lable at ScienceDirect

Journal of Environmental Management 92 (2011) 2698e2707

Contents lists avai

Journal of Environmental Management

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

The influence of the operational environment on the efficiency of water utilities

Pedro Carvalho*, Rui Cunha Marques 1

Center for Management Studies (CEG-IST), IST, Technical University of Lisbon, Av. Rovisco Pais, 1049-001 Lisbon, Portugal

a r t i c l e i n f o

Article history:Received 11 August 2010Received in revised form20 May 2011Accepted 3 June 2011Available online 1 July 2011

Keywords:Conditional efficiency measuresOperational environmentPerformanceOrder-mWater utilities

* Corresponding author.E-mail addresses: pedro.c.carvalho@ist.utl.pt (P. Car

(R.C. Marques).1 Tel.: þ351 218417729; fax: þ351 218417979.

0301-4797/$ e see front matter � 2011 Elsevier Ltd.doi:10.1016/j.jenvman.2011.06.008

a b s t r a c t

Adjusting for the operational environment in studies of performance measurement is very important,otherwise the analysis may lead to unrealistic scores, especially when its influence on costs is high, suchas in the water utilities. In this paper, we study the influence of exogenous variables on the water utilitiesperformance by applying conditional efficiency measures based on the order-m method and its proba-bilistic formulation. We use a sample of 66 water utilities operating between 2002 and 2008, repre-senting about 70% of the Portuguese population. Our research suggests that inefficiency of Portuguesewater utilities is substantial for some utilities: several exogenous variables might influence it consid-erably. For example, regulation has a positive influence on efficiency but when drinking water supply andwastewater services are provided by the same utility or when the wholesale and retail activities areprovided together, the performance is lower. The effect of ownership is inconclusive and the variablesresidential customers, water source, peak factor, and density of customers have a mixed influence onperformance which varies according to their scores.

� 2011 Elsevier Ltd. All rights reserved.

1. Introduction

One of the most critical problems in benchmarking is that mosttechniques assess performance without accounting for the opera-tional environment where the decision units are working in (Friedet al., 2008). Thus, not considering environmental variables inefficiency analysis may lead to biased results, especially if thevariables have a strong influence on the production process. This isthe case in the water utilities where heterogeneity indeed doesmatter. Therefore, this paper aims to assess the efficiency ofPortuguese water utilities (which include drinking water supplyand wastewater collection and treatment) taking into account theoperational environment. The purpose of this benchmarking studyis to encourage efficiency improvements and strengthen thesustainability of the water sector in Portugal providing lessons forthe rest of theworld. Operational environment is here defined as allthe exogenous variables (also called environmental) that interfere,to a larger or lesser extent, in the performance of observations(water utilities in this research) and do not depend on theirmanagerial practices (Witte and Marques, 2010). Strictly speaking,exogenous variables influence the costs rather than the efficiency

valho), rui.marques@ist.utl.pt

All rights reserved.

(although the two concepts are close to each other) but if we do notconsider them these two concepts get mixed leading to a biasedanalysis. This is particularly important in the case of regulation ofwater utilities when benchmarking is included in the regulatoryprocess (e.g. performance-based regulation).

The paper develops a recent and robust non-parametric meth-odology suggested by Cazals et al. (2002), called order-m, whichallows for the inclusion of environmental variables in efficiencyestimation. It is well-known that several exogenous variablesinfluence the water utilities efficiency, such as the scale of opera-tions, ownership, source of raw water, population density, financialand water quality regulations, the relative importance of non-residential vs. residential demand, and the topography of theregion where water utilities operate (Conti, 2005; Renzetti andDupont, 2009). However, they are not always considered in theanalysis of efficiency. When this happens, not always are the mostsuitable methodologies followed (Badin et al., 2008). A moregeneral and attractive approach is to consider the probabilisticformulation of the production process proposed by Cazals et al.(2002). The order-m methodology, being a partial frontier method,only uses part of the sample (m water utilities) to compute effi-ciency scores, therefore it is a less sensitive method to extreme dataand outliers and more robust than the traditional non-parametricmethods, such as data envelopment analysis (DEA) and the non-convex free disposal hull (FDH), known as full frontier methods(Fried et al., 2008). Based on this methodology, along with the oneproposed by Daraio and Simar (2005), we assessed the influence of

P. Carvalho, R.C. Marques / Journal of Environmental Management 92 (2011) 2698e2707 2699

exogenous variables on the performance of Portuguese waterutilities between 2002 and 2008.

The remainder of the paper is organized as follows. Section 2reviews the literature on the influence of the operational envi-ronment on water utilities efficiency. Section 3 explains themethodology adopted. Section 4 describes the case-study anddetermines the efficiency of the Portuguese water utilities andSection 5 examines the influence of the operational environmenton their efficiency. Finally, Section 6 presents the main conclusions.

2. Literature review

Berg and Marques (forthcoming) conducted a survey of perfor-mance about performance evaluation studies in the water utilitiesand found 190 papers published up to 2009. These studies addresspolicy issues, like the market structure (economies of scale andscope), the impact of ownership arrangements (public versusprivate), and the role of incentives and governance in promotingefficiency. These authors also identified the studies that assess theperformance of water utilities and that simultaneously consider theoperational environment. Exogenous variables are external to thewater utilities and are not under the control of the managers,but can influence the production process and consequently theirperformance.

Although the operational environment is one of the most rele-vant issues in performance analysis, only about 35% of the non-parametric studies published until now (42% of the total) includeit (Berg and Marques, forthcoming). Its non-inclusion may lead tobiased results, mainly in situations where these variables havea huge influence on the production process. Suppose, for example,that a given exogenous variable provides an operational environ-ment favorable to a water utility A, offering it the possibility ofgetting more outputs with less inputs, compared to other waterutilities that are not subject to that favorable operational environ-ment and that, in fact, are more efficient than water utility A. Aperformance analysis which does not consider that exogenousvariable can lead us to consider water utility A as more efficientthan the others, which is not right. These incorrect analyses canhave harmful effects, as poorly performing utilities might berewarded and truly efficient ones punished, thus frustrating goodmanagers and reducing the legitimacy of the process. Therefore, itis vital to identify and understand the influence of various exoge-nous variables on the performance of water utilities, so that theirinefficiency can be mitigated.

There are several exogenous variables with influence onwater utilities efficiency. The literature highlights the importantrole of customer or population density, proportion of residentialcustomers, water source and/or its water quality, peak factor,topography and climate of the region (Renzetti and Dupont,2009), the size of operation, ownership (Picazo-Tadeo et al.,2009b) and financial and water quality regulations (Conti,2005) and the water restrictions and the degree of managerialefficiency (Byrnes et al., 2010), among other variables. Abbott andCohen (2009) also report that exogenous variables related to theenvironmental issues, such as water conservation and reclaimedwater, and the relationship between water supply and urbanplanning might have impact on the efficiency of water utilities.Note, however, that these exogenous variables are not undesir-able variables which, a priori, are controlled by the managers(Chung et al., 1997).

Regarding the influence of customer (population) density, theliterature seems to prove the existence of economies of density, i.e.,higher levels of efficiency if the water utilities operate in areas ofhigh population density (Mann and Mikesell, 1976; Antoniolliand Filippini, 2001; Picazo-Tadeo et al., 2009a). However, the

literature provides other studies with different conclusions. Tupperand Resende (2004) found empirical evidence of the existence ofeconomies of density in the drinking water supply but not in thesewage collection, whereas the results of García and Thomas (2001)were inconclusive.

Residential customers has been one of the most importantexogenous variables included in the literature on performanceevaluation of the water utilities. In general, higher percentages ofthis variable correspond to higher costs and, therefore, to lowerefficiencies. For example, Anwandter and Ozuna (2002) investi-gated the efficiency of the water utilities in Mexico and consideredthe variable percentage of non-residential customers as an exoge-nous variable. The authors noticed a significant positive effect ofthis variable on the efficiency of water utilities. However, there islittle consensus on its effect on wastewater utilities (Byrnes et al.,2009).

The water source is another important exogenous variable.Although the literature suggests that the use of a high proportion ofsurface water may imply more advanced chemical treatments topurify water (Aubert and Reynaud, 2005), not always has itsinfluence been proved as positive on the water utilities efficiency(Aubert and Reynaud, 2005; Marques and Monteiro, 2004). Forexample, the latter authors measured the efficiency of the Portu-guese water utilities and evaluated the effect of the groundwatersource which was not relevant but Shih et al. (2006) studied the USwater utilities between 1995 and 2000 and concluded thatgroundwater systems are less costly than surface water systems.

The literature has also studied the influence of the peak factoron water utilities efficiency. Woodbury and Dollery (2004), forexample, to assess the performance of New South Wales municipalwater utilities in Australia, incorporated the seasonal variation inwater demand. The results seem to indicate that the peak factor hasa positive influence on the water utilities performance. Picazo-Tadeo et al. (2009a) found that the water utilities, in Andalucía,Spain, which provide water for touristic municipalities showgreater efficiency scores than the ones that provide water to non-touristic areas, pointing to the conclusion that water utilities withhigher peak factors tend to have lower costs, for example, byachieving some scale or density economies.

Several studies have focused on the existence of scope econo-mies in the water utilities. They have examined whether or not theprovision of two or more services simultaneously by a single entityinvolves less costs than the provision of several services separatelyby different water utilities. Many of these studies show costreductions when the scope of water utilities provided increases(Abbott and Cohen, 2009; Hunt and Lynk, 1995). However, otherstudies found the opposite (see Saal and Parker (2000) and Stoneand Webster Consultants (2004)). Most of these papers onlyconsider the drinking water distribution and the sewage collection(Fraquelli et al., 2004). Other authors, such as Piacenza andVannoni (2004), are exceptions including other services as gasand electricity.

Vertical integration has also been analyzed in the literature. Thestudies focusing on this issue seem to suggest the existence ofscope economies in joint retail and wholesale production (verticalintegration), specially for the smallest water utilities (Abbott andCohen, 2009). Hayes (1987), for example, provided evidence ofthis in the US, but only for the smallest water utilities. Urakami(2007) examined the Japanese water supply industry and noticedthe existence of economies of vertical integration between thewater intake-purification and water distribution stages, particu-larly for water utilities with lower purchasing water ratios and Shihet al. (2006), who studied the US water utilities, found that ‘in-house’ production (ground or surface water) is cheaper thanpurchasing water from other utilities.

Fig. 1. Order-m efficient frontier illustration.

P. Carvalho, R.C. Marques / Journal of Environmental Management 92 (2011) 2698e27072700

Ownership is one of the exogenous variables that mostly calledthe attention of researchers, leading to a great number of paperspublished investigating the influence of this variable on the effi-ciency of water utilities (see Marques, 2008 for a comprehensiveliterature review). Some papers provided empirical evidence thatpublic water utilities display better performance (Bhattacharyyaet al., 1994; Bruggink, 1982; Shih et al., 2006) while other studiesconclude that private water utilities outperform the public ones(Bhattacharyya et al., 1995; Crain and Zardkoohi, 1978). In contrast,other authors did not find conclusive evidence that one regime ofownership outperforms the other, such as Byrnes et al. (1986),Feigenbaum and Teeples (1983) and Souza et al. (2007).

Another variable widely discussed in the literature is the regu-lation and its effect on efficiency (Shirley, 2006). The majority ofstudies in the literature analyze the US and the UK experiences.Most of them argue that regulation has a positive influence onwater services efficiency. For example, Aubert and Reynaud (2005)show that the type of regulation adopted, such as rate of return orprice cap, influences the water services efficiency in the US. Theliterature demonstrates that the efficiency and productivity of theUK water industry has improved after regulation had been imple-mented in the 1990s (Saal and Parker, 2000, 2004). More recently,Saal et al. (2007) also observed that environmental regulation hasled to a technical change, allowing for the enhancement of newtechnologies and new production processes.

3. Conditional probabilities

The non-parametric methods are very popular and useful forbenchmarking purposes since they do not require a priori anyassumption to represent the production (cost) technology (attain-able set) neither do they consider too many assumptions, as withthe parametric methods, and are relatively easy to explain todifferent stakeholders. Some of their disadvantages are the highsensitivity to the outliers, the absence or, at least, the greaterdifficulty of making statistical inferences based on the results andmodel specification and the difficulty of dealing with the opera-tional environment [see, for example, Coelli et al. (2005) or Friedet al. (2008)].

By determining the DEA efficiency scores and considering thedrawbacks pointed out, we applied the order-m method (Cazalset al., 2002). This one, belonging to the partial frontier methods,uses only part of the sample (m water utilities) to determine effi-ciency scores. Consequently, they are less sensitive to extreme dataand outliers and are more robust than the traditional non-parametric techniques of FDH and DEA (also known as full fron-tier methods). Furthermore, the partial frontier methods do not facethe problem of “curse of dimensionality” as the full frontier ones.This issue requires the use of a large number of observations toavoid inaccurate results (Daraio and Simar, 2007). Fig. 1 provides anexample of an order-m frontier, trying to show that it is based onthe full frontier FDH, although some observations might be above it(efficiency can take values higher than 1). Besides all theseadvantages, the order-m also allows for an easy inclusion of exog-enous variables to determine efficiency.

Until recently mainly two groups of procedures that includeexogenous variables were available in the literature of efficiencyanalysis, respectively the one-stage and two-stage methods.However, both of them have shortcomings (Daraio and Simar,2007). The exogenous variables in the first stage, whichcontribute to define the attainable set, are introduced into themodel as (non-discretionary) inputs or outputs (Färe et al., 1994)depending on whether they are favorable or adverse to perfor-mance, respectively. However, it is often difficult to know at thebeginning what is the real influence of the exogenous variables on

the performance. In addition, they are also considered restrictiveassumptions on the free disposability and the convexity (if DEA isused) of the corresponding attainable extended production set(Daraio and Simar, 2007).

Concerning the two-stage methodology, first the efficiencyscores are estimated with the traditional non-parametric methods(usually DEA) and then their relationship with the exogenousvariables is evaluated through Tobit or censored regressions (seeSimar and Wilson (2007)). However, these methodologies as wellas similar ones, like the three-stage or the four-stage analysis (Friedet al., 2002) or even the recent methodologies proposed by Simarand Wilson (2007) and by Park et al. (2008), based on bootstrapregression, have drawbacks, such as requiring the restrictiveseparability condition between the inputeoutput space and thespace of exogenous variables (Daraio and Simar, 2007). Thus, thereare great advantages in using the methodology proposed by Cazalset al. (2002). According to these authors, the production processcan be defined by the joint distribution function of inputs andoutputs and the efficiencies can be obtained from a conditionaldistribution function resulting from the decomposition of that jointdistribution function.

Based on the work of Cazals et al. (2002), Daraio and Simar(2005) proposed an alternative probabilistic formulation to theproduction process. According to them, the production process canbe described by the joint probability measure of (X, Y) on <p

þ � <qþ,

being characterized by the probability function HXY(x, y):

HXY ðx; yÞ ¼ ProbðX � x; Y � yÞ (1)

In a context of input orientation, such probability function canbe decomposed, using Bayes’s rule, in two other probability func-tions, in a conditional distribution function of X, FX=Y ðxjyÞ, and ina survivor function of Y, SY ðyÞ:

HXY ðx;yÞ ¼ ProbðX � xjY � yÞProbðY � yÞ ¼ FXjYðxjyÞSYðyÞ (2)

In the current study we considered an input orientation inwaterutilities performance evaluation, since the objective of theseservices is to rationalize the quantity of inputs for a given level ofoutputs.

Following Daraio and Simar (2005), the input oriented efficiencyscores (in a radial sense) can be defined in terms of the support ofthese probabilities as:

qðx; yÞ ¼ infnq��FXjY ðqxjyÞ>0

o¼ inffqjHXY ðqx; yÞ>0g (3)

Table 1Management model of the ‘retail’ segment.

Management model Quantity (no.) Population (103) Population (%)

Private concessionaires 26 1,798 17%Municipal mixed company 6 386 4%Municipal owned company 13 1,315 13%PublicePublic partnership 2 323 3%Semi-autonomous utility 24 2,234 22%Municipal service 207 4,486 43%State-owned company 1 565 5%

P. Carvalho, R.C. Marques / Journal of Environmental Management 92 (2011) 2698e2707 2701

The estimator of this input efficiency score for a given point (x, y)coincides with the FDH estimator of q(x, y):

q̂ðx; yÞ ¼ infnq��F̂XjY ðqxjyÞ>0

o¼ q̂FDHðx; yÞ (4)

Yet, in the order-m input oriented method, the efficiencymeasurement considers only m water utilities drawn randomlyfrom the population according to FXjY ð$jyÞ, that is, mwater utilitiesthat produce at least the output level, in opposition to the proce-dure of the FDH method where all water utilities are considered.

The order-m input efficiency measure is defined, according toDaraio and Simar (2005), in the following way:

q̂m;nðx; yÞ ¼ZN

0

�1� F̂XjY;nðuxjyÞ

�mdu (5)

where: F̂XjY ;nðuxjyÞ ¼ Pni¼1 IðXi � ux;Yi � yÞ=Pn

i¼1 IðYi � yÞ andI(k) is the indicator function that takes the value of I(k) ¼ 1 if k istrue or I(k) ¼ 0 otherwise.

Thus, the order-m efficiency score can be viewed as the expec-tation of the minimal input efficiency score of the water utility(x, y), when compared to m water utilities randomly drawn fromthe population of water utilities producing more outputs than thelevel y.

The inclusion of the exogenous variables is very easy, beingenough to limit the production process to a given value of theexogenous variable (usually referred to as Z), that is:

HXYðx; yÞ ¼ ProbðX � xjY � y; Z ¼ zÞProbðY � yjZ ¼ zÞ¼ FXjY ;Zðxjy; zÞSY jZðyjzÞ (6)

Obtaining conditional efficiencies involves the estimation ofa non-standard conditional distribution function, which requiresthe use of smoothing techniques for the exogenous variables. Suchsmoothing techniques still require the choice of a kernel functionand the determination of a bandwidth. In this research, we used theEpanechnikov kernel function and the likelihood cross validationbased on K-Nearest Neighbor method to obtain the optimal band-widths. Thus, following Daraio and Simar (2005), the conditionalefficiencies of the order-m (input oriented) approach, for a givenvalue of Z ¼ z, can be determined as:

q̂mðx; yjzÞ ¼ZN

0

�1� F̂XjY ;Z;nðuxjy; zÞ

�mdu (7)

where: F̂XjY;Z;nðuxjy; zÞ ¼ Pni¼1 IðXi � ux;Yi � yÞKððZ � ZiÞ=hÞ=Pn

i¼1 IðYi � yÞKððZ � ZiÞ=hÞ and I(k) is the indicator function thattakes the value of I(k)¼ 1 if k is true or I(k)¼ 0 otherwise and K($) isthe kernel function and h the appropriate bandwidth.

To analyze the influence of exogenous variables on theproduction process a non-parametric smoothed regression of theratios between the order-m conditional efficiencies and theunconditional efficiencies, Qz

m ¼ q̂m;nðx; yjzÞ=q̂m;nðx; yÞ, on Z wasemployed (Daraio and Simar, 2005). The influence of exogenousvariables on the production process is interpreted according to thedevelopment of the non-parametric smoothed regression over thevalues of the exogenous variable. In an input orientation context,when the non-parametric smoothed regression shows a growingtrend, the exogenous variable is unfavorable to efficiency if thevalues are increasing and when the regression has a decreasingslope the exogenous variable is favorable to efficiency.

4. Performance of Portuguese water utilities

4.1. The water sector in Portugal

The water utilities (drinking water supply and wastewatercollection and treatment) are under the responsibility of themunicipalities. However, the Portuguese management model isquite different from other countries in Europe like France or Spain,since the drinking water supply and the wastewater sectors are notvertically integrated, that is, as a rule, there are different actorssupplying the ‘wholesale’ segment (intake, treatment and trans-portation in water and transportation and treatment in waste-water) and the ‘retail’ segment (water distribution and wastewatercollection).

In addition, there is a sector-specific regulator (with economicand quality of service regulatory functions), an unusual circum-stance in Europe (in EU 15 there are only economic regulators in theUK and Italy). This regulatory agency, until recently, only hadfunctions concerning the concessionaire companies with eitherpublic or private ownership. Even so, it had a sound role in thesector. By applying its model of sunshine regulation, consisting incomparing, discussing and publicizing the performance of waterutilities, the companies are encouraged to improve their perfor-mance. It uses reference values (targets) as benchmarks forperformance indicators that the regulator considers likely to bereached by each utility. However, exogenous variables are onlyqualitatively considered for each operator. As far as the quality ofservice is concerned, its results have been quite positive (Marquesand Simões, 2008) and, as a result, the regulatory responsibilitieshave recently been extended to the entire sector, including the non-concessionaire companies which are the majority.

Concerning the Portuguese water market structure, in the‘wholesale’ segment there are 21 water utilities (3 corresponding todrinking water, 6 to wastewater and 12 to water and wastewater)which encompass 233 municipalities (out of 278 in the mainland).From these, one is a private concessionaire (contractual public-private partnershipe ePPP) and the remaining are public conces-sionaires (a partnership between a State owned company andmunicipalities). In the retail segment, there are 279 retail compa-nies, including private concessionaires (contractual PPPs), a state-owned company (Epal, which supplies Lisbon), municipal compa-nies [with private participation (institutionalized PPP) or not],publicepublic partnerships between the State and municipalities,semi-autonomous utilities (municipality bodies with administra-tive and financial autonomy) and municipal services (without anymanagerial autonomy). Table 1 presents the market structure in thePortuguese water sector ‘retail’ segment.

In addition to the regulation and unbundling of the water sector,other two noteworthy features should be stressed. The first one isrelated to the importance of the Central State as player in the watersector, controlling almost all the wholesale segment, and havinga relevant position in the retail segment (three companiessupplying 31 municipalities). The second one refers to the growing

Table 3Major statistical features of DEA efficiency results.

Model Average St. Deviation Minimum Median Efficient waterutilities (no.)

P. Carvalho, R.C. Marques / Journal of Environmental Management 92 (2011) 2698e27072702

importance of the private sector participation. Although in a slowpace, the privatization is assuming a meaningful role year afteryear. Currently 21% of the population is already supplied by privateoperators.

CRS 0.610 0.193 0.354 0.570 5VRS 0.776 0.211 0.378 0.820 20SE 0.799 0.161 0.373 0.809 5

Table 4Major statistical features of DEA efficiency results for the period 2002e2008.

Model Average St. Deviation Minimum Median Efficient waterutilities (no.)

CRS 0.580 0.185 0.218 0.524 15VRS 0.720 0.192 0.219 0.725 51SE 0.820 0.173 0.369 0.883 15

4.2. Sample and model adopted

In this study, we analyzed 66 Portuguesewater utilities between2002 and 2008 which encompass almost 70% of resident pop-ulation. These utilities focus mainly on the retail segment and mostof them include other activities beyond drinking water supply, suchas wastewater and, to a less extent, urban waste. We think thattechnological change was not a determinant for efficiency in thisperiod (and in this sector in particular).

The model specification includes three inputs and two outputs.The inputs correspond to the staff cost, other operation andmaintenance expenses (other OPEX) and capital expenses (CAPEX).As financial data are homogenous and audited among water utili-ties, CAPEX is determined by the sum of depreciation and interestpaid. The outputs adopted are the volume of water delivered (sumof volume of water billed and volume of treated wastewater)expressed in m3 and the number of customers (sum of water andwastewater customers). We also adopted the ‘consensual’ inputorientation model in the water sector (Berg and Marques,forthcoming). Table 2 systematizes the major statistical featuresof the model used.

All data were obtained from the annual account reports pub-lished by the water utilities and from the annual sector reports ofthe regulator.

4.3. Efficiency of Portuguese water utilities

By the use of non-parametric methods, we first measured thewater utilities efficiency through the DEA technique, using constantreturns to scale (CRS) and variable returns to scale (VRS) technol-ogies. Then, we applied the order-m methodology.

For the year 2008, 5 (8%) water utilities out of 66 are efficientunder the CRS model and 20 (30%) under the VRS model. Con-cerning the efficiency scores, the VRS model displays an averagevalue of 0.776, whereas the CRS model highlights an average scoreof 0.610. These results mean that on average the water utilitiescould reduce their inputs 22.4% in the VRS model (and 39.0% in theCRS model) to become efficient.

The water utilities are about 20.1% inefficient owing to theirscale diseconomies. This means that, in general, they could save20.1% of the inputs consumed if demand conditions were such thatthey could operate at an optimal scale. Moreover, the sample isdominated by decreasing returns to scale, representing about 68%of the water utilities studied. Tables 3 and 4 show the statisticalfeatures of DEA efficiency scores obtained for the year 2008 and forthe period 2002e2008, respectively.

According to the order-m approach, for the period 2002e2008,efficiency varies between 3.13 and 0.64, for a value of m ¼ 60

Table 2Statistical features of the model specification adopted.

INPUTS

Staff cost (103 V) Other OPEX (103 V) CAP

Average 3609 5951 31St. Deviation 5456 7973 57Minimum 87 148Maximum 43,856 47,436 45,3Median 1787 3170 16

(as shown in Fig. 2). This means that the most efficient water utilitywith an efficiency score of 3.13 uses 2.13 times less inputs(proportional reduction) than the expected value of the minimuminput level of m ¼ 60 other water utilities, drawn from the pop-ulation of water utilities producing a level of output greater than orequal to its level of outputs. In contrast, the least efficient waterutility with an efficiency of 0.64, uses 36% more inputs (radialextension) than the expected value of the minimum input level ofm ¼ 60 other water utilities, drawn from the population of firmsproducing a level of output greater than or equal to its level ofoutputs. Fig. 2 shows the order-m efficiency results. We testedseveral values of m and adopted m ¼ 60 since there was a reason-able stabilization for a percentage of efficiency scores greater than 1(Daraio and Simar, 2007).

5. How do exogenous variables affect performance?

In order to evaluate the influence of the operational environ-ment on the water utilities performance we used the methodologyproposed by Daraio and Simar (2005). As far as we know, it is thefirst time that this methodology is applied to the water sector,avoiding the problems of the previous studies on this issue whichadopt one or two-stage methods. This methodology consists inapplying a non-parametric smoothed regression of the ratiosbetween the order-m conditional efficiencies and the uncondi-tional efficiencies, Qz

m ¼ q̂m;nðx; yjzÞ=q̂m;nðx; yÞ, on Z. In a context ofinput orientation, the influence of exogenous variables on theproduction process is interpreted according to the pattern of thenon-parametric smoothed regression through the values of theexogenous variable. Therefore, if regression presents a growingscope while the exogenous variable values increase, it indicatesthat the exogenous variable is unfavorable to efficiency and ifregression decreases it means that the exogenous variable isfavorable to efficiency.

As exogenous variables, we considered the scope of activitiesprovided, vertical integration, ownership, regulation, purchased

OUTPUTS

EX (103 V) Volume of water billed (103 m3) Customers (no.)

74 11,176 77,27909 27,464 85,40518 271 511588 223,116 371,12226 4749 42,091

Fig. 2. Order-m efficiency results for a value of m ¼ 60 (period 2002e2008).

P. Carvalho, R.C. Marques / Journal of Environmental Management 92 (2011) 2698e2707 2703

water, surface water, peak factor, customer density and residen-tial customers. Our sample is relatively homogenous since itincludes only Portuguese water utilities, therefore avoiding otherimportant variables such as water law, policy and administration(Araral, 2010) or climatic effects and water restrictions (Byrnes

Table 5Influence of the exogenous variables studied on efficiency.

Explanatory variables Expectatio

Scope (1 e only water j2 e onlywater and wastewater)

Ambiguou

(1 e only water j3 e water, wastewater andurban waste)(2 e only water and wastewater j3 e water,wastewater and urban waste)

Wholesale/retail (water)(0 e wholesale and retailj1 e only retail)

Ambiguou

Wholesale/retail (wastewater) (0 e wholesaleand retail j1 e only retail)

Ownership (0 e public j1 e partial or total private) Ambiguou

Regulation (0 e non-regulated j1 e regulated) þ

% Purchased water Variable

% of surface water provided% surface water source

e

Customer density e water utilities (customers/km) e

Customer density e wastewater utilities (customers/km) e

Peak factor þ

% Residential customers e

Kernel regression significance test: ***0.001; **0.01; *0.05; �0.1.

et al., 2009). Concerning the variable scope of activities, weconsidered it as a dummy variable depending on the activitiesprovided, taking the value of 1 for the water, the value of 2 forthe water and wastewater and the value of 3 for the water,wastewater and urban waste. Likewise, we regarded the variables

n Empirical conclusion p-value

s According to the results in Appendix 1 the provision of watersupply andwastewater utilities simultaneously is unfavorable toperformance when compared with the services that onlyprovide the water supply for a confidence level of 95%. However,between the water utilities, the water and wastewater utilitiesand the water, wastewater and urban waste utilities weobserved the existence of scope economies in Portugal, that is,there is a positive influence on the joint provision of watersupply, wastewater and urban waste activities.

0.045*

0.030*

0.040*

s Appendix 1 shows that it is favorable to performance to beresponsible only for the retail component rather than managingthe wholesale and retail simultaneously, both for the waterutilities and wastewater utilities, for a level of confidence of 95%.These results prove that specialization is favorable to the waterutilities in Portugal.

0.033*

0.043*

s In Appendix 1 we observed that the fact that water utilities arepartially or totally public or private services does not influencethe production process for a level of 95%.

0.013*

We found that regulation has a positive influence on theefficiency of water services in Portugal.

0. 043*

According to our results (Appendix 1) for a confidence level of95% there is a positive influence on the Portuguese waterutilities performance to purchase water near 20% or 80%.However, for percentages above 80% the influence is negative.There is also a negative influence between 20 and 40% of waterpurchased.

0.043*

Our results show that for a confidence level of 95% until about70% of surface water supplied there is virtually no influence onperformance. From about 80% on to about 95% there is a positiveand a negative influence on percentages of surface waterbetween 70% and 80% and for percentages higher than 95%approximately.

0.038*

According to the results obtained both for water andwastewaterutilities, there is a negative influence up to 50 customers per kmwhile there is approximately a null influence to higher customerdensities.

0.045*0.075�

Appendix 1 also illustrates that for a level of 95% there isa negative influence for peak factors up to 1.2 and for peakfactors above 1.4. There is a positive influence when the peakfactor gests close to 1.4

0.048*

Regarding the environmental variable domestic customers,Appendix 1 shows a slight negative influence on performanceuntil about 86%. However, our results show a contradiction tothe literature for percentages higher than 86% of residentialcustomers, from which there is a slight positive influence until90%. The results correspond to a confidence level of 95%.

0.035*

P. Carvalho, R.C. Marques / Journal of Environmental Management 92 (2011) 2698e27072704

vertical integration and ownership as dummy variables. Thevariable vertical integration has the value of 0 for the wholesaleand retail utilities, and the value of 1 for the retail utilities.Ownership distinguishes the private water utilities from publiclyowned companies. This variable takes the value of 0 for thepublic management and of 1 for the total or partial privatemanagement. The variable regulation assumes the value of 0 forthe non-regulated services and the value of 1 for the regulatedservices. The variable purchased water represents the percentageof water imported comparing with the total water delivered. Thevariable surface water, on its turn, represents the percentage ofsurface water in the total water delivered. The variable peakfactor is the ratio between the maximum monthly deliveredwater volume and the average monthly delivered water volume.The variable customer density corresponds to the ratio betweenthe number of customers and the pipe length. Finally, the vari-able residential customers corresponds to the percentage ofresidential customers supplied. Table 5 summarizes the variablesadopted in this analysis, the expectations about their influenceon performance, the results obtained and the corresponding p-values associated with each non-parametric regression analysis.Appendix 1 illustrates the results graphically.

In general, the results corresponded to our expectations. Forinstance, concerning the variable scope of services provided, it wasexpected that scope diseconomies would exist between waterutilities and water and wastewater utilities in Portugal becauseprevious research obtained that kind of results (Correia andMarques, 2011). However, we observed the existence of scopeeconomies between the water utilities, water and wastewaterutilities and water, wastewater and urban waste utilities inPortugal. This fact is in line with other studies that founda reduction of costs with the increase of the scope of water utilities(Abbott and Cohen, 2009; Fraquelli et al., 2004; Hunt and Lynk,1995). We also noted that there were no economies in jointretail and wholesale production (vertical integration). This resultwas not the one most expected since the few studies available inthe literature have observed the opposite, especially in thesmallest water utilities (Abbott and Cohen, 2009; Hayes, 1987;Urakami, 2007).

There is a positive influence of the variable percentage of waterpurchased if utilities purchase about 20% or 80% of water. Theresults support the previous conclusions on the variable verticalintegration, since it is more favorable for the water utilities to beonly responsible for the retail component buying approximately80% of water. However, there is still a positive influence on theperformance of water utilities in Portugal that buy percentages ofwater up to 20%, suggesting that there are also economies injointly managing the retail and wholesale components up to thatpercentage.

As in most studies, our analysis is inconclusive concerning thevariable ownership. Indeed, the results point out that neitherprivate nor public utilities outperform.

In line with the literature (Aubert and Reynaud, 2005; Saal andParker, 2004), we found that regulation has a positive influence onthe Portuguese water utilities.

The literature points out the peak factor as a relevant exogenousvariable (Picazo-Tadeo et al., 2009a; Woodbury and Dollery, 2004).We found that there is a positive influence on the performance ofthe water utilities in Portugal for peak factors between 1.2 and 1.4but negative otherwise. This matches the literature since extremevalues are usually harmful for efficiency.

The variable customer density was expected to have a negativeinfluence on the performance of water utilities in Portugal for lowdensities, that is, for high dispersion of population, as reported inliterature (Antoniolli and Filippini, 2001; Mann and Mikesell, 1976;

Picazo-Tadeo et al., 2009a). However, this study indicates that thereis only a negative influence for low densities until 50 customers perkm of pipes both for water and wastewater utilities.

Regarding the exogenous variable percentage of domesticcustomers, we observed, a negative influence on performance up to86% of domestic customers similarly to the literature results.However, we found a slight positive effect for percentages of resi-dential customers between 86% and 90%, which contradicts theliterature (Byrnes et al., 2009).

Finally, we did not expect the results for the variable surfacewater supplied (Byrnes et al., 2010). The literature shows that usinga high proportion of surface water may imply more advancedchemical treatments in order to purify water (Aubert and Reynaud,2005) rather than using groundwater (Shih et al., 2006). However,we have just seen this negative influence on percentages between70% and 80% and for percentages higher than 95%, noting that forpercentages between 80% and 95% there is a positive influence andfor percentages below 70% there is no influence on the waterutilities performance in Portugal. This could have happenedbecause most utilities buy water from other utilities and so thevariable water source (and, therefore, the treatment cost) is lesssignificant.

6. Conclusions

This research is a timely contribution to the literature asa benchmarking study, using a very updated and complete sample.In addition, it highlights the influence of various exogenous vari-ables on performance through the application of the recent androbust non-parametric methods proposed by Cazals et al. (2002)and Daraio and Simar (2005). A further contribution is the Portu-guese case-study which might be very useful for other Europeancountries and others under their influence (Central and SouthAmerica and African countries).

Several policy implications can be drawn concerning theinfluence of the operational environment on water utilitiesperformance. Considering that in Portugal there is a regulatorwhich applies benchmarking and sunshine regulation, theinclusion of exogenous variables in the regulatory process is evenmore crucial. Therefore, the first implication concerns the role ofregulator which should take into account the negative, positiveor neutral influence on performance of variables such as the peakfactor, percentage of water purchased, customer density,percentage of domestic customers, ownership, regulation,vertical integration and percentage of surface water supplied onbenchmarks defined in its regulatory process. The secondimplication is related to policy makers and the government.Portugal went through a reform of the water sector in the 1990swhich allowed for the privatization and unbundling water andwastewater activities (wholesale and retail activities began to beprovided by different operators) intending to optimize themarket structure and to increase the efficiency of water utilities.Concerning vertical integration and privatization, our researchproved that this reform was not successful and therefore itshould be reviewed or changed by the government. Finally,a third implication is directed to the operators. Voluntary exer-cises of benchmarking are common in the Portuguese waterindustry, although most of the comparisons are based on oper-ational activities. Therefore, water utilities should take care notto compare ‘apples’ with ‘oranges’.

To sum up, the importance of the exogenous variables are welljustified in this research and indeed employing benchmarking,which is a very relevant tool, without accounting for the opera-tional environment will certainly lead to biased results, especiallyin water utilities where it always has a strong influence.

Appendix 1. Non-parametric regression for the exogenous variables studied.

P. Carvalho, R.C. Marques / Journal of Environmental Management 92 (2011) 2698e2707 2705

P. Carvalho, R.C. Marques / Journal of Environmental Management 92 (2011) 2698e27072706

P. Carvalho, R.C. Marques / Journal of Environmental Management 92 (2011) 2698e2707 2707

References

Abbott, M., Cohen, B., 2009. Productivity and efficiency in the water industry. Util.Policy 17, 233e244.

Antoniolli, D., Filippini, M., 2001. The use of variable cost function in the regulationof the Italian water industry. Util. Policy 10, 181e187.

Anwandter, L., Ozuna, T., 2002. Can public sector reforms improve the efficiency ofpublic water utilities? Environ. Dev. Econ. 7, 687e700.

Araral, E., 2010. Improving effectiveness and efficiency in the water sector: insti-tutions, infrastructure and indicators. Water Policy 12, 1e7.

Aubert, C., Reynaud, A., 2005. The impact of regulation on cost efficiency: anempirical analysis of Wisconsin water utilities. J. Prod. Anal. 23, 383e409.

Badin, L., Daraio, C., Simar, L., 2008. Optimal Bandwidth Selection for ConditionalEfficiency Measures: A Data-driven Approach. LEM Papers Series. Laboratory ofEconomics and Management, Sant’Anna School of Advanced Studies.

Berg, S., Marques, R., Quantitative studies ofwater and sanitation utilities: a literaturesurvey. Water Policy, forthcoming. doi:10.2166/wp.2011.041.

Bhattacharyya, A., Harris, T., Narayanan, R., Raffiee, K., 1995. Technical inefficiency ofrural water utilities. J. Agr. Resour. Econ. 20, 373e391.

Bhattacharyya, A., Parker, E., Raffiee, K., 1994. An examination of the effect ofownership on the relative efficiency of public and private water utilities. LandEcon. 70, 197e209.

Bruggink, T., 1982. Public versus regulated private enterprise in the municipal waterindustry: a comparison of operating costs. Q. Rev. Econ. Bus 22, 111e125.

Byrnes, J., Crase, L., Dollery, B., Villano, R., 2009. An analysis of the relative efficiencyof wastewater utilities in non-metropolitan New South Wales and Victoria.Australas. J. Reg. Stud. 15, 153e169.

Byrnes, J., Crase, L., Villano, R., Dollery, B., 2010. The relative economic efficiency ofurban water utilities in regional New South Wales and Victoria. Resour. EnergyEcon. 32, 439e455.

Byrnes, P., Grosskopf, S., Hayes, K., 1986. Efficiency and ownership: further evidence.Rev. Econ. Stat. 68, 337e341.

Cazals, C., Florens, J., Simar, L., 2002. Nonparametric frontier estimation: a robustapproach. J. Econom. 106, 1e25.

Chung, Y.R., Färe, R., Grosskopf, S., 1997. Productivity and undesirable outputs:a directional distance function approach. J. Environ. Manag. 51, 229e240.

Coelli, T., Rao, P., O’Donnell, C., Battese, G., 2005. An Introduction to Efficiency andProductivity Analysis. Springer, New York.

Conti, M., 2005. Ownership relative efficiency in the water industry: a survey of theinternational empirical evidence. Econ. Int. 58, 273e306.

Correia, T., Marques, R., 2011. Performance of Portuguese water utilities: how doownership, size, diversification and vertical integration relate to efficiency?Water Policy 13, 343e361.

Crain, W.M., Zardkoohi, A., 1978. A test of the property-rights theory of the firm:water utilities in the United States. J. Law Econ. 21, 395e408.

Daraio, C., Simar, L., 2005. Introducing environmental variables in nonparametricfrontier models: a probabilistic approach. J. Prod. Anal. 24, 93e121.

Daraio, C., Simar, L., 2007. Advanced Robust and Nonparametric Methods in Effi-ciency Analysis. Methodology and Applications. Springer, New York.

Färe, R., Grosskopf, S., Lovell, K., 1994. Productions Frontiers. Cambridge UniversityPress, Cambridge.

Feigenbaum, S., Teeples, R., 1983. Public versus private water delivery: a hedoniccost approach. Rev. Econ. Stat. 65, 672e678.

Fraquelli, G., Piacenza, M., Vannoni, D., 2004. Scope and scale economies in multi-utilities: evidence from gas, water and electricity combinations. Appl. Econ. 36,2045e2057.

Fried, H., Lovell, K., Schmidt, S., 2008. The Measurement of Productive Efficiency andProductivity Change. Oxford University Press, New York.

Fried, H., Lovell, K., Schmidt, S., Yaisawarng, S., 2002. Accounting for environmentaleffects and statistical noise in Data Envelopment Analysis. J. Prod. Anal. 17,157e174.

García, S., Thomas, A., 2001. The structure of municipal water supply costs: appli-cation to a panel of French local communities. J. Prod. Anal. 16, 5e29.

Hayes, K., 1987. Cost structure of the water utility industry. Appl. Econ. 19,417e425.

Hunt, L., Lynk, E., 1995. Privatization and efficiency in the UK water industry: anempirical analysis. Oxf. Bull. Econ. Stat. 57, 371e388.

Mann, P., Mikesell, J., 1976. Ownership and water system operation. Water Resour.Bull. 12, 995e1004.

Marques, R., 2008. Comparing private and public performance of Portuguese waterutilities. Water Policy 10, 25e42.

Marques, R., Monteiro, A., 2004. Benchmarking the economic performance ofPortuguese water and sewerage services. In: Emrouznejad, A., Podinovski, V.(Eds.), Data Envelopment Analysis and Performance Measurement. WarwickPrint, Coventry, pp. 65e72.

Marques, R., Simões, P., 2008. Does the sunshine regulatory approach work?Governance and regulation model of the urban waste services in Portugal.Resour. Conserv. Recycl. 52, 1040e1049.

Park, B., Simar, L., Zelenyuk, V., 2008. Local likelihood estimation of truncatedregression and its partial derivatives: theory and application. J. Econom. 146,185e198.

Piacenza, M., Vannoni, D., 2004. Choosing among alternative cost function speci-fications: an application to Italian multi-utilities. Econ. Lett. 82, 415e422.

Picazo-Tadeo, A., Sáez-Fernández, F., González-Gómez, F., 2009a. The role of envi-ronmental factors in water utilities technical efficiency. Empirical evidencefrom Spanish companies. Appl. Econ. 41, 615e628.

Picazo-Tadeo, A., González-Gómez, F., Sáez-Fernández, F., 2009b. Accounting foroperating environments in measuring water utilities’ managerial efficiency.Serv. Ind. J. 29, 761e773.

Renzetti, S., Dupont, D., 2009. Measuring the technical efficiency of municipal watersuppliers: the role of environmental factors. Land Econ. 85, 627e636.

Saal, D., Parker, D., 2000. The impact of privatization and regulation on the waterand sewerage industry in England and Wales: a translog cost function model.Managerial Decis. Econ. 21, 253e268.

Saal, D., Parker, D., 2004. The comparative impact of privatization and regulation onproductivity growth in the English and Welsh water and sewerage industry,1985e99. Int. J. Regul. Gov. 4, 139e170.

Saal, D., Parker, D., Weyman-Jones, T., 2007. Determining the contribution oftechnical, efficiency and scale change to productivity growth in the privatizedEnglish and Welsh water and sewerage industry: 1985e2000. J. Prod. Anal. 28,127e139.

Shih, J., Harrington, W., Pizer, W., Gillington, K., 2006. Economies of scale incommunity water systems. J. Am. Water Works Associ. 98, 100e108.

Shirley, M., 2006. Urban water reform: what we know, what we need to know. In:Paper Prepared for the Annual Meeting of the International Society for NewInstitutional Economics (ISINE) in Boulder, Colorado, September 21e24.

Simar, L., Wilson, P., 2007. Estimation and inference in two-stage, semi-parametricmodels of production processes. J. Econometrics 136, 31e64.

Souza, G., Faria, R., Moreira, T., 2007. Estimating the relative efficiency of Brazilianpublicly and privately owned water utilities: a stochastic cost frontier approach.J. Am. Water Resour. Associ. 43, 1237e1244.

Stone and Webster Consultants, 2004. Investigation into Evidence for Economies ofScale in the Water and Sewerage Industry in England and Wales. Final Report.Office of Water Services, Birmingham.

Tupper, H., Resende, M., 2004. Efficiency and regulatory issues in the Brazilianwaterand sewerage sector: an empirical study. Util. Policy 12, 29e40.

Urakami, T., 2007. Economies of vertical integration in the Japanese water supplyindustry. Jahrb. für Regionalwissenschaft 27, 129e141.

Witte, K., Marques, R., 2010. Incorporating heterogeneity in non-parametricmodels: a methodological comparison. Int. J. Oper. Res. 9, 188e204.

Woodbury, K., Dollery, B., 2004. Efficiency measurement in Australian localgovernment: the case of New South Wales municipal water utilities. Rev. PolicyRes. 21, 615e636.