make versus buy in trucking: asset ownership, job design, and … · 2004-07-26 · asset...

Make Versus Buy in Trucking

Asset Ownership Job Design and Information

By GEORGE P BAKER AND THOMAS N HUBBARD

Explaining patterns of asset ownership is a central goal of both organizationaleconomics and industrial organization We develop a model of asset ownership intrucking which we test by examining how the adoption of different classes ofon-board computers (OBCs) between 1987 and 1997 in uenced whether shippersuse their own trucks for hauls or contract with for-hire carriers We nd that OBCsrsquoincentive-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 re ect the importance of both incompletecontracts and of job 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 bothorganizational economics and industrial organi-zation because it provides insights on rmboundaries and industry structure Majorprogress towards this goal was provided bySanford Grossman and Oliver Hartrsquos seminalpaper in 1986 which argues that asset owner-ship confers on owners residual rights of controlthat give them power and thus incentives todevote effort to value-increasing activities Inthis view rmsrsquo boundaries are determined bythe optimal allocation of these residual rights ofcontrol Bengt Holmstrom and Paul Milgrom(1994) however argue that rmsrsquo boundariesre ect trade-offs in which asset ownership in-teracts with job design and other organizationaldecisions If so rmsrsquo boundaries may re ectfactors that do not appear in Grossman andHartrsquos (1986) theory including those that affect

the optimal allocation of tasks across individu-als In 1999 Holmstrom offered a critique of theproperty rights view in which he argues that itfails to explain why rms rather than individu-als own assets He extends the insight from the1994 paper to argue that rms own assets pre-cisely because this mutes the incentives thatcome with individual asset ownership allowingthe rm to operate as a ldquosubeconomyrdquo that canmore precisely balance incentives and imple-ment more complex multitask job designs

In this paper we argue that the pattern ofasset ownership in truckingmdashin particular thedecision by shippers about whether to use theirinternal eet of trucks for a haul or contract withfor-hire carriersmdashre ects not only the factorsidenti ed in Grossman and Hartrsquos theory butalso those highlighted in Holmstrom and Mil-grom (1994) Consistent with the former own-ership patterns re ect trade-offs that arise fromproviding intermediaries strong incentives toidentify pro table uses for trucks Consistentwith the latter ownership patterns also re ectissues of job design ie the degree to whichdrivers simply drive trucks or provide a morecomplex combination of transportation andservice Job design matters because ldquoservice-intensiverdquo trucking hinders intermediariesrsquo abil-ity to nd pro table uses for the truck Shipperownership of trucks mutes incentives and favorsservice-intensive trucking in which driversrsquojobs involve more than just driving trucks

We develop a model that combines these

Baker Harvard Business School Boston MA 02163and NBER Hubbard University of Chicago GraduateSchool of Business Chicago IL 60637 and NBER Wewould like to thank all those we talked to at trucking rmsand private eets for allowing us to visit their rms anddiscuss the issues we investigate in this paper Thanks toAnn Merchant for research assistance and Oliver HartBengt Holmstrom Paul Milgrom Canice Prendergast Pres-ton McAfee MinSoo Park an anonymous referee andmany seminar participants for comments We gratefullyacknowledge support from NSF Grant No SES-9975413the NBERSloan Pin Factory Project and the Harvard Busi-ness School Division of Research

551

theoretical insights The model generates twosets of comparative static predictions One setof predictions is consistent with well-knowncross-sectional patterns in the industry Theseinclude the prediction that service-intensivetrucking is more likely to be performed byprivate than for-hire eets and that pri-vate eets are differentially more likely toadopt incentive-improving technologies whilefor-hire carriers are more likely to adoptcoordination-improving technologies

The other set of predictions concerns howchanges in the informational environment affectownership We test this second set of predic-tions using data from the 1987 1992 and 1997Truck Inventory and Use Surveys which con-tain detailed truck-level information abouttrucksrsquo characteristics ownership and use Inparticular we test predictions on how the dif-fusion of different types of on-board computers(OBCs) during the late 1980rsquos and early 1990rsquosalters the ldquomake versus buyrdquo decision for ship-pers We predict that the adoption of certaintypes of OBCs should lead indirectly to moreshipper ownership of trucks by lowering theagency costs associated with complex job de-signs We predict that the additional capabilitiesof other types of OBCsmdashthose that provide loca-tion information and real-time communicationmdashshould lead to less shipper ownership of trucksbecause these additional capabilities enhancethe comparative advantage of for-hire carriagewith respect to truck utilization and dispatch We nd evidence in favor of both of these predictions

Our results strongly suggest causal links be-tween informational and organizational changesin the trucking industry They show that own-ership patterns in trucking re ect the impor-tance of not only incomplete contracts (asstressed by Grossman and Hart 1986) but alsoof job design and measurement issues (likethose stressed in Holmstrom and Milgrom1994) These ndings thus shed important lighton theories of organizations They also make acontribution to the long-running debate abouthow information technology (IT) diffusion af-fects the boundaries of the rm1 We note that

information technology in general provides atleast two capabilitiesmdashimproved monitoring ofagents and improved coordination of activi-tiesmdashand that the organizational impact ofthese capabilities can differ (Michael C Jensenand William H Meckling 1992) In truckingimprovements in monitoring (and the attendantimprovement in incentives) lead to larger moreintegrated rms while improvements in coordi-nation (resulting in better asset utilization) leadto more diffuse asset ownership and smallerless integrated rms Whether these results gen-eralize to other settings remains an openquestion

In this paper we do not consider a third pos-sibility regarding truck ownership drivers mayown trucks We investigate driver ownership oftrucks in detail in another paper (Baker andHubbard 2000) In that paper we propose thatasset ownership strengthens driversrsquo incentivesto drive in ways that preserve trucksrsquo value butalso encourages them to engage in rent-seekingbehavior We then argue that OBC adoptionalters this trade-off by allowing companies (ei-ther for-hire carriers or private eets) to use themonitoring capabilities of OBCs to substitutefor asset ownership We show that OBCs lead toless driver ownership of trucks especially forhauls where rent-seeking is a potential problemWe ignore these issues in the present paperbecause we believe they are not salient to themake-or-buy decision Situations that are on themargin between for-hire carriage and privatecarriage are not those where owner-operatorsare used In general owner-operators are usedfor hauls that require little if any service provi-sion by the driver and for good reason Themultitasking problems with service provisionthat lead for-hire carriage to be inef cient rela-tive to private carriage are exacerbated whendrivers control trucks This is borne out by thefact that when shippers outsource hauls withnonnegligible service requirements they rarelyif ever do so by contracting with owner-operators2

The paper is organized as follows In the nextsection we describe the institutional setting that

1 See H J Leavitt and T L Whisler (1958) Thomas WMalone et al (1987) Erik Brynjolfsson and Lorin Hitt(1997)

2 In our empirical tests ldquofor-hire carriagerdquo includes driver-and carrier-owned trucks ownership structures where ship-pers do not own trucks Our results are unchanged when weleave out owner-operators altogether

552 THE AMERICAN ECONOMIC REVIEW JUNE 2003

we model de ning the players describing theirroles in the provision of trucking services andcharacterizing the contracting environment inwhich they operate In Section II we presentour model of job design and asset ownershipSection III describes OBCs and generates ourmain empirical propositions In Section IV wedescribe our data and present the main empiricalpatterns Section V contains our main empiricalresults regarding the relationships betweenOBC adoption and organizational change Sec-tion VI concludes

I Job Design Search Incentives and AssetOwnership in Trucking

This section describes the institutional frame-work drawing heavily from what we learned ina series of site visits and interviews We de-scribe the basic trade-offs involved in job de-sign and asset ownership decisions and explainwhy these decisions might be related Through-out the section we will refer to several differentparties Drivers are individuals who drive trucksand may have other customer-service-orientedtasks Shippers are rms or divisions with de-mands to move cargo from one place to anotherCarriers are rms or divisions that supply trans-portation services Carriers that supply servicesusing trucks owned by shippers are private car-riers (ie shippersrsquo internal eets) Carriers thatsupply services using their own rather than ship-persrsquo trucks are for-hire carriers Brokers arethird-party informational intermediaries

A Driver Job Design Driving andService Provision

Drivers can engage in two sorts of activitiesdriving the truck and performing nondrivingservice activities3 De ning driversrsquo jobs to in-clude nondriving activities lets carriers offerhigh service options in which their customerscan ask drivers to do things such as help unloadthe truck and sort and store the cargo This givescustomers exibility in how many of their ownworkers they allocate to such tasks and can im-prove the division of labor in the short run because

deliveries might take place when the opportunitycost of customersrsquo workersrsquo time is high

The bene t of giving drivers service respon-sibilities varies systematically across hauls withthe characteristics of the cargo There are rarelysuch bene ts when they haul bulk goods such asgravel ores or grain in large part because nohandling is required upon delivery when trucksreach their destination drivers dump the cargowhere the recipient wants it Giving driversservice responsibilities is also generally unpro-ductive when trucks haul goods for which han-dling requires special equipment For examplespecial machinesmdashwhich drivers generally areeither unable or not trusted to usemdashare usuallynecessary to move very heavy goods (large rollsof paper sheet metal) As a consequence driv-ers generally just drive trucks when they haulbulk or unwieldy goods

In contrast giving drivers service responsi-bilities can be valuable when trucks haul otherclasses of goods such as packaged goods orhazardous cargo Packaged goods can be carriedby hand or transported with standard equipmentsuch as hand trucks conveyor belts or forkliftsHandling hazardous cargo such as petroleum orchemicals requires certi cation which driversgenerally must have to haul such cargo legallyGiving drivers service responsibilities dimin-ishes the extent recipients must have certi edpersonnel As a consequence drivers often haveservice responsibilities when trucks haul pack-aged goods or hazardous cargo

A drawback to giving drivers additional re-sponsibilities is that agency costs are higher4

Carriers always face the problem of motivatingdrivers to pick up and deliver goods on time anddrive in ways that preserve trucksrsquo value Whendriversrsquo jobs involve service they also face theproblem of motivating drivers to allocate theirtime ef ciently between driving and service

Motivating drivers to pick up and delivergoods on time is straightforward because it isrelatively easy to evaluate driversrsquo performancein this dimension The distances traveled andthe return time at the end of the run are knownCarriers also normally have good information

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

4 Following Jensen and Meckling (1976) agency costshere include both monitoring costs and the ldquoresidual lossrdquoattributable to nonoptimal decisions

553VOL 93 NO 3 BAKER AND HUBBARD ASSET OWNERSHIP IN TRUCKING

regarding whether drivers arrive late to interme-diate stopsmdashangry customers call them whenthey domdashand have some information about theimpact of factors outside of driversrsquo controlsuch as traf c and weather conditions Thuswhen driversrsquo jobs involve only driving fromlocation to location the main agency problemthat remains is inducing them to drive wellbecause this is what remains noncontractible

Incentive problems are more complicatedwhen driversrsquo jobs include service activities Asis generally the case in multitasking problemsincentives must attend both to overall effortlevels and the allocation of effort across tasksIn this case the incentive problem created bymultitasking is that carriers now must inducedrivers to allocate effort between driving andservice appropriately Simple distance and ar-rival time data provide little indication of thefraction of time drivers spend driving versusdoing other things Some common service ac-tivities such as cargo-handling are strenuous5

Drivers with service responsibilities have anincentive to misallocate their effort for exam-ple by taking more time handling cargo thenmaking it up by driving faster between stopsCarriers may respond to this in the spirit ofHolmstrom and Milgrom (1991 1994) andBaker (1992) by weakening driversrsquo incentiveswith respect to other tasks For example theybalance incentives by de-emphasizing on-timearrivals or allowing more slack in schedules Ingeneral agency costs are higher when drivershave more responsibilities because of somecombination of lower overall effort levels and aworse allocation of effort across tasks

B Market Clearing Load Matchingand Search

The demand for trucking services and thesupply of truck capacity are highly differen-tiated Shippersrsquo demands are speci c withrespect to time location and equipment re-quirements Likewise truck capacity is idiosyn-cratic with respect to its geographic location and

the characteristics of the trailer Capacity utili-zation in the industry depends crucially on howef ciently supply and demandmdashtrucks andhaulsmdashare matched Trucks and hauls arematched in a highly decentralized manner inwhich shippers carriers and third-party brokerssearch for good matches

The matching problem is particularly dif cultin trucking because individual shippers rarelyhave demands that ll trucks for both legs of around-trip For this reason once carriers receiveservice orders from shippers they then searchfor complementary hauls When individualshipments are too small to ll a truck searchtakes the form of identifying other shippers withsimilar demands When demands are unidirec-tional search is directed at identifying shipperswith demands that would ll the truck for thereturn trip (the ldquobackhaulrdquo)

Dispatchers and brokers play a crucial role inidentifying complementary hauls and arrangingmatches Dispatchers work for carriers andseek to match hauls to trucks within their car-rierrsquos eet Brokers seek to match hauls totrucks owned by other parties These partiesacquire knowledge about city-pair demand in atwo-stage process they make long-run invest-ments in learning general demand patterns (egwho the demanders are) then learn detailed ldquoonthe spotrdquo information about short-run demandsby contacting shippersrsquo traf c managers period-ically throughout the day

Search for complementary hauls in the shortrun tends to be more re ned and hence produc-tive the more precisely parties can forecastwhen trucks will come free This in turn leadsto better matches between trucks and hauls Forexample backhauls may begin sooner afterldquofronthaulsrdquo end and trucks may arrive to beloaded closer to when shippers want themThus a second drawback to giving drivers ser-vice responsibilities on a haul is that serviceinterferes with search for the following haultrucksrsquo availability is more predictable follow-ing lowservice than high-service hauls6

5 Drivers whose jobs involve taking a fully loaded trailerand delivering the goods to various destinations handle upto 40000 pounds of cargo per day Handling requires hand-lifting when trucks deliver to places without loadingdocksmdashsuch as most retail outlets

6 In interviews eet managers and dispatchers indicatedto us that forecasting how long deliveries take is mucheasier when drivers have fewer service responsibilitiesThey indicated that they could forecast how long a no-service delivery of a truckload of packaged goods wouldtake within a half-hour window but could only forecast how


C Asset Ownership and Incentives

Shippersrsquo make-or-buy decision correspondsto whether they use a truck from their internal eet or an external eet for a haul Industryparticipants distinguish between private andfor-hire carriage by who has control rights overthe truck7 Below we discuss how and why assetownership affects incentives

Ownership rights over trucks matter becausecontracts are incomplete with respect to trucksrsquoschedules In particular shippers and carriers donot write fully contingent contracts with respect totrucksrsquo schedules because the relevant contingen-cies are costly to identify ex ante and verify expost To see this consider one class of schedulingdecisions how long a truck should wait at theloading dock to be loaded A fully contingentcontract would stipulate how long trucks shouldwait as a function of all relevant states of theworld including especially those factors affectingthe bene ts of delay and individual trucksrsquo oppor-tunity cost Many of these factors are known onlyto shippers andor carriers and are dif cult toverify by outsiders It is thus prohibitively costlyto make contracts contingent on them Schedule-setting is therefore a residual right of control thatis by de nition held by the truckrsquos owner8

The contractual incompleteness surroundingtruck scheduling leads to the main consequenceof the allocation of ownership rights In privatecarriage shippers own trucks if they want toalter trucksrsquo schedules in ways that do not vio-late existing agreements they can do so Theycan unilaterally require that a truck picking upor delivering goods wait for example In for-

hire carriage shippers do not own trucks Ifshippers want to change trucksrsquo schedules theymust negotiate this with carriers

The possibility that schedules will have to berenegotiated leads to familiar sorts of transac-tions costs in for-hire carriage Both partieshave an incentive to improve their bargainingposition and thus engage in rent-seeking behav-ior9 For shippers this takes the form of iden-tifying other carriers who could serve them onshort notice for carriers this takes the form ofidentifying other local shippers with similardemandsmdash nding substitute hauls Exploringback-up plans expends real resources and iscostly In private carriage by contrast disputesmay arise between shippers and their private eetsrsquo dispatchers (or shippers and brokers) butidentifying other ways to use trucks does notimprove dispatchersrsquo or brokersrsquo bargaining po-sition because they cannot threaten to use trucksfor other hauls Neither private eet dispatchersnor brokers have incentives to identify substi-tute hauls for rent-seeking purposes

While rent-seeking tends to be greater underfor-hire carriage truck utilization also tends tobe higher One reason has to do with rmsrsquoincentives to obtain market information andsearch for complementary hauls Firms cansearch more effectively for complementaryhauls in the short run if they have previouslymade investments (in the form of customer re-lationships and general knowledge of demand)in particular markets Shippers for-hire carriersand brokers can all potentially make such in-vestments But because these investments aremore valuable to those who are frequently look-ing for backhauls individual shippers will tendto only make signi cant investments on city-pairs where their trucks haul high volumes ofgoods regularly On other routes they will investless have less information about demand andtherefore search less productively in the shortrun than for-hire carriers or brokers who ex-ploit increasing returns by utilizing knowledgeacross many shippersrsquo hauls Intermediariesthus have a comparative advantage in ndingcomplementary hauls in many circumstances

long a high-service delivery would take within a two- tothree-hour window

7 Trucks in private eets are sometimes leased are some-times driven by short-term employees and sometimes haulother shippersrsquo goods (such as on backhauls) The distinc-tion between private and for-hire carriage thus does notcorrespond to residual claimancy the length of labor con-tracts or exclusivity of use

8 In practice it is common for contracts between ship-pers and carriers to have clauses that penalize shippers whenthey delay trucks The penalties however are not statedependent and thus are set intentionally high to deter ship-pers from delaying trucks in states of the world wheretrucksrsquo shadow value is high Parties realize that renegoti-ation is likely to be ef cient when trucksrsquo shadow value islow creating a situation that is analytically similar to thosewhere schedules are noncontractible

9 See Grossman and Hart (1986) Milgrom and JohnRoberts (1990) Baker and Hubbard (2000) argue that thisincentive is also central for understanding why truck driverstend not to own the trucks they operate


This alone need not imply that truck utilizationis necessarily lower under private carriagesince shippers could rely on brokers to ndhauls However brokers have weaker incen-tives to nd particularly good matches becausethey do not own trucks and are thus less able toappropriate as large a share of the value thatthey create The combination of strong incentivesto learn about demand and strong incentives to nd good matches for particular trucks leadsmatches to be better and thus truck utilization tobe higher under for-hire than private carriage

Another reason why truck utilization tends tobe higher in for-hire carriage is that drivers aregenerally assigned fewer service responsibili-ties Trucks spend more time on the road and asnoted above load matching is easier when driv-ersrsquo responsibilities are narrow

The next section develops a model of assetownership and job design that captures the in-stitutional features described above and ana-lyzes organizational relationships formallyThis model generates comparative static pre-dictions that explain several important cross-sectional patterns in the industry It alsogenerates predictions regarding how changes inthe informational environment should affect themake-or-buy decision Later in the paper wetake these predictions to the data

II A Model of Asset Ownership and Job Design

The model combines elements of Holmstromand Milgrom (1991 1994) and Grossman andHart (1986) We embed multitask models ofdriver job design and dispatcher effort toward nding hauls into a setting in which noncon-tractible truck scheduling problems make assetownership important The timing follows Ini-tially a shipperrsquos ldquofronthaulrdquo and a matchingtruck are assumed to exist we do not model theprocess of matching fronthauls to trucks Thishaul may be one for which the value of serviceis high or low We assume that parties cannotwrite a complete contract with respect to thishaul ex ante Organizational form is then cho-sen at this point asset ownership and driversrsquojob design are determined Next search forcomplementary backhauls (and possibly substi-tute fronthauls) occurs Depending on assetownership and the organizational form choseneither a carrier or a broker chooses how much to

search for hauls that complement or substitutefor the shipperrsquos haul Parties then bargain thisdetermines which haul the truck is used for andhow the surplus is split Finally productiontakes place (including provision of service bythe driver) and payoffs are realized

Complementarities between job design andasset ownership are critical to the results andare a central feature of our model To highlightthis relationship and simplify the exposition wedevelop a model rst of driver job design thenoverlay the shipperrsquos ldquomake-or-buyrdquo decisionWhen shippers own trucks this corresponds toldquomakerdquo when they do not this corresponds toldquobuyrdquo For simplicity we assume that under theldquomakerdquo option shippers use brokers to ndbackhauls rather than nd them themselves10

We begin with a model of driver job design


Let s be the scope of the driverrsquos activities andm be the marginal product of this scope11 Forsome hauls and shippers service activities arevaluable (high m) and for some they are lessvaluable Motivating high service levels is costlysince it involves monitoring the mix of activitiesthat the driver is performing Let s be a parameterthat captures the ability of the carrier to monitorthe driverrsquos ef ciency in performing high-serviceactivities the higher is s the lower is the marginalcost of monitoring We specify V the value ofusing the truck and driver for the shipperrsquos haul as

(1) V 5 V 1 ms 2 M~s s

where V is a xed quantity s is the scope of thedriverrsquos activities m is the marginal product ofthis scope s is the degree to which the carrier canmonitor driver activities and M(s s) is agencycosts We assume M1 0 M2 0 M12 0

10 We present a more general model in which privatecarriers might use their own dispatchers to search for back-hauls in an earlier version of the paper (Baker and Hubbard2002) None of the comparative statics presented belowdiffer in this more general model

11 Our equation of scope with service levels re ects an(unmodeled) assumption that some signi cant amount ofdriving is always part of the driverrsquos job the driver is neverdoing mostly service Thus more service involves a greatermix of activities


Given this setup the optimal amount of scopein the driverrsquos job depends on the costs andbene ts of such scope Assuming an interiorsolution optimal job design sets scope such thatm 5 M1(s s) Raising the marginal productof scope (raising m) or raising the rmrsquos abilityto monitor driver activities (raising s) raises theoptimal amount of scope We assume that thisexpression is invertible so that we can expressthe result as s 5 f(m s)

B Load Matching

Following the discussion in Section I weassume that search for complementary haulsadds value Value is increasing in search levelsbecause more effort produces better matchesWe also assume that the marginal productivityof search is reduced when drivers are assignedmore service-oriented activities

We specify the value added of search forcomplementary hauls as

(2) ~g1 2 use 1

where e1 is the effort toward nding comple-mentary hauls and g1 is the marginal product ofthis effort for hauls involving no service ucaptures the extent to which high service levelsreduce the marginal product of search u 0We also assume u g1f(m s) this regular-ity condition ensures that the marginal bene t ofsearching for complementary hauls is positiveat the optimum12 We specify the cost of search-ing for complementary hauls as C1(e1) 5 e1

2 2The expression for V now including the

value of the complementary haul becomes

(3) V 5 V 1 ~g1 2 use1 1 ms 2 M~s s

C Bargaining Truck Ownership andResidual Rights of Control

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

terms of trade These activities yield potential usesof the truck that are close substitutes for the ship-perrsquos haul For simplicity we assume that thissearch is over alternatives that involve the samelevel of driver service but that using the truck anddriver for the alternative is always less valuablethan using them for the rst shipperrsquos haul (per-haps because the alternative haulrsquos origin is moredistant) Assume that the value created when thetruck is used for an alternative shipperrsquos haul is

(4) P 5 g2 e2 1 ~g1 2 use 1 1 ms 2 M~s s

where e2 represents effort toward nding alter-native hauls and g2 represents the marginal pro-ductivity of this effort13 This formulationassumes that e1 the effort that the dispatcherexpends toward nding hauls that complementthe rst shipperrsquos hauls is equally valuable forthe alternative shipperrsquos hauls (eg the back-haul she nds would complement either out-bound haul) We specify the cost of searchingfor substitute hauls as C2(e2) 5 e2

2 2We can now calculate the amount of search

when carriers or brokers search for hauls Weassume that when shippers bargain with eitherfor-hire carriers or brokers over the surplusthey split the difference between the value ofthe haul and the value of the carrierrsquos or bro-kerrsquos outside alternative A for-hire carrierrsquosoutside option is equal to P the value of usingthe truck for an alternative shipperrsquos haul Abroker does not have this outside option be-cause it does not own trucks We thereforenormalize brokersrsquo outside option to zero14

A for-hire carrier chooses e1 and e2 to maximize

(5) ~V 1 P2 2 12

e12 2 1

2e2

2

5 ~V 1 2~g1 2 use1 1 2ms 2 2M~s s

1 g2 e2 )2 2 12

e12 2 1

2e2

2

12 This guarantees that g1 2 us is nonnegative in theresults below The condition ensures that bene ts of serviceare never so high so that the direct bene ts of searching forcomplementary hauls are overwhelmed by its indirect costs

13 Thus V is the value of the rst shipperrsquos haul (net ofservice) and g2e2 is the value of the alternative haul Weassume V 12 g2

214 This is a simpli cation brokers might get some value

from searching for substitute fronthauls What is requiredfor the model is that the marginal returns to searching forsubstitute fronthauls are lower for brokers than for carrierswhich seems reasonable given the difference in residualcontrol rights


This yields search effort equal to

(6) e1F 5 ~g1 2 us e2

F 5 12

g2

If search is completed by a for-hire carrier itwill search both for hauls that complement andsubstitute for the shipperrsquos Total value whichequals V less search costs under this organiza-tional alternative is

(7) TVF 5 V 1 12

~g1 2 us2

2 18

g22 1 ms 2 M~s s

A broker chooses e1 and e2 to maximize

(8) V2 2 12

e12 2 1

2e 2

2

5 ~V 1 ~g1 2 use1 1 ms 2 M~s s2

2 12

e12 2 1

2e2

2

yielding effort of

(9) e1P 5 1

2~g1 2 us e2

P 5 0

Effort levels are lower under private carriagebrokers search less intensively for comple-ments and not at all for substitutes Total valueunder private carriage is

(10)

TVP 5 V 1 38

~g1 2 us2 1 ms 2 M~s s

D Ef cient Organizational Forms JobDesign and Asset Ownership

In order to compare the total value created by pri-vate carriage versus for-hire carriage we introducean index variable d that indicates asset ownershipd 5 1 indicates for-hire carriage d 5 0 privatecarriage Total value as a function of s and d is

(11) TV~s d 5 V 1 18

~3 1 d~g1 2 us2

2 18

dg22 1 ms 2 M~s s

PROPOSITION 1 TV(s d) is supermodularin (2s 2m d 2s g1 2g2) on the domainwhere s $ 0 d [ 0 1 and 0 u g1f(m s)

PROOFSupermodularity requires that TV has nonde-

creasing differences in (2s 2m d 2s g12g2) this is equivalent to nonnegative cross-derivatives when TV is continuously twice-differentiable (Donald M Topkis 1978) Allterms except the second term are supermodularin (2s 2m d 2s g1 2g2) on this domainby inspection The second term is supermodularif g1 2 us $ 0 which is guaranteed if u g1f(m s) The sum of supermodular func-tions is supermodular

This result allows us to apply a theorem fromTopkis (1978) (see also Theorem 5 of Milgromand Christina Shannon 1994) and generate aset of monotone comparative statics that we cantest with data on asset ownership and technol-ogy adoption

PROPOSITION 2 2s and d are monotonenondecreasing in (2m 2s g1 2g2) on the do-main where s $ 0 d [ 0 1 and 0 u g1f(m s) 2s and d are (weak) complements

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

One simple prediction is that s and d should beinversely correlated that is high service should beassociated with shipper ownership of trucks Thisis consistent with the stylized fact that drivers inprivate eets engage in more service-related ac-tivities than drivers in for-hire eets It is alsoconsistent with the assertion made by many ship-pers that attainment of better service is why theychoose private carriage over for-hire carriage15

A second prediction is that d should be high

15 ldquo[T]here are some good reasons why private carriageremains attractive to companies Service is the key consid-eration Many companies claim they require a private eetto provide the high levels of service their customers expectlsquoThere are companies that decided to outsource their entire eet yet came running back to private eets when theservice was not what they expectedrsquo says [John McQuaid ofthe National Private Truck Council]rdquo (Jim Thomas 1998)


when g1 is high that is for-hire carriage shouldbe more prevalent when effort toward identify-ing complementary hauls is particularly valu-able This is consistent with the stylized factthat for-hire carriage tends to be used more forsmall shipments and long-distance shipmentsthan large and short-distance shipments [SeeBureau of the Census (1999b) and Hubbard(2001a) for empirical evidence]

A third cross-sectional prediction concernsthe adoption of different types of on-board com-puters As we discuss in more detail belowOBCs have different informational capabilitiesCertain simple devices (called trip recorders)allow eet owners to monitor the actions ofdrivers ex post more advanced devices (elec-tronic vehicle management systems calledEVMS) also allow them to track trucksrsquo loca-tion in real time The model predicts that thevalue of these different informational capabili-ties should differ between private and for-hirecarriage increasing the contractibility of driv-ersrsquo actions should be more valuable in privatecarriage while capabilities that raise the returnsto searching for complementary hauls should bemore valuable in for-hire carriage16 As a con-sequence private eets should be differentiallylikely to adopt trip recorders and for-hire carri-ers should be relatively more likely to adoptEVMS Hubbard (2000) tests this predictionand nds exactly this pattern He shows that in1992 adoption rates for trip recorders andEVMS were (respectively) 88 percent and 58percent for private carriers and 65 percent and154 percent for for-hire carriers This evidenceprovides important support for the model thatwe propose suggesting the relevance of theagency and ownership issues that we highlight

Our main empirical tests however examinerelationships between informational improve-ments enabled by the adoption of on-boardcomputers (OBCs) and changes in ownershipThese exploit the predictions that increasing g1should lead rms to (weakly) increase d andincreasing s should lead rms to (weakly) de-crease d If the productivity of searching forcomplementary hauls ( g1) increases as a result

of improved information technology thisshould lead to two changes a shift from privateto for-hire carriage and a decrease in the scopeof driversrsquo activities If rmsrsquo ability to monitorthe allocation of driversrsquo effort (s) increasesthis should lead directly to increases in thescope of driversrsquo activities and indirectly tomore shipper ownership of trucks

Proposition 2 implies that sometimes changesin the modelrsquos parameters may not result inchanges in the optimal organizational structureOne case is of particular interest to us If m 50 it is optimal to set s 5 0 because there is nobene t from giving drivers service responsibil-ities If this is true the total value function is

(12) TV~s d 5 V 1 18

~3 1 dg12 2 1

8dg2

2

If m 5 0 TV is independent of s there is nomultitasking and no multitasking-relatedagency problem Therefore if m 5 0 changesin rmsrsquo ability to monitor the allocation ofdriversrsquo effort (s) should have no effect on assetownership

The following section describes OBCs inmore detail and generates empirical proposi-tions relating OBC adoption to ownershipchanges

III On-Board Computers andOrganizational Change

Two types of OBCs began to diffuse in thetrucking industry in the late 1980rsquos trip record-ers and electronic vehicle management systems(EVMS)17 Trip recorders measure trucksrsquo op-eration They record when trucks are turned onand off their speed sudden accelerations ordecelerations and various engine performancestatistics (eg fault codes) Dispatchers and eet managers receive the information trip re-corders collect when drivers return to their baseat the end of a trip Drivers give dispatchers a oppy disk or a similar device Dispatchersupload the information onto a computer whichprocesses the information and provides reportsThese reports indicate how drivers operated the

16 Increasing s raises total value more when 2s andthus its complement d is low than high and increasing g1

does so more when d is high than low

17 See also Baker and Hubbard (2000) and Hubbard(2000)


truck for example how quickly they drovehow long they allowed trucks to idle andwhether there were any nonscheduled stopsThey also indicate how long drivers spent ateach stop

EVMS record the same information trip re-corders do but provide three additional capa-bilities One is that they record trucksrsquogeographic location often using satellite track-ing systems Another is that they can transmitany information they collect to dispatchers inreal time Dispatchers can thus know wheretrucks are at any point in time Third theyprovide dispatchers a way of initiating commu-nication with drivers For example dispatcherscan send a text message that updates driversrsquoschedule If the message is complicated dis-patchers can send a message that asks drivers tocall in This is a signi cant advance over thesystem rms have traditionally used to commu-nicate with drivers who are outside radio range(about 25 miles) Traditionally rms requiredrivers to call in every three or four hours Thisrequires drivers to frequently pull over stopand nd a phone even though much of thetime neither dispatchers nor drivers have newinformation to communicate Without EVMSdispatchers often nd it hard to verify trucksrsquolocation and must wait for distant driversto call in before they can communicateinstructions

As Hubbard (2000) relates there is an eco-nomically important distinction between thesetwo devices Trip recorders are useful for im-proving incentives because they provide veri- able information about how trucks wereoperated Importantly for this paper they mon-itor how long drivers spent driving and howlong they spent performing other tasks thishelps mitigate the agency problems associatedwith more complex job designs18 Trip record-ers are not generally useful for improvingresource-allocation decisions (ldquocoordinationrdquo)They do not improve dispatchersrsquo ability tomatch trucks to hauls in the very short runbecause they do not supply information in atimely enough fashion They are generally not

used to improve routing decisions made overthe longer runmdashfor example by helping bench-mark routesmdashbecause rms usually can obtaininformation about such things as how longroutes take by other less costly means19

In contrast EVMS are useful for improvingboth incentives and coordination Their addi-tional capabilities help dispatchers match trucksto hauls better thereby increasing capacity uti-lization Real-time information about trucksrsquo lo-cation helps them schedule backhauls moreef ciently for example20 These capabilitiesalso enable them to communicate schedulechanges to drivers in real time Dispatchers canquickly reroute trucks in response to changes inmarket conditions For example suppose atruck on the road is half-full If a dispatcher can nd a shipper with cargo that can ll the truckhe can send a message to the driver asking himto make an additional pick-up and delivery

We next discuss our main empirical proposi-tions which predict how OBC adoption shouldaffect truck ownership These propositions arebased on the premise that trip recorder adoptionincreases s and EVMS adoption increases boths and g1

P1 Overall trip recorder adoption should leadto more shipper ownership of trucks

OBCsrsquo incentive-improving capabilities al-low carriers to better monitor how drivers allo-cate time and thus effort across tasks Triprecorder adoption thus raises s which by Prop-osition 2 increases the optimal choice of s anddecreases d carrier ownership of trucks shoulddecrease We cannot test whether trip recorderadoption increases s because the data do not

18 An advertised bene t of trip recorders is their abilityto help monitor drivers in this way For example Atrolclaims that its devices can ldquotell you how effective yourdrivers are in managing their timerdquo (wwwatrolcom)

19 Many rms use software packages to help dispatchersschedule trucks These packages often use informationEVMS collect (for example trucksrsquo location) but rarely usethe information trip recorders collect

20 Trade press articles and advertisements emphasizethis An example of a quote from a driver ldquoDispatch knowswhere I am and where Irsquom headed so before I even get to mydestination they can plan ahead Quite often I get a loadoffering over my Qualcomm system before Irsquom evenemptyrdquo (wwwqualcommcom) Empirical evidence ofEVMSrsquo impact on capacity utilization is in Hubbard (2003)who nds that EVMS has increased loaded miles amongadopters by 13 percent as of 1997


contain information on the scope of driversrsquoactivities but we can test whether it leads tomore shipper ownership of trucks

P2 EVMS adoption should lead to less of anincrease in shipper ownership of trucks thantrip recorder adoption and may lead to lessshipper ownership of trucks

EVMSrsquo coordination-improving capabilitiesmake dispatchersrsquo search more productive andthus raise g1 Knowing where trucks are allowsdispatchers to better anticipate when trucks willcome free and hence helps them re ne theirsearch Being able to initiate communicationswith drivers while they are in their cab en-ables them to better exploit the opportunitiesthey identify For example they can quicklyreallocate drivers and trucks across haulsin response to new opportunities BecauseEVMS contain both incentive- and coordination-improving capabilities EVMS adoption shouldincrease both s and g1 and thus has a theo-retically ambiguous impact on asset owner-ship However because EVMS adoptionincreases s in the same way trip recorderadoption does EVMS adoption should movehauls less toward private carriage than triprecorder adoption

P3 Trip recorder adoption should increaseshipper ownership of trucks more when driversrsquocargo-handling activities are potentially pro-ductive than when they are not productive Itshould not affect whether shippers own truckswhen driversrsquo handling activities are not pro-ductive

Trip recorder adoption should not lead toownership changes when m 5 0 for examplefor hauls of bulk goods or goods that requirepeople other than drivers to load and unload Itshould lead to ownership changes when m 0From above this should be the case when truckshaul packaged goods especially when they pickup or deliver to small outlets It should also bethe case when trucks haul goods for whichhandling requires certi cation such as petro-leum or chemicals However it should notbe true for hauls of bulk goods or goods thatcannot be lifted or transported with standardequipment

IV Data

The data are from the 1987 1992 and 1997Truck Inventory and Use Surveys (TIUS)21

The TIUS is a mail-out survey of trucks takenby the Census as part of the Census of Trans-portation The Census sends forms to a randomsample of truck owners These forms ask ques-tions about individual trucksrsquo characteristicsTruck owners report the truckrsquos type (pick-upvan tractor-trailer etc) make model andmany other characteristics The TIUS also askshow trucks are equipped including whetherthey have trip recorders or EVMS installed andhow they are used Owners report how far fromhome individual trucks generally operated thetype of trailer to which they were typicallyattached the class of product they generallyhauled the state in which they were based andwhether they were used for for-hire or privatecarriage Publicly available data from the sur-vey do not identify trucksrsquo owners because ofcon dentiality restrictions This paper uses onlyobservations of truck-tractors (the front halvesof tractor-trailers) and excludes those that weregenerally operated off-road carried householdgoods (ie moving trucks) or were attached totrailers that do not haul goods (eg trailers withlarge winches permanently attached) Eliminat-ing these observations leaves 21236 32015and 18856 observations of tractor-trailers in1987 1992 and 1997 respectively This is over85 percent of the tractor-trailers in the originalsamples

Figure 1 shows private carriage shares ineach of the three years In each of these yearsthe overall share is about 50 percent and ishigher for shorter hauls than longer ones Theoverall share uctuated during this period in-creasing from 501 percent to 546 percent be-tween 1987 and 1992 then falling back to 517percent in 1997 The time trends differ for haulsof different lengths The private carriage shareincreased for all distances between 1987 and1992 It increased for short hauls but declinedfor medium and long hauls between 1992 and1997 This paperrsquos empirical tests examine how

21 See Bureau of the Census (1995 1999a) Baker andHubbard (2000) and Hubbard (2000 2001) for more on theTIUS The 1997 survey is actually called the Vehicle In-ventory and Use Survey


these changes relate to the diffusion of on-boardcomputers

It is useful to put these changes in perspec-tive within the industry they are consideredhistorically signi cant changes in ownershippatterns22 Other studies indicate that the pri-vate carriage share was remarkably constantbetween the early 1970rsquos and late 1980rsquos Evenderegulation which initiated large structuralchanges within the for-hire sector of the indus-try appears to have changed make-or-buy de-cisions in the aggregate by no more than 2percentage points23 We suspect that this re-

ects that neither haul characteristics nor the in-formational environment changed much duringthis period The aggregate changes in ownershippatterns depicted in Figure 1mdashthe increase of35 percentage points in the private carriageshare between 1987 and 1992 and the decreaseof 29 percentage points between 1992 and1997mdashappear larger than those that immedi-ately followed deregulation or any other periodin the previous 25 years

Figure 2 summarizes patterns of OBC adop-tion over time and across distances There arethree important patterns First adoption ofOBCs was rapid during our sample period In1987 a negligible number of tractor-trailers hadan OBC installed we treat this number as zerothroughout In 1992 19 percent of tractor-trailers had an OBC installed by 1997 adoptionincreased to 34 percent Second adoption wasgreater for trucks used for longer hauls By

22 For example Standards and Poorrsquos DRI remarks uponthe ldquorapid shift to for-hire trucking observed in the early tomid-1990srdquo (American Trucking Associations 1999)

23 Using different measures than those in Figure 1 (pri-vate eetsrsquo volume share of intercity trucking shipments)the Eno Transportation Foundation estimates that the pri-vate carriage share was approximately 59 percent through-out the 1970rsquos and fell to 577 percent immediatelyfollowing deregulationmdasha decline of only 13 percentagepointsmdashbefore climbing back to 603 percent in 1992 Itestimates that by 1997 this gure had fallen by 35 percent-

age points to 568 percent and has continued to fall sincethen (Rosalyn A Wilson 2001)

FIGURE 1 PRIVATE CARRIAGE SHARE BY YEAR DISTANCE OF HAUL


1997 half of long-haul trucks had an OBCinstalled Third the composition of OBCs dif-fered in the early and late part of our sampleperiod Trip recorders made up nearly half ofOBC adoption between 1987 and 1992 butthere was no net adoption of trip recorders be-tween 1992 and 1997 Evidence from the tradepress and interviews suggests that this re ectstwo offsetting factors new trip recorder adop-tion and upgrades from trip recorders to EVMSIn contrast adoption of EVMS accelerated Theshare of trucks with EVMS more than doubledbetween 1992 and 1997 Thus broad patterns inthe data suggest a correlation between techno-logical and organizational change the move-ment from for-hire to private carriage between1987 and 1992 was during a time when triprecorder adoption was relatively high and themovement from private to for-hire carriagebetween 1992 and 1997 was during a timewhen OBC adoption was disproportionatelyEVMS

A Cohort Data

The bulk of our empirical analysis uses co-horts rather than individual trucks as the unit ofobservation This allows us to exploit the timedimension of the data and use rst-differencingto control for unobserved time-invariant factorsthat affect OBC use and the make-or-buy deci-sion independently As in our earlier work wede ne cohorts narrowly basing them on state-product-trailer-distance combinations an exam-ple is ldquotrucks based in New Jersey haulingchemicals in tank trucks long distancesrdquo Thereare 2773 cohorts with a positive number ofobservations in 1987 1992 and 1997 Aboutthree-quarters of our original observations are inthese cohorts

The characteristics of the trucks in the origi-nal and cohort samples are similar with twoexceptions One is that the cohort sample tendsnot to contain trucks that are predominantlyattached to uncommon trailers such as auto

FIGURE 2 OBC ADOPTION BY YEAR DISTANCE OF HAUL


trailers logging trailers and specialized plat-form types An implication is that the cohortsample contains a higher fraction of long-haultrucks than the population because hauls usingspecialized trailers tend to be short The other isthat conditional on distance trucks attached torefrigerated vans make up a disproportionateshare of the cohort sample about 20 percentrather than their 10 percent share in the originalsample The reason for this is refrigerated vansalmost exclusively haul a single product classprocessed food Refrigerated van cohorts tendto be larger and are less likely to have zeroobservations than cohorts associated with trail-ers that haul multiple product classes

Table 1 contains summary statistics for thecohort sample Cohorts tend to be based onrelatively few observations due to our narrowcohort de nition the number of observationsper cohort is less than ten in each year24 Theaverage private carriage share is about 50 per-cent and average OBC adoption rates are similarto those in Figure 2 Table 2 provides evidenceof relationships between technological adoptionand organizational change at the cohort level

The top panel uses cohorts with positive obser-vations in both 1987 and 1992 The rst rowindicates that averaging across cohorts the pri-vate carriage share increased from 049 to 050between 1987 and 1992 The next three rowssplit the cohort sample according to OBC adop-tion On average the private carriage sharestayed the same for cohorts with low OBCadoption Among cohorts with high OBC adop-tion the private carriage share increased forthose where trip recorder adoption was high butdecreased slightly for those where EVMS adop-tion was high The bottom panel reports resultsfrom a similar exercise that analyzes patternsbetween 1992 and 1997 The private carriageshare decreased for the low OBC and highEVMS adoption cohorts (slightly more for thelatter) but increased for high trip recorder adop-tion cohorts

In sum relationships between OBC adoptionand organizational change differ for trip record-ers and EVMS Cohorts with high trip recorderadoption moved toward private carriage morethan cohorts with low OBC adoption Cohortswith high EVMS adoption moved toward for-hire carriage slightly more than those with lowOBC adoption but this difference is very smallNevertheless the fact that cohorts with highEVMS adoption did not move toward private car-riage is interesting in light of the fact that EVMSenable the same contractual improvements triprecorders do This suggests EVMSrsquo resource-allocation-improving capabilitiesmdashwhich trip

24 In earlier versions of this paper we reported estimatesof our main speci cations using the subsample of cohortswhere the private and for-hire carriage shares are positive ineach year The average cohort size is about double in thissubsample but observations in these cohorts make up onlyabout 35 percent of the original sample We showed that ourmain results do not change

TABLE 1mdashCOHORT SUMMARY STATISTICS

Positive number of observations in

1987 1992 1997 1987 1992 1992 1997

Cohorts 2773 3908 4101Observationscohort 1987 57 47Observationscohort 1992 83 66 64Observationscohort 1997 51 41Private carriage share 1987 050 049Private carriage share 1992 050 050 052Private carriage share 1997 048 050Trip recorder adoption 1987ndash1992 008 008 008EVMS adoption 1987ndash1992 008 009 011Trip recorder adoption 1992ndash1997 2001 000EVMS adoption 1992ndash1997 016 016

Notes Averages are computed using weights where weight 5 (numobs87expanf87 1numobs92expanf92 1 numobs97expanf97)3 for samples using all three years Analogousweights are used for samples that use only two of the three years


recorders do not havemdashhave organizational im-plications that offset those of their incentive-improving ones

V Results

Our base speci cation takes the form

y it 5 x itb 1 fi 1 laquo it

where yit is the for-hire carriage share in cohorti at time t xit includes a vector of explanatoryvariables and fi and laquoit represent unobservedtime-invariant and time-varying variables thataffect optimal organizational form The vari-ables of interest in xit are OBC the share oftrucks with either class of OBC installed andEVMS the share of trucks with EVMS in-stalled The coef cient on OBC therefore picksup the relationship between OBCsrsquo incentive-improving capabilities and asset ownership andthat on EVMS picks up the relationship betweenEVMSrsquo coordination-improving capabilities andasset ownership The control variables in xit aresimilar to those in Table 5 in Hubbard (2001)They include a full set of dummy variables thatindicate the cohortrsquos trailer type (dry van re-frigerated van tank truck etc) a dummy thatequals one if the cohort is of trucks haulingmixed cargo and ln(trailer density) Trailer den-

sity is the number of trucks in the state attachedto a given trailer type normalized by the statersquosurbanized area and is a proxy for local marketthickness for hauls using a particular trailertype We allow the coef cients on the dry vanand auto trailer dummies to vary across years toaccount for secular changes in contractual formover time (see Hubbard 1998)

Most of our results will be from rst-difference speci cations

y it 2 y i~t 2 1 5 ~x it 2 xi~t 2 1 b 1 hit

where hit 5 laquoit 2 laquoi(t2 1) First-differencingmitigates an important class of endogeneityproblems that would appear in cross-sectionalanalysis For example suppose that when ship-ping patterns are regular private carriage tendsto be used more (perhaps because intermediar-iesrsquo efforts are less valuable) and trip recordersare disproportionatelyvaluable relative to EVMSThis would lead private carriage and trip re-corder use to be correlated in the cross sectioneven if trip recorders adoption did not causetruck ownership to change First-differencingeffectively allows us to control for unobservedtime-invariant variables that could affect OBCadoption and organizational form indepen-dently and base inferences on relationships be-tween adoption and changes in asset ownership

TABLE 2mdashPRIVATE CARRIAGE SHARE AND OBC ADOPTION

Cohort Data

Mean private carriage share OBC adoption 1987ndash1992

N1987 1992 Change Trip recorder EVMS

All cohorts 049 050 001 008 009 3908Low OBC adoption cohorts 054 054 000 002 002 2972High TR adoption cohorts 050 054 004 034 006 470High EVMS adoption cohorts 033 032 2001 005 036 466




Notes The top (bottom) panel includes all cohorts with a positive number of observations in both 1987 and 1992 (1992and 1997) Low OBC adoption cohorts are those where OBC adoption was less than 015 High TR adoption cohortsare those where OBC adoption was greater than 015 and TR adoption was greater than EVMS adoption High EVMSadoption cohorts are those where OBC adoption was greater than 015 and EVMS adoption was greater than TRadoption


rather than levels25 Thus if haulsrsquo unobservedregularity is constant over time rst-differencingeliminates this as a possible endogeneity prob-lem For simplicity our initial discussion of theresults will assume hit to be independent ofadoption that is we assume changes in unob-served haul characteristics to be independent ofadoption Later we will relax this assumptionand present instrumental variables estimates ofthe rst-difference speci cations

Table 3 contains two sets of regression esti-mates that use cohort data from 1987 1992 and1997 The rst column presents levels esti-mates the second presents rst-difference esti-mates The coef cients on OBC are negativeand signi cant and those on EVMS are positiveand signi cant in both columns They are about40 percent lower in absolute value when mov-ing from the levels to the rst-difference esti-mates but remain statistically signi cant andeconomically important

These estimates supply evidence consistentwith our main propositions Consistent with P1trip recorder adoption is associated with move-ment from for-hire to private carriage Consis-tent with P2 EVMS adoption is less associatedwith such movement Assuming these re ect

causal relationships they indicate that OBCsrsquoincentive- and coordination-improving capabil-ities affect the make-or-buy margin differentlyTheir incentive-improving capabilities movehauls from ldquobuyrdquo to ldquomakerdquo their coordination-improving capabilities move them from ldquomakerdquoto ldquobuyrdquo The former shifts truck ownershipfrom for-hire carriers to shippers the latter fromshippers to for-hire carriers

The rst-difference estimates imply that allelse equal cohorts with a 20-percentage-pointhigher trip recorder adoption rate experienced a2-percentage-point greater increase in their pri-vate carriage share Likewise moving 20 per-cent of trucks from trip recorders to EVMScorresponds to a 3-percentage-point increase inthe for-hire carriage share These magnitudessuggest the OBCsrsquo incentive- and coordination-improving capabilities have an economicallyimportant impact on make-or-buy decisions inthe aggregate in light of the industryrsquos recenthistory We discuss magnitudes at more lengthin a subsection below

Table 4 breaks these results down furtherThe top panel reports rst-difference estimatesusing all cohorts and subsamples of short-medium- and long-haul cohorts The coef -cient on OBC is negative for all three sub-samples and statistically signi cant formedium- and long-haul cohorts The EVMScoef cient is positive and signi cant for allthree subsamples and of about the same mag-nitude The basic relationships between adop-tion and changes in asset ownership hold acrosshauls of different distances

The bottom part of the table estimates thecoef cients using only 1987 and 1992 thenonly 1992 and 1997 data In the earlier periodthe pattern of a negative coef cient on OBCand a positive one on EVMS only appearswithin the long-haul subsample In contrastthis pattern appears strongly and consistentlyin the later period In the later period thecoef cients on OBC are negative in each sub-sample and statistically signi cant for theshort- and medium-haul subsample Those onEVMS are positive and signi cant in eachsubsample The cross-year differences are in-teresting because they are consistent withwidespread speculation that organizationalchanges tend to lag IT adoption even whenthey are complementary

25 The control variables in the rst-difference speci ca-tions only include the dry van and auto trailer dummies andln(trailer density) because none of the coef cients on theother controls vary over time Changes in unobserved cohortcharacteristics may either re ect true changes or samplingerror See Angus Deaton (1985)

TABLE 3mdashOBC ADOPTION AND ASSET OWNERSHIP

Dependent variable For-hire carriage share

Levels estimates First-differences

OBC 2 0144 2 0090(0021) (0024)

EVMS 0239 0149(0024) (0028)

Notes SUR estimates Sample includes all cohorts withpositive number of observations in 1987 1992 and 1997N 5 2773 Cohorts are weighted using Censusrsquo weightingfactors times number of observations Speci cations includetrailer dummies mixed cargo dummy distance dummiesand ln(trailer density) as controls and allow the coef cienton the auto trailer and van dummy to vary across years toaccount for secular changes


A Interactions Multitasking Tests

In the model trip recorders affect optimalasset ownership indirectly by lowering theagency costs associated with multitasking If sothen the OBC coef cient should only be nega-tive for hauls where driversrsquo cargo-handlingeffort is potentially productive This is the basisof P3 above To examine this we create inter-actions between OBC adoption and product cat-egories One set is between adoption and adummy variable that equals one if the cohorthauls processed food or mixed cargo Truckshauling processed food or mixed cargo tend todeliver packaged goods to retail outlets Driv-ersrsquo cargo handling efforts are potentially morevaluable when they haul these goods than otherbulkier goods The other set is between adop-tion and a dummy variable that equals one if thecohort hauls petroleum or chemicals cargo forwhich handling requires certi cation We there-fore test whether the OBC coef cient is morenegative for these ldquomultitaskingrdquo cohorts thanothers

Table 5 summarizes the results The rst col-

umn uses data from all three years The coef -cient on OBC alone is small and statisticallyinsigni cant There is no relationship betweenOBC adoption and asset ownership when truckshaul goods in the omitted category which con-tains raw materials and bulky goods26 The in-teractions on OBC(food or mixed cargo) andon OBC(petroleum or chemicals) are bothnegative and the former is statistically signi -cant The other two columns report estimatesusing only two of the years The OBC owneffects are statistically zero in both periodsBoth interactions are negative and signi cant inthe late period of the sample The estimatesprovide support for P3 and are important evi-dence that OBCsrsquo incentive-improving capabil-ities affect asset ownership through job designThere is no evidence that incentive improve-ments affect the make-or-buy decision for hauls

26 The most prevalent product classes in the omittedcategory are fresh farm products building materials ma-chinery and lumber and wood products About 70 percentof cohorts are in the omitted category about 30 percent arein the ldquomultitaskingrdquo categories


All Short hauls Medium hauls Long hauls

Dependent variables Change in for-hire carriage shares 1987ndash1992 1992ndash1997

OBC 2 0090 20043 2 0103 2 0113(0024) (0049) (0040) (0040)

EVMS 0149 0183 0147 0159(0028) (0068) (0052) (0043)

N 2773 736 1019 1018

Dependent variable Change in for-hire carriage share 1987ndash1992

OBC 20043 0226 20087 2 0111(0031) (0066) (0052) (0052)

EVMS 0048 20138 20017 0157(0043) (0124) (0019) (0062)

N 3908 1115 1405 1388Dependent variable Change in for-hire carriage share 1992ndash1997

OBC 2 0087 2 0152 2 0116 20058(0026) (0055) (0047) (0040)

EVMS 0171 0125 0193 0157(0029) (0068) (0054) (0042)

N 4101 1097 1531 1473

Notes SUR estimates Includes all cohorts with positive number of observations in eachrelevant year Cohorts weighted by Censusrsquo weighting factors times number of observationsSpeci cations include change in trailer density and auto trailer and van dummies as controlsand allow the coef cient on the auto trailer and van dummy to vary across years to accountfor secular changes


where driversrsquo handling efforts are rarely pro-ductive When driversrsquo handling efforts tend tobe productive hauls for which trip recorderswere adopted moved from for-hire to privatecarriage

The rst-difference estimates are thus consis-tent with all three of our theoretical proposi-tions The following subsection providesadditional evidence regarding whether they in-deed re ect causal relationships between adop-tion and organizational change

B Instrumental Variables Estimates

As noted above alternative interpretations ofthe rst-difference estimates center on thepremise that OBC adoption and organizationalchanges might be independently affected bysome omitted factor For example if haulsrsquoregularity changes over time and increasesmore in some cohorts than others this couldlead independently to more private carriage andto trip recorder adoption This would makeadoption econometrically endogenous in the rst-difference estimates Similarly unobservedfactors may also explain why EVMS adoptionmight be correlated with movements towardfor-hire carriage Suppose unobserved shippercharacteristics change over time some shippers

may both establish more sophisticated logisticspractices and begin to value shipment tracingcapabilities This could lead independently toless private carriage (because driversrsquo handlingeffort is less valuable) and increased EVMSadoption

Below we present instrumental variables es-timates of the rst-difference speci cationsFactors that affect OBC adoption but do notdirectly affect organizational form are good in-struments We use four main instruments thefraction of miles trucks are operated outside oftheir base state the number of weeks per yeartrucks in the state are in use and dummy vari-ables that equal one if the truck is based in awestern state (ie west of Missouri) or in aNew England state These are computed at thecohort level hence the rst two of these arecohort-level averages Fraction of miles out ofstate affects OBC adoption because driversmust keep track of how many miles trucks areoperated in each state State fuel taxes are paidon this basis OBCs let drivers enter in thisinformation on a keypad and lower data entryand processing costs when trucksrsquo owners cal-culate the tax they owe each state This is morevaluable for trucks that spend more time outsideof their base state because they cross state linesmore State averages for number of weeks in

TABLE 5mdashOBC ADOPTION AND ASSET OWNERSHIPmdashINTERACTIONS

Multitasking Tests

Dependent variables Change in for-hire carriage share1987 1992 1997 1987ndash1992 1992ndash1997

OBC 20010 0033 0009(0036) (0046) (0034)

EVMS 20066 20002 0088(0040) (0057) (0039)

OBC(food or mixed cargo) 2 0141 2 0131 2 0241(0054) (0069) (0060)

EVMS(food or mixed cargo) 0142 0065 0164(0062) (0094) (0064)

OBC(petroleum or chemicals) 20111 20091 2 0184(0074) (0101) (0076)

EVMS(petroleum or chemicals) 0156 0021 0166(0092) (0171) (0089)

N 2773 3908 4101



use differ considerably across states rangingbetween 35 and 45 weeks and re ect differ-ences in the cyclicality of truck shipmentsMuch of this variation is likely climate relatedas the bottom ve states are Montana Wyo-ming North Dakota Alaska and South DakotaTrucks are idled more weeks per year andOBCsrsquo bene ts are correspondingly lower inareas where shipments are highly cyclical Weassume that statewide averages in the number ofweeks trucks are in use are unaffected by OBCadoption27 The two regional dummies are in-cluded because it is traditionally more dif cultfor drivers to contact dispatchers quickly in lessdensely populated areas This is one reason whyadoption tends to be above average in the Westbut below average in New England We usethese four variables and their interactions asinstruments for OBC and EVMS Table A1reports estimates from four simple ldquo rst-stagerdquospeci cations that regress cohort-level trip re-corder and EVMS adoption in the early and latesample periods on the four instruments and avector of controls

Table 6 contains estimates from speci ca-tions analogous to that in the rst column ofTable 5 but which restrict the interaction termsto be the same across the four ldquomultitaskingrdquoproduct types Looking at the rst column thecoef cient on OBC is nearly zero as beforethere is no evidence that trip recorders affectasset ownership for the ldquononmultitaskingrdquo co-horts The same is true for the EVMSMultinteraction there is no evidence that OBCsrsquocoordination-improving capabilitiesrsquo impactdiffers with whether multitasking is poten-tially productive The negative coef cient onOBCMult suggests that trip recorders movehauls toward private carriage more for the mul-titasking cohorts than the nonmultitasking onesThe positive coef cient on EVMS suggests thatEVMSrsquo coordination-improving capabilitiesmove hauls from private to for-hire carriageHowever these coef cients are not statisticallysigni cantly different from zero because theyare not estimated precisely The noisiness of the

estimates re ects that while three of our instru-ments are predictors of OBC adoption in gen-eral none of them shift trip recorder adoptionbut not EVMS adoption (See Table A1) Thismakes it hard to distinguish between the orga-nizational effects of OBCsrsquo incentive- andcoordination-improving capabilities in the in-strumental variables estimates

The bottom part of the table contains estimatesof (OBC1EVMS) and (OBC1EVMS)Multwhich re ect EVMSrsquo overall organizational im-pact We can estimate these more precisely con-sistent with our earlier results EVMS adoptionpushes hauls toward for-hire carriage but does soless for the multitasking cohorts than the nonmul-titasking cohorts The right column restricts thecoef cient on OBC to zero for the nonmultitask-ing cohorts a restriction suggested by the near-zero point estimates on this coef cient in Table5 The sign and signi cance of the remaining threecoef cients are the same as in the rst-differenceestimates

In sum our instrumental variables estimatesprovide no evidence that our rst-differenceestimates re ect noncausal relationships Al-

27 Individual trucks with OBCs do tend to be used moreweeks than those without them because trucks withoutOBCs are more likely to be idled when demand is low ButOBC adoption should have a minimal impact on number ofweeks averaged across all trucks in a state


GMM-IV Estimates

Dependent variables Change in for-hire carriage share1987ndash1992 1992ndash1997

OBC 0097(0351)

EVMS 0391 0502(0408) (0096)

OBCMult 20421 2 0351(0297) (0146)

EVMSMult 0098 0020(0326) (0158)

p-value for test of OID restrictions 0952 0964

(OBC1EVMS) 0488 0502(0105) (0095)

(OBC1EVMS)Mult 2 0322 2 0330(0143) (0141)

Notes Instruments include percent of miles out of basestate number of weeks in use per year West dummy NewEngland dummy and interactions among these Includes allcohorts with positive number of observations in each rele-vant year Cohorts weighted by Censusrsquo weighting factorstimes number of observations Speci cations includechange in trailer density and auto trailer and van dummiesas controls and allow the coef cient on the auto trailer andvan dummy to vary across years to account for secularchanges


though we cannot statistically distinguish be-tween the impact of OBCsrsquo incentive- andcoordination-related capabilities to the same de-gree the qualitative patterns that we are able toidentify are similar to those in our earlierresults

C Magnitudes

Although not the main focus of the study theestimates also indicate the degree to whichoverall changes in the private carriage sharebetween 1987 and 1997 were due to the diffu-sion of OBCs Table 7 summarizes our analysisThe top line reports the actual private carriageshares in our sample in each of the three yearsThe bottom part of the table reports the esti-mated shares absent OBC diffusion computedusing the simple and GMM-IV rst-differenceestimates from the right columns of Table 5 andthe right column of Table 6 respectively Thesimple rst-difference estimates suggest thatOBCs had little overall impact between 1987and 1992mdashthe diffusion of trip recorders andEVMS had offsetting effectsmdashbut caused about1 percentage point of the overall 29-percentage-point decline between 1992 and 1997 TheGMM-IV point estimates imply that OBCsrsquo over-all impact was much larger They indicate thatabsent OBC diffusion the private carriage sharewould have continued to increase to almost 60percent by 1997 One interpretation is that theorganizational impact of EVMSrsquo coordination-improving effects worked against a broad increasein 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 noisyWe are more con dent in stating the qualitativeconclusion that overall OBC diffusion played asigni cant and possibly large role in inducingshippers to outsource more during the 1990rsquos

VI Conclusion

In this paper we combine recent theoreticalwork from organizational economics with a de-tailed and disaggregated data set to gain insightabout the interaction between asset ownershipjob design and information Of particular im-portance is our ability to distinguish betweeninformational changes that lead either to bettermonitoring of agents or to better coordinationof activities We believe that our resultsmdashthatimproved monitoring technologies lead ship-pers to integrate into trucking while technolo-gies that improve coordination lead to moreoutsourcing of trucking servicesmdashhighlight therespective advantages of rms and markets inthe economy and thus shed light on their rolesin the operation and development of economicsystems

In describing and explaining the developmentof nineteenth-century capitalism Alfred PChandler (1977) and others have argued that thedevelopment of new communication technolo-gies (eg the telegraph) enabled the growth oflarge integrated rms Large transportationmanufacturing and retailing rms were impos-sible without a technology that enabled manag-ers to coordinate large-scale economic activityYet we have found exactly the opposite effect inlate twentieth-century trucking a new commu-nications technology that improved coordina-tion led to smaller less integrated rms Whythe difference

We believe that our new results arise becausewe distinguish between informational changesthat improve coordination from those that im-prove incentives Such a distinction is rare inempirical work on organizations Yet it is es-sential to understanding the true role of rmsand markets as competing mechanisms for or-ganizing economic activity F A Hayek (1945)argued that the true value of the market-basedprice system is its ability to utilize dispersedinformation about resources and coordinatetheir use in a way that no centrally plannedeconomy (or rm) ever could Given this com-

TABLE 7mdashOBC DIFFUSION AND THE PRIVATE

CARRIAGE SHARE

True share

1987 1992 1997

050 055 052

Estimated share absentOBC adoption

1987 1992 1997

Table 5 (right columns) 050 054 053Table 6 (right column) 050 060


parative advantage of markets over rms it isnot surprising that a technological change thatmitigates the Hayekian coordination problemshould lead to a greater relative improvement inthe ef ciency of markets

Holmstrom and Milgrom (1994) and Holm-strom (1999) by contrast argue that the trueadvantage of rms over markets is their abilityto craft delicately balanced incentives for agentsengaged in multiple activities in a way that thestrong incentives generated by markets and as-set ownership cannot Given this comparativeadvantage for rms it is again not surprisingthat a technological change that mitigates con-tracting problems should lead to a greater rela-tive improvement in the ef ciency of rms

Information costs are at the core of nearlyall economic theories of organizations Thusall of these theories predict that changes ininformation technology that change the costof contracting and communication will affectthe organization of economic activity We nd that the answer to the question ldquoHas ITadoption led to larger or smaller rms intruckingrdquo is ldquoYesrdquo and show how the orga-nizational implications of ITrsquos incentive- andcoordination-improving capabilities system-atically differ Future research will furtherinform debates regarding the organizationalimplications of changes in information costsby investigating whether this systematic dif-ference in trucking is general

REFERENCES

American Trucking Associations US freighttransportation forecast to 2007 AlexandriaVA American Trucking Associations 1999

Baker George P ldquoIncentive Contracts and Per-formance Measurementrdquo Journal of PoliticalEconomy June 1992 100(3) pp 598ndash614

Baker George P and Hubbard Thomas N ldquoCon-tractibility and Asset Ownership On-BoardComputers and Governance in US Truck-ingrdquo National Bureau of Economic Research(Cambridge MA) Working Paper No 76342000

ldquoMake Versus Buy in Trucking AssetOwnership Job Design and Informationrdquo

National Bureau of Economic Research(Cambridge MA) Working Paper No 87272002

Brynjolfsson Erik and Hitt Lorin ldquoInformationTechnology and Organizational DesignSome Evidence From Micro Datardquo Workingpaper Stanford University 1997

Bureau of the Census 1992 Census of transpor-tation Truck inventory and use surveyWashington DC US Government PrintingOf ce 1995

1997 Census of transportation Ve-hicle inventory and use survey Washing-ton DC US Government Printing Of ce1999a

1997 Census of transportation

TABLE A1mdashFIRST-STAGE ESTIMATES ADOPTION EQUATIONS

Dependent variableTR adoption1987ndash1992

TR adoption1992ndash1997

EVMS adoption1987ndash1992


West 0039 20012 0037 0002(0008) (0012) (0009) (0015)

New England 0015 20021 20024 0012(0028) (0038) (0029) (0050)

Percent out of base state 0061 20023 0127 0091(0020) (0025) (0021) (0032)

Weeks in use 0004 20002 0007 20001(0002) (0002) (0002) (0003)

N 5 2773

Notes All speci cations include distance trailer and product class dummies and the privatecarriage share Includes all cohorts with positive number of observations in each relevant yearCohorts weighted by Censusrsquo weighting factors times number of observations


Commodity ow survey Washington DCUS Government Printing Of ce 1999b

Chandler Alfred P The visible hand The man-agerial revolution in American businessCambridge MA Belknap 1977

Deaton Angus ldquoPanel Data from Time Series ofCross-Sectionsrdquo Journal of EconometricsOctoberndashNovember 1985 30(1ndash2) pp 109ndash26

Grossman Sanford and Hart Oliver ldquoThe Costsand Bene ts of Ownership A Theory of Ver-tical and Lateral Integrationrdquo Journal of Po-litical Economy August 1986 94(4) pp691ndash719

Hayek F A ldquoThe Use of Knowledge in Soci-etyrdquo American Economic Review September1945 35(4) pp 519ndash30

Holmstrom Bengt ldquoThe Firm as a Sub-economyrdquo Journal of Law Economics andOrganization April 1999 15(1) pp 74ndash102

Holmstrom Bengt and Milgrom Paul ldquoMultitaskPrincipalndashAgent Analyses Incentive Con-tracts Asset Ownership and Job DesignrdquoJournal of Law Economics and Organiza-tion 1991 Spec Iss 7 pp 24ndash52

ldquoThe Firm as an Incentive SystemrdquoAmerican Economic Review September1994 84(4) pp 972ndash91

Hubbard Thomas N ldquoGovernance Structure inthe Deregulated Trucking Industryrdquo Work-ing paper UCLA 1998

ldquoThe Demand for Monitoring Tech-nologies The Case of Truckingrdquo QuarterlyJournal of Economics May 2000 115(2) pp533ndash76

ldquoContractual Form and Market Thick-ness in Truckingrdquo RAND Journal of Eco-nomics Summer 2001 32(2) pp 369ndash86

ldquoInformation Decisions and Produc-tivity On-Board Computers and Capacity

Utilization in Truckingrdquo American EconomicReview 2003 (forthcoming)

Jensen Michael C and Meckling William HldquoTheory of the Firm Managerial BehaviorAgency Costs and Ownership StructurerdquoJournal of Financial Economics October1976 3(4) pp 305ndash60

ldquoSpeci c and General Knowledge andOrganizational Structurerdquo in Lars Werin andHans Wijkander eds Contract economicsCambridge Blackwell 1992 pp 251ndash91

Leavitt H J and Whisler T L ldquoManagement inthe 1980srdquo Harvard Business Review No-vember 1958 36(6) pp 41ndash48

Malone Thomas W Yates JoAnne and Ben-jamin Robert I ldquoElectronic Markets andElectronic Hierarchiesrdquo Communications ofthe ACM June 1987 30(6) pp 484ndash97

Milgrom Paul and Roberts John ldquoBargainingCosts In uence Costs and the Organizationof Economic Activityrdquo in James E Alt andKenneth A Shepsle eds Perspectives onpositive political economy Cambridge Cam-bridge University Press 1990 pp 57ndash89

Milgrom Paul and Shannon Christina ldquoMono-tone Comparative Staticsrdquo EconometricaJanuary 1994 62(2) pp 157ndash80

Ouellet Lawrence J Pedal to the metal Thework lives of truckers Philadelphia PATemple University Press 1994

Thomas Jim ldquoPrivate Fleets Down But NotOutrdquo Logistics Management and Distribu-tion Report July 1998 37(7) pp 31ndash32

Topkis Donald M ldquoMinimizing a SubmodularFunction on a Latticerdquo Operations ResearchMarchndashApril 1978 26(2) pp 305ndash21

Wilson Rosalyn A Transportation in America2000 with historical compendium 1939ndash1999Washington DC Eno Transportation Foun-dation 2001


theoretical insights The model generates twosets of comparative static predictions One setof predictions is consistent with well-knowncross-sectional patterns in the industry Theseinclude the prediction that service-intensivetrucking is more likely to be performed byprivate than for-hire eets and that pri-vate eets are differentially more likely toadopt incentive-improving technologies whilefor-hire carriers are more likely to adoptcoordination-improving technologies

The other set of predictions concerns howchanges in the informational environment affectownership We test this second set of predic-tions using data from the 1987 1992 and 1997Truck Inventory and Use Surveys which con-tain detailed truck-level information abouttrucksrsquo characteristics ownership and use Inparticular we test predictions on how the dif-fusion of different types of on-board computers(OBCs) during the late 1980rsquos and early 1990rsquosalters the ldquomake versus buyrdquo decision for ship-pers We predict that the adoption of certaintypes of OBCs should lead indirectly to moreshipper ownership of trucks by lowering theagency costs associated with complex job de-signs We predict that the additional capabilitiesof other types of OBCsmdashthose that provide loca-tion information and real-time communicationmdashshould lead to less shipper ownership of trucksbecause these additional capabilities enhancethe comparative advantage of for-hire carriagewith respect to truck utilization and dispatch We nd evidence in favor of both of these predictions

Our results strongly suggest causal links be-tween informational and organizational changesin the trucking industry They show that own-ership patterns in trucking re ect the impor-tance of not only incomplete contracts (asstressed by Grossman and Hart 1986) but alsoof job design and measurement issues (likethose stressed in Holmstrom and Milgrom1994) These ndings thus shed important lighton theories of organizations They also make acontribution to the long-running debate abouthow information technology (IT) diffusion af-fects the boundaries of the rm1 We note that

information technology in general provides atleast two capabilitiesmdashimproved monitoring ofagents and improved coordination of activi-tiesmdashand that the organizational impact ofthese capabilities can differ (Michael C Jensenand William H Meckling 1992) In truckingimprovements in monitoring (and the attendantimprovement in incentives) lead to larger moreintegrated rms while improvements in coordi-nation (resulting in better asset utilization) leadto more diffuse asset ownership and smallerless integrated rms Whether these results gen-eralize to other settings remains an openquestion

In this paper we do not consider a third pos-sibility regarding truck ownership drivers mayown trucks We investigate driver ownership oftrucks in detail in another paper (Baker andHubbard 2000) In that paper we propose thatasset ownership strengthens driversrsquo incentivesto drive in ways that preserve trucksrsquo value butalso encourages them to engage in rent-seekingbehavior We then argue that OBC adoptionalters this trade-off by allowing companies (ei-ther for-hire carriers or private eets) to use themonitoring capabilities of OBCs to substitutefor asset ownership We show that OBCs lead toless driver ownership of trucks especially forhauls where rent-seeking is a potential problemWe ignore these issues in the present paperbecause we believe they are not salient to themake-or-buy decision Situations that are on themargin between for-hire carriage and privatecarriage are not those where owner-operatorsare used In general owner-operators are usedfor hauls that require little if any service provi-sion by the driver and for good reason Themultitasking problems with service provisionthat lead for-hire carriage to be inef cient rela-tive to private carriage are exacerbated whendrivers control trucks This is borne out by thefact that when shippers outsource hauls withnonnegligible service requirements they rarelyif ever do so by contracting with owner-operators2

The paper is organized as follows In the nextsection we describe the institutional setting that

1 See H J Leavitt and T L Whisler (1958) Thomas WMalone et al (1987) Erik Brynjolfsson and Lorin Hitt(1997)

2 In our empirical tests ldquofor-hire carriagerdquo includes driver-and carrier-owned trucks ownership structures where ship-pers do not own trucks Our results are unchanged when weleave out owner-operators altogether


















































(1) V 5 V 1 ms 2 M~s s






B Load Matching



(2) ~g1 2 use 1




(3) V 5 V 1 ~g1 2 use1 1 ms 2 M~s s




(4) P 5 g2 e2 1 ~g1 2 use 1 1 ms 2 M~s s





(5) ~V 1 P2 2 12

e12 2 1

2e2

2

5 ~V 1 2~g1 2 use1 1 2ms 2 2M~s s

1 g2 e2 )2 2 12

e12 2 1

2e2

2







(6) e1F 5 ~g1 2 us e2

F 5 12

g2


(7) TVF 5 V 1 12

~g1 2 us2

2 18

g22 1 ms 2 M~s s


(8) V2 2 12

e12 2 1

2e 2

2

5 ~V 1 ~g1 2 use1 1 ms 2 M~s s2

2 12

e12 2 1

2e2

2

yielding effort of

(9) e1P 5 1

2~g1 2 us e2

P 5 0


(10)

TVP 5 V 1 38

~g1 2 us2 1 ms 2 M~s s



(11) TV~s d 5 V 1 18

~3 1 d~g1 2 us2

2 18

dg22 1 ms 2 M~s s
















(12) TV~s d 5 V 1 18

~3 1 dg12 2 1

8dg2

2


























IV Data
















A Cohort Data












1987 1992 1997 1987 1992 1992 1997





V Results









Cohort Data




















OBC 2 0144 2 0090(0021) (0024)

EVMS 0239 0149(0024) (0028)











OBC 2 0090 20043 2 0103 2 0113(0024) (0049) (0040) (0040)

EVMS 0149 0183 0147 0159(0028) (0068) (0052) (0043)

N 2773 736 1019 1018


OBC 20043 0226 20087 2 0111(0031) (0066) (0052) (0052)

EVMS 0048 20138 20017 0157(0043) (0124) (0019) (0062)


OBC 2 0087 2 0152 2 0116 20058(0026) (0055) (0047) (0040)

EVMS 0171 0125 0193 0157(0029) (0068) (0054) (0042)

N 4101 1097 1531 1473










Multitasking Tests


OBC 20010 0033 0009(0036) (0046) (0034)

EVMS 20066 20002 0088(0040) (0057) (0039)





N 2773 3908 4101










GMM-IV Estimates


OBC 0097(0351)

EVMS 0391 0502(0408) (0096)

OBCMult 20421 2 0351(0297) (0146)

EVMSMult 0098 0020(0326) (0158)


(OBC1EVMS) 0488 0502(0105) (0095)

(OBC1EVMS)Mult 2 0322 2 0330(0143) (0141)




C Magnitudes




VI Conclusion





CARRIAGE SHARE

True share

1987 1992 1997

050 055 052


1987 1992 1997






REFERENCES















West 0039 20012 0037 0002(0008) (0012) (0009) (0015)

New England 0015 20021 20024 0012(0028) (0038) (0029) (0050)


Weeks in use 0004 20002 0007 20001(0002) (0002) (0002) (0003)

N 5 2773











































































(1) V 5 V 1 ms 2 M~s s






B Load Matching



(2) ~g1 2 use 1




(3) V 5 V 1 ~g1 2 use1 1 ms 2 M~s s




(4) P 5 g2 e2 1 ~g1 2 use 1 1 ms 2 M~s s





(5) ~V 1 P2 2 12

e12 2 1

2e2

2

5 ~V 1 2~g1 2 use1 1 2ms 2 2M~s s

1 g2 e2 )2 2 12

e12 2 1

2e2

2







(6) e1F 5 ~g1 2 us e2

F 5 12

g2


(7) TVF 5 V 1 12

~g1 2 us2

2 18

g22 1 ms 2 M~s s


(8) V2 2 12

e12 2 1

2e 2

2

5 ~V 1 ~g1 2 use1 1 ms 2 M~s s2

2 12

e12 2 1

2e2

2

yielding effort of

(9) e1P 5 1

2~g1 2 us e2

P 5 0


(10)

TVP 5 V 1 38

~g1 2 us2 1 ms 2 M~s s



(11) TV~s d 5 V 1 18

~3 1 d~g1 2 us2

2 18

dg22 1 ms 2 M~s s
















(12) TV~s d 5 V 1 18

~3 1 dg12 2 1

8dg2

2


























IV Data
















A Cohort Data












1987 1992 1997 1987 1992 1992 1997





V Results









Cohort Data




















OBC 2 0144 2 0090(0021) (0024)

EVMS 0239 0149(0024) (0028)











OBC 2 0090 20043 2 0103 2 0113(0024) (0049) (0040) (0040)

EVMS 0149 0183 0147 0159(0028) (0068) (0052) (0043)

N 2773 736 1019 1018


OBC 20043 0226 20087 2 0111(0031) (0066) (0052) (0052)

EVMS 0048 20138 20017 0157(0043) (0124) (0019) (0062)


OBC 2 0087 2 0152 2 0116 20058(0026) (0055) (0047) (0040)

EVMS 0171 0125 0193 0157(0029) (0068) (0054) (0042)

N 4101 1097 1531 1473










Multitasking Tests


OBC 20010 0033 0009(0036) (0046) (0034)

EVMS 20066 20002 0088(0040) (0057) (0039)





N 2773 3908 4101










GMM-IV Estimates


OBC 0097(0351)

EVMS 0391 0502(0408) (0096)

OBCMult 20421 2 0351(0297) (0146)

EVMSMult 0098 0020(0326) (0158)


(OBC1EVMS) 0488 0502(0105) (0095)

(OBC1EVMS)Mult 2 0322 2 0330(0143) (0141)




C Magnitudes




VI Conclusion





CARRIAGE SHARE

True share

1987 1992 1997

050 055 052


1987 1992 1997






REFERENCES















West 0039 20012 0037 0002(0008) (0012) (0009) (0015)

New England 0015 20021 20024 0012(0028) (0038) (0029) (0050)


Weeks in use 0004 20002 0007 20001(0002) (0002) (0002) (0003)

N 5 2773




























B Load Matching



(2) ~g1 2 use 1




(3) V 5 V 1 ~g1 2 use1 1 ms 2 M~s s




(4) P 5 g2 e2 1 ~g1 2 use 1 1 ms 2 M~s s





(5) ~V 1 P2 2 12

e12 2 1

2e2

2

5 ~V 1 2~g1 2 use1 1 2ms 2 2M~s s

1 g2 e2 )2 2 12

e12 2 1

2e2

2







(6) e1F 5 ~g1 2 us e2

F 5 12

g2


(7) TVF 5 V 1 12

~g1 2 us2

2 18

g22 1 ms 2 M~s s


(8) V2 2 12

e12 2 1

2e 2

2

5 ~V 1 ~g1 2 use1 1 ms 2 M~s s2

2 12

e12 2 1

2e2

2

yielding effort of

(9) e1P 5 1

2~g1 2 us e2

P 5 0


(10)

TVP 5 V 1 38

~g1 2 us2 1 ms 2 M~s s



(11) TV~s d 5 V 1 18

~3 1 d~g1 2 us2

2 18

dg22 1 ms 2 M~s s
















(12) TV~s d 5 V 1 18

~3 1 dg12 2 1

8dg2

2


























IV Data
















A Cohort Data












1987 1992 1997 1987 1992 1992 1997





V Results









Cohort Data




















OBC 2 0144 2 0090(0021) (0024)

EVMS 0239 0149(0024) (0028)











OBC 2 0090 20043 2 0103 2 0113(0024) (0049) (0040) (0040)

EVMS 0149 0183 0147 0159(0028) (0068) (0052) (0043)

N 2773 736 1019 1018


OBC 20043 0226 20087 2 0111(0031) (0066) (0052) (0052)

EVMS 0048 20138 20017 0157(0043) (0124) (0019) (0062)


OBC 2 0087 2 0152 2 0116 20058(0026) (0055) (0047) (0040)

EVMS 0171 0125 0193 0157(0029) (0068) (0054) (0042)

N 4101 1097 1531 1473










Multitasking Tests


OBC 20010 0033 0009(0036) (0046) (0034)

EVMS 20066 20002 0088(0040) (0057) (0039)





N 2773 3908 4101










GMM-IV Estimates


OBC 0097(0351)

EVMS 0391 0502(0408) (0096)

OBCMult 20421 2 0351(0297) (0146)

EVMSMult 0098 0020(0326) (0158)


(OBC1EVMS) 0488 0502(0105) (0095)

(OBC1EVMS)Mult 2 0322 2 0330(0143) (0141)




C Magnitudes




VI Conclusion





CARRIAGE SHARE

True share

1987 1992 1997

050 055 052


1987 1992 1997






REFERENCES















West 0039 20012 0037 0002(0008) (0012) (0009) (0015)

New England 0015 20021 20024 0012(0028) (0038) (0029) (0050)


Weeks in use 0004 20002 0007 20001(0002) (0002) (0002) (0003)

N 5 2773




























(6) e1F 5 ~g1 2 us e2

F 5 12

g2


(7) TVF 5 V 1 12

~g1 2 us2

2 18

g22 1 ms 2 M~s s


(8) V2 2 12

e12 2 1

2e 2

2

5 ~V 1 ~g1 2 use1 1 ms 2 M~s s2

2 12

e12 2 1

2e2

2

yielding effort of

(9) e1P 5 1

2~g1 2 us e2

P 5 0


(10)

TVP 5 V 1 38

~g1 2 us2 1 ms 2 M~s s



(11) TV~s d 5 V 1 18

~3 1 d~g1 2 us2

2 18

dg22 1 ms 2 M~s s
















(12) TV~s d 5 V 1 18

~3 1 dg12 2 1

8dg2

2


























IV Data
















A Cohort Data












1987 1992 1997 1987 1992 1992 1997





V Results









Cohort Data




















OBC 2 0144 2 0090(0021) (0024)

EVMS 0239 0149(0024) (0028)











OBC 2 0090 20043 2 0103 2 0113(0024) (0049) (0040) (0040)

EVMS 0149 0183 0147 0159(0028) (0068) (0052) (0043)

N 2773 736 1019 1018


OBC 20043 0226 20087 2 0111(0031) (0066) (0052) (0052)

EVMS 0048 20138 20017 0157(0043) (0124) (0019) (0062)


OBC 2 0087 2 0152 2 0116 20058(0026) (0055) (0047) (0040)

EVMS 0171 0125 0193 0157(0029) (0068) (0054) (0042)

N 4101 1097 1531 1473










Multitasking Tests


OBC 20010 0033 0009(0036) (0046) (0034)

EVMS 20066 20002 0088(0040) (0057) (0039)





N 2773 3908 4101










GMM-IV Estimates


OBC 0097(0351)

EVMS 0391 0502(0408) (0096)

OBCMult 20421 2 0351(0297) (0146)

EVMSMult 0098 0020(0326) (0158)


(OBC1EVMS) 0488 0502(0105) (0095)

(OBC1EVMS)Mult 2 0322 2 0330(0143) (0141)




C Magnitudes




VI Conclusion





CARRIAGE SHARE

True share

1987 1992 1997

050 055 052


1987 1992 1997






REFERENCES















West 0039 20012 0037 0002(0008) (0012) (0009) (0015)

New England 0015 20021 20024 0012(0028) (0038) (0029) (0050)


Weeks in use 0004 20002 0007 20001(0002) (0002) (0002) (0003)

N 5 2773
































(12) TV~s d 5 V 1 18

~3 1 dg12 2 1

8dg2

2


























IV Data
















A Cohort Data












1987 1992 1997 1987 1992 1992 1997





V Results









Cohort Data




















OBC 2 0144 2 0090(0021) (0024)

EVMS 0239 0149(0024) (0028)











OBC 2 0090 20043 2 0103 2 0113(0024) (0049) (0040) (0040)

EVMS 0149 0183 0147 0159(0028) (0068) (0052) (0043)

N 2773 736 1019 1018


OBC 20043 0226 20087 2 0111(0031) (0066) (0052) (0052)

EVMS 0048 20138 20017 0157(0043) (0124) (0019) (0062)


OBC 2 0087 2 0152 2 0116 20058(0026) (0055) (0047) (0040)

EVMS 0171 0125 0193 0157(0029) (0068) (0054) (0042)

N 4101 1097 1531 1473










Multitasking Tests


OBC 20010 0033 0009(0036) (0046) (0034)

EVMS 20066 20002 0088(0040) (0057) (0039)





N 2773 3908 4101










GMM-IV Estimates


OBC 0097(0351)

EVMS 0391 0502(0408) (0096)

OBCMult 20421 2 0351(0297) (0146)

EVMSMult 0098 0020(0326) (0158)


(OBC1EVMS) 0488 0502(0105) (0095)

(OBC1EVMS)Mult 2 0322 2 0330(0143) (0141)




C Magnitudes




VI Conclusion





CARRIAGE SHARE

True share

1987 1992 1997

050 055 052


1987 1992 1997






REFERENCES















West 0039 20012 0037 0002(0008) (0012) (0009) (0015)

New England 0015 20021 20024 0012(0028) (0038) (0029) (0050)


Weeks in use 0004 20002 0007 20001(0002) (0002) (0002) (0003)

N 5 2773












































IV Data
















A Cohort Data












1987 1992 1997 1987 1992 1992 1997





V Results









Cohort Data




















OBC 2 0144 2 0090(0021) (0024)

EVMS 0239 0149(0024) (0028)











OBC 2 0090 20043 2 0103 2 0113(0024) (0049) (0040) (0040)

EVMS 0149 0183 0147 0159(0028) (0068) (0052) (0043)

N 2773 736 1019 1018


OBC 20043 0226 20087 2 0111(0031) (0066) (0052) (0052)

EVMS 0048 20138 20017 0157(0043) (0124) (0019) (0062)


OBC 2 0087 2 0152 2 0116 20058(0026) (0055) (0047) (0040)

EVMS 0171 0125 0193 0157(0029) (0068) (0054) (0042)

N 4101 1097 1531 1473










Multitasking Tests


OBC 20010 0033 0009(0036) (0046) (0034)

EVMS 20066 20002 0088(0040) (0057) (0039)





N 2773 3908 4101










GMM-IV Estimates


OBC 0097(0351)

EVMS 0391 0502(0408) (0096)

OBCMult 20421 2 0351(0297) (0146)

EVMSMult 0098 0020(0326) (0158)


(OBC1EVMS) 0488 0502(0105) (0095)

(OBC1EVMS)Mult 2 0322 2 0330(0143) (0141)




C Magnitudes




VI Conclusion





CARRIAGE SHARE

True share

1987 1992 1997

050 055 052


1987 1992 1997






REFERENCES















West 0039 20012 0037 0002(0008) (0012) (0009) (0015)

New England 0015 20021 20024 0012(0028) (0038) (0029) (0050)


Weeks in use 0004 20002 0007 20001(0002) (0002) (0002) (0003)

N 5 2773





































A Cohort Data












1987 1992 1997 1987 1992 1992 1997





V Results









Cohort Data




















OBC 2 0144 2 0090(0021) (0024)

EVMS 0239 0149(0024) (0028)











OBC 2 0090 20043 2 0103 2 0113(0024) (0049) (0040) (0040)

EVMS 0149 0183 0147 0159(0028) (0068) (0052) (0043)

N 2773 736 1019 1018


OBC 20043 0226 20087 2 0111(0031) (0066) (0052) (0052)

EVMS 0048 20138 20017 0157(0043) (0124) (0019) (0062)


OBC 2 0087 2 0152 2 0116 20058(0026) (0055) (0047) (0040)

EVMS 0171 0125 0193 0157(0029) (0068) (0054) (0042)

N 4101 1097 1531 1473










Multitasking Tests


OBC 20010 0033 0009(0036) (0046) (0034)

EVMS 20066 20002 0088(0040) (0057) (0039)





N 2773 3908 4101










GMM-IV Estimates


OBC 0097(0351)

EVMS 0391 0502(0408) (0096)

OBCMult 20421 2 0351(0297) (0146)

EVMSMult 0098 0020(0326) (0158)


(OBC1EVMS) 0488 0502(0105) (0095)

(OBC1EVMS)Mult 2 0322 2 0330(0143) (0141)




C Magnitudes




VI Conclusion





CARRIAGE SHARE

True share

1987 1992 1997

050 055 052


1987 1992 1997






REFERENCES















West 0039 20012 0037 0002(0008) (0012) (0009) (0015)

New England 0015 20021 20024 0012(0028) (0038) (0029) (0050)


Weeks in use 0004 20002 0007 20001(0002) (0002) (0002) (0003)

N 5 2773


































1987 1992 1997 1987 1992 1992 1997





V Results









Cohort Data




















OBC 2 0144 2 0090(0021) (0024)

EVMS 0239 0149(0024) (0028)











OBC 2 0090 20043 2 0103 2 0113(0024) (0049) (0040) (0040)

EVMS 0149 0183 0147 0159(0028) (0068) (0052) (0043)

N 2773 736 1019 1018


OBC 20043 0226 20087 2 0111(0031) (0066) (0052) (0052)

EVMS 0048 20138 20017 0157(0043) (0124) (0019) (0062)


OBC 2 0087 2 0152 2 0116 20058(0026) (0055) (0047) (0040)

EVMS 0171 0125 0193 0157(0029) (0068) (0054) (0042)

N 4101 1097 1531 1473










Multitasking Tests


OBC 20010 0033 0009(0036) (0046) (0034)

EVMS 20066 20002 0088(0040) (0057) (0039)





N 2773 3908 4101










GMM-IV Estimates


OBC 0097(0351)

EVMS 0391 0502(0408) (0096)

OBCMult 20421 2 0351(0297) (0146)

EVMSMult 0098 0020(0326) (0158)


(OBC1EVMS) 0488 0502(0105) (0095)

(OBC1EVMS)Mult 2 0322 2 0330(0143) (0141)




C Magnitudes




VI Conclusion





CARRIAGE SHARE

True share

1987 1992 1997

050 055 052


1987 1992 1997






REFERENCES















West 0039 20012 0037 0002(0008) (0012) (0009) (0015)

New England 0015 20021 20024 0012(0028) (0038) (0029) (0050)


Weeks in use 0004 20002 0007 20001(0002) (0002) (0002) (0003)

N 5 2773




























V Results









Cohort Data




















OBC 2 0144 2 0090(0021) (0024)

EVMS 0239 0149(0024) (0028)











OBC 2 0090 20043 2 0103 2 0113(0024) (0049) (0040) (0040)

EVMS 0149 0183 0147 0159(0028) (0068) (0052) (0043)

N 2773 736 1019 1018


OBC 20043 0226 20087 2 0111(0031) (0066) (0052) (0052)

EVMS 0048 20138 20017 0157(0043) (0124) (0019) (0062)


OBC 2 0087 2 0152 2 0116 20058(0026) (0055) (0047) (0040)

EVMS 0171 0125 0193 0157(0029) (0068) (0054) (0042)

N 4101 1097 1531 1473










Multitasking Tests


OBC 20010 0033 0009(0036) (0046) (0034)

EVMS 20066 20002 0088(0040) (0057) (0039)





N 2773 3908 4101










GMM-IV Estimates


OBC 0097(0351)

EVMS 0391 0502(0408) (0096)

OBCMult 20421 2 0351(0297) (0146)

EVMSMult 0098 0020(0326) (0158)


(OBC1EVMS) 0488 0502(0105) (0095)

(OBC1EVMS)Mult 2 0322 2 0330(0143) (0141)




C Magnitudes




VI Conclusion





CARRIAGE SHARE

True share

1987 1992 1997

050 055 052


1987 1992 1997






REFERENCES















West 0039 20012 0037 0002(0008) (0012) (0009) (0015)

New England 0015 20021 20024 0012(0028) (0038) (0029) (0050)


Weeks in use 0004 20002 0007 20001(0002) (0002) (0002) (0003)

N 5 2773






































OBC 2 0144 2 0090(0021) (0024)

EVMS 0239 0149(0024) (0028)











OBC 2 0090 20043 2 0103 2 0113(0024) (0049) (0040) (0040)

EVMS 0149 0183 0147 0159(0028) (0068) (0052) (0043)

N 2773 736 1019 1018


OBC 20043 0226 20087 2 0111(0031) (0066) (0052) (0052)

EVMS 0048 20138 20017 0157(0043) (0124) (0019) (0062)


OBC 2 0087 2 0152 2 0116 20058(0026) (0055) (0047) (0040)

EVMS 0171 0125 0193 0157(0029) (0068) (0054) (0042)

N 4101 1097 1531 1473










Multitasking Tests


OBC 20010 0033 0009(0036) (0046) (0034)

EVMS 20066 20002 0088(0040) (0057) (0039)





N 2773 3908 4101










GMM-IV Estimates


OBC 0097(0351)

EVMS 0391 0502(0408) (0096)

OBCMult 20421 2 0351(0297) (0146)

EVMSMult 0098 0020(0326) (0158)


(OBC1EVMS) 0488 0502(0105) (0095)

(OBC1EVMS)Mult 2 0322 2 0330(0143) (0141)




C Magnitudes




VI Conclusion





CARRIAGE SHARE

True share

1987 1992 1997

050 055 052


1987 1992 1997






REFERENCES















West 0039 20012 0037 0002(0008) (0012) (0009) (0015)

New England 0015 20021 20024 0012(0028) (0038) (0029) (0050)


Weeks in use 0004 20002 0007 20001(0002) (0002) (0002) (0003)

N 5 2773



































OBC 2 0090 20043 2 0103 2 0113(0024) (0049) (0040) (0040)

EVMS 0149 0183 0147 0159(0028) (0068) (0052) (0043)

N 2773 736 1019 1018


OBC 20043 0226 20087 2 0111(0031) (0066) (0052) (0052)

EVMS 0048 20138 20017 0157(0043) (0124) (0019) (0062)


OBC 2 0087 2 0152 2 0116 20058(0026) (0055) (0047) (0040)

EVMS 0171 0125 0193 0157(0029) (0068) (0054) (0042)

N 4101 1097 1531 1473










Multitasking Tests


OBC 20010 0033 0009(0036) (0046) (0034)

EVMS 20066 20002 0088(0040) (0057) (0039)





N 2773 3908 4101










GMM-IV Estimates


OBC 0097(0351)

EVMS 0391 0502(0408) (0096)

OBCMult 20421 2 0351(0297) (0146)

EVMSMult 0098 0020(0326) (0158)


(OBC1EVMS) 0488 0502(0105) (0095)

(OBC1EVMS)Mult 2 0322 2 0330(0143) (0141)




C Magnitudes




VI Conclusion





CARRIAGE SHARE

True share

1987 1992 1997

050 055 052


1987 1992 1997






REFERENCES















West 0039 20012 0037 0002(0008) (0012) (0009) (0015)

New England 0015 20021 20024 0012(0028) (0038) (0029) (0050)


Weeks in use 0004 20002 0007 20001(0002) (0002) (0002) (0003)

N 5 2773


































Multitasking Tests


OBC 20010 0033 0009(0036) (0046) (0034)

EVMS 20066 20002 0088(0040) (0057) (0039)





N 2773 3908 4101










GMM-IV Estimates


OBC 0097(0351)

EVMS 0391 0502(0408) (0096)

OBCMult 20421 2 0351(0297) (0146)

EVMSMult 0098 0020(0326) (0158)


(OBC1EVMS) 0488 0502(0105) (0095)

(OBC1EVMS)Mult 2 0322 2 0330(0143) (0141)




C Magnitudes




VI Conclusion





CARRIAGE SHARE

True share

1987 1992 1997

050 055 052


1987 1992 1997






REFERENCES















West 0039 20012 0037 0002(0008) (0012) (0009) (0015)

New England 0015 20021 20024 0012(0028) (0038) (0029) (0050)


Weeks in use 0004 20002 0007 20001(0002) (0002) (0002) (0003)

N 5 2773


































GMM-IV Estimates


OBC 0097(0351)

EVMS 0391 0502(0408) (0096)

OBCMult 20421 2 0351(0297) (0146)

EVMSMult 0098 0020(0326) (0158)


(OBC1EVMS) 0488 0502(0105) (0095)

(OBC1EVMS)Mult 2 0322 2 0330(0143) (0141)




C Magnitudes




VI Conclusion





CARRIAGE SHARE

True share

1987 1992 1997

050 055 052


1987 1992 1997






REFERENCES















West 0039 20012 0037 0002(0008) (0012) (0009) (0015)

New England 0015 20021 20024 0012(0028) (0038) (0029) (0050)


Weeks in use 0004 20002 0007 20001(0002) (0002) (0002) (0003)

N 5 2773




























C Magnitudes




VI Conclusion





CARRIAGE SHARE

True share

1987 1992 1997

050 055 052


1987 1992 1997






REFERENCES















West 0039 20012 0037 0002(0008) (0012) (0009) (0015)

New England 0015 20021 20024 0012(0028) (0038) (0029) (0050)


Weeks in use 0004 20002 0007 20001(0002) (0002) (0002) (0003)

N 5 2773






























REFERENCES















West 0039 20012 0037 0002(0008) (0012) (0009) (0015)

New England 0015 20021 20024 0012(0028) (0038) (0029) (0050)


Weeks in use 0004 20002 0007 20001(0002) (0002) (0002) (0003)

N 5 2773



























make versus buy in trucking: asset ownership, job design, and … · 2004-07-26 · asset...

Documents