We enhance the existing literature by leveraging the Commodity Flow Survey (CFS) data from the United States Census Bureau (2020) to reliably estimate the density of long-haul trucking across various regions and analyze the operator-hours necessary for different routes. Our main analysis focuses on utilizing freight data alongside routing and operator-hour algorithms to determine the proportion of operator-hours that could be impacted by automation. To enrich this quantitative analysis, we conducted a limited number of interviews with stakeholders in long-haul trucking to gain insights into the practicality of a transfer-hub deployment model. Our mixed-methods approach follows the triangulation method (Jick, 1979), which is effective for examining socio-technical transitions and emerging technologies. We discuss our methodologies further in the sections that follow.
Data
The Commodity Flow Survey is a prominent dataset for transportation research and planning, compiled every five years by the U.S. Bureau of Transportation Statistics, U.S. Census Bureau, and U.S. Department of Commerce. The most recent version, CFS 2017, includes a sample of 5.9 million shipments from approximately 60,000 participating establishments (United States Census Bureau, 2020). We exclude inter-modal shipments and concentrate on deliveries made via for-hire and private trucks. Our analysis specifically examines shipments covering distances greater than 150 miles, as these are commonly recognized as long-haul (FMCSA, 2020; Viscelli, 2018). This subset comprises nearly 1.5 million trucking shipments, detailing origin and destination states, shipment distances, weights, and financial quarters. Additionally, the dataset includes a weighting factor, allowing for the estimation of total shipments of that type within the broader population.
Routing
We utilize the Google Maps API and the GGMAP package (Kahle and Wickham, 2013) in R to calculate highway and urban splits for each shipment. For routing purposes, we categorize shipments into two types: intrastate (within the same state) and inter-state (crossing state lines). We employ different methodologies to compute the highway and urban splits based on whether shipments are intrastate or inter-state, and whether they include specified Metropolitan Statistical Areas (MSAs). If MSAs are absent, we approximate the origin or destination based on the closest available location.
For inter-state shipments, we primarily use the origin and destination MSAs from the CFS dataset in Google Maps, estimating highway and urban distance ratios, which we then apply to actual shipment distances. For cases requiring it, we estimate using state centroids. For intrastate shipments, we apply similar methods, using highway and urban splits derived from the Freight Analysis Framework dataset (Bureau of Transportation Statistics, 2012), segmenting roads based on average speeds.
Operator-hours Calculation
The final calculation incorporates urban and highway operator-hours, assuming an equal split at both ends of the journey with highway travel in between. We adhere to an 11-hour daily driving limit as per hour of service (HOS) regulations (FMCSA, 2020) and calculate the required operator-hours for both travel legs. Using the calculated times and the weighting factor, we establish total operator-hours and differentiate between highway and urban operator-hours.
Interviews
To validate the assumptions of the transfer-hub model, we conducted semi-structured interviews with key stakeholders in the trucking industry. Using purposeful sampling (Robinson, 2014), informed by prior work on automated vehicles, we identified relevant questions and targeted participants. Our sample included stakeholders from trucking automated startups (2), drivers (5), logistics operators (1), and labor unions (1), with a deliberate emphasis on driver perspectives. This approach helps in addressing operational challenges for the transfer-hub model and assessing the feasibility of automation as perceived by drivers. Accordingly, these interviews were integrated into our triangulation method to complement our quantitative analysis, providing insights into the assumptions and deployment modes suggested in the paper. The complete list of interviewees is detailed in Table 2.
