In the long-haul trucking industry, the turnover rate for drivers has been consistently near or higher than 100% for many years (Fournier, Lamontagne, & Gagnon, 2012; LeMay, Williams, & Garver, 2009). There are many complexly-interacting contributors to this high rate, including competition among industry members for the short supply of qualified drivers, characteristics of driving assignments, family needs, and interpersonal relationships among coworkers. Several costs to trucking companies are associated with a driver quitting their current position, including lost profit, underutilization of equipment, training costs, insurance investments, and administration costs. The overall turnover cost to the industry is estimated to be around $2.8 billion per year (Morrow et al., 2005), which is eventually passed on to consumers. The current and expected increase in the shortage of drivers provides additional motivation to investigate the high turnover problem.
Several past studies have looked into the possible causes of high driver turnover, although only within consideration of current trucking operations. In addition to only considering the problem within current operations, most of these studies focused only on a particular subset of the possible issues.
A driver survey was created as part of this research effort to take into account a broader range of issues and with a more generalizable approach to allow for consideration of new operational paradigms. In addition to gathering survey results, a key activity in this research effort was to develop a model that can accurately predict the likelihood that a truck driver will quit or stay at their job. When we reach a point when a driver’s decision to stay or quit can be accurately predicted based on the explored issues, it will reveal which combination of issues most strongly influences this decision. Once these issues are discovered, they can inform the development of new operational paradigms that may lead to a decrease in the high turnover problem.
A thorough literature review informed the creation of the content of the survey. This was followed by multiple revisions based on feedback from an in-person focus group with truckers, a test launch with a sample of students (non-drivers), and a second test launch with a sample of local drivers. The survey was then administered and 308 valid responses were obtained.
The final survey consisted of 84 questions covering demographic information, training and education, job preferences, job nature, management practices, relationship with supervisor, job risks and benefits, work history, job seeking experience, and alternate job opportunities. One survey question which asked: “How do you feel today about the likelihood you'll stay with your current company for the foreseeable future?” resulted in the dependent variable (range of 0.00 to 1.00) used in the analysis.
The data was analyzed using the free statistical software environment, R. The first attempt to create a prediction model involved performing multiple linear regression with most of the independent variables and the stated dependent variable. From this analysis, the variables deemed significant included the following: how routes are communicated to the driver (in-person, by phone/radio, by text/email, or other), how often pre-planned routes are given to the driver, what the driver thinks about their work amount, how the driver feels about being affiliated with their company, how well the driver feels their supervisor represents them, how satisfied the driver is with their pay, the driver’s flexibility in being able to take time off work, how interesting the driver considers their job to be, and how frequently the driver interacts with other drivers by radio.
This resulting linear model had an adjusted R-squared value of 0.4658. Further data analysis involving other variable selection methods and the creation of complex models will be performed in hopes of obtaining a more accurate model. Future analysis may also include clustering the drivers into groups based on their characteristics and preferences and then seeing if the variables that predict the job plans of the drivers in these groups are different.