewed include companies that have their own trucking fleets, contract with a specific trucking organization to operate a dedicated fleet on their behalf, work with a core group of selected carriers, and contract with multiple carriers as needed. All of the companies interviewed use more than one truck, with fleet sizes varying up to thousands of power units and trailers. No individual owner operators were interviewed. The companies interviewed include: Transportation Providers: • Best Transportation • Con-Way • Halls Fast Freight • New England Motor Freight • NFI • Schneider Shippers: • Fiat Chrysler • H-E-B • Macy’s • Miller Coors • Whirlpool Major Interview Findings • A variety of practices are employed to manage reliability risks, ranging from added resources and padded schedules, to charges assessed when assets are tied up for periods longer than an established norm. Notably, delay costs at the origin and destination have been monetized into a penalty system primarily affecting the vendors. In general, the practices are associated with keeping performance by the shipper or carrier within an acceptable range and do not address consequential damages. For example, carrier delay charges seek to recover the cost of providing the driver and equipment and not the opportunity cost of missed loads, with the implication that the carrier is able to mitigate opportunity costs by adding drivers and equipment, and expects to be reimbursed. To the extent this is inadequate, the carrier stops doing business with the vendor—just as the vendor will stop doing business with a carrier who cannot find ways (such as adding assets to absorb longer, buffered cycle times) to meet reliability standards. • Imposing strict delivery windows appears to be a common if not the most common method of trying to manage and mitigate uncertainty, and time windows are often built into delivery contracts. Where deliveries are not made within that window, penalties (similar to port demurrage charges) are incurred. A two-hour delivery window is reported

5 as fairly standard. This is an important data finding to incorporate into reliability cost calculations. • The delivery window constraints can be imposed on the inbound truck trip or can and often are assessed against the receiver; the latter is typically a maximum time to load or unload on the dock. • The delivery appointments can vary based on the firm and the commodity. Some appoint- ments can be “Must Arrive by Dates” (MABD), noting a specific date. In other cases, delivery appointment windows may be specified to a one- or two-hour time slot on a specific day. Retail companies and production lines tended to have the more specific P&D appointment windows. Shipments that are less time sensitive may have a MABD appointment. • Interviewees noted that unloading time delays can vary based on the commodity and type of customer. Excessive unloading times tie up trailers and drivers that can be used for other movements. Some interviewees noted that carriers may have “unloading allowance agree- ments” to address the amount of time that drivers and equipment is held at a drop-off location. • Trucking companies have increased buffer “in-transit” times for customers in congested, unpredictable travel time corridors so as to avoid such penalties. Buffer costs increase the transportation charges as the trucking companies need to be compensated for the time that their equipment and drivers are deployed for a customer. When the compensation is consistently below the cost of providing the service, trucking companies have stopped serving the customer, vendor, and the corridor or area. • At pick-up locations, some trucking companies have sought increased secured truck parking space at the vendor so that their drivers can come off-peak to pick up fully loaded trailers when notified of their availability electronically. This solution requires that trailers be left at the vendor location, which can represent an increased capital cost and reduce availability of equipment. However, this solution reduces unpredictability at the pick-up point. A fixed cost solution such as this will probably require enough volume at a given pick-up point to achieve economies of scale needed to make this option economically attractive. • Late delivery or pick up penalties are an indicator of economic costs, but it should be care- fully noted that such charges are not market based in a strict sense, and may not reflect very well the “underlying” economic or financial costs of late delivery caused by unreliable conditions. Charges can be set above or below economic values where market distor- tions exist. And in fact, supply chains are not generally fully or even close to competitive but rather are often dominated by large shippers or consignees (beneficial owners of the freight, or BFOs) who may dominate in a particular supply chain. Truck Freight Reliability Valuation Model Based on findings from the study’s surveys and interviews, supplemented by other trucking and freight reliability research, a picture emerges of shipper and trucking industry behavior in response to variable levels of truck travel time uncertainty. The behavioral responses, choice sets, and actual imposed costs identified in these findings combine to reveal substan- tial insight into the economic value/cost of uncertainty to truck freight users and service providers. The data and findings have led to developing a valuation methodology—as well as an accompanying Excel based spreadsheet analysis model, the Truck Freight Reliability Valuation Model. Through a cost simulation of shipper behavior given particular travel con- ditions, trip characteristics, and supply chain parameters, the model estimates per (loaded) truck trip economic costs of travel time uncertainty for different levels of trip time vari- ability within a given truck freight corridor. These per truck trip costs can then be applied to disaggregated or total truck flow volumes in a freight corridor to derive the daily and annual additional truck freight economic costs imposed due to travel time uncertainty. The

6total economic value of improvements in travel time variability from highway investments or operational improvements may be applied in benefit-cost analysis; application of the per trip model to actual truck corridors to derive per trip and annual economic cost penalties is demonstrated in the case studies in Chapter 5. The method of deriving economic value has elements in common with revealed prefer- ence analysis, albeit not in rigorous statistical form. It takes the previously reported survey responses to questions about what shippers and service providers report doing when faced with travel time uncertainty—what strategies are employed as mitigation (effectively, as “insurance”) against late arrivals, and to what level of uncertainty shippers are willing to use those strategies to virtually guarantee on-time delivery—to infer the perceived costs of uncertainty. The estimates also reflect information obtained from the shipper and trucker interviews. The costs of such mitigation strategies form the basis of economic value in the model. In particular, the model uses this approach to estimate the economic costs of travel time variability in a given truck corridor, where variability is measured by the Buffer Index. Basic Model Logic The definition of travel time reliability used in the model refers to on-time performance of trucks to their delivery destination. It is measured based on the probability distribution of travel times and is bounded by percentile yardsticks. It is a relative measure that compares the spread between the average time and a given percentile. For example, when 95% of all trips arrive up to one hour later than scheduled delivery times, unreliability would be much greater than when 95% of all trips arrive up to 15 minutes late. In foregoing discussions, the term “uncertainty” is used interchangeably with unreliability, to improve clarity of meaning. For a given level of travel uncertainty, as measured by the ratio of the 95th percentile to the 50th percentile (the median) of the trip time distribution (i.e., the Buffer Time Index), the model estimates, first, the expected cost of a hypothetical case in which shippers are assumed to accept the risk of late delivery and absorb (or pass on) the additional costs of delay. This value is then compared to the costs incurred when shippers build in buffer time and effec- tively limit the chances of late deliveries to the most infrequent outlier cases. The use of applied buffer time is assumed in the model to be the weighted average of buffer time values added to schedules by shippers and truckers, as obtained from the surveys. Surveys revealed that buffering behavior under conditions of uncertainty cluster around the 95th percentile travel time marker for on-time delivery for time sensitive cargo. While Buffer Time Index values are often defined with respect to an average, or mean, travel time, the median value is used in the model to derive probability values, since it allows for a more accurate assessment of skewed distributions, which characterize travel time distributions. The first value, the median value of delay given travel time variability without buffering or other forms of mitigation, represents a “comparator” against which the costs of shippers’ stated buffer time behavior is measured. This second value, the cost incurred after adding buffer time to the scheduled (median) travel time, represents a form of revealed economic value that shippers and truckers place on reliability. Where the costs of buffering (including the direct cost of transportation plus the occasional higher expected costs when the delivery window is still exceeded) are greater than the expected cost without buffering, a residual economic value is implied. That value may be viewed as an additional unreliability cost pre- mium that is not accounted for by direct transport costs, dock penalties, and assumed cargo related “late to point of delivery” costs. Costs Captured by the Model The valuation methodology identifies relevant costs to shippers and truckers from delivery uncertainty. As noted, not all costs can be fully captured in the model, but the methodology

7 described here tries to capture the major truck related delay costs within the supply chain, as follows: • Directly Variable Truck Transportation Cost. This is the variable cost per hour of operating and maintaining the truck. The American Transportation Research Institute (ATRI) publishes updated estimates of the operational costs of trucking each year; the last year available at the time of the study was 2014. Values need to be updated annually as the model is applied. • On-Dock Penalties. The surveys showed that in close to 70% of cases, customers and in some cases truck service providers stipulate penalties for late delivery. The interviews further highlighted this practice in trucking, and indicated on-dock penalties ranging from a few hundred dollars for delivering outside the stipulated time window to $500 per truckload. These penalties are sometimes assessed against the truck service providers by consignees or shippers for arriving late, or they may be assessed by trucking companies in strong market positions when trucks are not loaded within a given time frame at the distribution facility. Other research tends to confirm that late delivery or “detention” penalties are widespread in the trucking industry, and that the typical “free time” before charges are imposed is two hours, with charges above free time ranging from $50 to $90 per hour • Cargo Related Supply Chain Cost. This bundle of costs is most similar to the “Inven- tory Cost” category often cited in freight cost. At the margin, these costs cover a number of specific supply chain attributes, such as cost of capital incurred from delays in get- ting intermediate inputs to production facilities, opportunity cost of delayed final sales, administration and management, insurance, product spoilage, reduced production effi- ciencies, etc. Each of these costs are expressed as “expected values,” based on the trip time distribu- tions implied by Buffer Index values or derived directly from disaggregate travel time data. Expected values reflect the central tendency of the assumed travel time distribution above the median trip time value for a log normal distribution. In effect, it is the most likely delay above the median travel time for the distribution. This form is specified based on assumed right skewed travel time frequency distributions—a reasonable assumption and useful methodological technique which focuses on late rather than early arrivals. While arriving early is desirable it entails smaller positive benefits, as early arrival times will be much closer to the median value (they will generally not be that much less than the median travel time given a rightward skewed distribution toward late arrivals). Moreover, unloading crews may not be available to work until the appointed arrival time is hit. Log normal distributions of this form have been used in other truck reliability analyses. Applicability to Truck Freight Corridors The per truck trip costs can be applied to disaggregated or total truck flow volumes in a freight corridor to derive the daily and annual additional truck freight economic costs imposed due to travel time uncertainty. The total economic value of improvements in travel time variability from highway investments or operational improvements may be applied in benefit cost analysis; application of the per trip model to actual truck corridors to derive daily and annual economic cost penalties is demonstrated in the case studies in Chapter 5. The model provides a useful framework for estimating the economic value of truck freight reliability. However, this model is a simplified first step in a two stage research program that would result in more robust and fine grained valuation factors; there are significant limita- tions to the model presently. However, the model does provide a reasonable quasi-revealed preference approach to valuation and leads to insights.

8Case Study Demonstrations Use of the Truck Freight Reliability Valuation Model is demonstrated through two case studies: the Georgia I-75/I-16 Corridor and Texas El Paso Ports of Entry. These two were selected because they provide a representative cross section of trucking situations, including port access, metropolitan area distribution, and cross border trade. In addition, sufficient data are readily available from online publications and through direct contacts with study participants to allow for the model demonstrations to be conducted within the study time frame. The case studies illustrate the use of the model across varied supply chain conditions. The case studies demonstrate that the model can be applied to aggregate corridor level data and to disaggregate data at the segment level. Results of both case studies indicate signifi- cant additional economic costs when travel times are highly unreliable. In the Georgia corridor case, additional annual costs due to variability in trip times amounted to about $48 million per year, given current levels of truck traffic and highway congestion. The El Paso case study focuses very specifically on northbound trucks moving across one of the crossings—the Bridge of the Americas. That analysis estimated additional annual costs of about $8.5 million. In this case, costs were significantly lower than might have been expected because the share of trucks crossing empty—i.e., without a load—comprised about half of the northbound trips. Costs were also lower because of downward adjustments for Mexican truck operating costs. Proposed Phase II Research Plan Consistent with the original intent of the truck freight reliability valuation research state- ment, a stated preference analysis to develop more fine grained and statistically grounded estimates of the value of truck freight reliability is proposed. The objective will be to conduct follow-up stated preference survey research, applied to a significantly larger and more stratified sample than obtained in this study, to obtain statisti- cally valid functional relationships between the perceived level of trip time variability on the one hand, and the costs of actions taken to mitigate trip time uncertainty, such as buffering or other strategies identified in this study. Key Research Questions The research and empirical analysis and modeling should focus on the following key questions: • What are the costs of employing various mitigation strategies? The Truck Freight Reli- ability Valuation Model produced for NCHRP Project 08-99 provides a good estimation framework and tool for those costs and how those costs vary as a function of varying levels of truck trip time uncertainty, as measured by the trip time distributions. The log normal specification might be reviewed if time permits, to determine if other functional forms could provide better fits to trip time data. • What is the “rate” of cost tradeoffs made by shippers in employing such strategies for a given truck trip time distribution? This question relates to the functional form of the relationship between the direct cost given trip time variability and the costs of mitigation. • NCHRP Project 08-99 surveys found considerable variability in on-time performance tar- gets, from time certain, to one or two hours, to twelve or more hours. Unfortunately, because of the small response rate, the survey results did not allow for the identification of significant differences in these on-time goals by types of shipper, supply chain characteristics, or com- modities. That information is clearly required, and should be explored in depth in the next

9 phase of research. What factors explain these differences? Shipper characteristics/types of shipper, supply chain characteristics, or commodities are all possibilities. • What variables explain the specific mitigation strategies employed by shippers and truck service providers, such as commodity type, trip purpose (i.e., line haul, connection to port or intermodal facilities), and how do these strategies and their costs vary at different levels of trip time uncertainty? As noted above, additional analysis is needed to explore and explain cases where more aggressive and—in most cases—more costly mitigating strate- gies are employed; our surveys and interviews uncovered a range of strategies in addition to adding buffer time, such as driver teaming, real-time trip tracking, geo-fencing, price surcharges for very congested service areas, and in the long run, relocation of distribution or production centers. Recommended Scope of Work With these questions in mind, a stated preference analysis, entailing surveys and calibration of econometric models, is proposed, comprised of the following steps/tasks: Phase I Task 1: Develop modeling hypotheses: A first step will be to develop hypotheses of shipper and truck service provider behavior in response to trip time uncertainty. The results of the current NCHRP 08-99 study, combined with the other research cited throughout this report, are expected to provide a basis for hypotheses, which can be empirically tested through stated preference analysis and econometric modeling. Task 2: Design the survey: The survey instrument needs to be carefully designed. Expertise in conducting stated preference surveys, to obtain valid insights into tradeoffs between the cost of uncertainty and the cost of employing mitigation strategies is needed. Frequently, such surveys have focused on time vs. cost tradeoffs (e.g., how much of a toll are users willing to pay to avoid congestion and delay at various levels.) This study is slightly different, in that modifying truck routes (such as to a toll facility or a longer but less uncertain route) is only one strategy employed. The possibility of posing simple toll vs. uncertainty tradeoffs may be considered, but its limitations should be carefully reviewed. Moreover, the relevant tradeoff is not between cost and travel time, but between cost and variability in trip times. In this study, considerable attention was given to framing the notion of travel time uncertainty in the survey instrument in ways understandable to truckers and supply chain professionals. The research team found that terms such as “unpredictable delay” of “percent on-time delivery” were useful. Task 3: Develop the sampling frame and survey approach: A large enough sample is needed to conduct stratified analysis, for factors such as commodity type and value, supply chain and type of trip and market served, etc. The NCHRP Project 08-99 survey yielded fewer than 200 respondents; this demonstrated how difficult it can be to obtain very high response rates for surveys of this type. The survey expert for this study, Tompkins, finds that getting large responses is difficult—indeed increasingly difficult—as shippers and truckers and others in the freight industry are becoming “surveyed out.” Phase II Task 4: Conduct the survey: The survey will be administered based on the Task 3 plan. Task 5: Develop the statistical framework: A standard econometric approach, used by travel demand modelers, is to develop the models within a utility maximization framework. The study should specify the functional form of the relationships for testing, based on the hypotheses developed in Task 1.

10 The techniques applied to model development of this nature are well established, and the skills required to calibrate models of this type are quite ubiquitous throughout the transportation industry. Task 6: Conduct additional case study demonstrations of the updated methodology: Two case studies should be conducted. The possibility of using the same case studies as pre- sented in NCHRP 08-99 could be considered. Task 7: Integrate the econometric results into the Truck Freight Reliability Valuation Model: The ultimate goal is to develop the valuation tool for project planning and prioriti- zation purposes. The approach to modifying the Truck Freight Reliability Valuation Model should be developed. The revisions should address the technical limitations in the current model, identified above. Time and Resources The study should be completed within 1 year to 18 months. A budget of $300K is suggested.