Download - Measuring Efficiency Improvement in Brazilian Trucking: A Distance Friction Minimization Approach with Fixed Factors

Transcript

Measurement 54 (2014) 166–177

Contents lists available at ScienceDirect

Measurement

journal homepage: www.elsevier .com/ locate /measurement

Measuring efficiency improvement in Brazilian trucking: ADistance Friction Minimization approach with fixed factors

http://dx.doi.org/10.1016/j.measurement.2014.04.0130263-2241/� 2014 Elsevier Ltd. All rights reserved.

⇑ Corresponding author. Tel.: +55 21 25989896.E-mail addresses: [email protected] (P. Wanke), [email protected].

pt (C.P. Barros), [email protected] (O. Figueiredo).1 Tel.: +351 213 016115; fax: +351 213 925 912.2 Tel.: +55 21 25989800.

Peter Wanke a,⇑, Carlos P. Barros b,c,1, Otavio Figueiredo d,2

a Center for Studies in Logistics, Infrastructure and Management, COPPEAD Graduate Business School, Federal University of Rio de Janeiro, Rua Paschoal Lemme,355, Rio de Janeiro CEP.: 21949-900, Brazilb Instituto Superior de Economia e Gestão, Technical University of Lisbon Rua Miguel Lupi, 20, 1249-078 Lisbon, Portugalc UECE (Research Unit on Complexity and Economics), Portugald COPPEAD Graduate Business School, Federal University of Rio de Janeiro, Rua Paschoal Lemme, 355, Rio de Janeiro CEP.: 21949-900, Brazil

a r t i c l e i n f o

Article history:Received 30 December 2013Received in revised form 5 March 2014Accepted 18 April 2014Available online 28 April 2014

Keywords:Longitudinal studyTrucking industryBrazilDFMFixed factorsEfficiency driversUnbalanced panel

a b s t r a c t

This paper investigates paths for improving efficiency in the Brazilian motor carrierindustry, which has undergone significant transformations since the economy deregulationin the mid 1990s. The main research objective is to determine whether or not differenttypes of cargoes and geographic regions serviced significantly impact trucking efficiencylevels by applying a Distance Friction Minimization approach with fixed factors. Resultssupport the evidence regarding a heterogeneous impact of cargo mix and route mix oninput reducing and output increasing potentials. Managerial impacts in terms of fleetsubcontracting and increased focus on specific transport demands are also addressed.

� 2014 Elsevier Ltd. All rights reserved.

1. Introduction

The motor carrier industry occupies an importantposition in the movement of goods and services [43]. Givenits advantages in the areas of accessibility to points oforigin and final destination, and the relatively low capitalrequirements for industry entry, motor carriers haveovershadowed other transportation modes in terms ofmarket share, employment, and the number of firms [6].Bolstered by the Plano Real economic plan and post-1994economic stability [31], the trucking industry began togarner more attention in Brazil, one of the so-called‘‘emerging countries’’ or ‘‘BRICs’’ [100], an acronym for

Brazil, Russia, India, and China. Although Brazil has experi-enced significant changes in terms of market competitive-ness since 1994, it is still a country strongly dependent onits motor carriers. Approximately two thirds of Brazilianfirms’ transport-related expenditure is spent on truckingservices [21]. As such, motor carriers must continually beon the lookout for new ways to stay competitive [68], withefficiency evaluation techniques serving a fundamentalrole in this pursuit.

A powerful tool for measuring efficiency is DataEnvelopment Analysis (DEA), developed over 30 years ago[23,49]. Its main characteristic is the capacity to simulta-neously process multiple inputs and outputs, therebyaiding managers in decision-making [1,50]. The DEAfrontier model remains widely used in transportation/logistics efficiency research in general, probably becauseit has been successfully applied to a wide number ofdifferent planning situations, such as third-party logistics

P. Wanke et al. / Measurement 54 (2014) 166–177 167

(e.g. [75,57,58,42,77,103,94], airline industry (e.g. [76,13,17,92],airports (e.g. [55,67,8,14,34,80,96,97], road passengertransport (e.g. [66,64], container terminals (e.g.[91,89,56,51,69,95].

The DEA Distance Friction Minimization (DFM) model,however, is one of the most recent solutions to face thismajor DEA drawback: there are, in principle, an infinitenumber of improvements paths a DMU could take in orderto reach the efficient frontier [36]. The DFM model, whichwas developed by Suzuki et al. [85], serves to improve theperformance of a DMU by identifying the most appropriatemovement towards the efficiency frontier surface. In thisapproach, a generalized distance function, based on aEuclidean distance metric in weighted spaces, is proposedto assist a DMU to improve its performance by an appro-priate movement towards the efficiency frontier surface.Such approach offers a refreshing perspective on efficiencyenhancement by employing a weighted projection func-tion. This can address both input reduction and outputaugmentation options. More recently, Suzuki and Nijkamp[86] extended the DFM approach to circumstanceswhere decision makers are faced with fixed factors ornon-discretionary variables, following the seminal ideasof Banker and Morey [11] model.

The Brazilian trucking industry is the focus of thispaper. Its objective is to identify the major drivers forincreasing outputs and reducing inputs from 2002 to2011, assessing whether or not different types of cargoesand geographic regions serviced present significant impact.To this end, a review of the literature was carried out, bothto provide some background to the sector and to supportthe DFM approach adopted. More precisely, the determina-tion of input/output improvement potentials under thepresence of non-discretionary variables was followed bya generalized least-squares models using unbalancedpanel, thus allowing the effect of different types of cargoesand geographic regions serviced on sector managerial effi-ciency to be estimated.

The remainder of the article comprises six sections. Sec-tion 2 presents a brief background on the Brazilian truckingindustry; also presented are the scant previous studies thatapplied DEA to the motor carrier industry in other countries,as well as the major research proposition considered here.Section 3 supports the most fundamental methodologicalissues concerning data analysis, providing not only a back-ground on DFM, but also giving more detailed insights onfixed factors within the ambit of the motor carrier industry.In Section 4, the data are analyzed and the results discussedin the context of several relevant issues, such as, the selec-tion of variables and the possibility of variable reduction.Section 5 addresses managerial implications, especially interms of opportunities of fleet subcontracting and increasedfocus on specific transport demands. Lastly, Section 6presents the conclusions of the study.

2. Literature review

2.1. Background on the Brazilian trucking industry

Historically, the integration of Brazil was based on theconstruction of highways, while the construction of

railroads and development of waterways was insteadaimed at meeting specific projects for out flowing cargo,particularly exports from the ports. Currently, the pavedhighway network is around five times larger than the rail-road network. When considering all types of highways, thesize is over 50 times as large as the rail system. The water-way system, on the other hand, is underused due to a lackof needed investment to improve navigability [30].

The highway system is responsible for about 56% of thetotal tonnage-per-mile moved in Brazil, according to ANTT[62] surveys. The enduring market share held by roadtransportation in Brazil is due to low price and a lack ofcomparable, equally reliable modes of transportation, par-ticularly rail and waterborne services. Evidence collectedby Wanke and Fleury [93] indicate that there are 2926 reg-istered motor carrier companies with more than 20employees operating in Brazil, and more than 100,000self-employed drivers and single-driver trucking compa-nies. In a country where the minimum monthly wage isonly about US$300, highway transportation is often linkedto certain aggressive and potentially unfair businesspractices, including excessive/uninterrupted driving hours,excessive speed, and uncontrolled vehicle/cargo weight[32].

2.2. Efficiency in the trucking industry

Although efficiency studies in the trucking industry arescarce, differently from other applications in logistics andtransportation, a number of methods to evaluate motorcarrier efficiency have been tested and used over thecourse of time [98]. Basically, these methods can bedivided in two major groups, following Bogetoft and Otto[18]: parametric and non-parametric. They are discussednext.

Parametric models are characterized by prior definitionexcept for a finite set of unknown parameters that areestimated from data. This group highlights the techniqueknown as Stochastic Frontier Analysis (SFA), directly linkedto econometric theory. It is worth commenting that inter-national, peer-reviewed papers dealing with the applica-tion of SFA in the trucking industry are even scarcer. Asearch at Proquest Database with the terms ‘‘truckingindustry’’ or ‘‘motor carrier’’ and ‘‘stochastic frontier’’returned only one paper [53]. This paper compared costefficiency in the US motor carrier industry, both beforeand after deregulation in 1980. Unionization was foundto be the most important determinant of inefficiency inboth periods, with its negative impact being stronger inthe deregulated industry.

On the other hand, non-parametric models are charac-terized by being much less restricted, since there is noneed for prior parameter definition. DEA, which has itsroots in mathematical programming, is an example of atechnique for this group. Specifically concerning the truck-ing industry, only few papers were found after performinga search at Proquest Database with the terms ‘‘truckingindustry’’ or ‘‘motor carrier’’ and ‘‘data envelopment anal-ysis’’. Odeck and Hjalmarsson [65] and Hjalmarsson andOdeck [41] initially demonstrated the usefulness of DEAas a tool for evaluating the efficiency of trucks involved

168 P. Wanke et al. / Measurement 54 (2014) 166–177

in road construction and maintenance production pro-cesses in Norway. Meja and Corsi [54] conducted DEA anal-ysis to assess a motor carrier’s safety process, showing howthis technique might be useful to carriers, regulators, andshippers. Poli and Scheraga [71] ran DEA models to identifythe causes of inefficiencies in maintenance strategies. Theirimpact on firms’ quality rating performance was also ana-lyzed. More recently, Weber and Weber [99] used DEA toestimate efficiency and productivity measures in the UStrucking and warehousing industry during the years1994–2000, accounting for both desirable and undesirableoutputs (fatalities). The authors found that traditionaltechniques of estimating efficiency which ignore trafficfatalities produce biased estimates on efficiency levelsand total factor productivity growth.

In a broader sense, these previous researches on truck-ing efficiency are pretty much aligned with the overalldescription provided by Yu [102] on general-purpose effi-ciency studies. Often, these studies seek to: (i) evaluateoperational efficiencies; (ii) illustrate how efficiency mea-sures can be useful for monitoring operations; (iii) identifycharacteristics or contextual variables that may explaindifferences in operational efficiency; (iv) assess size-impact on efficiency levels; and (v) measure productionfunction slacks.

Yu [102], however, also identifies two major questionswith respect to performance measurement that constitutebroader research gap opportunities. First, the question asto whether and how expansion/reduction of the capacitylevel should be implemented based on slack calculations;second, the subsequent question on the most appropriatepolicy implementation venues for performance improve-ment and the role played by contextual variables on theincreasing outputs/reducing inputs potentials. Thisresearch provide insights regarding both questions withinthe ambit of the trucking industry.

There is some consecrated anecdotal evidence regard-ing the efficiency drivers in the productive process ofmotor carrier companies (see [35,26,19] for examples),whose relative impacts still deserve to be further explored[44,2] in terms of production slacks. They serve as startingpoints are related to cargo and route mix decisions, leadingto the following major proposition of this paper:

2.3. Major proposition

Different types of cargoes and geographic regions ser-viced significantly impact the increasing outputs andreducing inputs potentials of trucking companies. Themajor ideas behind this major proposition relate to opera-tional specificities that may be embedded within each dif-ferent cargo or route type. For instance, containerized andbulk cargoes are well-known not only for being handledmore efficiently, but also for assuring a better utilizationof the available load space of the vehicle. The same basiceconomic principles apply to the geographic regionsserviced: the farther away they lie, the more likely theprevalence of productivity gains in distribution. Thesegains are also verified in geographic regions with higherlevels of cargo density (kilos to be collected per squarekilometer).

3. Methodological fundaments

3.1. DEA and DFM background

DEA is a non-parametric model first introduced byCharnes et al. [22] [9,24,92]. It is based on linear program-ming (LP) and is used to address the problem of calculatingrelative efficiency for a group of DMUs by using multiplemeasures of inputs and outputs. Given a set of DMUs,inputs, and outputs, DEA determines for each DMU a mea-sure of efficiency obtained as a ratio of weighted outputs toweighted inputs.Consider a set of n observations on theDMUs. Each observation, DMUj (j = 1,..., n) uses m inputsxij (i = 1,..., m) to produce s outputs yrj (r = 1,..., s). DMUo rep-resents one of the n DMUs under evaluation, and xio and yro

are the ith input and rth output for DMUo, respectively.Eq. (1) presents the envelopment model for the CRS fron-tier type, where e is a non-Archimedian element and s�iand sþr account, respectively, for the input and output slackvariables [15] and [105].

min /� eXm

i¼1

s�i þXs

r¼1

sþr

!

s:t:Xn

j¼1

kjxij þ s�i ¼ /xio; 8i

Xn

j¼1

kjyrj � sþr ¼ yro; 8r

kj P 0; 8j

ð1Þ

The slack analysis and the choice of the most adequateprojection model are crucial issues in DEA, both havingreceived significant attention over the past years – referto Amirteimoori et al. [7] and Esmaeilzadeh and Hadi-Ven-cheh [29] and the references therein. In this sense, the DFMwas originally proposed by Suzuki et al. [85], based on theCCR model, in order to provide an appropriate projectionmodel to improve efficiency in DEA. More precisely, DFMsearches for the point on the efficiency frontier that is asclose as possible to the DMU’s inputs and outputs vector.This type of distance minimization has also been proposedin recent papers of Slavova [82] and Beyko et al. [16] as amethod of resemblance. In general terms, it is based on ageneralized distance friction function and serves toimprove the performance of a DMU by identifying the mostappropriate movement towards the efficiency frontier sur-face. A suitable form of multidimensional projection func-tions is given by a Multiple Objective QuadraticProgramming (MOQP) model using a Euclidean distance,while it may also address both input reduction and outputaugmentation [85]. Their approach is given next.

min Frx ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiX

i

v�i xio � v�i dxio

� �2s

ð2Þ

min Fry ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiX

r

u�r yro � u�r dyro

� �2s

ð3Þ

P. Wanke et al. / Measurement 54 (2014) 166–177 169

s:t:Xi

v�i ðxio � dxioÞ ¼ 2h�=ð1þ h�Þ; ð4Þ

Xr

u�r ðyro � dyroÞ ¼ 2h�=ð1þ h�Þ; ð5Þ

xio � dxio P 0; ð6Þ

dxio P 0; ð7Þ

dyro P 0; ð8Þ

where Frx and Fry are the distance friction functions to besolved using MOQP, and dx

io and dyro are, respectively, the

reduction distances for xio and yro. Constraints (4) and (5)refer to the target values of input reduction and outputaugmentation. For a complete discussion on the fairnessin the distribution of contributions from the input andthe output side to achieve efficiency, readers should followSuzuki et al. [85].

3.2. Fixed factors or non-discretionary variables

Discretionary models of DEA assume that all inputs andoutputs can be varied at the discretion of management orother users. In any realistic situation, however, there mayexist ‘‘exogenously fixed’’ or non-discretionary factors thatare beyond the control of a DMU’s management, whichalso need to be considered [52,28]. Banker and Morey[11] developed the first model for evaluating DEAefficiency with ’’exogenously fixed’’ inputs and outputs informs like ’’age of store’’ in an analysis of a network offast-food restaurants [72,73,74,60].

Some examples of non-discretionary factors in the DEAliterature are the number of competitors in the branches ofa restaurant chain, snowfall or weather in evaluating theefficiency of maintenance units, soil characteristics andtopography in different farms, age of facilities in differentuniversities, the populations of wards in evaluating therelative efficiency of public libraries. In logistics applica-tions, such as in airports, capital is frequently treated as anon-discretionary variable over which management haslittle to no control [5].

Suzuki and Nijkamp [86] presented the followingversion of the DFM model, in which both input or outputfixed factors (FF) are taken into account:

min Frx ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiXi2D

ðv�i xio � v�i dxioÞ

2s

ð9Þ

min Fry ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiXr2D

ðu�r yro � u�r dyroÞ

2s

ð10Þ

s:t:Xi2D

v�i ðxio�dxioÞþ

Xi2ND

v�i xio¼1� 1�h�ð Þ 1�P

i2NDv�

ixio

� �1�P

i2NDv�

ixio

� �þ h��

Pr2ND

u�r yro

� � ;ð11Þ

Xr2D

u�r ðyro�dysoÞþ

Xr2ND

u�r yro ¼1�ð1�h�Þ 1�

Pr2NDu�r yro

� �1�

Pi2NDv�i xio

� �þ h� �

Pr2NDu�r yro

� � ;ð12Þ

xio � dxio P 0; ð13Þ

dxio P 0; ð14Þ

dyro P 0; ð15Þ

where the symbols i e D and r e D refer to the set of discre-tionary inputs and outputs; conversely, the symbols i e NDand r e ND refer to the set of non-discretionary inputs andoutputs. There are some methods to solve the modelpresented in Eqs. (9)–(15): Simple Additive Weighting(SAW), lexicographic method, interactive methods etc[84] and [45]. In this paper, the SAW method was used,mostly due its simplicity, and a new objective functionwas rewritten summing up Eqs. (9) and (10) with equalweights.

3.3. Treating statistical variations on reduction distances

The approaches to the statistical treatment of the vari-ations in the measurements produced using DEA haveevolved over the course of the years; see, for example,Banker [10] and Simar and Wilson [81]. As a depiction ofthis evolution, Cooper et al. [25] point to the growing num-ber of studies that combine DEA measurements, obtainedin a first stage, with multivariate data analysis, such asregression analysis, in a second stage, when these mea-surements are incorporated in the form of the dependentvariable. According to Fried et al. [33], such two-stageDEA approaches are an important recognition that envi-ronmental factors or contextual variables can significantlyinfluence efficiency levels. The authors also show thatmanagerial competence (or incompetence) is insufficientto explain individual variations in efficiency, given that,environmental factors, contextual variables, or evenstatistical noise can exert some influence over measuredperformance. The adequate control for these impacts maysuggest possible paths for a DMU to become more efficient(see, for example, [83,46,88].

Turner et al. [89] advocate the use of censoredregression models on DEA measurements. In general, thebasic model for a censored regression is similar to thatfor OLS; however, the former assumes a truncated normaldistribution in lieu of a normal distribution and employsmaximum likelihood estimation [37]. Banker and Natara-jan [12] showed that DEA-based procedures usingcensored regression in the second stage perform as wellas the best of the parametric methods in the estimationof the impact of contextual variables on efficiency.

Generally speaking, panel data models allow the exam-ination of fixed or random effects of a specific firm or oftime periods on efficiency scores [70]. According to Guja-rati [38], in the random effects model is assumed thatthe intercept of a DMU is a random drawing from a muchlarger population with a constant mean value. On the otherhand, in the fixed effects model, the intercept in theregression model is allowed to differ among individuals

170 P. Wanke et al. / Measurement 54 (2014) 166–177

in recognition of the fact each DMU may have some specialcharacteristics of its own. To take into account the differingintercepts, one can use dummy variables.

As far as this research concerns, the generalized least-squares model on the balanced panel data was carriedout, using the random effects model to test for significantdifferences in increasing output and reducing input poten-tials, given a set of cargo and geographic region relatedvariables. As suggested by Croissant and Millo [27], theHausman test was performed to assess the suitability of arandom effect to the detriment of a fixed effect model.

More precisely, for the random effects models—accord-ing to Greene [37], the model most frequently used—, thebasic assumptions are: the random effect ui is the samefor all periods and should not be correlated with otherregressors; the angular coefficients are the same for allgroups and periods; and eit, the stochastic component ofthe model, does not correlate across periods. The func-tional form of the random effects model is given by:

yit ¼ Xitbþ ui þ eit ð16Þ

where i denotes the group or individual; t denotes the timeperiod; yit denotes the dependent variable; and Xit denotesthe vector of independent variables.

4. Data analysis and discussion of results

This section is structured in two subsections, detailednext. The first subsection presents the discretionary andnon-discretionary inputs/outputs, and contextual variablesconsidered in the DFM approach with fixed factors in lightof several modeling issues, such as, their adequacy in termsof representing a typical motor carrier company and theirvalidity as data obtained from secondary sources. In itsturn, the second subsection encompasses the steps takenthroughout the treatment of variations on reducing inputand increasing output potentials vis-à-vis heterogeneousimpact of cargo mix and route mix. Not only censoredregression results derived are presented; but also the useof data reduction techniques – such as principal compo-nent analysis – in conjunction with DEA is discussed withrespect to contextual variables.

4.1. The data

The data used in this research was obtained from theranking of the largest Brazilian trucking companies listedin Transporte Moderno/Maiores e Melhores (2002–2011), aspecialized Brazilian magazine focused on the motorcarrier industry. They provide information on differentinputs (fleet size-own, fleet size-subcontracted, numberof employees, and number of branches to collect cargoes)and outputs (fuel consumption (in liters per year), totalcargo transported (in tonnes per year) and distancetravelled (in kilometers per year)). It should be noted thatthe original variable set was cleaned up, rejecting thevariables that were not collected for all the individualslisted in the panel. The list of variables used in this studyand relevant descriptive statistics are given in Table 1.

4.1.1. Using secondary dataConducting a secondary analysis of existing data

saved the time and resources needed to collect primarydata. However, the benefits of saving time and effortmust be weighed against the limitations due to the levelof data and the lack of specificity of the data for thesecondary project [78,20]. All the data collected fromTransporte Moderno/Maiores e Melhores are objective mea-sures based on explicit criteria, represented by metric(inputs and outputs) and nominal scales (contextual vari-ables, which are further detailed below). As single-itemindicators of objective measures, data can be valid andreliable indicators of the variables under consideration[101], although the choice of the analytical method isclearly limited by the given structure of the data [20].The match between the data structure and the analyticalmethod should be, therefore, cross-checked. The dataimplications of the methods chosen, in light of the majorproposition of this research, were previously explored inSection 3.

Although the data set provided by Transporte Moderno/Maiores e Melhores might not have been collected in thecontext of a theoretical model, a theoretical model can stillbe identified and applied to the research process and datathat are theoretically consistent can be identified [104,59].The importance of this step in secondary analysis cannotbe underestimated [78]. As with any quantitative methodof research, the selection of the variables to be studiedmust first involve combining through the model to identifycritical concepts. In this sense, secondary data relies morestrongly on indirect analyses and interpretations of its data[20]. Lastly, the theoretical concepts are then matchedwith appropriate variables that form the data set.

4.1.2. Assessing inputs/outputs, and contextual variablesCorrelation analyses were performed, indicating signif-

icant positive relationships between the five inputs and thetwo outputs variables, which are, therefore, isotonic andjustified to be included in the model [92]. Similarly, inorder to check on the possibility of reducing the numberof inputs and outputs to be considered in the analysis, acorrelation analysis was performed again, testing eachgroup. Table 2 shows the correlation coefficients of thepairs of inputs and outputs, which were calculated from497 observations – comprising 271 individuals distributedover the course of course of ten years – taken in aggregate.Since their serial correlations are relatively low, all inputsand outputs were retained in the analysis.

The ranking also provided information on contextualvariables for each company, such as cargo diversity (liquid,chemical and petrochemical, container, refrigerated,fragile, large, live animals, expedited, etc.) and thegeographical scope of the transport operation (geographicBrazilian regions [South, Southeast, North, Northeast, andCentral-West] and different South-American countries[Argentina, Paraguay, Uruguay, Chile, Bolivia, Peru, Vene-zuela, Ecuador, and Colombia]). All of these variables werecoded as dummies. It should be noted that expedited cargois intrinsically related to expedited shipping, which is, by

Table 1Sample descriptive statistics.

Years Descriptivestatistics

Inputs Outputs

Fixed Discretionary Discretionary Fixed

Number ofbranches tocollect cargoes

Number ofemployees

Fleet size-own(in number oftrucks)

Fleet size-subcontracted (innumber of trucks)

Total cargotransported (intonnes per year)

Distancetravelled (inkm per year)

Fuelconsumption(in liters peryear)

2002 Average 9.41 392.22 269.50 128.95 340.736.00 4.968.280.00 1.440.423.00Minimum 1.00 5.00 3.00 1.00 720.00 213.446.00 30.987.00Maximum 76.00 3700.00 1384.00 960.00 3.405.073.00 33.600.000.00 6.860.272.00Standarddeviation

13.02 632.58 348.24 2.42 656.893.00 5.915.383.00 1.540.575.00

Coefficientof variation

1.38 1.61 1.29 0.02 1.93 1.19 1.07

2003 Average 7.87 441.30 232.20 125.91 347.323.00 5.764.192.00 2.004.271.00Minimum 1.00 4.00 8.00 2.00 700.00 120.000.00 50.000.00Maximum 50.00 671.00 1471.00 800.00 3.000.000.00 29.355.000.00 7.336.400.00Standarddeviation

9.82 874.36 295.94 2.42 606.866.00 6.049.949.00 1.930.239.00

Coefficientof variation

1.25 1.98 1.27 0.02 1.62 1.05 0.96

2004 Average 10.60 549.30 323.50 184.93 509.493.00 7.365.249.00 2.147.228.00Minimum 1.00 9.00 19.00 3.00 509.493.00 7.365.249.00 237.000.00Maximum 48.00 4803.00 3556.00 3000.00 4.260.000.00 94.000.000.00 7.500.000.00Standarddeviation

13.43 1084.03 615.95 2.82 821.547.00 6.904.079.00 1.745.053.00

Coefficientof variation

1.27 1.97 1.90 0.015 1.61 0.94 0.81

2005 Average 21.87 393.90 230.60 113.55 432.391.00 7.486.782.00 2.282.560.00Minimum 1.00 25.00 26.00 2.00 19.737.00 120.000.00 46.000.00Maximum 328.00 2650.00 2400.00 1000 2.768.000.00 32.450.000.00 9.240.000.00Standarddeviation

50.94 558.41 378.78 2.42 539.611.00 7.352.713.00 2.159.510.00

Coefficientof variation

2.33 1.42 1.64 0.021 1.25 0.98 0.95

2006 Average 17.08 447.50 294.50 130.18 555.658.00 9.472.186.00 3.234.237.00Minimum 1.00 21.00 34.00 18.00 6.240.00 120.000.00 46.000.00Maximum 328.00 3880.00 1349.00 891.00 4.000.000.00 42.000.000.00 16.000.000.00Standarddeviation

43.86 699.80 331.31 2.36 733.615.00 10.725.693.26 4.185.720.00

Coefficientof variation

2.57 1.56 1.13 0.018 1.32 1.41 1.28

2007 Average 14.00 675.50 380.30 142.76 636.134.00 11.659.751.00 3.989.372.00Minimum 1.00 40.00 23.00 5.00 5.000.00 140.000.00 20.000.00Maximum 86.00 8.300.00 2209.00 809.00 4.320.000.00 51.430.040.00 22.360.887.00Standarddeviation

18.85 1299.25 428.75 2.38 935.693.00 11.726.196.56 4.726.036.00

Coefficientof variation

1.35 1.92 1.13 0.017 1.13 1.01 1.18

2008 Average 18.16 561.50 282.50 110.37 610.794.00 9.493.914.00 3.440.464.00Minimum 1.00 20.00 16.00 1.00 5.000.00 140.000.00 20.000.00Maximum 137.00 4800.00 1759.00 1018.00 4.320.000.00 50.688.000.00 21.120.000.00Standarddeviation

25.42 87.285 360.57 2.37 901.719.00 10.712.995.15 4.391.230.00

Coefficientof variation

1.40 1.55 1.28 0.022 1.48 1.13 1.28

2009 Average 15.04 870.00 432.40 236.65 635.746.00 8.456.410.00 2.581.136.00Minimum 1.00 29.00 38.00 2.00 12.000.00 893.518.00 200.774.00Maximum 130.00 6500.00 3482.00 3111.00 7.200.000.00 41.141.150.00 18.500.000.00Standarddeviation

23.10 1341.98 597.56 2.80 1.153.590.00 8.846.398.00 3.177.212.00

Coefficientof variation

1.54 1.574 1.38 0.012 1.81 1.54 1.21

(continued on next page)

P. Wanke et al. / Measurement 54 (2014) 166–177 171

Table 1 (continued)

Years Descriptivestatistics

Inputs Outputs

Fixed Discretionary Discretionary Fixed

Number ofbranches tocollect cargoes

Number ofemployees

Fleet size-own(in number oftrucks)

Fleet size-subcontracted (innumber of trucks)

Total cargotransported (intonnes per year)

Distancetravelled (inkm per year)

Fuelconsumption(in liters peryear)

2010 Average 18.31 949.00 493.50 190.94 235.500.00 10.563.522.00 3.189.029.00Minimum 1.00 20.00 20.00 2.00 9.123.00 83.000.00 38.000.00Maximum 130.00 6200.00 3276.00 1270 497.356.00 58.777.778.00 20.100.000.00Standarddeviation

26.77 1716.21 760.83 2.57 1.440.972.00 14.145.061.00 4.342.799.00

Coefficientof variation

0.46 1.81 1.54 0.013 2.67 1.34 1.36

2011 Average 18.048 1044 518.88 283.98 762.895.20 10.147.692.00 3.097.363.20Minimum 1.2 34.8 45.6 2.4 14.400.00 1.072.221.60 240.928.80Maximum 156 7800.00 4178.40 3733.20 8.640.000.00 49.369.380.00 22.200.000.00Standarddeviation

27.72 1610.38 717.072 3.36 1.384.308.00 10.615.677.60 3.812.654.40

Coefficientof variation

1.848 1.8888 1.656 0.0144 2.172 1.848 1.452

Table 2Correlation coefficient matrix.

Variables Inputs Outputs

Number ofbranches

Number ofemployees

Fleetsize-own

Fleetsize-subcontracted

Fuelconsumption

Total cargotransported

Distancetravelled

InputsNumber of branches 1.00Number of employees 0.41 1.00Fleet size-own 0.20 0.24 1.00Fleet size-sucontracted 0.21 0.27 0.36 1.00

OutputsFuel consumption 0.51 0.64 0.54 0.39 1.00Total cargo transported 0.52 0.70 0.48 0.48 0.24 1.00 0.15Distance travelled 0.26 0.60 0.41 0.51 0.12 0.18 1.00

172 P. Wanke et al. / Measurement 54 (2014) 166–177

definition, the process of shipping at a faster rate thannormal.

4.1.3. Defining discretionary inputs/outputsThe definition on the discretionary/non-discretionary

inputs and outputs observed a resource based strategy onown assets [63]. Thus, only the subcontracted fleet (onthe input side), as well as the demand served (on the out-put side, proxied by the total cargo transported and dis-tance travelled) were considered discretionary variablesin the DFM model. Not only managers can decide to breakout contracts with third parties, but also they can increasetheir focus on specific cargo demands that may be reflectedon the distance travelled and the total volume of cargotransported each year.

The remainder inputs related to branches, own fleet andnumber of employees were considered to be fixed. In fact,it takes a few years to plan, design, and implement a truck-ing service network. Fuel consumption was considered as anon-discretionary output so as to reflect a by-product ofthe trucking service network [79], especially with respectto the decisions regarding the type of vehicles that formthe own fleet of trucks, which cannot be altered in theshort term.

4.2. Results

4.2.1. Determining DEA scores and DFM slacksInitially, traditional CCR and BCC DEA models were exe-

cuted ten times, i.e., once for each year for the period2002–2011. Their results are presented in Table 3. Then,the DFM model with fixed factors was also executed tentimes. Table 4 shows the descriptive statistics of the reduc-ing input and increasing output potentials for this timeperiod.

4.2.2. Reducing contextual variables into main factorsDifferent types of cargoes and geographic regions

serviced were then considered as regressors in order toidentify the opportunities for efficiency improvement inBrazilian motor carriers. Such characteristics are thecontrol variables of this study, since they comprise neitherprocess inputs nor products, but, rather, their attributes, ina total of 28 contextual variables. These variables areterminal, i.e., they assume the value of one if theobservation has the mentioned characteristic and zerootherwise. It is understood that k � 1 dummy variablesare required to represent a variable with k categories

Table 3Descriptive statistics of DEA scores by year.

Corrected estimates Year 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

VRS Average 78% 88% 60% 90% 80% 85% 94% 90% 93% 83%Minimum 23% 46% 15% 47% 26% 29% 56% 22% 47% 15%Maximum 100% 100% 100% 100% 100% 100% 100% 100% 100% 100%Standard deviation 21.05% 13.73% 25.75% 14.08% 21.81% 19.18% 11.89% 16.65% 12.24% 20.56%Coefficient of variation 27.08% 15.53% 42.78% 15.67% 27.10% 22.49% 12.72% 18.50% 13.17% 24.65%# of efficient DMUs 11 9 4 11 10 9 19 4 14 9% of efficient DMUs 12% 24% 10% 24% 16% 17% 44% 9% 44% 20%

Total DMUs 92 38 40 45 61 54 43 47 32 45

Table 4Descriptive statistics of reducing input and increasing output potentials by year.

Year Mean Standard deviation

Potential forreducingsubcontracted fleet(# of vehicles)

Potential forincreasing distancetravelled (in km peryear)

Potential for increasingtotal cargo transported(in tonnes per year)

Potential forreducingsubcontracted fleet(# of vehicles)

Potential forincreasing distancetravelled (in km peryear)

Potential for increasingtotal cargo transported(in tonnes per year)

2002 63.56 2.348.440.49 186.509.42 92.56 3.247.965.29 301.194.202003 68.70 2.416.641.33 152.255.74 117.19 3.009.926.65 192.133.582004 78.83 3.660.348.96 98.915.58 175.55 4.033.862.45 219.513.432005 55.30 3.619.973.85 218.403.22 98.87 3.964.570.50 271.919.132006 66.60 4.184.602.01 264.896.47 92.54 5.227.763.24 298.608.452007 78.72 5.035.038.30 303.586.71 104.26 5.728.314.01 395.363.542008 60.58 4.469.482.69 295.651.62 133.23 5.595.311.35 365.085.632009 114.64 3.280.917.96 373.587.76 235.23 3.509.262.28 573.244.362010 95.27 5.237.329.06 288.642.08 140.81 7.472.453.14 733.053.822011 98.36 4.220.354.34 312.134.18 150.17 6.352.345.24 718.321.90

Table 5Types of cargoes: rotated component matrix.

Rotated component matrix for the types of cargoes (variance explained = 55.22%)

Types of cargoes C1 – liquid ordry bulk cargo

C2 – generalmiddle sized cargo

C3 – expressand live cargo

C4 – generallarge cargo

C5 – fragilecargo

Liquid 0.56 (0.14) (0.25) (0.41) 0.33Chemical and Petrochemical 0.52 (0.02) 0.29 0.48 0.15Dry bulk 0.51 0.15 0.08 0.03 0.27Dry LTL 0.12 0.45 0.48 0.11 (0.23)Vehicles 0.31 0.68 (0.10) (0.19) 0.07Furniture (0.22) 0.72 0.15 0.14 0.13Expedited (0.08) 0.28 0.61 0.24 0.04Live animals 0.02 (0.15) 0.81 (0.22) 0.06General 0.08 (0.21) (0.11) 0.59 0.00Coils 0.04 0.30 0.08 0.69 0.04Parcel 0.06 (0.03) (0.10) (0.03) 0.72Sensitive 0.07 0.12 0.13 0.10 0.73Refrigerated 0.43 0.34 0.28 0.21 (0.21)Container (0.62) 0.07 0.11 (0.08) 0.04Kaiser–Meyer–Olkin Measure of Sampling Adequacy 0.658Bartlett’s Test of Sphericity Approx. Chi-Square 722.08

Sig. 0.00

Bold values indicate significant at 0.05.

P. Wanke et al. / Measurement 54 (2014) 166–177 173

[48]. The base-category is the absence a given characteris-tic – type of cargo or geographic region serviced.

However, given the large number of potential contex-tual variables to be considered in the second stage, datareduction techniques assume particular relevance here[39,88]. Several authors have used some of these tech-niques in conjunction with DEA. Adler and Golany [4]

and Adler and Berechman [3], for instance, employed prin-cipal component analysis. The use of factor analysis wasproposed by Vargas and Bricker [90] and implemented inJenkins and Anderson [47] and Nadimi and Jolai [61].Specifically, factor analysis is an appropriate procedurefor data reduction based on observed variables and onexisting theoretical constructs [40].

Table 6Geographic regions: rotated component matrix.

Rotated component matrix for the geographic regions serviced (variance explained = 69.51%)

Geographic regions R1 – Andean countriesthat share borderswith Brazil

R2 – Andean countriesthat do not share borderswith Brazil

R3 – inland Brazilplus largestmercosur countries

R4 – coastalBrazil

R5 – smallestmercosurcountries

Colombia 0.91 0.16 0.04 (0.03) (0.08)Peru 0.76 0.21 0.01 (0.02) 0.36Chile 0.18 0.79 0.14 0.09 0.10Ecuador 0.06 0.90 0.11 0.04 0.00Brazil – North (0.18) 0.14 0.60 0.06 (0.02)Brazil – Central-West 0.01 (0.09) 0.80 (0.05) 0.11Argentina 0.12 0.16 0.78 0.19 0.05Bolivia 0.25 0.05 0.60 0.41 (0.07)Brazil – South (0.10) 0.04 0.08 0.83 (0.02)Brazil – Southeast (0.05) 0.04 0.01 0.84 0.02Brazil – Northeast 0.07 0.10 0.30 0.51 (0.07)Paraguay 0.07 0.11 0.08 (0.05) 0.90Uruguay 0.43 0.23 (0.01) (0.01) 0.53Venezuela 0.35 0.45 (0.02) 0.07 0.16Kaiser–Meyer–Olkin measure of sampling adequacy 0.75Bartlett’s test of sphericity Approx. Chi-Square 2836.80

Sig. 0.00

Bold values indicate significant at 0.05.

174 P. Wanke et al. / Measurement 54 (2014) 166–177

In order to make the types of cargoes concept opera-tional, a factor analysis with Varimax standardized rota-tion was conducted. Table 5 presents the five factorsrelated to the type of cargoes. According to Tabachnickand Fidell [87], only factor loads greater than 0.50 meritanalyses, and in such cases, the variable is said to representa good factor measure.

As regards the impact of geographical regions on effi-ciency, the same approach was conducted with differentgeographic Brazilian regions (South, Southeast, Northeast,Central-West, and North) and different South-Americancountries (Argentina, Paraguay, Uruguay, Chile, Bolivia,Peru, Venezuela, Ecuador, and Colombia) serviced bythese trucking companies. Similarly, all of these arebinary variables, available, for each company, at theTransporte Moderno database. Results are presented inTable 6.

Table 7Censored regression results for reducing input and increasing output potentials.

Log of potential for reducing subcontracted fleet Log of potential for intravelled

Adj. R-Squared: 0.254 Adj. R-Squared: 0.689F-statistic: 15.105 on 10 and 430 DF, p-value:<2.22e�16

F-statistic: 210.65 onp-value: <2.22e�16

Estimate Std.error

t-value Pr(>|t|) Estimate Std.error

t-v

Intercept 2.95 0.18 15.66 2.2e�16 ⁄⁄⁄ 14.12 0.19 71C1 0.09 0.08 1.10 0.27 �0.09 0.09 �1C2 0.21 0.08 2.60 0.01 ⁄⁄ �0.14 0.09 �1C3 0.08 0.06 1.35 0.17 �0.05 0.07 �0C4 �0.00 0.09 �0.04 0.96 0.25 0.09 2C5 0.14 0.06 2.15 0.03 ⁄ 0.07 0.07 1R1 0.01 0.06 0.22 0.82 0.02 0.06 0R2 �0.03 0.08 �0.42 0.67 �0.08 0.07 �1R3 �0.02 0.11 �0.22 0.81 �0.32 0.11 �2R4 0.20 0.07 2.94 0.00 ⁄⁄ 0.70 0.07 9R5 0.03 0.06 0.61 0.54 0.08 0.06 1

Signif. codes: 0 ‘⁄⁄⁄’ 0.001 ‘⁄⁄’ 0.01 ‘⁄’ 0.05 ‘.’ 0.1 ‘ ’ 1.

4.2.3. Regressing slacks into main factorsAt last, the generalized least-squares censored regres-

sion on the unbalanced panel data, using the randomeffects model, was carried out. Readers should recall thesuitability of the random effects model in light of the majorproposition of this research: the major underlying assump-tion is that the error terms are a random drawing from amuch larger population [38]. With respect to the accept-able level of significance, the customary cut-off value of0.05 was adopted. Table 7 shows the regression resultsfor each one of these ten factors.

The results presented in Table 7 confirm the significantimpact of some of the different types of cargoes and geo-graphic regions serviced on the potentials for reducinginputs and increasing outputs. With respect to the poten-tial for reducing subcontracted fleet, opportunities emergewhere higher levels of specialization or differentiation are

creasing distance Log of potential for increasing total cargotransportedAdj. R-Squared : 0.689

10 and 430 DF, F-statistic: 99.87 on 10 and 430 DF, p-value:<2.22e�16

alue Pr(>|t|) Estimate Std.error

t-value Pr(>|t|)

.32 <2e�16 ⁄⁄⁄ 11.04 0.21 52.04 <2.2e�16 ⁄⁄⁄

.00 0.31 �0.09 0.09 �0.97 0.33

.55 0.12 �0.04 0.09 0.44 0.65

.79 0.42 �0.17 0.07 �2.41 0.01 ⁄

.61 0.00 ⁄⁄ 0.12 0.10 1.19 0.23

.11 0.26 0.06 0.07 0.90 0.36

.37 0.70 0.01 0.07 0.24 0.80

.06 0.28 �0.02 0.08 �0.24 0.80

.78 0.00 ⁄⁄ 0.05 0.12 0.42 0.67

.19 0.00 ⁄⁄⁄ 0.61 0.08 7.52 3.01e�13 ⁄⁄⁄

.39 0.16 0.02 0.06 0.39 0.69

P. Wanke et al. / Measurement 54 (2014) 166–177 175

required in cargo transportation, such as in the case ofFragile cargo (C5), or where volumes may be not so attrac-tive for subcontracting, as in the case of General middlesized cargo (C2). Transportation in regions with highereconomic density, such as Coast Brazil (R4) may also favormore opportunities for lowering subcontracted fleet, dueto higher volumes, eventually shorter distances in high-performance transportation, and better possibility offreight returns.

As regards the potential for increasing the distance trav-elled, truck-load transportation altogether with regionscharacterized by higher economic density merit attention.This is the case of General large cargo (C4) and Coastal Bra-zil (R4) regions, both presenting a positive impact. On theother hand, regions where productivity is low in terms ofthe number of trips per year should be avoided, as in thecase of Inland Brazil plus largest Mercosur countries (R3).

At last, the potential for increasing the total cargo trans-ported depends positively on the focus on regions withhigher economic density, such as Coastal Brazil (R4). Onthe other hand, specialized low-density cargoes shouldbe avoided, as in the case of Express and live cargo (C3).

5. Conclusions and managerial implications

As competition in the trucking industry has intensifiedover the last decade, today’s Brazilian motor carriers arefaced with daunting challenges of continuously improvingtheir efficiency levels and competitiveness. Under thisstudy, a DFM approach with fixed factors was used toassess the efficiency drivers of the Brazilian motor carrierindustry from 2002 to 2011. The results show evidencethat the types of cargoes carried and the geographicregions serviced play a significant role in affecting poten-tials for reducing inputs and increasing outputs. In otherwords, the results not only suggest that certain decisionsas to cargo mix or route mix may favor a more efficientoperation but also that the measurement of these impactscan play a relevant role in decision-making processes.

More precisely, the contribution of this paper is two-fold, shedding some light on how efficiency measurementcan provide some guidance to the managerial decision-making process in the trucking industry. On the theoreticalside, a valuable scale for the measurement of managerialefficiency was built and validated, representing an indextowards the most productive cargo/route mix. On the otherhand, the managerial implication of this possibility of mea-suring the efficiency levels is that motor carriers can use itas a basis for establishing future action plans. For instance,should a given motor carrier service a given geographicregion? This may imply not only different concerns interms of service offer, but also in terms of what productsshould be carried.

The results produced here also have a number of policyimplications for an industry still very fragmented amongseveral companies, despite the economic growth wit-nessed in recent years and pressures for higher quality ser-vices. Since most trucking companies are still relativelysmall, consolidation of supplementary operations – ‘whatmatch of cargoes/routes mix from different motor carriers

does increase efficiency?’ – appears to be the natural pathto achieve gains in efficiency. It follows, then, that mergersand acquisitions in the Brazilian trucking industry couldincrease in the near future, accelerating a trend that beganover a decade ago, as the measures delivered in this papermay help determining the best match of cargo types andgeographical scope.

In a broader sense, motor carrier managers may use thegeneral results of this research as guidance for future stepstowards higher levels outputs or reduced inputs. ‘Whatroute mix/cargo mix should be explored?’ is an exampleof a question that may direct motor carriers through ashorter path to higher efficiency levels, helping them inestablishing a course of action going forward.

Acknowledgements

The authors would like to thank the editors and thereviewers for their helpful comments on this paper. Thisresearch was supported FAPERJ (Fundação Carlos ChagasFilho de Amparo à Pesquisa do Estado do Rio de Janeiro)– Project ID E-26/103.286/2011.

References

[1] J.H. Ablanedo-Rosas, A.J. Ruiz-Torres, Benchmarking of Mexicanports with data envelopment analysis, Int. J. Shipping TransportLogistics 1 (3) (2009) 276–294.

[2] B. Adenso-Diaz, S. Lozano, P. Moreno, Analysis of the synergies ofmerging multi-company transportation needs, Transport. A:Transport Sci. (2013), http://dx.doi.org/10.1080/23249935.2013.797518.

[3] N. Adler, J. Berechman, Measuring airport quality from the airlines’viewpoint: an application of data envelopment analysis, Transp.Policy 8 (3) (2001) 171–181.

[4] N. Adler, B. Golany, Evaluation of deregulated airline networksusing data envelopment analysis combined with principalcomponent analysis with an application to Western Europe, Eur. J.Oper. Res. 132 (2) (2001) 260–273.

[5] N. Adler, V. Liebert, E. Yazhemsky, Benchmarking airports from amanagerial perspective, Omega 41 (2) (2013) 442–458.

[6] A.J. Allen, S. Shaik, J.K. Estrada, An assessment of the efficiency ofagribusiness trucking companies: a data envelopment analysisapproach, in: Southern Agricultural Economics Association 2005Proceedings of the Annual Meetings in Little Rock, Arkansas, USA,2005.

[7] A. Amirteimoori, A. Emrouznejad, L. Khoshandam, Classifyingflexible measures in data envelopment analysis: a slack-basedmeasure, Measurement 46 (10) (2013) 4100–4107.

[8] A. Assaf, Bootstrapped scale efficiency measures of UK airports, J.Air Transport Manage. 16 (2010) 42–44.

[9] R.D. Banker, A. Charnes, W.W. Cooper, Some models for estimatingtechnical and scale inefficiencies in data envelopment analysis,Manage. Sci. 30 (9) (1984) 1078–1092.

[10] R.D. Banker, Maximum likelihood, consistency and DEA: astatistical foundation, Manage. Sci. 39 (10) (1993) 1265–1273.

[11] R.D. Banker, R.C. Morey, Efficiency analysis for exogenously fixedinputs and outputs, Oper. Res. 34 (4) (1986) 513–521.

[12] R.D. Banker, R. Natarajan, Evaluating contextual variables affectingproductivity using data envelopment analysis, Oper. Res. 56 (1)(2008) 48–58.

[13] C. Barbot, Á. Costa, E. Sochirca, Airlines performance in the newmarket context: a comparative productivity and efficiency analysis,J. Air Transport Manage. 14 (2008) 270–274.

[14] T.D. Barros, T.G. Ramos, J.C. Mello, L.A. Meza, Avaliação dos atrasosem transporte aéreo com um modelo DEA, Produção 20 (4) (2010)601–611.

[15] M. Bazargan, B. Vasigh, Size versus efficiency: a case study of UScommercial airports, J. Air Transport Manage. 9 (3) (2003) 187–193.

[16] I.W. Beyko, L.Ju. Bodachivsjka, T.W. Korobko, The problems ofmulticriteria estimations with incomplete data with the use of theaggregated Slavova criteria in problems of stochastic and discrete

176 P. Wanke et al. / Measurement 54 (2014) 166–177

optimization. In: International Ukrainian-Polish Workshop, Kaniv,2005. pp. 111–112.

[17] D. Bhadra, Race to the bottom or swimming upstream:Performance analysis of US airlines, J. Air Transport Manage. 15(2009) 227–235.

[18] P. Bogetoft, L. Otto, Benchmarking with DEA, SFA, and R, Springer,New York, 2010.

[19] D.J. Bowersox, D.J. Closs, Logistical Management: the IntegratedSupply Chain Process, McGraw-Hill, New York, NY, 1996.

[20] C. Busse, A procedure for secondary data analysis: innovation bylogistics service providers, J. Supply Chain Manage. 46 (4) (2010)44–58.

[21] CEL, Panorama Logístico CEL/COPPEAD – Terceirização Logística noBrasil, COPPEAD/UFRJ, Rio de Janeiro, RJ, 2009.

[22] A. Charnes, W.W. Cooper, E. Rhodes, Measuring the efficiency ofdecision -making units, Eur. J. Oper. Res. 2 (6) (1978) 429–444.

[23] W.D. Cook, L.M. Seiford, Data envelopment analysis (DEA) – thirtyyears on, Eur. J. Oper. Res. 192 (2009) 1–17.

[24] W.W. Cooper, S. Li, L.M. Seiford, R.M. Thrall, J. Zhu, Sensitivity andstability analysis in DEA: some recent developments, J. Prod. Anal.15 (3) (2001) 217–246.

[25] W.W. Cooper, L.M. Seiford, K. Tone, Data Envelopment Analysis: aComprehensive Text With Models Applications References andDEA-Solver Software, Springer, New York, NY, 2007.

[26] J.J. Coyle, E.J. Bardi, R.A. Novack, Transportation, West PublishingCompany, St. Paul, MN, 1994.

[27] Y. Croissant, G. Millo, Panel data econometrics in R: the plmpackage, 2012. Available at: <http://cran.r-project.org/web/packages/plm/vignettes/plm.pdf> (accessed May 2nd 2014).

[28] M.J. Ebadi, S. Shahraki, Determination of scale elasticity in theexistence of non-discretionary factors in performance analysis,Knowl.-Based Syst. 23 (5) (2010) 434–439.

[29] A. Esmaeilzadeh, A. Hadi-Vencheh, A super-efficiency model formeasuring aggregative efficiency of multi-period productionsystems, Measurement 46 (10) (2013) 3988–3993.

[30] K.F. Figueiredo, P.F. Fleury, P. Wanke, Logística e Gerenciamento daCadeia de Suprimentos: Planejamento do Fluxo e dos Recursos,Atlas, São Paulo, SP, 2003.

[31] P.F. Fleury, A.F.M. Ribeiro, A indústria de provedores de serviçoslogísticos no Brasil, in: P.F. Fleury, K.F. Figueiredo, P.F. Wanke(Eds.), Logística e Gerenciamento da Cadeia de Suprimentos:Planejamento do Fluxo e dos Recursos, Atlas, São Paulo, SP, 2003,pp. 302–312.

[32] P.F. Fleury, M.F. Hijjar, Logistics overview in Brazil 2008, availableat: <http://www.guiadotrc.com.br/logistica/Logistics_Overview_in_Brazil_2008.pdf>. (accessed May 2nd 2014).

[33] H.O. Fried, C.A.K. Lovell, S.S. Schmidt, S. Yaisawarng, Accounting forenvironmental effects and statistical noise in data envelopmentanalysis, J. Prod. Anal. 17 (1) (2002) 157–174.

[34] A.B.M. Fonseca, J.C. Mello, E.G. Gomes, L.A. Meza, Uniformization offrontiers in non-radial ZSG-DEA models: an application to airportrevenues, Pesquisa Oper. 30 (1) (2010) 175–193.

[35] R. Gardner, Transportation, Twenty-First Century Books, New York,NY, 1994.

[36] B. Golany, An interactive MOLP procedure for the extension of DEAto effectiveness analysis, J. Oper. Res. Soc. 39 (1988) 725–734.

[37] W.H. Greene, LIMDEP Version 9.0 – Econometric Modeling Guide,Econometric Software Inc., New York, NY, 2007.

[38] D.M. Gujarati, Basic Econometrics, fourth edition., McGraw-Hill,New York, 2003.

[39] A. Hadi-Vencheh, Z. GhelejBeigi, K. Gholami, On the input/outputreduction in efficiency measurement, Measurement 50 (2013) 244–249.

[40] J.F. Hair, R.E. Anderson, R.L. Tatham, W.C. Black, AnáliseMultivariada de Dados, Bookman, Porto Alegre, RS, 2005.

[41] L. Hjalmarsson, J. Odeck, Efficiency of trucks in road constructionand maintenance. an evaluation with data envelopment analysis,Comput. Oper. Res. 23 (4) (1996) 393–404.

[42] A. Hamdan, K. Rogers, Evaluating the efficiency of 3PL logisticsoperations, Int. J. Prod. Econ. 113 (1) (2007) 235–244.

[43] H. Hernandez, S. Peeta, A carrier collaboration problem for less-than-truckload carriers: characteristics and carrier collaborationmodel, Transport. A: Transport Sci. 10 (4) (2014) 327–349.

[44] J. Holguín-Veras, C.A.T. Cruz, X. Ban, On the comparativeperformance of urban delivery vehicle classes, Transportmetrica(2011), http://dx.doi.org/10.1080/18128602.2010.523029.

[45] C.L. Hwang, A.S.M. Masud, Multiple Objective Decision Making –Methods and Applications: A State-of-the-art survey, Springer,Berlin, 1979.

[46] Z. Ismail, J.C. Tai, K.K. Kong, K.H. Law, S.M. Shirazi, R. Karim, Usingdata envelopment analysis in comparing the environmentalperformance and technical efficiency of selected companies intheir global petroleum operations, Measurement 46 (9) (2013)3401–3413.

[47] L. Jenkins, L. Anderson, A multivariate statistical approach toreducing the number of variables in data envelopment analysis,Eur. J. Oper. Res. 147 (1) (2003) 51–61.

[48] D.M. Levine, D.F. Stephan, T.C. Krehbiel, M.L. Berenson, Estatística:Teoria e Aplicações, LTC, São Paulo, SP, 2007.

[49] M.C. Lai, H.C. Huang, W.D. Wang, Designing a knowledge-basedsystem for benchmarking: A DEA approach, Knowl-Based Syst. 24(5) (2011) 662–671.

[50] L.W. Lan, E.T.J. Lin, Measuring railway performance withadjustment of environmental effects, data noise and slacks,Transportmetrica (2009) 161–189.

[51] L.C. Lin, L.A. Tseng, Application of DEA and SFA on the measurementof operating efficiencies for 27 international container ports, in:Proceedings of the Eastern Asia Society for Transportation Studies,Bangkok, Thailand, 2005, pp. 592–607.

[52] F. Lotfi, G. Jahanshahloo, Non-discretionary factors and imprecisedata in DEA, Int. J. Math. Anal. 1 (5) (2007) 237–246.

[53] B.S. McMullen, M.K. Lee, Cost efficiency in the US motor carrierindustry before and after deregulation: a stochastic frontierapproach, J. Transport Econ. Policy 33 (3) (1999) 303–317.

[54] M.M. Meja, T.M. Corsi, Assessing motor carrier potential forimproving safety process, Transport. J. 38 (4) (1999) 36–50.

[55] J.C. Melo, Análise de envoltória de dados no estudo da eficiência edos benchmarks para companhias aéreas brasileiras, Pesquisa Oper.23 (2) (2003) 325–345.

[56] H. Min, B.I. Park, Evaluating the inter-temporal efficiency trends ofinternational container terminals using data envelopment analysis,Int. J. Integr. Supply Manage. 1 (3) (2005) 258–277.

[57] H. Min, S.J. Joo, Benchmarking the operational efficiency of thirdparty logistics providers using data envelopment analysis, SupplyChain Manage.: Int. J. 11 (3) (2006) 259–265.

[58] H. Min, S.J. Joo, Benchmarking third-party logistics providers usingdata envelopment analysis: an update, Benchmarking: Int. J. 16 (5)(2009) 572–587.

[59] H.J. Moriarty, J.A. Deatrick, M.M. Mahon, S.L. Feetham, R.M. Carroll,M.P. Shepard, A.J. Orsi, Issues to consider when choosing and usinglarge national databases for research of families, West. J. Nurs. Res.21 (2) (1999) 143–153.

[60] M. Muniz, J. Paradi, J. Ruggiero, Z. Yang, Evaluating alternative DEAmodels used to control for non-discretionary inputs, Comput. Oper.Res. 33 (2006) 1173–1183.

[61] R. Nadimi, F. Jolai, Joint use of Factor Analysis (FA) and DataEnvelopment Analysis (DEA) for ranking of Data EnvelopmentAnalysis, Int. J. Math. Phys. Eng. Sci. 2 (4) (2008) 218–222.

[62] National Agency for Terrestrial Transportation, available at:<www.antt.gov.br> (accessed March 1st 2012).

[63] A.G.N. Novaes, Rapid-transit efficiency analysis with the assurance-region DEA method, Pesquisa Oper. 21 (2) (2011) 179–197.

[64] A.G.N. Novaes, S.F. Silveira, H.C. Medeiros, Efficiency andproductivity analysis of the interstate bus transportation industryin Brazil, Pesquisa Oper. 30 (2) (2010) 465–485.

[65] J. Odeck, L. Hjalmarsson, The performance of trucks – an evaluationusing data envelopment analysis, Transport. Planning Technol. 20(1) (1996) 49–66.

[66] J. Odeck, A. Alkadi, Evaluating efficiency in the Norwegian busindustry using data envelopment analysis, Transportation 28 (3)(2001) 211–232.

[67] R.R. Pacheco, E. Fernandes, Managerial efficiency of Brazilianairports, Transp. Res. Part A 37 (8) (2003) 667–680.

[68] P.M. Panayides, Effects of organizational learning in third-partylogistics, J. Business Logistics 28 (2) (2007) 133–157.

[69] P.M. Panayides, C.N. Maxoulis, T.F. Wang, K. Ng, A critical analysis ofDEA applications to seaport economic efficiency measurement,Transport Rev. 29 (2) (2009) 183–206.

[70] H.M. Park, Linear regression models for panel data using SAS, Stata,LIMDEP, and SPSS’’, 2005, available at: <www.indiana.edu/~statmath> (accessed 15 April 2005).

[71] P.M. Poli, C.A. Scheraga, A quality assessment of motor carriermaintenance strategies: an application of data envelopmentanalysis, Q. J. Business Econ. 40 (1) (2001) 25–43.

[72] S.C. Ray, Resource-use efficiency in public schools: a study ofconnecticut data, Manage. Sci. 37 (1991) 1620–1628.

[73] J. Roggiero, On the measurement of technical efficiency in thepublic sector, Eur. J. Oper. Res. 90 (1996) 553–565.

P. Wanke et al. / Measurement 54 (2014) 166–177 177

[74] J. Roggiero, Non- discretionary inputs in data envelopment analysis,Eur. J. Oper. Res. 111 (1998) 461–469.

[75] A.D. Ross, C. Droge, An analysis of operations efficiency in large-scale distribution systems, J. Oper. Manage. 21 (6) (2004) 673–688.

[76] M. Schefczyk, Operational performance of airlines: an extension oftraditional measurement paradigms, Strateg. Manag. J. 14 (4)(1993) 301–317.

[77] L.F. Senra, L.C. Nanci, J.C. Mello, L.A. Meza, Estudo sobre métodos deseleção de variáveis em DEA, Pesquisa Oper. 27 (2) (2007) 191–207.

[78] M.P. Shepard, R.M. Carroll, M.M. Mahon, H.J. Moriarty, S.L. Feetham,J.A. Deatrick, A.J. Orsi, Conceptual and pragmatic considerations inconducting a secondary analysis – an example from research offamilies, West. J. Nurs. Res. 21 (2) (1999) 154–167.

[79] S. Shokari, S.F. Ghaderi, Optimizing consumption and emission ingas fuel consuming power applying DEA model, 2013. Accessed at:<http://icrepq.com/icrepq-08/379-Shokravi.pdf> (accessed May 2nd,2014).

[80] J.Q. Silveira, L.A. Meza, J.C. Mello, Identificação de benchmarks eanti-benchmarks para companhias aéreas usando modelos DEA efronteira invertida, Produção 22 (4) (2012) 788–795.

[81] L. Simar, P.W. Wilson, Estimation and inference in two-stage,semiparametric models of production processes, J. Econ. 136 (1)(2007) 31–64.

[82] T. Slavova, The SOW-index for benchmark of European stock-exchanges, In: Proceedings of the International ScientificMethodological Conference. Modern Problems of theMathematical Modeling, the Forecasting and the Optimization,Kiev, Kamjanecj – Podiljsjk, 2006, pp. 70–76.

[83] G.S. Souza, E.G. Gomes, M.C. Magalhães, A.F.D. Avila, Economicefficiency of Embrapa‘s research centers and the influence ofcontextual variables, Pesquisa Oper. 27 (1) (2007) 15–26.

[84] R.E. Steuer, Multiple criteria optimization: Theory, Computation,and Application, Wiley, New York, 1986.

[85] S. Suzuki, P. Nijkamp, P. Rietveld, E. Pels, A distance frictionminimization approach in data envelopment analysis: acomparative study on airport efficiency, Eur. J. Oper. Res. 207(2010) 1104–1115.

[86] S. Suzuki, P. Nijkamp, A stepwise efficiency improvement DEAmodel for airport operations with fixed production factors, in: ERSAConference Papers ersa11p1065, European Regional ScienceAssociation. Accessed at: <http://ideas.repec.org/p/wiw/wiwrsa/ersa11p1065.html#biblio> (in April 18th, 2013).

[87] B.G. Tabachnick, L.S. Fidell, Using Multivariate Statistics, Allyn &Bacon, Boston, 2001.

[88] M. Toloo, The most efficient unit without explicit inputs: anextended MILP–DEA model, Measurement 46 (9) (2013) 3628–3634.

[89] H. Turner, R. Windle, M. Dressner, North American containerportproductivity: 1984–1997, Transp. Res. Part E 40 (4) (2004) 339–356.

[90] Vargas, C. and Bricker, D. (2000), ‘‘Combining DEA and factoranalysis to improve evaluation of academic departments given

uncertainty about the output constructs’’, working paper,Department of Industrial Engineering, University of Iowa, IowaCity, April.

[91] T.F. Wang, D.W. Song, K. Cullinane, The applicability of dataenvelopment analysis to efficiency measurement of containerports, in: International Association of Maritime EconomistsProceedings of the IAME Panamá 2002 Conference, 2002.Accessed at: <www.eclac.cl/transporte/perfil/iame.../wang_et_al.doc> (in May 2nd, 2014).

[92] W.-K. Wang, W.-M. Lu, C.-J. Tsa, The relationship between airlineperformance and corporate governance amongst US Listedcompanies, J. Air Transport Manage. 17 (2) (2011) 148–152.

[93] P.F. Wanke, P.F. Fleury, Transporte de Cargas no Brasil: EstudoExploratório das Principais Variáveis Relacionadas aos DiferentesModais e às suas Estruturas de Custos’’ in De Negri, J.B. and Kubota,L.C. (Org.), Estrutura e Dinâmica do Setor de Serviços no Brasil, IPEA,Brasília, 2006.

[94] P.F. Wanke, C.R. Affonso, Determinantes da eficiência de escala nosetor brasileiro de operadores logísticos, Produção 21 (1) (2011)53–63.

[95] Peter F. Wanke, R.F. Barbastefano, M.F. Hijjar, Determinants ofefficiency at major Brazilian port terminals, Transport Rev. 31 (5)(2011) 653–677.

[96] P. Wanke, Efficiency of Brazil’s airports: evidences frombootstrapped DEA and FDH estimates, J. Air Transport Manage. 23(2012) 47–53.

[97] P. Wanke, Capacity shortfall and efficiency determinants inBrazilian airports: evidence from bootstrapped DEA estimates,Socio-Econ. Planning Sci. 46 (3) (2012) 216–229.

[98] P. Wanke, Evaluating efficiency in the Brazilian trucking industry,Produção 23 (3) (2013) 508–524.

[99] M.M. Weber, W.L. Weber, Productivity and efficiency in thetrucking industry: accounting for traffic fatalities, Int. J. Phys.Dist. Logistics Manage. 34 (1) (2004) 39–61.

[100] D. Wilson, R. Purushothaman, Dreaming with BRICs: The path to2050, Global Economic Paper No: 99.2003, 2013. Available at:<http://www.gs.com> (accessed 13 October 2010).

[101] J.M. Youngblut, G.R. Casper, Focus on psychometrics: single-itemindicators in nursing research, Res. Nursing Health 16 (6) (1993)459–465.

[102] M.M. Yu, Assessment of airport performance using the SBM–NDEAmodel, Omega 38 (6) (2010) 440–452.

[103] G. Zhou, H. Min, C. Xu, Z. Cao, Evaluating the comparative efficiencyof Chinese third-party logistics providers using data envelopmentanalysis, Int. J. Phys. Dist. Logistics Manage. 38 (4) (2008) 262–279.

[104] N. Zill, M. Daly, Researching the Family: A Guide to Survey andStatistical Data in US Families, U.S. Department of Health andHuman Services, Washington, D.C., 2003.

[105] J. Zhu, Quantitative Models for Performance Evaluation andBenchmarking: Data Envelopment Analysis with Spreadsheetsand DEA Excel Solver, Springer, New York, NY, 2003.