Stuck in the slow lane: undoing traffic composition biases in the measurement of trucking productivity.

• Author: Boyer, Kenneth D.

1. Introduction

   Deregulation and the logistics revolution have transformed the American economic landscape over the last 30 years. At the heart of these changes has been an apparently dynamic trucking industry, rapidly adapting to the freedoms of deregulation and adopting the most modern technology to drive down costs and increase productivity.

   This picture of a dynamic industry, using information technology and achieving impressive cost reductions, is reminiscent of the way the U.S. railroad industry was initially viewed in the post-World War II era. Soaring railroad ton-miles per worker-hour in the 1940s and 1950s occurred at the time that the industry was replacing steam with diesel locomotives. Costs per ton-mile declined rapidly. The idea that the sector had rapidly improving labor productivity affected labor negotiations within the railroad industry and public policy toward it. But a famous study by academic economists (Task Force on Railroad Productivity 1973) demonstrated that the rapid productivity improvements were mostly an illusion, the result of a change in traffic composition rather than the adoption of new production techniques. In the 1950s and 1960s rail traffic began to look like what we see today--a preponderance of long-distance movements of bulk commodities like coal, grain, and chemicals--where ton-miles per worker are high. Prior to that, railroad traffic was a heterogeneous mix of short and long-distance movements of both light and heavy goods. Such traffic is inherently more labor intensive to handle. Beginning in the late 1940s and continuing for at least five decades, measured railroad productivity per worker-hour rose as traffic that was costly to handle left for the highways. The observed productivity increases were driven by this change in the traffic mix and were independent of any changes in operational efficiency of the railroads. (1)

   This paper argues that a similar scenario played out in the trucking industry over the last 25 years of the 20th century, as the industry increased its proportion of traffic that is relatively cheap to handle. As in the case of the railroads a half century ago, changes in traffic composition have again inflated the apparent productivity changes in a transportation industry.

   In this paper, we analyze microdata from the quinquennial Vehicle Inventory and Use Survey (VIUS; U.S. Census Bureau 1977-1997) to show that failing to account for systematic changes in the composition of truck traffic has a significant impact on the magnitude of measured trucking industry productivity. Because trucks vary so widely, we focus on the single most common type of heavy trucking equipment: the standard enclosed van pulled by a heavy truck tractor. Over a 15-year period following trucking deregulation for which comparable operational data are available in the VIUS (U.S. Census Bureau 1982-1997), we estimate how the annual ton-miles per truck-and-driver combination--what we will call physical trucking productivity--varies with the operational characteristics of these rigs. The results permit us to compare the fleet-level average productivity of like vehicles over time, partly controlling for changes in traffic composition, something that is impossible to do if output is measured using a simple aggregation of ton-miles, or if productivity is estimated from the financial accounts of trucking firms.

   The focus on vehicles rather than firms is an engineering approach to measuring production functions. Using this approach, we will argue that the traffic mix changes we document are endogenous. This in turn suggests that the causality between measured productivity changes over time and cost changes in trucking is largely from costs to measured productivity, rather than from higher productivity to lower costs. There was a large increase in long-distance relative to short-distance trucking over the period from 1977 to 1997, as shown in Table 1 below (see also Burks, Monaco, and Myers-Kuykindall 2004a). This shift was driven primarily by a decline in the price for trucking services that occurred during this period. The price of trucking services declined as a result of a reduction of regulatory rents and sharp cuts over this period in the price of inputs, notably drivers' wages and fuel prices (Rose 1985: Belzer 1995; Belman and Monaco 2001; Monaco and Brooks 2001).

   As Table 1 shows, ton-miles per truck are the highest in long-haul markets. Long-haul trucks run many more miles per year than do those involved with short-haul transportation because they spend a higher proportion of their total operational time traveling at high speeds on intercity highways. They also tend to stay relatively fully loaded. The cost of loading short-haul and long-haul units is similar, but the cost savings from keeping a long-haul movement fully loaded is higher. The demand for trucking services is derived from the geographic distribution of demands for the products that trucks haul, and short-haul lanes were already relatively intensely served, so when the price of all trucking services dropped, long-haul service is exactly where the demand for trucking grew most rapidly. The result has been an increasing proportion of long-distance traffic, which in turn created the appearance that labor productivity had risen faster than was actually the case.

   This bias in productivity measures due to endogenous changes in output composition is relatively common when working with aggregate statistics of heterogeneous populations. Blundell and Stoker (2005) describe the biases associated with ignoring aggregation issues and offer some techniques for mitigating the bias. While attention to aggregation issues is routine in labor economics, it is relatively uncommon in the transportation literature, the Task Force on Railroad Productivity (1973) being a notable exception.

   We will show in this paper that physical productivity in trucking has indeed increased since 1982, but at a slower rate than in the rest of the economy as a whole. Two primary sources of increased productivity are (i) an increase in miles per truck, due in part to increases in speed limits, and (ii) the increasing length of truck trailers. Longer trailers have increased the interior volume, partly offsetting a trend to less densely loaded trailers.

   The remainder of the paper is structured as follows. Section 2 discusses the significance of the argument about productivity in trucking. Section 3 lays out our method for calculating the productivity of driver-and-vehicle combinations from the submeasures of miles per year, tons per loaded movement, and the likelihood of having a loaded movement. In this section, we also briefly describe the data and our method for measuring traffic composition and for separating composition effects from true changes in physical productivity. Section 4 provides the results of our measurement of trends in productivity submeasures and traffic composition in the trucking industry at the end of the 20th century. Section 5 summarizes the argument and offers some caveats and conclusions. The Appendix describes the VIUS data (U.S. Census Bureau 1977-1997) and how it was used, in more detail.

2. The Significance of Productivity in Trucking

   There are two reasons to take the issue of productivity in trucking seriously. One reason is that this sector has been the subject of recent analyses that attempt to draw conclusions about the long-run growth of productivity in the economy as a whole, and about the productivity impact of innovation in information technology. Both papers described below fail to control for changes in traffic composition, and this throws doubt on the larger conclusions they wish to draw. A second reason that trucking productivity is itself of considerable interest is because trucking's output makes up a modest but measurable share of gross domestic product (GDP) and because of the internal diversity of the sector. The deep relationship this diversity has to the development of modern supply chains is little understood outside the industry and its principal customers, the managers of corporate logistics in the U.S. economy.

   Examples of the importance of controlling for traffic composition when measuring transportation productivity are provided by two recent papers that attempt to draw lessons from trucking productivity changes for the broader economy. Fernald (1999) hypothesized that the unusually high growth rates in the American economy in the period 1953-1973 could be due to the building of the Interstate highway system. He finds support for this hypothesis in the fact that the transportation industries, of which trucking is by far the largest part, are estimated to have had large and positive growth in total factor productivity during this period, large enough to be a statistical outlier. However, if the measurement of productivity changes in the transportation industries is biased as a result of traffic composition effects--the hypothesis of this paper--then his fundamental measurement of the magnitude of the effect of government investment is brought into question. (2)

   Hubbard (2003) uses data from the same underlying source as our work and finds that in the years 1992 and 1997 the use of on-board computers is associated with cross-sectional differences in the annual miles of trucks that are in the low double digits. (3) However, Hubbard's inference depends critically on the interpretation of the determinants of the number of months per year that a truck is used, a factor that is not independent of the composition of traffic carried by the truck. (4) In addition, by examining only annual miles he ignores the output dimension of weight, which--as we will show below--varies significantly across industry segments, and which therefore provides a potential source of bias. Hubbard's work received considerable notice because it was the first to find significant physical productivity improvements as a...