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.

16 A stated-preference survey was conducted to obtain data on how shippers and motor carriers make trade-offs between travel time reliability and costs. On the basis of the literature reviewed, this was determined to be the best approach for developing robust estimates of truck VOR in this study. The literature on stated-preference studies was reviewed carefully to develop a survey that followed best practices and had the right structure and complexity. This included reviewing previous studies on the value of reliability (VOR) and value of time (VOT) in both freight and passenger contexts and the broader literature on best practices in survey design. Appendix A describes some of the studies reviewed and provides the rationale for the design adopted. 3.1 Overview The goal of the survey was to maximize the amount and quality of information collected given the cognitive and time constraints of respondents. The survey had to be made as short and simple as possible to encourage participation while producing enough information to estimate models with the parameters of interest. Table 3-1 outlines the different sections of the survey. The survey started by describing the objective of the study in general terms without overly focusing on reliability, so as to prevent biasing in the responses. It then asked respondents to self-select into one of three categories—motor carrier, shipper with transportation, or shipper without transportation—so that the phrasing of subsequent questions could be altered to make them more relevant. Respondents could also select “other,” which was then recoded into the three types on the basis of contextual information or removed from the sample if it was not relevant. The core of the survey was a set of eight stated-preference questions that asked respondents to choose between two hypothetical transportation alternatives with different travel time, cost, and reliability levels. Respondents were asked to select their preferred alternative by following the same criteria they use to make decisions in reality, especially with regard to how unreli- ability impacts their operations. As described later in this chapter, the hypothetical alterna- tives presented were constructed to maximize the usefulness of the responses in estimating the VOR. To improve the realism of the stated-preference questions, the alternatives were pivoted around the characteristics of a recent typical shipment described by respondents. These charac- teristics, in terms of length, content, weight, and cost, were used as a baseline for the alternatives. In addition to making the alternatives more familiar, this approach also helps in exploring a C H A P T E R 3 Stated-Preference Survey

Stated-Preference Survey 17 wider range of situations. Respondents were also asked about the intended receiver and whether intermodal transfers were involved. Context questions were then asked that could serve as controls in the modeling. This included asking for the title and responsibilities of respondents, in order to assess whether the right people took the survey. This section also asked about company size and industry [as identified by the North American Industry Classification System (NAICS) code]. In the debrief section, respon- dents were asked to rate the relevance of the stated-preference questions and provide general comments on the survey or reliability in general. Table 3-2 shows the attribute levels used in the stated-preference questions. These levels were selected to explore the range of trade-offs of interest in this study, particularly those involving reliability. Previous studies have used comparable levels (de Jong et al. 2014, Jin and Shams 2016). Table 3-3 shows how different reliability levels were communicated to respondents. In contrast to previous studies, reliability was described by using percentiles expressed as 1 out of x trips arriving late. Not only is this phrasing familiar to individuals in the freight sector, but it is compatible with the use of travel time indices in freight planning, particularly at the Section Questions/Content Screens Typical Duration Introduction Describe objective and sponsor. Link to additional information. 1 30 s Self-selection Select one: motor carrier, shipper with transportation, or shipper without transportation. Describe ideal respondent. Provide overview of survey sections. 2 1 min Recent typical shipment Describe recent typical shipment: receiver type, commodity, length, average duration, size, transportation cost, and other characteristics. 7 3.5 min Stated-preference questions Answer guide. Eight stated-preference questions pivoted off recent typical shipment characteristics. 9 8 min Context questions Describe respondent’s title and responsibilities. Select company size and industry. 4 1 min Debrief Rate relevance of stated-preference questions. Open ended comments. 2 1 min Total 25 15 min Table 3-1. Stated-preference survey overview. Attribute Attribute Level Attribute Level Label Cost –20%, –10%, reference, +10%, and +20% $/shipment Travel time –20%, –10%, reference, +10%, and +20% Hours Travel time reliability Very low, low, medium, high, very high See Table 3-3 Table 3-2. Stated-preference attributes and levels.

18 Estimating the Value of Truck Travel Time Reliability 95th percentile level (corresponding to 1 out of 20 trips). This measure also aligns better with modeling, as described in Section 4.1.1. 3.2 Experimental Design Constructing the stated-preference alternatives is a critical step with wide-ranging implica- tions for modeling, in that it affects the type and magnitude of the results obtained. A main- effects orthogonal design was adopted for this study, because only main-effects are relevant for estimating the VOR. In this design, attribute levels are uncorrelated with each other and only certain combinations of attributes are considered. The main-effects orthogonal design resulted in a shorter survey than if full factorial or fractional factorial designs had been used. Each respondent faced eight choice questions that were randomly selected without replace- ment from Table 3-4. This set of potential questions was constructed by first generating an orthogonal main-effects plan for Alternative A and then using Burgess (2007) to generate the plan for Alternative B. This scheme produced a set of choice questions that was as short as pos- sible while retaining the advantageous statistical properties of an orthogonal design. Several of the questions were trivial, in that one alternative was better along all attributes than the other one. These questions were left in the survey to preserve orthogonality and to test whether respondents answered questions logically. 3.3 Survey Administration The survey was programmed in a website to be clear, concise, and appealing to respondents. Figure 3-1 shows one of the choice questions presented to respondents (to a shipper with trans- portation). In this case, the respondent needed to move electronic goods 500 miles by truck and had to choose between a cheaper alternative with poor reliability and a more expensive alterna- tive with better reliability but longer average travel time. Achieving a sufficient response rate is one of the most challenging aspects of stated-preference surveys. Many previous studies have failed to attract enough responses to estimate even a basic model. A comprehensive strategy was implemented to maximize the number of responses. Dif- ferent sources of contact information were mined, and different ways of contacting potential Reliability Level Presentation Very high 1 out of 5 trips arrives late by more than 15 min 1 out of 20 trips arrives late by more than 30 min 1 out of 50 trips arrives late by more than 38 min High 1 out of 5 trips arrives late by more than 30 min 1 out of 20 trips arrives late by more than 1:00 h 1 out of 50 trips arrives late by more than 1:30 h Medium 1 out of 5 trips arrives late by more than 1:00 h 1 out of 20 trips arrives late by more than 2:00 h 1 out of 50 trips arrives late by more than 2:40 h Low 1 out of 5 trips arrives late by more than 1:30 h 1 out of 20 trips arrives late by more than 3:30 h 1 out of 50 trips arrives late by more than 4:30 h Very low 1 out of 5 trips arrives late by more than 3:00 h 1 out of 20 trips arrives late by more than 7:30 h 1 out of 50 trips arrives late by more than 10:30 h Table 3-3. Stated-preference reliability presentation.

Figure 3-1. Online survey screen. Choice Alternative A Alternative B Trivial? Cost (%) Travel Time (%) Travel Time Reliability Cost (%) Travel Time (%) Travel Time Reliability 1 –20 Base Medium 20 20 Very high 2 –20 –10 Low 20 10 High 3 –10 Base Very low –20 20 Medium 4 Base –20 Low –10 Base High 5 10 Base Low Base 20 High 6 Base 20 Very low –10 –10 Medium Yes 7 Base Base High –10 20 Very low 8 10 10 Medium Base –20 Very high Yes 9 20 10 Very low 10 –20 Medium Yes 10 –10 10 Low –20 –20 High Yes 11 Base 10 Very high –10 –20 Low 12 10 –20 Very high Base Base Low 13 20 –10 High 10 10 Very low 14 20 Base Very high 10 20 Low 15 –10 20 Medium –20 –10 Very high Yes 16 20 20 Low 10 –10 High Yes 17 –20 10 High 20 –20 Very low 18 –20 20 Very high 20 –10 Low 19 –20 –20 Very low 20 Base Medium 20 10 20 High Base –10 Very low 21 20 –20 Medium 10 Base Very high 22 –0 –10 Very high –20 10 Low 23 Base –10 Medium –10 10 Very high 24 10 –10 Very low Base 10 Medium 25 –10 –20 High –20 Base Very low Table 3-4. Set of potential choice questions.

20 Estimating the Value of Truck Travel Time Reliability respondents were tried. Table 3-5 summarizes these efforts. The three types of participants were contacted with methods and messages tailored to their situation. The methods included the following: • Mass e-mail. Reliable e-mail addresses of motor carriers and shippers were obtained from two sources. Shipper e-mails were obtained from Tompkins’s lists of shippers and 3PL providers. Contact e-mails were tailored on the basis of the position of the recipient (C-level executive, director of logistics, or analyst) so that the survey reached the appropriate person. E-mail addresses for motor carriers were obtained from the Federal Motor Carrier Safety Adminis- tration census file. Most responses came from this source. • Phone contact. The Dun & Bradstreet commercial data set was queried for companies with in-house transportation positions. Out of this list, more than 100 companies were contacted by phone by professional interviewers to elicit their participation. A small number of respon- dents ultimately followed through and completed the survey. Even though this was the most time-consuming part of the data collection effort, it produced the least survey responses. • Association forward. Nineteen trade associations around the United States that work with shippers or motor carriers were contacted to help increase the visibility of the survey. Of these, four agreed to distribute the survey to their members. 3.4 Responses The survey was completed by 1,142 respondents who provided answers to 18,096 hypothetical choice questions. To improve the quality of the sample, a minimal number of cleaning actions was undertaken. First, respondents who did not appear to have taken the survey seriously, in that they gave nonsensical responses, were removed from the sample (15 respondents). Respondents who did not appear to be involved in freight transportation were also removed (24 respondents). Most of these involved small trucks used for service calls or buses used for passenger transit. Cutoffs were also put in place for the characteristics of the recent typical shipment. Responses were removed if trucking costs were reported as being higher than $15,000, average speeds were reported as being higher than 100 mph or lower than 3 mph, or shipment weights were reported as being greater than 50 tons. This removed an additional 121 respondents from the sample. Models estimated on the complete sample, with outliers and nonsensical answers, fit the data poorly and provided results that were less interpretable than those provided by the cleaned sample. Most responses came from self-identified motor carriers, which represented 59 percent of the sample. Shippers without transportation represented 11 percent of the sample, and ship- pers with transportation represented the remaining 30 percent. Obtaining a significant number of responses from each of these types was a key accomplishment of the survey. As can be seen in Figure 3-2, most respondents belonged to companies with just 1 to 4 employees, although a significant sample was also achieved for larger companies. Most of the smaller companies were classified as motor carriers, which likely represents owner-operated trucks. Type of Participant Shipper With Transportation Shipper Without Transportation Motor Carrier Costs considered Supply-chain and trucking Only supply chain Only trucking Sampling strategy Direct phone contact E-mailing commercial list Forward by trade association E-mail commercial lists Forward by trade association E-mail commercial lists Forward by trade association Table 3-5. Sampling strategy.

Stated-Preference Survey 21 0 50 100 150 200 250 300 350 400 450 1-4 5-24 25-100 100-1000 1000+ N um be r of R es po nd en ts Company Size (# employees) Motor Carrier Shipper w/o Transportation Shipper w Transportation Figure 3-2. Respondents by company size and type.