Make versus Buy in Trucking: Asset Ownership, Job Design, and Information

by George P. Baker, Thomas N. Hubbard
Citation
Title:
Make versus Buy in Trucking: Asset Ownership, Job Design, and Information
Author:
George P. Baker, Thomas N. Hubbard
Year: 
2003
Publication: 
The American Economic Review
Volume: 
93
Issue: 
3
Start Page: 
551
End Page: 
572
Publisher: 
Language: 
English
URL: 
Select license: 
Select License
DOI: 
PMID: 
ISSN: 
Abstract:

Make Versus Buy in Trucking: Asset Ownership, Job Design, and Information

Explaining pattems of asset ownership is a central goal of both organizational economics and industrial organization. We develop a model of asset ownership in trucking, which we test by examining how the adoption of diferent classes of on-board computers (OBCs) between 1987 and 1997 influenced whether shippers use their own trucks for hauls or contract with for-hire carriers. Wefind that OBCs' incentive-improving features pushed hauls toward private carriage, but their re- source-allocation-improving features pushed them toward for-hire carriage. We con- clude that ownership patterns in trucking reject the importance of both incomplete contracts and ofjob design and measurement issues. (JEL D23, L14, L22, L23, L92)

Understanding the patterns of asset owner- ship in the economy is a central goal of both organizational economics and industrial organi- zation because it provides insights on firm boundaries and industry structure. Major progress towards this goal was provided by Sanford Grossman and Oliver Hart's seminal paper in 1986, which argues that asset owner- ship confers on owners residual rights of control that give them power and thus incentives to devote effort to value-increasing activities. In this view, firms' boundaries are determined by the optimal allocation of these residual rights of control. Bengt Holmstrom and Paul Milgrom (1994), however, argue that firms' boundaries reflect trade-offs in which asset ownership in- teracts with job design and other organizational decisions. If so, firms' boundaries may reflect factors that do not appear in Grossman and Hart's (1986) theory, including those that affect

* Baker: Harvard Business School, Boston, MA 02163, and NBER; Hubbard: University of Chicago Graduate School of Business, Chicago, IL 60637, and NBER. We would like to thank all those we talked to at trucking firms and private fleets for allowing us to visit their firms and discuss the issues we investigate in this paper. Thanks to Ann Merchant for research assistance and Oliver Hart, Bengt Holmstrom, Paul Milgrom, Canice Prendergast, Pres- ton McAfee, MinSoo Park, an anonymous referee, and many seminar participants for comments. We gratefully acknowledge support from NSF Grant No. SES-9975413, the NBERISloan Pin Factory Project, and the Harvard Busi- ness School Division of Research.

the optimal allocation of tasks across individu- als. In 1999, Holmstrom offered a critique of the property rights view in which he argues that it fails to explain why firms rather than individu- als own assets. He extends the insight from the 1994 paper to argue that firms own assets pre- cisely because this mutes the incentives that come with individual asset ownership, allowing the firm to operate as a "subeconomy" that can more precisely balance incentives and imple- ment more complex multitask job designs.

In this paper, we argue that the pattern of asset ownership in trucking-in particular the decision by shippers about whether to use their internal fleet of trucks for a haul or contract with for-hire camers-reflects not only the factors identified in Grossman and Hart's theory, but also those highlighted in Holmstrom and Mil- grom (1994). Consistent with the former, own- ership patterns reflect trade-offs that arise from providing intermediaries strong incentives to identify profitable uses for trucks. Consistent with the latter, ownership patterns also reflect issues of job design: i.e., the degree to which drivers simply drive trucks, or provide a more complex combination of transportation and service. Job design matters because "service- intensive" trucking hinders intermediaries' abil- ity to find profitable uses for the truck. Shipper ownership of trucks mutes incentives and favors service-intensive trucking in which drivers' jobs involve more than just driving trucks.

We develop a model that combines these

551

theoretical insights. The model generates two sets of comparative static predictions. One set of predictions is consistent with well-known cross-sectional patterns in the industry. These include the prediction that service-intensive trucking is more likely to be performed by private than for-hire fleets, and that private fleets are differentially more likely to adopt incentive-improving technologies, while for-hire carriers are more likely to adopt coordination-improving technologies.

The other set of predictions concerns how changes in the informational environment affect ownership. We test this second set of predic- tions using data from the 1987, 1992, and 1997 Truck Inventory and Use Surveys, which con- tain detailed truck-level information about trucks' characteristics, ownership, and use. In particular, we test predictions on how the dif- fusion of different types of on-board computers (OBCs) during the late 1980's and early 1990's alters the "make versus buy" decision for ship- pers. We predict that the adoption of certain types of OBCs should lead indirectly to more shipper ownership of trucks, by lowering the agency costs associated with complex job de- signs. We predict that the additional capabilities of other types of OBCs-those that provide loca- tion information and real-time communication- should lead to less shipper ownership of trucks, because these additional capabilities enhance the comparative advantage of for-hire carriage with respect to truck utilization and dispatch. We find evidence in favor of both of these predictions.

Our results strongly suggest causal links be- tween informational and organizational changes in the trucking industry. They show that own- ership patterns in trucking reflect the impor- tance of not only incomplete contracts (as stressed by Grossman and Hart, 1986), but also of job design and measurement issues (like those stressed in Holmstrom and Milgrom, 1994). These findings thus shed important light on theories of organizations. They also make a contribution to the long-running debate about how information technology (IT) diffusion af- fects the boundaries of the firm.' We note that

See H. J. Leavitt and T. L. Whisler (1958); Thomas W. Malone et al. (1987); Erik Brynjolfsson and Lorin Hitt (1997).

information technology in general provides at least two capabilities-improved monitoring of agents and improved coordination of activities-and that the organizational impact of these capabilities can differ (Michael C. Jensen and William H. Meckling, 1992). In trucking, improvements in monitoring (and the attendant improvement in incentives) lead to larger, more integrated firms, while improvements in coordi- nation (resulting in better asset utilization) lead to more diffuse asset ownership and smaller, less integrated firms. Whether these results gen- eralize to other settings remains an open question.

In this paper we do not consider a third pos- sibility regarding truck ownership: drivers may own trucks. We investigate driver ownership of trucks in detail in another paper (Baker and Hubbard, 2000). In that paper, we propose that asset ownership strengthens drivers' incentives to drive in ways that preserve trucks' value, but also encourages them to engage in rent-seelung behavior. We then argue that OBC adoption alters this trade-off by allowing companies (ei- ther for-hire carriers or private fleets) to use the monitoring capabilities of OBCs to substitute for asset ownership. We show that OBCs lead to less driver ownership of trucks, especially for hauls where rent-seeking is a potential problem. We ignore these issues in the present paper because we believe they are not salient to the make-or-buy decision. Situations that are on the margin between for-hire carriage and private carriage are not those where owner-operators are used. In general, owner-operators are used for hauls that require little if any service provi- sion by the driver, and for good reason. The multitasking problems with service provision that lead for-hire carriage to be inefficient rela- tive to private caniage are exacerbated when drivers control trucks. This is borne out by the fact that when shlppers outsource hauls with nonnegligible service requirements, they rarely, if ever, do so by contracting with owner-operators.'

The paper is organized as follows. In the next section, we describe the institutional setting that

In our empirical tests, "for-hire carriage" includes driver- and carrier-owned trucks, ownership structures where ship- pers do not own trucks. Our results are unchanged when we leave out owner-operators altogether.

we model, defining the players, describing their roles in the provision of trucking services, and characterizing the contracting environment in which they operate. In Section 11, we present our model of job design and asset ownership. Section III describes OBCs and generates our main empirical propositions. In Section IV, we describe our data and present the main empirical patterns. Section V contains our main empirical results regarding the relationships between OBC adoption and organizational change. Sec- tion VI concludes.

I. Job Design, Search Incentives, and Asset Ownership in Trucking

This section describes the institutional frame- work, drawing heavily from what we learned in a series of site visits and interviews. We de- scribe the basic trade-offs involved in job de- sign and asset ownership decisions and explain why these decisions might be related. Through- out the section, we will refer to several different parties. Drivers are individuals who drive trucks and may have other customer-service-oriented tasks. Shippers are firms or divisions with de- mands to move cargo from one place to another. Carriers are firms or divisions that supply trans- portation services. Carriers that supply services using trucks owned by shippers are private car- riers (i.e., shippers' internal fleets). Carriers that supply services using their own rather than ship- pers' trucks are for-hire carriers. Brokers are third-party informational intermediaries.

A. Driver Job Design: Driving and
Service Provision

Drivers can engage in two sorts of activities: driving the truck and performing nondriving service a~tivities.~

Defining drivers' jobs to in- clude nondriving activities lets carriers offer high service options in which their customers can ask drivers to do things such as help unload the truck and sort and store the cargo. This gives customers flexibility in how many of their own workers they allocate to such tasks, and can improve the division of labor in the short run because

See Lawrence J. Ouellet (1994) for a detailed descrip- tion of incentives and the organization of work in trucking.

deliveries might take place when the opportunity cost of customers' workers' time is high.

The benefit of giving drivers service respon- sibilities varies systematically across hauls with the characteristics of the cargo. There are rarely such benefits when they haul bulk goods such as gravel, ores, or grain, in large part because no handling is required upon delivery: when trucks reach their destination, drivers dump the cargo where the recipient wants it. Giving drivers service responsibilities is also generally unpro- ductive when trucks haul goods for which han- dling requires special equipment. For example, special machines-which drivers generally are either unable or not trusted to use-are usually necessary to move very heavy goods (large rolls of paper, sheet metal). As a consequence, driv- ers generally just drive trucks when they haul bulk or unwieldy goods.

In contrast, giving drivers service responsi- bilities can be valuable when trucks haul other classes of goods, such as packaged goods or hazardous cargo. Packaged goods can be carried by hand or transported with standard equipment such as hand trucks, conveyor belts, or forklifts. Handling hazardous cargo such as petroleum or chemicals requires certification, which drivers generally must have to haul such cargo legally. Giving drivers service responsibilities dimin- ishes the extent recipients must have certified personnel. As a consequence, drivers often have service responsibilities when trucks haul pack- aged goods or hazardous cargo.

A drawback to giving drivers additional re- sponsibilities is that agency costs are higher.4 Carriers always face the problem of motivating drivers to pick up and deliver goods on time and drive in ways that preserve trucks' value. When drivers' jobs involve service, they also face the problem of motivating drivers to allocate their time efficiently between driving and service.

Motivating drivers to pick up and deliver goods on time is straightforward because it is relatively easy to evaluate drivers' performance in this dimension. The distances traveled and the return time at the end of the run are known. Carriers also normally have good information

Following Jensen and Meckling (1976), agency costs here include both monitoring costs and the "residual loss" attributable to nonoptimal decisions.

regarding whether drivers arrive late to interme- diate stops-angry customers call them when they do--and have some information about the impact of factors outside of drivers' control, such as traffic and weather conditions. Thus, when drivers' jobs involve only driving from location to location, the main agency problem that remains is inducing them to drive well because this is what remains noncontractible.

Incentive problems are more complicated when drivers' jobs include service activities. As is generally the case in multitashng problems, incentives must attend both to overall effort levels and the allocation of effort across tasks. In this case, the incentive problem created by multitasking is that carriers now must induce drivers to allocate effort between driving and service appropriately. Simple distance and ar- rival time data provide little indication of the fraction of time drivers spend driving versus doing other things. Some common service ac- tivities such as cargo-handling are ~trenuous.~ Drivers with service responsibilities have an incentive to rnisallocate their effort: for exam- ple, by thng more time handling cargo, then malung it up by driving faster between stops. Carriers may respond to this, in the spirit of Holmstrom and Milgrom (1991, 1994) and Baker (1992), by weakening drivers' incentives with respect to other tasks. For example, they balance incentives by de-emphasizing on-time arrivals or allowing more slack in schedules. In general, agency costs are higher when drivers have more responsibilities because of some combination of lower overall effort levels and a worse allocation of effort across tasks.

B. Market Clearing: Load Matching and Search

The demand for trucking services and the supply of truck capacity are highly differen- tiated. Shippers' demands are specific with respect to time, location, and equipment re- quirements. Likewise, truck capacity is idiosyn- cratic with respect to its geographic location and

Drivers whose jobs involve taking a fully loaded trailer and delivering the goods to various destinations handle up to 40,000 pounds of cargo per day. Handling requires hand- lifting when trucks deliver to places without loading docks-such as most retail outlets.

the characteristics of the trailer. Capacity utili- zation in the industry depends crucially on how efficiently supply and demand-trucks and hauls-are matched. Trucks and hauls are matched in a highly decentralized manner in which shippers, carriers, and third-party brokers search for good matches.

The matching problem is particularly difficult in trucking because individual shippers rarely have demands that fill trucks for both legs of a round-trip. For this reason, once carriers receive service orders from shippers, they then search for complementary hauls. When individual shipments are too small to fill a truck, search takes the form of identifying other shippers with similar demands. When demands are unidirec- tional, search is directed at identifying shippers with demands that would fill the truck for the return trip (the "backhaul").

Dispatchers and brokers play a crucial role in identifying complementary hauls and arranging matches. Dispatchers work for carriers, and seek to match hauls to trucks within their car- rier's fleet. Brokers seek to match hauls to trucks owned by other parties. These parties acquire knowledge about city-pair demand in a two-stage process: they make long-run invest- ments in learning general demand patterns (e.g., who the demanders are), then learn detailed "on the spot" information about short-run demands by contacting shippers' traffic managers period- ically throughout the day.

Search for complementary hauls in the short run tends to be more refined, and hence produc- tive, the more precisely parties can forecast when trucks will come free. This, in turn, leads to better matches between trucks and hauls. For example, backhauls may begin sooner after "fronthauls" end, and trucks may arrive to be loaded closer to when shippers want them. Thus, a second drawback to giving dnvers ser- vice responsibilities on a haul is that service interferes with search for the following haul; trucks' availability is more predictable follow- ing lowservice than high-service hauls.6

In interviews, fleet managers and dispatchers indicated to us that forecasting how long deliveries take is much easier when drivers have fewer service responsibilities. They indicated that they could forecast how long a no-service delivery of a truckload of packaged goods would take within a half-hour window, but could only forecast how

C. Asset Ownership and Incentives

Shippers' make-or-buy decision corresponds to whether they use a truck from their internal fleet or an external fleet for a haul. Industry participants distinguish between private and for-hire carriage by who has control rights over the truck.' Below we discuss how and why asset ownership affects incentives.

Ownership rights over trucks matter because contracts are incomplete with respect to trucks' schedules. In particular, shippers and caniers do not write fully contingent contracts with respect to trucks' schedules because the relevant contingen- cies are costly to idenw ex ante and venfy ex post. To see this, consider one class of scheduling decisions: how long a truck should wait at the loading dock to be loaded. A fully contingent contract would stipulate how long trucks should wait as a function of all relevant states of the world, including especially those factors affecting the benefits of delay and individual trucks' oppor- tunity cost. Many of these factors are known only to shippers andlor caniers and are difficult to venfy by outsiders. It is thus prohibitively costly to make contracts contingent on them. Schedule- setting is therefore a residual right of control that is, by definition, held by the truck's owner?

The contractual incompleteness surrounding truck scheduling leads to the main consequence of the allocation of ownership rights. In private carriage, shippers own trucks: if they want to alter trucks' schedules in ways that do not vio- late existing agreements, they can do so. They can unilaterally require that a truck picking up or delivering goods wait, for example. In for-

long a high-service delivery would take within a two- to three-hour window.

'Trucks in private fleets are sometimes leased, are some- times driven by short-term employees, and sometimes haul other shippers' goods (such as on backhauls). The distinc- tion between private and for-hire camage thus does not correspond to residual clairnancy, the length of labor con- tracts, or exclusivity of use.

In practice, it is common for contracts between ship- pers and carriers to have clauses that penalize shippers when they delay trucks. The penalties, however, are not state dependent, and thus are set intentionally high to deter ship- pers from delaying trucks in states of the world where trucks' shadow value is high. Parties realize that renegoti- ation is likely to be efficient when trucks' shadow value is low, creating a situation that is analytically similar to those where schedules are noncontractible.

hire carriage, shippers do not own trucks. If shippers want to change trucks' schedules, they must negotiate this with carriers.

The possibility that schedules will have to be renegotiated leads to familiar sorts of transac- tions costs in for-hire carriage. Both parties have an incentive to improve their bargaining position, and thus engage in rent-seeking behav- i~r.~

For shippers, this takes the form of iden- tifying other carriers who could serve them on short notice; for carriers, this takes the form of identifying other local shippers with similar demands-finding substitute hauls. Exploring back-up plans expends real resources, and is costly. In private carriage, by contrast, disputes may arise between shippers and their private fleets' dispatchers (or shippers and brokers), but identifying other ways to use trucks does not improve dispatchers' or brokers' bargaining po- sition because they cannot threaten to use trucks for other hauls. Neither private fleet dispatchers nor brokers have incentives to identify substi- tute hauls for rent-seeking purposes.

While rent-seeking tends to be greater under for-hire carriage, truck utilization also tends to be higher. One reason has to do with firms' incentives to obtain market information and search for complementary hauls. Firms can search more effectively for complementary hauls in the short run if they have previously made investments (in the form of customer re- lationships and general knowledge of demand) in particular markets. Shippers, for-hire carriers, and brokers can all potentially make such in- vestments. But because these investments are more valuable to those who are frequently look- ing for backhauls, individual shippers will tend to only make significant investments on city- pairs where their trucks haul high volumes of goods regularly. On other routes they will invest less, have less information about demand, and therefore search less productively in the short run than for-hire carriers or brokers, who ex- ploit increasing returns by utilizing knowledge across many shippers' hauls. Intermediaries thus have a comparative advantage in finding complementary hauls in many circumstances.

See Grossman and Hart (1986); Milgrom and John Roberts (1990). Baker and Hubbard (2000) argue that this incentive is also central for understanding why auck drivers tend not to own the trucks they operate.

This alone need not imply that truck utilization is necessarily lower under private carriage, since shippers could rely on brokers to find hauls. However, brokers have weaker incen- tives to find particularly good matches, because they do not own trucks and are thus less able to appropriate as large a share of the value that they create. The combination of strong incentives to learn about demand and strong incentives to find good matches for particular trucks leads matches to be better, and thus truck utilization to be lugher, under for-hire than private carriage.

Another reason why truck utilization tends to be higher in for-hire carriage is that drivers are generally assigned fewer service responsibili- ties. Trucks spend more time on the road and, as noted above, load matching is easier when driv- ers' responsibilities are narrow.

The next section develops a model of asset ownership and job design that captures the in- stitutional features described above and analyzes organizational relationships formally. This model generates comparative static pre- dictions that explain several important cross- sectional patterns in the industry. It also generates predictions regarding how changes in the informational environment should affect the make-or-buy decision. Later in the paper, we take these predictions to the data.

11. A Model .of Asset Ownership and Job Design

The model combines elements of Holmstrom and Milgrom (1991, 1994) and Grossman and Hart (1986). We embed multitask models of driver job design and dispatcher effort toward finding hauls into a setting in which noncon- tractible truck scheduling problems make asset ownership important. The timing follows. Ini- tially, a shipper's "fronthaul" and a matching truck are assumed to exist: we do not model the process of matching fronthauls to trucks. This haul may be one for which the value of service is high or low. We assume that parties cannot write a complete contract with respect to this haul ex ante. Organizational form is then cho- sen; at this point, asset ownership and drivers' job design are determined. Next, search for complementary backhauls (and possibly substi- tute fronthauls) occurs. Depending on asset ownership and the organizational form chosen, either a carrier or a broker chooses how much to search for hauls that complement or substitute for the shipper's haul. Parties then bargain; this determines which haul the truck is used for and how the surplus is split. Finally, production takes place (including provision of service by the driver) and payoffs are realized.

Complementarities between job design and asset ownership are critical to the results, and are a central feature of our model. To highlight this relationship and simplify the exposition, we develop a model first of driver job design, then overlay the shipper's "make-or-buy" decision. When shippers own trucks, this corresponds to "make"; when they do not, this corresponds to "buy." For simplicity, we assume that under the "make" option, shippers use brokers to find backhauls rather than find them themsel~es.'~ We begin with a model of driver job design.

A. Driver Job Design: Driving and Service Provision

Let s be the scope of the driver's activities, and m be the marginal product of this scope." For some hauls and shippers, service activities are valuable (high m), and for some they are less valuable. Motivating high service levels is costly, since it involves monitoring the mix of activities that the driver is performing. Let abe a parameter that captures the ability of the carrier to monitor the driver's efficiency in performing high-service activities: the higher is a,the lower is the marginal cost of monitoring. We speclfy V, the value of using the truck and driver for the shipper's haul, as:

where Ij is a fixed quantity, s is the scope of the dnver's activities, m is the margmal product of this scope, u is the degree to which the carrier can monitor driver activities, and M(s, a) is agency costs. We assume M, > 0, M2 < 0, M12< 0.

'O We present a more general model in which private carriers might use their own dispatchers to search for back- hauls in an earlier version of the paper (Baker and Hubbard, 2002). None of the comparative statics presented below differ in this more general model.

l1 Our equation of scope with service levels reflects an (unmodeled) assumption that some significant amount of driving is always part of the driver's job: the driver is never doing mostly service. Thus, more service involves a greater mix of activities.

Given this setup, the optimal amount of scope in the driver's job depends on the costs and benefits of such scope. Assuming an interior solution, optimal job design sets scope such that m = M,(s*,a). Raising the marginal product of scope (raising m)or raising the firm's ability to monitor driver activities (raising a)raises the optimal amount of scope. We assume that this expression is invertible, so that we can express the result as s* = +(m, a).

B. Load Matching

Following the discussion in Section I, we assume that search for complementary hauls adds value. Value is increasing in search levels because more effort produces better matches. We also assume that the marginal productivity of search is reduced when drivers are assigned more service-oriented activities.

We specify the value added of search for complementary hauls as:

where el is the effort toward finding comple- mentary hauls and g ,is the marginal product of thls effort for hauls involving no service. 0 captures the extent to which high service levels reduce the marginal product of search, 0 3 0. We also assume 0 < gll+(m, a); this regular- ity condition ensures that the marginal benefit of searching for complementary hauls is positive at the optim~m.'~

We specify the cost of search- ing for complementary hauls as C,(el) = e:/2. The expression for V, now including the value of the complementary haul, becomes:

C. Bargaining, Truck Ownership, and
Residual Rights of Control

The timing of the model is such that carriers and brokers can search for alternative uses of the truck before they negotiate with shippers over the

This guarantees that g, -Os* is nonnegative in the results below. The condition ensures that benefits of service are never so high so that the direct benefits of searching for complementary hauls are overwhelmed by its indirect costs.

terms of trade. These activities yield potential uses of the truck that are close substitutes for the ship- per's haul. For simplicity, we assume that this search is over alternatives that involve the same level of driver service, but that using the truck and driver for the alternative is always less valuable than using them for the first shipper's haul (per- haps because the alternative haul's origin is more distant). Assume that the value created when the truck is used for an alternative shipper's haul is:

where e2 represents effort toward finding alter- native hauls and g2 represents the marginal pro- ductivity of this effort.13 This formulation assumes that el, the effort that the dispatcher expends toward finding hauls that complement the first shipper's hauls, is equally valuable for the alternative shipper's hauls (e.g., the back- haul she finds would complement either out- bound haul.). We specify the cost of searching for substitute hauls as C2(e2) = ei12.

We can now calculate the amount of search when carriers or brokers search for hauls. We assume that when shippers bargain with either for-hire carriers or brokers over the surplus, they split the difference between the value of the haul and the value of the carrier's or bro- ker's outside alternative. A for-hire carrier's outside option is equal to P, the value of using the truck for an alternative shipper's haul. A broker does not have this outside option, be- cause it does nor own trucks. We therefore normalize brokers' outside option to zero.14

A for-hire carrier chooses el and e2 to maximize:

l3Thus, T/I is the value of the first shipper's haul (net of service) and g,e, is the value of the alternative haul. We assume y > l/2 g:.

l4 This is a simplification: brokers might get some value from searching for substitute fronthauls. What is required for the model is that the marginal returns to searching for substitute fronthauls are lower for brokers than for carriers, which seems reasonable given the difference in residual control rights.

This yields search effort equal to:

If search is completed by a for-hire carrier, it will search both for hauls that complement and substitute for the shipper's. Total value, which equals V less search costs, under this organiza- tional alternative is:

A broker chooses el and e, to maximize:

yielding effort of:

Effort levels are lower under private carriage; brokers search less intensively for complements, and not at all for substitutes. Total value under private carriage is:

3

TVp= T/'+ ;(g, -0s)'+ ms -M(s, o).

D. Eficient Organizational Forms: Job Design and Asset Ownership

In order to COW the total value created by private carriage versus for-hire carriage, we introduce an index variable, S, that indicates asset ownership. S = 1 indicates for-hire carriage, S = 0 private carriage. Total value as a function of s and S is:

PROPOSITION 1: TV(s, 6) is supermodular in (-s, -m, 6, -a, g,, -g2) on the domain where s r 0, 6 E (0, 11, and 0 < 0 <

g1/4(m, a).

PROOF:

Supermodularity requires that TV has nonde- creasing differences in (-s, -m, 6, -a, g,, -g2); this is equivalent to nonnegative cross- derivatives when TV is continuously twice- differentiable. (Donald M. Topkis, 1978). All terms except the second term are supermodular in (-s, -m, 6, -IT, g,, -g2) on this domain by inspection. The second term is supermodular if g, -0s r 0, which is guaranteed if 0 < g,l+(m, a). The sum of supermodular func- tions is supermodular.

This result allows us to apply a theorem from Topkis (1978) (see also Theorem 5 of Milgrom and Christina Shannon, 1994), and generate a set of monotone comparative statics that we can test with data on asset ownership and technol- ogy adoption.

PROPOSITION 2: -s* and 6* are monotone nondecrearing in (-m, -a, g,, -g2) on the domain where s r 0, 6 E {0, 11, and0 < 0 < g,lc$(m, a). -s'* and S* are (weak) complements.

Propositions 1 and 2 generate predictions that are consistent with several well-known cross- sectional patterns in the industry.

One simple prediction is that s and 6 should be inversely correlated: that is, high service should be associated with shipper ownership of trucks. This is consistent with the stylized fact that drivers in private fleets engage in more service-related ac- tivities than drivers in for-hire fleets. It is also consistent with the assertion made by many shippers that attainment of better service is why they choose private carriage over for-hire carriage.''

A second prediction is that S should be high

15 G.

[Tlhere are some good reasons why private carriage remains attractive to companies. Service is the key consid- eration. Many companies claim they require a private fleet to provide the high levels of service their customers expect. 'There are companies that decided to outsource their entire fleet, yet came running back to private fleets when the service was not what they expected,' says [John McQuaid of the National Private Truck Council]" (Jim Thomas, 1998).

when g is high: that is, for-hire carriage should be more prevalent when effort toward identify- ing complementary hauls is particularly valu- able. This is consistent with the stylized fact that for-hire carriage tends to be used more for small shipments and long-distance shipments than large and short-distance shipments. [See Bureau of the Census (1999b) and Hubbard (2001a) for empirical evidence.]

A third cross-sectional prediction concerns the adoption of different types of on-board com- puters. As we discuss in more detail below, OBCs have different informational capabilities. Certain simple devices (called trip recorders) allow fleet owners to monitor the actions of drivers ex post; more advanced devices (elec- tronic vehicle management systems, called EVMS) also allow them to track trucks' loca- tion in real time. The model predicts that the value of these different informational capabili- ties should differ between private and for-hire carriage: increasing the contractibility of driv- ers' actions should be more valuable in private carriage, while capabilities that raise the returns to searching for complementary hauls should be more valuable in for-hire carriage.16 As a con- sequence, private fleets should be differentially likely to adopt trip recorders and for-hire carri- ers should be relatively more likely to adopt EVMS. Hubbard (2000) tests this prediction and finds exactly this pattern. He shows that, in 1992, adoption rates for trip recorders and EVMS were (respectively) 8.8 percent and 5.8 percent for private carriers, and 6.5 percent and

15.4 percent for for-hire carriers. This evidence provides important support for the model that we propose, suggesting the relevance of the agency and ownership issues that we highlight.

Our main empirical tests, however, examine relationships between informational improve- ments enabled by the adoption of on-board computers (OBCs) and changes in ownership. These exploit the predictions that increasing g, should lead firms to (weakly) increase 6, and increasing a should lead firms to (weakly) de- crease 6. If the productivity of searching for complementary hauls (g ,) increases as a result

l6 Increasing u raises total value more when -s, and thus its complement 6, is low than hlgh, and increasing g, does so more when 6 is high than low.

of improved information technology, this should lead to two changes: a shift from private to for-hire carriage and a decrease in the scope of drivers' activities. If firms' ability to monitor the allocation of drivers' effort (o) increases, this should lead directly to increases in the scope of drivers' activities and indirectly to more shipper ownership of trucks.

Proposition 2 implies that sometimes changes in the model's parameters may not result in changes in the optimal organizational structure. One case is of particular interest to us. If m = 0, it is optimal to set s = 0 because there is no benefit from giving drivers service responsibil- ities. If this is true, the total value function is:

If m = 0, TV is independent of a: there is no multitasking, and no multitasking-related agency problem. Therefore, if m = 0, changes in firms' ability to monitor the allocation of drivers' effort (o) should have no effect on asset ownership.

The following section describes OBCs in more detail and generates empirical proposi- tions relating OBC adoption to ownership changes.

III. On-Board Computers and
Organizational Change

Two types of OBCs began to diffuse in the trucking industry in the late 1980's: trip record- ers and electronic vehicle management systems (EVMS)." Trip recorders measure trucks' op- eration. They record when trucks are turned on and off, their speed, sudden accelerations or decelerations, and various engine performance statistics (e.g., fault codes). Dispatchers and fleet managers receive the information trip re- corders collect when drivers return to their base at the end of a trip. Drivers give dispatchers a floppy disk or a similar device. Dispatchers upload the information onto a computer, which processes the information and provides reports. These reports indicate how drivers operated the

l7 See also Baker and Hubbard (2000) and Hubbard (2000).

truck; for example, how quickly they drove, how long they allowed trucks to idle, and whether there were any nonscheduled stops. They also indicate how long drivers spent at each stop.

EVMS record the same information trip re- corders do, but provide three additional capa- bilities. One is that they record trucks' geographic location, often using satellite track- ing systems. Another is that they can transmit any information they collect to dispatchers in real time. Dispatchers can thus know where trucks are at any point in time. Third, they provide dispatchers a way of initiating commu- nication with drivers. For example, dispatchers can send a text message that updates drivers' schedule. If the message is complicated, dis- patchers can send a message that asks drivers to call in. This is a significant advance over the system firms have traditionally used to commu- nicate with drivers who are outside radio range (about 25 miles). Traditionally, firms require drivers to call in every three or four hours. This requires drivers to frequently pull over, stop, and find a phone, even though much of the time neither dispatchers nor drivers have new information to communicate. Without EVMS, dispatchers often find it hard to verify trucks' location and must wait for distant drivers to call in before they can communicate instructions.

As Hubbard (2000) relates, there is an eco- nomically important distinction between these two devices. Trip recorders are useful for im- proving incentives, because they provide veri- fiable information about how trucks were operated. Importantly for this paper, they mon- itor how long drivers spent driving and how long they spent performing other tasks: this helps mitigate the agency problems associated with more complex job designs.'' Trip record- ers are not generally useful for improving resource-allocation decisions ("coordination"). They do not improve dispatchers' ability to match trucks to hauls in the very short run because they do not supply information in a timely enough fashion. They are generally not

"An advertised benefit of hip recorders is their ability to help monitor drivers in this way. For example, Atrol claims that its devices can "tell you how effective your drivers are in managing their time" (www.atrol.com).

used to improve routing decisions made over the longer run-for example, by helping bench- mark routes-because firms usually can obtain information about such things as how long routes take by other, less costly means.19

In contrast, EVMS are useful for improving both incentives and coordination. Their addi- tional capabilities help dispatchers match trucks to hauls better, thereby increasing capacity uti- lization. Real-time information about trucks' lo- cation helps them schedule backhauls more efficiently, for example.20 These capabilities also enable them to communicate schedule changes to drivers in real time. Dispatchers can quickly reroute trucks in response to changes in market conditions. For example, suppose a truck on the road is half-full. If a dispatcher can find a shipper with cargo that can fill the truck, he can send a message to the driver asking him to make an additional pick-up and delivery.

We next discuss our main empirical proposi- tions, which predict how OBC adoption should affect truck ownership. These propositions are based on the premise that trip recorder adoption increases a and EVMS adoption increases both u and g,.

PI: Overall, trip recorder adoption should lead to more shipper ownership of trucks.

OBCs' incentive-improving capabilities al- low carriers to better monitor how drivers allo- cate time, and thus effort, across tasks. Trip recorder adoption thus raises u,which by Prop- osition 2 increases the optimal choice of s and decreases 6; carrier ownership of trucks should decrease. We cannot test whether trip recorder adoption increases s because the data do not

l9 Many firms use software packages to help dispatchers schedule trucks. These packages often use information EVMS collect (for example, trucks' location), but rarely use the information hip recorders collect.

20 Trade press articles and advertisements emphasize this. An example of a quote from a driver: "Dispatch knows where I am and where I'm headed so before I even get to my destination, they can plan ahead. Quite often I get a load offering over my Qualcomm system before I'm even empty" (www.qualcomm.com). Empirical evidence of EVMS' impact on capacity utilization is in Hubbard (2003). who finds that EVMS has increased loaded miles among adopters by 13 percent as of 1997.

contain information on the scope of drivers' activities, but we can test whether it leads to more shipper ownership of trucks.

P2: EVMS adoption should lead to less of an increase in shipper ownership of trucks than trip recorder adoption, and may lead to less shipper ownership of trucks.

EVMS' coordination-improving capabilities make dispatchers' search more productive, and thus raise g ,. Knowing where trucks are allows dispatchers to better anticipate when trucks will come free, and hence helps them refine their search. Being able to initiate communications with drivers while they are in their cab en- ables them to better exploit the opportunities they identify. For example, they can quickly reallocate drivers and trucks across hauls in response to new opportunities. Because EVMS contain both incentive- and coordination- improving capabilities, EVMS adoption should increase both a and g, and thus has a theo- retically ambiguous impact on asset owner- ship. However, because EVMS adoption increases a in the same way trip recorder adoption does, EVMS adoption should move hauls less toward private carriage than trip recorder adoption.

P3: Trip recorder adoption should increase shipper ownership of trucks more when drivers' cargo-handling activities are potentially pro- ductive than when they are not productive. It should not affect whether shippers own trucks when drivers' handling activities are not pro- ductive.

Trip recorder adoption should not lead to ownership changes when m = 0: for example, for hauls of bulk goods or goods that require people other than drivers to load and unload. It should lead to ownership changes when m > 0. From above, this should be the case when trucks haul packaged goods, especially when they pick up or deliver to small outlets. It should also be the case when trucks haul goods for which handling requires certification, such as petro- leum or chemicals. However, it should not be true for hauls of bulk goods or goods that cannot be lifted or transported with standard equipment.

IV. Data

The data are from the 1987, 1992, and 1997 Truck Inventory and Use Surveys (TIUS)." The TIUS is a mail-out survey of trucks taken by the Census as part of the Census of Trans- portation. The Census sends forms to a random sample of truck owners. These forms ask ques- tions about individual trucks' characteristics. Truck owners report the truck's type (pick-up, van, tractor-trailer, etc.), make, model, and many other characteristics. The TIUS also asks how trucks are equipped, including whether they have trip recorders or EVMS installed, and how they are used. Owners report how far from home individual trucks generally operated, the type of trailer to which they were typically attached, the class of product they generally hauled, the state in which they were based, and whether they were used for for-hire or private carriage. Publicly available data from the sur- vey do not identify trucks' owners because of confidentiality restrictions. This paper uses only observations of truck-tractors (the front halves of tractor-trailers) and excludes those that were generally operated off-road, carried household goods (i.e., moving trucks), or were attached to trailers that do not haul goods (e.g., trailers with large winches permanently attached). Eliminat- ing these observations leaves 21,236, 32,015, and 18,856 observations of tractor-trailers in 1987, 1992, and 1997 respectively. This is over 85 percent of the tractor-trailers in the original samples.

Figure 1 shows private carriage shares in each of the three years. In each of these years, the overall share is about 50 percent and is higher for shorter hauls than longer ones. The overall share fluctuated during this period, in- creasing from 50.1 percent to 54.6 percent be- tween 1987 and 1992, then falling back to 51.7 percent in 1997. The time trends differ for hauls of different lengths. The private carriage share increased for all distances between 1987 and 1992. It increased for short hauls but declined for medium and long hauls between 1992 and 1997. This paper's empirical tests examine how

See Bureau of the Census (1995, 1999a), Baker and Hubbard (2000), and Hubbard (2000,2001) for more on the TIUS. The 1997 survey is actually called the Vehicle Inventory and Use Survey.

Positive number of observations in:

Cohorts Observations/cohort, 1987 Observations/cohort, 1992 Observations/cohort, 1997 Private caniage share, 1987 Private carriage share, 1992 Private carriage share, 1997 Trip recorder adoption, 1987- EVMS adoption, 1987-1992 Trip recorder adoption, 1992- EVMS adoption, 1992-1997

Notes: Averages are computed using weights, where weight = (numobs87*expanf87 + numobs92*expanf92 + numobs97*expanf97)/3 for samples using all three years. Analogous weights are used for samples that use only two of the three years.

trailers, logging trailers, and specialized plat- form types. An implication is that the cohort sample contains a higher fraction of long-haul trucks than the population because hauls using specialized trailers tend to be short. The other is that, conditional on distance, trucks attached to refrigerated vans make up a disproportionate share of the cohort sample: about 20 percent rather than their 10 percent share in the original sample. The reason for this is refrigerated vans almost exclusively haul a single product class: processed food. Refrigerated van cohorts tend to be larger and are less likely to have zero observations than cohorts associated with trail- ers that haul multiple product classes.

Table 1 contains summary statistics for the cohort sample. Cohorts tend to be based on relatively few obseryations due to our narrow cohort definition: the number of observations per cohort is less than ten in each year.24 The average private carriage share is about 50 per- cent and average OBC adoption rates are similar to those in Figure 2. Table 2 provides evidence of relationships between technological adoption and organizational change at the cohort level.

24 In earlier versions of this paper, we reported estimates of our main specifications using the subsample of cohorts where the private and for-hire carriage shares are positive in each year. The average cohort size is about double in this subsample, but observations in these cohorts make up only about 35 percent of the original sample. We showed that our main results do not change.

The top panel uses cohorts with positive obser- vations in both 1987 and 1992. The first row indicates that averaging across cohorts, the pri- vate carriage share increased from 0.49 to 0.50 between 1987 and 1992. The next three rows split the cohort sample according to OBC adop- tion. On average, the private carriage share stayed the same for cohorts with low OBC adoption. Among cohorts with high OBC adop- tion, the private carriage share increased for those where trip recorder adoption was high but decreased slightly for those where EVMS adop- tion was high. The bottom panel reports results from a similar exercise that analyzes patterns between 1992 and 1997. The private carriage share decreased for the low OBC and high EVMS adoption cohorts (slightly more for the latter), but increased for high trip recorder adop- tion cohorts.

In sum, relationships between OBC adoption and organizational change differ for trip record- ers and EVMS. Cohorts with high trip recorder adoption moved toward private carriage more than cohorts with low OBC adoption. Cohorts with high EVMS adoption moved toward for- hire caniage slightly more than those with low OBC adoption, but this difference is very small. Nevertheless, the fact that cohorts with high EVMS adoption did not move toward private car- riage is interesting in light of the fact that EVMS enable the same contractual improvements trip recorders do. This suggests EVMS' resource- allocation-improving capabilities-which trip

TABLE2-PRIVATE CARRIAGESHAREAND OBC ADOFTION
Cohort Data
Mean private carriage share OBC adoption, 1987-1992 1987 1992 Change Trip recorder EVMS N

Mean private carriage share OBC adoption, 1992-1997 1992 1997 Change Trip recorder EVMS N

Notes: The top (bottom) panel includes all cohorts with a positive number of observations in both 1987 and 1992 (1992 and 1997). Low OBC adoption cohorts are those where OBC adoption was less than 0.15. High TR adoption cohorts are those where OBC adoption was greater than 0.15, and TR adoption was greater than EVMS adoption. High EVMS adoption cohorts are those where OBC adoption was greater than 0.15, and EVMS adoption was greater than TR adoption.

recorders do not have-have organizational im- sity is the number of trucks in the state attached plications that offset those of their incentive- to a given trailer type, normalized by the state's improving ones. urbanized area, and is a proxy for local market

thickness for hauls using a particular trailer

V. Results type. We allow the coefficients on the dry van and auto trailer dummies to vary across years to Our base specification takes the form: account for secular changes in contractual form over time (see Hubbard, 1998). Yit = xitP f 4i + &it Most of our results will be from first-

difference specifications: where y,, is the for-hire carriage share in cohort i at time t, xi, includes a vector of explanatory variables, and c$i and represent unobserved time-invariant and time-varying variables that where qit = -E~(,-First-differencing affect optimal organizational form. The vari- mitigates an important class of endogeneity ables of interest in xi, are OBC, the share of problems that would appear in cross-sectional trucks with either class of OBC installed, and analysis. For example, suppose that when ship- EVMS, the share of trucks with EVMS in- ping patterns are regular, private carriage tends stalled. The coefficient on OBC therefore picks to be used more (perhaps because intermediar- up the relationship between OBCs' incentive- ies' efforts are less valuable) and trip recorders improving capabilities and asset ownership and are dispropomonately valuable relative to EVMS. that on EVMS picks up the relationship between This would lead private carriage and trip re- EVMS' coordination-improving capabilities and corder use to be correlated in the cross section asset ownership. The control variables in xi, are even if trip recorders adoption did not cause similar to those in Table 5 in Hubbard (2001). truck ownership to change. First-differencing They include a full set of dummy variables that effectively allows us to control for unobserved indicate the cohort's trailer type (dry van, re- time-invariant variables that could affect OBC frigerated van, tank truck, etc.), a dummy that adoption and organizational form indepenequals one if the cohort is of trucks hauling dently, and base inferences on relationships be- mixed cargo, and ln(trai1er density). Trailer den- tween adoption and changes in asset ownership

TABLE3--0BC ADOFTIONAM) ASSETOWNERSHIP

Nores: SUR estimates. Sample includes all cohorts with positive number of observations in 1987, 1992, and 1997; N = 2,773. Cohorts are weighted using Census' weighting factors times number of observations. Specifications include trailer dummies, mixed cargo dummy, distance dummies, and ln(trai1er density) as controls, and allow the coefficient on the auto trailer and van dummy to vary across years to account for secular changes.

rather than levels.25 Thus, if hauls' unobserved regularity is constant over time, first-differencing eliminates this as a possible endogeneity prob- lem. For simplicity, our initial discussion of the results will assume qir to be independent of adoption: that is, we assume changes in unob- served haul characteristics to be independent of adoption. Later we will relax this assumption and present instrumental variables estimates of the first-difference specifications.

Table 3 contains two sets of regression esti- mates that use cohort data from 1987, 1992, and 1997. The first column presents levels esti- mates; the second presents first-difference esti- mates. The coefficients on OBC are negative and significant and those on EVMS are positive and significant in both columns. They are about 40 percent lower in absolute value when mov- ing from the levels to the first-difference esti- mates, but remain statistically significant and economically important.

These estimates supply evidence consistent with our main propositions. Consistent with PI, trip recorder adoption is associated with move- ment from for-hire to private carriage. Consis- tent with P2, EVMS adoption is less associated with such movement. Assuming these reflect

25 The control variables in the first-difference specifica- tions only include the dry van and auto trailer dummies and ln(trai1er density) because none of the coefficients on the other controls vary over time. Changes in unobserved cohort characteristics may either reflect true changes or sampling error. See Angus Deaton (1985).

causal relationships, they indicate that OBCs' incentive- and coordination-improving capabil- ities affect the make-or-buy margin differently. Their incentive-improving capabilities move hauls from "buy" to "make"; their coordination- improving capabilities move them from "make" to "buy." The former shifts truck ownership from for-hire carriers to shippers; the latter from shippers to for-hire carriers.

The first-difference estimates imply that all else equal, cohorts with a 20-percentage-point higher trip recorder adoption rate experienced a 2-percentage-point greater increase in their pri- vate carriage share. Likewise, moving 20 per- cent of trucks from trip recorders to EVMS corresponds to a 3-percentage-point increase in the for-hire carriage share. These magnitudes suggest the OBCs' incentive- and coordination- improving capabilities have an economically important impact on make-or-buy decisions in the aggregate, in light of the industry's recent history. We discuss magnitudes at more length in a subsection below.

Table 4 breaks these results down further. The top panel reports first-difference estimates using all cohorts and subsarnples of short-, medium-, and long-haul cohorts. The coeffi- cient on OBC is negative for all three sub- samples and statistically significant for medium- and long-haul cohorts. The EVMS coefficient is positive and significant for all three subsarnples, and of about the same mag- nitude. The basic relationships between adop- tion and changes in asset ownership hold across hauls of different distances.

The bottom part of the table estimates the coefficients using only 1987 and 1992, then only 1992 and 1997 data. In the earlier period, the pattern of a negative coefficient on OBC and a positive one on EVMS only appears within the long-haul subsample. In contrast, this pattern appears strongly and consistently in the later period. In the later period, the coefficients on OBC are negative in each sub- sample, and statistically significant for the short- and medium-haul subsample. Those on EVMS are positive and significant in each subsample. The cross-year differences are in- teresting because they are consistent with widespread speculation that organizational changes tend to lag IT adoption, even when they are complementary.

TABLEABC ADO~ON

AND ASSETOWNERSHIP All Short hauls Medium hauls Long hauls

Dependent variables: Change in for-hire carriage shares, 1987-1992, 1992-1997

Notes: SUR estimates. Includes all cohorts with positive number of observations in each relevant year. Cohorts weighted by Census' weighting factors times number of observations. Specifications include change in trailer density and auto trailer and van dummies as controls, and allow the coefficient on the auto trailer and van dummy to vary across years to account

for secular changes.

A. Interactions: Multitasking Tests

In the model, trip recorders affect optimal asset ownership indirectly, by lowering the agency costs associated with multitasking. If so, then the OBC coefficient should only be nega- tive for hauls where drivers' cargo-handling effort is potentially productive. This is the basis of P3 above. To examine this, we create inter- actions between OBC adoption and product cat- egories. One set is between adoption and a dummy variable that equals one if the cohort hauls processed food or mixed cargo. Trucks hauling processed food or mixed cargo tend to deliver packaged goods to retail outlets. Driv- ers' cargo handling efforts are potentially more valuable when they haul these goods than other, bulkier goods. The other set is between adop- tion and a dummy variable that equals one if the cohort hauls petroleum or chemicals, cargo for which handling requires certification. We there- fore test whether the OBC coefficient is more negative for these "multitasking" cohorts than others.

Table 5 summarizes the results. The first col- umn uses data from all three years. The coeffi- cient on OBC alone is small and statistically insignificant. There is no relationship between OBC adoption and asset ownership when trucks haul goods in the omitted category, which con- tains raw materials and bulky goods.26 The in- teractions on OBC*(food or mixed cargo) and on OBC*(petroleum or chemicals) are both negative, and the former is statistically signifi- cant. The other two columns report estimates using only two of the years. The OBC own effects are statistically zero in both periods. Both interactions are negative and significant in the late period of the sample. The estimates provide support for P3, and are important evi- dence that OBCs' incentive-improving capabil- ities affect asset ownership through job design. There is no evidence that incentive improve- ments affect the make-or-buy decision for hauls

26 The most prevalent product classes in the omitted category are fresh farm products, building materials, ma- chinery, and lumber and wood products. About 70 percent of cohorts are in the omitted category; about 30 percent are in the "multitasking" categories.

TABLE5-4BC ADOPTIONAND ASSET OWNERSHIP-INTERACTIONS Multitasking Tests

Dependent variables: Change in for-hire carriage share

1987, 1992, 1997 1987-1992 1992-1997

OBC EVMS OBC*(food or mixed cargo) EVMS*(food or mixed cargo) OBC*(petroleum or chemicals) EVMS*(petroleum or chemicals)

N

Notes: SUR estimates. Includes all cohorts with positive number of observations in each relevant year. Cohorts weighted by Census' weighting factors times number of observations. Specifications include change in trailer density and auto trailer and van dummies as controls, and allow the coefficient on the auto trailer and van dummy to vary across years to account

for secular changes.

where dnvers' handling efforts are rarely pro- ductive. When drivers' handling efforts tend to be productive, hauls for which trip recorders were adopted moved from for-hire to private carriage.

The first-difference estimates are thus consis- tent with all three of our theoretical proposi- tions. The following subsection provides additional evidence regarding whether they in- deed reflect causal relationships between adop- tion and organizational change.

B. Instrumental Variables Estimates

As noted above, alternative interpretations of the first-difference estimates center on the premise that OBC adoption and organizational changes might be independently affected by some omitted factor. For example, if hauls' regularity changes over time, and increases more in some cohorts than others, this could lead independently to more private carriage and to trip recorder adoption. This would make adoption econometrically endogenous in the first-difference estimates. Similarly, unobserved factors may also explain why EVMS adoption might be correlated with movements toward for-hire carriage. Suppose unobserved shipper characteristics change over time; some shippers may both establish more sophisticated logistics practices and begin to value shipment tracing capabilities. This could lead independently to less private carriage (because dnvers' handling effort is less valuable) and increased EVMS adoption.

Below we present instrumental variables es- timates of the first-difference specifications. Factors that affect OBC adoption but do not directly affect organizational form are good in- struments. We use four main instruments: the fraction of miles trucks are operated outside of their base state, the number of weeks per year trucks in the state are in use, and dummy vari- ables that equal one if the truck is based in a western state (i.e., west of Missouri), or in a New England state. These are computed at the cohort level; hence, the first two of these are cohort-level averages. Fraction of miles out of state affects OBC adoption because drivers must keep track of how many miles trucks are operated in each state. State fuel taxes are paid on this basis. OBCs let drivers enter in this information on a keypad and lower data entry and processing costs when trucks' owners cal- culate the tax they owe each state. This is more valuable for trucks that spend more time outside of their base state because they cross state lines more. State averages for number of weeks in use differ considerably across states, ranging between 35 and 45 weeks, and reflect differ- ences in the cyclicality of truck shipments. Much of this variation is likely climate related, as the bottom five states are Montana, Wyo- ming, North Dakota, Alaska, and South Dakota. Trucks are idled more weeks per year, and OBCs' benefits are correspondingly lower, in areas where shipments are highly cyclical. We assume that statewide averages in the number of weeks trucks are in use are unaffected by OBC adoption.27 The two regional dummies are in- cluded because it is traditionally more difficult for drivers to contact dispatchers quickly in less densely populated areas. This is one reason why adoption tends to be above average in the West but below average in New England. We use these four variables and their interactions as instruments for OBC and EVMS. Table A1 reports estimates from four simple "first-stage" specifications that regress cohort-level trip re- corder and EVMS adoption in the early and late sample periods on the four instruments and a vector of controls.

Table 6 contains estimates from specifica- tions analogous to that in the first column of Table 5, but which restrict the interaction terms to be the same across the four "multitasking" product types. Looking at the first column, the coefficient on OBC is nearly zero; as before, there is no evidence that trip recorders affect asset ownership for the "nonmultitasking" cohorts. The same is true for the EVMS*Mult interaction: there is no evidence that OBCs' coordination-improving capabilities' impact differs with whether multitasking is poten- tially productive. The negative coefficient on OBC*Mult suggests that trip recorders move hauls toward private carriage more for the mul- titasking cohorts than the nonmultitasking ones. The positive coefficient on EVMS suggests that EVMS' coordination-improving capabilities move hauls from private to for-hire carriage. However, these coefficients are not statistically significantly different from zero because they are not estimated precisely. The noisiness of the

''Individual trucks with OBCs do tend to be used more weeks than those without them, because hucks without OBCs are more likely to be idled when demand is low. But OBC adoption should have a minimal impact on number of weeks, averaged across all trucks in a state.

TABLECBC ADOPTIONAND ASSET OWNERSHIP GMM-IV Estimates

Dependent variables: Change in for-hire carriage share,

Notes: Instruments include percent of miles out of base state, number of weeks in use per year, West dummy, New England dummy, and interactions among these. Includes all cohorts with positive number of observations in each rele- vant year. Cohorts weighted by Census' weighting factors times number of observations. Specifications include change in trailer density and auto trailer and van dummies as controls, and allow the coefficient on the auto trailer and van dummy to vary across years to account for secular changes.

estimates reflects that while three of our instru- ments are predictors of OBC adoption in gen- eral, none of them shift trip recorder adoption but not EVMS adoption. (See Table Al.) This makes it hard to distinguish between the orga- nizational effects of OBCs' incentive-and coordination-improving capabilities in the in- strumental variables estimates.

The bottom part of the table contains estimates of (OBC+EVMS) and (OBC+EVMS)*Mult, which reflect EVMS' overall organizational irnpact. We can estimate these more precisely; con- sistent with our earlier results, EVMS adoption pushes hauls toward for-hire carriage, but does so less for the multitasking cohorts than the nonmul- titasking cohorts. The right column restricts the coefficient on OBC to zero for the nonmultitask- ing cohorts, a restriction suggested by the near- zero point estimates on this coefficient in Table

5. The sign and si@cance of the remaining three coefficients are the same as in the first-difference estimates.

In sum, our instrumental variables estimates provide no evidence that our first-difference estimates reflect noncausal relationships. Al-

True share

Estimated share, absent OBC adoption

Table 5 (right columns) 0.50 0.54 0.53 Table 6 (right column) 0.50 0.60

though we cannot statistically distinguish be- tween the impact of OBCs' incentive- and coordination-related capabilities to the same de- gree, the qualitative patterns that we are able to identify are similar to those in our earlier results.

C. Magnitudes

Although not the main focus of the study, the estimates also indicate the degree to which overall changes in the private carriage share between 1987 and 1997 were due to the diffu- sion of OBCs. Table 7 summarizes our analysis. The top line reports the actual private carriage shares in our sample in each of the three years. The bottom part of the table reports the esti- mated shares, absent OBC diffusion, computed using the simple and GMM-IV first-difference estimates from the right columns of Table 5 and the right column of Table 6, respectively. The simple first-difference estimates suggest that OBCs had little overall impact between 1987 and 1992-the diffusion of trip recorders and EVMS had offsetting effects- but caused about 1 percentage point of the overall 2.9-percentage- point decline between 1992 and 1997. The GMM-IV point estimates imply that OBCs' over- all impact was much larger. They indicate that absent OBC diffusion, the private carriage share would have continued to increase to almost 60 percent by 1997. One interpretation is that the organizational impact of EVMS' coordination-improving effects worked against a broad increase in the demand for high service levels.

We do not put a large weight on these quan- titative conclusions, in large part because the GMM-IV point estimates in Table 6 are noisy. We are more confident in stating the qualitative conclusion that overall, OBC diffusion played a significant and possibly large role in inducing shippers to outsource more during the 1990's.

VI. Conclusion

In this paper, we combine recent theoretical work from organizational economics with a de- tailed and disaggregated data set to gain insight about the interaction between asset ownership, job design, and information. Of particular im- portance is our ability to distinguish between informational changes that lead either to better monitoring of agents, or to better coordination of activities. We believe that our results-that improved monitoring technologies lead ship- pers to integrate into trucking, while technolo- gies that improve coordination lead to more outsourcing of trucking services-highlight the respective advantages of firms and markets in the economy, and thus shed light on their roles in the operation and development of economic systems.

In describing and explaining the development of nineteenth-century capitalism, Alfred P. Chandler (1977) and others have argued that the development of new communication technolo- gies (e.g., the telegraph) enabled the growth of large, integrated firms. Large transportation, manufacturing, and retailing firms were impos- sible without a technology that enabled manag- ers to coordinate large-scale economic activity. Yet we have found exactly the opposite effect in late twentieth-century trucking: a new commu- nications technology that improved coordina- tion led to smaller, less integrated firms. Why the difference?

We believe that our new results arise because we distinguish between informational changes that improve coordination from those that im- prove incentives. Such a distinction is rare in empirical work on organizations. Yet it is es- sential to understanding the true role of firms and markets as competing mechanisms for or- ganizing economic activity. F. A. Hayek (1945) argued that the true value of the market-based price system is its ability to utilize dispersed information about resources and coordinate their use in a way that no centrally planned economy (or firm) ever could. Given this com- parative advantage of markets over firms,it is not surprising that a technological change that mitigates the Hayekian coordination problem should lead to a greater relative improvement in the efficiency of markets.

Holrnstrom and Milgrom (1994) and Holm- strom (1999), by contrast, argue that the true advantage of firms over markets is their ability to craft delicately balanced incentives for agents engaged in multiple activities, in a way that the strong incentives generated by markets and as- set ownership cannot. Given this comparative advantage for firms, it is again not surprising that a technological change that mitigates con- tracting problems should lead to a greater rela- tive improvement in the efficiency of firms.

Information costs are at the core of nearly all economic theories of organizations. Thus, all of these theories predict that changes in information technology that change the cost of contracting and communication will affect the organization of economic activity. We find that the answer to the question: "Has IT adoption led to larger or smaller firms in trucking?'is "Yes," and show how the orga- nizational implications of IT'S incentive- and coordination-improving capabilities system- atically differ. Future research will further inform debates regarding the organizational implications of changes in information costs by investigating whether this systematic dif- ference in trucking is general.

TABLEA1-FIRST-STAGEESTIMATES: ADOITON EQUATIONS TR adoption TR adoption EVMS adoption EVMS adoption

Dependent variable: West New England Percent out of base state Weeks in use N = 2,773

Notes: All specifications include distance, trailer, and product class dummies and the private carriage share. Includes all cohorts with positive number of observations in each relevant year. Cohorts weighted by Census' weighting factors times nutnber of observations.

REFERENCES

American Trucking Associations. U.S.freight transportation forecast ... to 2007. Alexandria, VA: American Trucking Associations, 1999.

Baker, George P. "Incentive Contracts and Per- formance Measurement." Journal of Political Economy, June 1992, 100(3), pp. 598-614.

Baker, George P. andHubbard, Thomas N."Contractibility and Asset Ownership: On-Board Computers and Governance in U.S. Truck- ing." National Bureau of Economic Research (Cambridge, MA) Working Paper No. 7634, 2000.

. "Make Versus Buy in Trucking: Asset Ownership, Job Design, and Information."

National Bureau of Economic Research (Cambridge, MA) Working Paper No. 8727, 2002.

Bryqjolfsson, Erik and Hitt, Lorin. "Information Technology and Organizational Design: Some Evidence From Micro Data." Working paper, Stanford University, 1997.

Bureau of the Census. 1992 Census of transpor- tation: Truck inventory and use survey. Washington, DC: U.S. Government Printing Office, 1995.

. 1997 Census of transportation: Ve- hicle inventory and use survey. Washington, DC: U.S. Government Printing Office, 1999a.

. 1997 Census of transportation:

Commodity flow survey. Washington, DC:

U.S. Government Printing Office. 1999b.

Chandler, Alfred P. The visible hand: The man- agerial revolution in American business.

Cambridge, MA: Belknap, 1977.

Deaton, Angus. "Panel Data from Time Series of Cross-Sections." Journal of Econometrics, October-November 1985, 30(1-2), pp. 109

26.

Grossman, Sanford and Hart, Oliver. "The Costs and Benefits of Ownership: A Theory of Ver- tical and Lateral Integration." Journal of Po- litical Economy, August 1986, 94(4), pp. 691-719.

Hayek, F. A. "The Use of Knowledge in Soci- ety." American Economic Review, September 1945, 35(4), pp. 519-30.

Holmstrom, Bengt. "The Firm as a Sub-economy." Journal of Law, Economics, and Organization, April 1999, 15(1), pp. 74-102.

Holmstrom, Bengt and Milgrom, Paul. "Multitask Principal-Agent Analyses: Incentive Con- tracts, Asset Ownership, and Job Design." Journal of Law, Economics, and Organiza- tion, 1991, Spec. Iss., 7, pp. 24-52. . "The Firm as an Incentive System." American Economic Review, September 1994, 84(4), pp. 972-91.

Hubbard, Thomas N. "Governance Structure in the Deregulated Trucking Industry." Work- ing paper, UCLA, 1998. . "The Demand for Monitoring Tech- nologies: The Case of Trucking." Quarterly Journal of Economics, May 2000,11.5(2), pp. 533-76. ."Contractual Form and Market Thick- ness in Truclung." RAND Journal of Eco- nomics, Summer 2001, 32(2), pp. 369-86. . "Information, Decisions, and Produc- tivity: On-Board Computers and Capacity

Utilization in Truclung." American Economic Review, 2003 (forthcoming).

Jensen, Michael C. and MecMing, William H. "Theory of the Firm: Managerial Behavior, Agency Costs, and Ownership Structure." Journal of Financial Economics, October 1976, 3(4), pp. 305-60. ."Specific and General Knowledge, and Organizational Structure," in Lars Werin and Hans Wijkander, eds., Contract economics. Cambridge: Blackwell, 1992, pp. 25 1-91.

Leavitt, H. J. and Whisler, T. L. "Management in the 1980s." Harvard Business Review, November 1958, 36(6), pp. 41-48.

Malone, Thomas W.; Yates, JoAnne and Ben- jamin, Robert I. "Electronic Markets and Electronic Hierarchies." Communications of the ACM, June 1987, 30(6), pp. 484-97.

Milgrom, Paul and Roberts, John. "Bargaining Costs, Influence Costs, and the Organization of Economic Activity," in James E. Alt and Kenneth A. Shepsle, eds., Perspectives on positive political economy. Cambridge: Cam- bridge University Press, 1990, pp. 57-89.

Milgrom, Paul and Shannon, Christina. "Monotone Comparative Statics." Econometrica, January 1994, 62(2), pp. 157-80.

Ouellet, Lawrence J. Pedal to the metal: The work lives of truckers. Philadelphia, PA: Temple University Press, 1994.

Thomas, Jim. "Private Fleets: Down But Not Out." Logistics Management and Distribu- tion Report, July 1998, 37(7), pp. 3 1-32.

Topkis, Donald M. "Minimizing a Submodular Function on a Lattice." Operations Research, March-April 1978, 26(2), pp. 305-21.

Wilson, Rosalyn A. Transportation in America 2000, with historical compendium 1939-1999. Washington, DC: Eno Transportation Foun- dation, 2001.

Comments
  • Recommend Us