Int. J.

Production Economics 166 (2015) 64–71

Contents lists available at ScienceDirect

Int. J. Production Economics

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

Measurement of input-specific productivity growth

with an application to the construction industry in Spain and Portugal
M. Kapelko a,n, I.M. Horta b, A.S. Camanho b, A. Oude Lansink c
a
Institute of Applied Mathematics, Department of Logistics, Wroclaw University of Economics, Poland
b
Department of Industrial Engineering and Management, Faculdade de Engenharia, Universidade do Porto, Portugal
c
Business Economics Group, Wageningen University, The Netherlands

art ic l e i nf o a b s t r a c t

Article history: Decision making in companies requires an assessment of the efficiency and productivity of individual
Received 25 June 2014 inputs to provide insights into the scope for improvement of inputs' use. This paper estimates an input-
Accepted 30 March 2015 specific Luenberger productivity growth indicator that can be decomposed to identify the contributions
Available online 6 April 2015
of input-specific technological change, technical efficiency change and scale efficiency change. These
Keywords: components for a specific input sum up to the aggregated indicators which are then compared with the
Input-specific productivity growth traditional Luenberger indicator. The application focuses on panel data of Spanish and Portuguese
Directional distance function construction firms over the period 2002–2011, accounting for three inputs: materials, labor and capital.
Data Envelopment Analysis The results show that aggregated productivity change and its components computed from the input-
Construction industry
specific productivity indicator are different from those obtained using a traditional approach. The results
also indicate that productivity change is negative for labor and capital for construction firms in both
Spain and Portugal, while productivity change of materials is positive for Portugal and negative for Spain.
Productivity decline is worse for capital in the Spanish construction firms, and for labor in Portugal.
& 2015 Elsevier B.V. All rights reserved.

1. Introduction
The measurement of firms' productivity growth has been
extensively studied in the literature. Productivity analysis over time
can provide valuable insights into the evolution of an industry and
its degree of competitiveness. It is a useful tool to support the
design of firm's strategies and government policies towards the
improvement of industry performance over time.
A frequently employed approach in the literature to evaluate
productivity change over time is the Malmquist productivity index,
introduced by Caves et al. (1982) and enhanced by Färe et al. (1992).
The Malmquist index is calculated using ratios of Shephard distance
functions and can adopt either an input contraction or an output
expansion perspective. For those cases requiring simultaneous
adjustments of inputs and outputs or non-radial expansions or
contractions, Chambers et al. (1996) proposed the use of directional
distance functions to evaluate productivity change over time using
the Luenberger productivity growth indicator.
More recently, Mahlberg and Sahoo (2011) developed an enha-
nced Luenberger indicator using directional slacks-based measu-
res, as proposed by Fukuyama and Weber (2009). This indicator is
based on non-radial distance functions, and has the advantage of
n
Corresponding author. Tel.: þ 48 713680479; fax: þ 48 713680334.
E-mail address: magdalena.kapelko@ue.wroc.pl (M. Kapelko).
accounting for the existences of slacks in a context where both input
contractions and output expansions are considered. As it allows the
estimation of productivity change associated to individual inputs, it
provides input-specific estimates of productivity change. This app-
roach has the advantage of providing insights into the contributions
of individual inputs to productivity change. As input-specific pro-
ductivity change measures are recent, only a few applications are
reported in the literature (e.g., Skevas and Oude Lansink, 2014, which
used a Luenberger indicator, and Oude Lansink and Ondersteijn,
2006, which used a Malmquist index). Nevertheless, input-specific
efficiency is a topic well-established in the literature, with studies
mostly applied to the agricultural sector (Oude Lansink and Silva,
2003; D’Haese et al., 2009; Oude Lansink et al., 2002; Oude Lansink
and Bezlepkin, 2003).
This paper aims at analyzing changes in productivity of indivi-
dual inputs within the Iberian construction industry over the past
decade (2002–2011). For this purpose, we develop an input-
specific Luenberger productivity growth indicator, estimated using
a Russell-type measure in the context of directional distance fun-
ctions, following Färe and Grosskopf (2010). This approach allows
removing the slacks in all inputs for the estimation of the pro-
jection to the efficient frontier. The use of a multiplicative direc-
tional distance function model, with a directional vector specified
as being equal to the input levels of the decision making unit
(DMU) under evaluation, has the advantage of allowing the inter-
pretation of the inefficiency component associated to each input as

http://dx.doi.org/10.1016/j.ijpe.2015.03.030
0925-5273/& 2015 Elsevier B.V. All rights reserved.
 M. Kapelko et al. / Int. J. Production Economics 166 (2015) 64–71 65

a proportional change to the original input levels. We propose a and apply it to the construction industry of two European countries.
decomposition of the efficiency change component of the input- Hence, our measure provides more information regarding the sources
specific Luenberger indicator to obtain further information con- of input-specific productivity change.
cerning the sources of input-specific productivity change. We start from a production technology where we assume that
To the best of our knowledge, this study is the first to apply the inputs xti (i¼1,…,m) are used to produce outputs ytr (r ¼1,…,s) in
input-specific productivity growth indicator in the context of the year t. Adapting Färe and Grosskopf (2010) slacks-based measure
construction industry. Previous research of construction industry of efficiency in the directional distance ,
function context, the dir-
assesses the productivity growth of all inputs simultaneously and ectional input distance function (D T t ) in year t seeks to reduce the
relates to the use of the Malmquist index to evaluate productivity use of inputs xti :
change in the construction firms in China (Xue et al., 2008) and ( )
Australia (Li and Liu, 2010). Using a growth accounting framework, ,
 X
m
D T t xti ; ytr ; g txi ¼ sup βi : ðxti  βi g txi ; ytr Þ A P t ð1Þ
Abdel-Wahab and Vogl (2011) compare productivity growth of the i¼1
construction sectors in Germany, France, UK, USA and Japan, while
Ruddock and Ruddock (2011) analyze construction industry in the where g txi represents the component of the directional vector
UK. The study developed by Horta et al. (2013) compares the determining the direction in which input xti can be scaled, βi mea-
productivity change of the construction industry in three world sures the degree of inefficiency at time t of input i, and Pt rep-
regions (i.e. Asia, North America and Europe) using the Malmquist resents the production technology in time t that transforms inputs
index. Given the limited attention devoted to productivity change into outputs. The production technology is defined as
in the construction industry, a study of input-specific productivity 
 
change in Iberian countries can contribute to the literature by Pt ¼ xti ; ytr : xti can produce ytr ð2Þ
widening the scope of knowledge concerning the evolution of this
sector. Expression (1) seeks for the largest feasible contraction of
The construction industry plays a central role in the economy of inputs. The Russell type of model represented by expression (1)
Iberian countries. The sector accounts for 10% and 7% of the Gross sums the input inefficiencies, thus accounting simultaneously for
Domestic Product in Spain and Portugal, respectively. As a conse- the contribution of all inputs. Therefore, the approach we use is a
quence of the 2008 global economic crisis, the Portuguese and Spanish Russell type of measure in the directional distance function
construction industries faced a period of downturn. This caused a context, which removes all slacks in the inputs. This approach
slowdown of construction industry activity and the bankruptcy of has some desirable properties that motivate its usage. First of all,
many construction firms. Comparing to Portuguese construction firms, the Russell type of measure represents a solution for the problem
Spanish companies were more severely impacted by the crisis as of nonzero slack in efficiency measurement using DEA. Therefore,
reflected by the higher decrease in the relative output and employ- it measures inefficiency taking into account all sources of ineffi-
ment of construction industry in this country. An in-depth evaluation ciency (including slacks). Also, it supports the notion of a Pareto–
of productivity change in the Iberian construction industry is particu- Koopmans inefficiency that equates the efficiency with belonging
larly important to support firms in the definition of successful to the efficient subset, on the contrary to the Debreu–Farrell
strategies in order to prosper in the long-run, boosting the Iberian measures that require that efficient observations belong to the
economy. isoquant (Ferrier et al., 1994). Moreover, it allows measuring the
The remainder of this paper is organized as follows. Section 2 inefficiency of particular inputs, which is not the case for the most
describes the methodology used in this paper. Section 3 presents widely used measure, i.e. the Debreu–Farrell measure.1
the empirical application, including the description of the data set To compute input-specific productivity, four linear program-
and the discussion of the results. The last section concludes and ming (LP) models have to be solved for two consecutive years
points topics for future research. using Data Envelopment Analysis (DEA): two single periods LP
models and two cross-period LP models. All models assume
constant returns to scale. The LP models used to evaluate a firm
2. Input-specific productivity growth in the directional 0 in a sample with j¼1,…,n firms are
distance function context !
,
 X
m
D T t xi ; yr ; g xi ¼ maxβ1 ;λ1
t t t
βi
1
i j
ð3Þ
Input-specific productivity growth has its origins in the notion of i¼1
sub-vector efficiency that dates back to the work of Färe et al. (1994a)
that estimated technical efficiency measures for a subset of inputs X
n
rather than for the entire vector of inputs. The approach to input- s:t: λ1j ytrj Z ytr0 ; r ¼ 1; …; s
specific productivity we develop in this paper is similar to the app- j¼1

roach used by Oude Lansink and Ondersteijn (2006), Mahlberg and

Sahoo (2011) and Skevas and Oude Lansink (2014). It is different from X
n
λ1j xtij r xti0  β1i gtxi i ¼ 1; …; m
Oude Lansink and Ondersteijn (2006) and Mahlberg and Sahoo (2011) j¼1
because our study applies a Russell-type measure in the directional
distance function context following Färe and Grosskopf (2010), while !
the study of Oude Lansink and Ondersteijn (2006) is based on a Russell ,
 X
m

measure with a DEA model in the context of radial measures of

DTt þ 1 xti ; ytr ; g txi ¼ maxβ2 ;λ2
i j
β 2
i ð4Þ
i¼1
technical efficiency, and the study of Mahlberg and Sahoo (2011) is
based on the slacks-based measure of Fukuyama and Weber (2009)
with a directional distance function model. Our approach also differs 1
In the literature there is a theoretical debate on which efficiency measure
from Skevas and Oude Lansink (2014) as their approach is designed to should be used. Some authors favor the Russell measure, whereas others criticize it.
account for undesirable factors and assumes weak disposability of Recently, Russell and Schworm (2011) posit four axioms that an ideal efficiency
measure should satisfy and compare different efficiency indicators including
some inputs. Our approach also differs from aforementioned works in Russell measure, analyzing whether they satisfy these properties. The conclusion
the sense that we develop an extended decomposition of input- is that none of the measures can satisfy all four conditions simultaneously and
specific productivity growth accounting for scale efficiency change, hence the trade-off exists in selecting among indexes.
66 M. Kapelko et al. / Int. J. Production Economics 166 (2015) 64–71

X
n Input-specific scale efficiency change (LSECi) is then computed
s:t: λ2j ytrjþ 1 Z ytr0 ; r ¼ 1; …; s as
j¼1

LSEC i ¼ ðβi  β i Þ  ðβi  βi

1 4 1 VRS 4 V RS
X
n Þ ð13Þ
λ2j xtijþ 1 r xti0  β2i g txi i ¼ 1; …; m
j¼1 The sum of these two components (LEC Vi RS
and LSEC i ) is equal
to the LEC i .
!
,
 X
m Following Mahlberg and Sahoo (2011) the aggregate indicators
D T t xti þ 1 ; ytr þ 1 ; g xi
t þ1
¼ maxβ3 ;λ3 β3i
i j
ð5Þ of productivity change and their components can be derived by
i¼1
summing up their respective input-specific indicators. Therefore,
X
n aggregate Luenberger productivity change (AL), aggregate techno-
s:t: λ3j ytrj Z ytr0þ 1 ; r ¼ 1; …; s logical change (ALTC), aggregate technical efficiency change under
j¼1 VRS (ALEC V RS ) and aggregate scale efficiency change (ALSEC) can
X
n be computed as, respectively
λ3j xtij rxti0þ 1  β3i gtxiþ 1 i ¼ 1; …; m
j¼1 X
m
AL ¼ ALi ð14Þ
! i¼1
,
 X
m
D T t þ 1 xti þ 1 ; ytr þ 1 ; g txiþ 1 ¼ maxβ4 ;λ4
i j
β4i ð6Þ X
m
i¼1 ALTC ¼ LTC i ð15Þ
i¼1
X
n
s:t: λ4j ytrjþ 1 Z ytr0þ 1 ; r ¼ 1; …; s X
m
j¼1 ALEC VRS ¼ LEC Vi RS ð16Þ
X
n i¼1
λ 4 t þ1
j xij r xti0þ 1  β 4 t þ1
i g xi i ¼ 1; …; m
j¼1 X
m
ALSEC ¼ LSEC i ð17Þ
In Eqs. (3)–(6), the computed values of βi provide the max- i¼1
imum contraction of each input if a firm has to operate efficiently Note that positive (negative) values of input-specific and agg-
given the directional vectors specified. A separate intensity vector regate indicators indicate improvements (declines) in respective
λj is calculated for each model, indicating the role that each firm j indicators.
(j¼ 1,…,n) plays in determining the peers for the firm 0 under Note that the aggregate measures of productivity change com-
assessment. Positive values of λj identify the firms that determine puted using expressions (14)–(17) are based on Russell-type mea-
the production frontier of the corresponding model. sures in a directional distance function context. Due to their non-
The Luenberger indicator of input-specific productivity growth radial nature, they can give different estimates of productivity
for input i (i¼1,…,m), (Li) is change than those obtained using the traditional input-oriented
1  2  Luenberger indicator, as it is computed using directional distance
Li ¼  β i  βi þ β i  βi
4 1 3
ð7Þ
2 functions associated with radial projections towards the frontier.
This indicator can be additively decomposed into two compo- These differences between the aggregated Luenberger indicators
nents: input-specific efficiency change (LECi) and input-specific developed in this paper and the traditional Luenberger indicator
technological change (LTCi), as shown in can occur even when the directional vector used in the estimation
of the distance functions is identical.
LEC i ¼ β i  β i
1 4
ð8Þ

1 3. Empirical application
 ðβ i  β i þ β i  β i Þ
4 3 2 1
LTC i ¼ ð9Þ
2
This decomposition is similar to the one proposed by Färe et al. 3.1. Data
(1992) in the context of the Malmquist index.
Input-specific efficiency change LECi can be further decom- The data on construction firms in Spain and Portugal come from
posed to identify the contribution of scale efficiency change. We the SABI database (Bureau Van Dijk), which offers the longitudinal
propose a decomposition similar to the one proposed by Färe et al. data from Spanish and Portuguese firms' financial statements as well
(1994b) in the context of the Malmquist index. Input-specific scale as general information (such as firms' ownership structure, year of
efficiency change can be estimated by running the two single establishment or international activity). The sample population
period LP models, corresponding to model (3) with the addition of comprises firms for which the main activity is construction of res-
restriction (10), and model (6) with the addition of restriction (11): idential and non-residential buildings (NACE Rev. 2 code 412); this
guarantees the homogeneity of the sample. Along with the specia-
X
n
λ1j ¼ 1 ð10Þ lized construction activity, construction of residential and non-
j¼1 residential buildings is the most important subsector of Spanish
and Portuguese construction industry considering the number of
X
n
companies, turnover and number of employees (Eurostat, 2014).
λ4j ¼ 1 ð11Þ
Furthermore, this study excludes microfirms and focuses on small,
j¼1
medium and large firms, based on the European Union (EU) defini-
This will yield new solutions under variable returns to scale tion of these categories of firm's size.2 Excluding microfirms results in
(VRS) technology that we denote as β i and β i
1 V RS 4 VRS
. The input-
specific technical efficiency change under VRS (LEC VRS i ) is then 2
The EU definition of firms' size is based on two variables: number of
computed as
employees and operating revenues (see European Commission (2003)). Because
¼ βi  βi
1 V RS 4 VRS the data on the number of employees was not available for the majority of
LEC VRS
i ð12Þ
Portuguese firms, we decided to use operating revenues only. Therefore, the firms
 M. Kapelko et al. / Int. J. Production Economics 166 (2015) 64–71 67

Table 1
Descriptive statistics of input and output variables per size, 2002–2011, thousand euros, constant 2002 prices.

Variable Small Medium Large

Mean St. dev. Mean St. dev. Mean St. dev.

Spain
Materials 2234.353 1476.515 11,146.780 6707.975 125,191.200 250,143.600
Labor 709.118 531.616 2249.005 1777.936 23,257.980 47,247.560
Capital 1081.884 5906.033 6241.538 30,134.790 57,970.290 173,997.700
Operating revenues 3597.801 1651.689 16,823.100 8301.395 186,346.400 363,062.600
Portugal
Materials 1257.442 978.036 4678.664 3535.903 23,263.100 24,667.070
Labor 453.265 410.838 1995.038 2212.723 13,531.820 16,149.010
Capital 941.663 2886.061 4072.391 6830.964 24,340.890 33,511.860
Operating revenues 3669.988 1686.993 17,753.020 8987.404 113,001.400 104,349.800

a dataset with companies that are more comparable in terms of size.
Finally, firms with missing information, outliers3 and firms that left
the industry during the period 2002–2011 were excluded from the
sample. This enables us to characterize more precisely the evolution
of productivity for the active firms operating in Portugal and Spain.
The final dataset consists of 5706 and 965 construction firms that
operated at least two consecutive years over the 2002–2011 period in
Spain and Portugal, respectively. That amounts to almost 50% and
more than 20% of the entire populations of small, medium and large
firms engaged in the construction of residential and nonresidential
buildings in Spain and Portugal, respectively. In total the dataset
sums up to 22,384 observations of Spanish companies and 3583
observations of Portuguese companies (unbalanced panels).
Following previous efficiency studies of construction industry
(You and Zi, 2007; Kapelko and Oude Lansink, 2015), three inputs
– capital, labor and materials, and one output – operating revenues
were used in our analysis. Capital is proxied by the book value of
fixed assets and is directly taken from the firms' balance sheet
statement in SABI. It is measured in thousand euros at constant
prices of the year 2002 that are obtained by deflating the fixed assets
using the industrial price index for capital goods. Labor is measured
as employee costs and is directly taken from the profit and loss
account in SABI. This variable is measured in thousand euros and is
deflated using the price index of labor costs in construction industry
to obtain constant 2002 prices. Materials are represented by material
costs and are also directly obtained from the profit and loss account
in SABI. This variable is converted into constant 2002 prices by
applying the price indexes of materials of residential buildings. Ope-
rating revenues represent a profit and loss account item from SABI
and include total sales and other operating revenues. They are
deflated using the price index of residential buildings (2002¼ 100).
All price indices were taken from the Spanish and Portuguese sta-
tistical offices (various years). Table 1 presents the descriptive
statistics of the data used in this study for the samples of Spanish
and Portuguese construction firms for the whole period 2002–2011
for different size categories (small, medium and large).
Table 1 shows that there are considerable differences between
firms in both the Spanish and Portuguese samples, as indicated by
the relatively high values of standard deviations of input and output
(footnote continued)
that have an annual turnover not exceeding 2/10/50 million euros are defined as
micro/small/medium firms, respectively, while firms with an annual turnover
exceeding 50 million euros are defined as large.
3 4
Outliers were determined using the basic threshold statistical rule adapted to
efficiency analysis by focusing on the ratios of output to input. An observation was
defined as an outlier if the ratio of output over any of the inputs was outside the
interval of the median plus and minus two standard deviations. Alternative
methods of outliers' detection are described, for example, in Wilson (1993) or
Simar (2003).
variables relative to their respective means. Furthermore, there are
also considerable differences between the size categories in Portugal
and Spain. Across all size categories, Spanish firms in the sample in
comparison to Portuguese counterparts have, on average, higher
values of the inputs relative to output. These differences are espe-
cially considerable for material costs relative to operating revenues.
Regarding the differences between sizes, for Portugal the average
values of inputs relative to output tend to decrease with firm size,
with the exception of employee costs relative to operating revenues.
This ratio decreases from the small to medium size category, and
then slightly increases for firms in the large size category. In contrast,
for the Spanish sample, employee costs relative to operating reven-
ues decreases with firm size, material costs in relation to output
increases with size, while fixed assets relative to output first incr-
eases when moving from small to the medium size category, and
then decreases for large firms.
3.2. Results
Input-specific productivity growth indicators and their decom-
position were computed using the General Algebraic Modelling
System (GAMS). The calculations were undertaken for firms
between two consecutive years for Spanish and Portuguese sam-
ples separately. Note that the directional vectors for inputs at time
t (g txi ) and at time tþ 1 (g txiþ 1 ) applied in this study are the actual
input levels of the DMUs under assessment for time t and time
tþ1, respectively.4
It is well known that the mixed period directional distance
functions used in the computation of the Luenberger indicator
may yield infeasibilities, which frequency depends on the data
structure, specification of technology and the choice of directional
vector (Briec and Kerstens, 2009). The same remarks apply in the
computation of the input-specific Luenberger productivity indica-
tor. It is worth pointing out that the problem of infeasibility
pertains to almost all well-known productivity indicators. Briec
and Kerstens (2009) show that the infeasibilities cannot be
avoided for both nonparametric and parametric estimation meth-
ods and there are no easy solutions to this problem. They conclude
that the property of well-determinateness in index theory may
have to be abandoned. Recently, Aparicio et al. (2013) proposed an
approach to overcome the occurrence of infeasibilities in
In the literature there are no clear indications which values of directional
vector should be used and there are many different views on this matter. In this
paper we follow Russell and Schworm (2011) recommendation that restricts
directional vector to be positive as well as to have the same units as those of the
production vector. Note that the directional vector used in this paper gives a
proportional interpretation to input reductions.
68 M. Kapelko et al. / Int. J. Production Economics 166 (2015) 64–71

productivity change assessments. We do not use this approach as
it would prevent detecting deterioration of productivity levels due
to movements of the frontier to the interior of the production
possibility set over time. The infeasible observations of the mixed
period directional distance functions found in our calculations
accounted for 1296 observations for Spanish firms (5.8% of the
initial sample) and 320 observations for Portuguese firms (8.9% of
the initial sample).5 Our treatment of infeasibilities consisted of
removing them from the computation of averages of the input-
specific productivity growth indicators and their decompositions.
We compute the input-specific and aggregate Luenberger ind-
icator and their respective decompositions following the approach
developed in this paper and compare it with traditional aggre-
gated Luenberger computed following Chambers et al. (1996). The
statistical significance of the differences in input-specific and agg-
regate measures between Spain and Portugal are tested using the
adapted Li test proposed by Simar and Zelenyuk (2006) (deno-
ted further as S–Z test). This test adapts the nonparametric test of
the equality of two densities developed by Li (1996). Simar and
Zelenyuk (2006) propose the algorithm which is based on the
computation and bootstrapping the Li statistic using DEA mea-
sures, where values equal to unity (truncation of DEA measures)
are smoothed by adding a small noise. In our implementation of
this algorithm, the step of smoothing is not undertaken because
productivity change and its components are not truncated. The
test algorithm is run in R, using 1000 bootstrap replications.
Table 2 reports the decomposition of input-specific Luenberger
indicator (Li ) into input-specific technological change (LTC i ), techni-
cal efficiency change under VRS (LEC Vi RS ), and scale efficiency change
(LSEC i ) with regard to materials, labor and capital during the period
2002–2011 for Spanish and Portuguese construction companies
(arithmetic averages). Table 2 also shows the aggregate productivity
change (AL) and its decomposition into aggregate technological
change (ATC), aggregate technical efficiency change under VRS
(ALEC V RS ), and aggregate scale efficiency change (ALSEC). In addition,
the table reports the S–Z statistics. For all measures, the distributions
significantly differ between construction firms in Spain and Portugal
as indicated by the results of S–Z test (p-values equal to 0.000 in all
cases). The detailed results for input-specific and aggregate measures
by pairs of consecutive years for Spain and Portugal are presented in
Appendix Tables A1 and A2.
Table 2 shows that for Spanish construction firms in the sample
productivity change is negative for all inputs in the period 2002–
2011, and equal to  0.1% for materials,  0.4% for labor and  0.5%
for capital. The key factor driving the process of productivity decline
for materials and labor is technical efficiency decrease ( 6.8% for
materials and  4.1% for labor). The productivity decline of capital is
mostly driven by the negative technological change of  2.8%, which
suggests that the technology of capital used by Spanish construction
capital). This result suggests that Spanish construction firms have
succeeded on average in moving to a more optimal scale of using all
inputs, i.e. the scale that is consistent with constant returns to scale.
Portuguese construction firms experienced a negative productiv-
ity growth both for labor (0.8%) and capital ( 0.1%) during the
period under investigation. The productivity decline of labor is driven
by the negative contributions of efficiency change ( 3.1%) and
technological change ( 0.2%). These negative contributions were
undoing the positive contribution of scale efficiency change (2.6%).
The productivity decline of capital is entirely driven by negative
technical efficiency change ( 2%); technological change and scale
efficiency change contributed positively (by 1.2% and 0.8%, respec-
tively). Regarding materials, the results show that productivity grows
by 0.3% on average in the analyzed period, with technological
progress of 2.7% and positive scale efficiency change of 2.5% being
the driving factors. Technical efficiency change is found to be
negative similarly to other inputs ( 5%).
A closer look at the results of input-specific indicators reveals
that the lowest productivity growth is found for capital in Spanish
construction firms, and for labor in the Portuguese counterparts. In
construction firms in both countries, the same patterns are found
for the contributions of technical efficiency change and scale
efficiency change to productivity growth: all inputs analyzed
exhibit a negative impact with regard to technical efficiency
change indicator and a positive for scale efficiency change indi-
cator. The contribution of input-specific technological change to
productivity growth differs between countries and is negative for
capital in Spain and for employees in Portugal, while it is positive
for the other inputs in both countries. For Spanish construction
firms capital is the worst performing input (three out of four
indicators are negative), while labor is the worst performing input
in Portugal. These differences found between Spain and Portugal
can be explained in the following way. First, both Spanish and
Portuguese labor markets have traditionally been highly protected.
Labor productivity growth is lower in Portugal than in Spain,
because in recent years the Spanish labor market has gradually
liberalized after the economic crisis that hit this country. The labor
market reforms introduced in Spain enable firms to lay off
employees more easily or reduce salaries in times of financial
Table 2
Decomposition of input-specific and aggregate Luenberger indicator in construc-
tion industries in Spain and Portugal, 2002–2011.
Indicator Spain Portugal S–Z statistics
Materials
Lmaterials  0.001 0.003 109.893*
LTC materials 0.008 0.027 693.370*
LEC VRS  0.068  0.050 77.758*
materials

firms in the sample worsens in the period under investigation. LSEC materials 0.058 0.025 82.206*
Labor
However, technical efficiency change of capital also contributes Llabor  0.004  0.008 49.145*
negatively to productivity change of this input. The negative con- LTC labor 0.026  0.002 712.186*
tribution of technical efficiency change of all inputs indicates that LEC VRS  0.041  0.031  3.430*
labor
construction firms' efficiency in using materials, labor and capital has LSEC labor 0.011 0.026  5.470*
declined during the sample period, i.e. the gap between efficient and Capital
Lcapital  0.005  0.001  5.956*
inefficient firms with regard to all inputs has grown on average. On
LTC capital  0.028 0.012 612.966*
the other hand, the positive technological change for materials (0.8%)  0.009  0.020 23.873*
LEC VRS
capital
and labor (2.6%) suggests technological advances in these inputs over
LSEC capital 0.032 0.008 20.371*
the period under analysis in Spain. Furthermore, the results show Aggregate
that scale efficiency change contributes positively to productivity AL  0.010  0.006 79.933*
change of all inputs (5.8% for materials, 1.1% for labor and 3.2% for ALTC 0.006 0.037 625.714*
ALEC V RS  0.118  0.101 5.019*
ALSEC 0.101 0.059 22.990*
5
After removing the infeasible observations, the sample size we are left with is No obs. 21,088 3263
still quite large, representing the considerable fraction of the entire populations of
n
construction firms in both Spain and Portugal. Hence, our results can still be Denotes significant differences between Spain and Portugal at the critical
generalized. 1% level.
 M. Kapelko et al. / Int. J. Production Economics 166 (2015) 64–71
distress (Noblet and Regojo, 2012). On the contrary, Portugal still
maintains its high level of protection of labor market. This may
show up as a lower labor productivity decline of Spanish con-
struction firms relative to Portugal. It can also explain why
technological change with regard to labor is negative in the
Portuguese construction industry, while it is positive for Spanish
construction firms. On the other hand, the larger capital produc-
tivity decline in the construction industry in Spain as compared to
Portugal suggests a higher capacity underutilization on Spanish
firms, i.e. existing buildings and machinery are more heavily
underutilized in the Spanish construction industry. This can be
due to the fact that the Spanish construction firms were more
severely affected by the economic crisis compared to the Portu-
guese construction industry. The higher capacity underutilization
of Spanish construction firms can also explain why technological
change with regard to capital was negative in Spain, while being
positive for the Portuguese construction counterparts. Finally, we
can observe that the productivity of materials improved in
Portugal, whereas it declined in Spain. This suggests that Portu-
guese companies managed to use and acquire raw materials and
components for construction effectively. The more severe impact
of the economic crisis on the Spanish construction firms can also
explain why productivity change of materials is negative in the
Spanish, and positive in the Portuguese construction industry.
The results of aggregate productivity change and its compo-
nents in construction industry in both Spain and Portugal indicate
that over the 2002–2011 period, productivity change has
decreased (by 1% and 0.6%, respectively). The productivity decline
was mainly due to a negative contribution of technical efficiency
change (  11.8% and 10.1%, respectively), which was undoing
improvements in technological change (0.6% and 3.7%, respec-
tively) and scale efficiency change (10.1% and 5.9%). These patterns
suggest the technological advances, scale improvements and a
growth of the gap between inefficient and efficient Spanish and
Portuguese construction firms.
Table 3 summarizes the results of traditional Luenberger
indicator (L) and its decomposition into technological change
(LTC), technical efficiency change under VRS (LEC V RS ), and scale
efficiency change (LSEC).
As one can expect the results shown in Table 3 are slightly
different to these reported in Table 2, as those measures remove all
slacks in inputs. Although the patterns of the behavior of the
Luenberger indicators and their components observed in both
approaches are exactly the same, the average values of the
traditional Luenberger indicator for Spanish and Portuguese firms
are slightly different from the values of the aggregate Luenberger
indicator adopted in our paper. It might suggest the presence of
slacks in one of the time periods when estimating the productivity
growth and its components. Hence, the results might suggest the
importance of accounting for slacks when measuring productivity
growth and taking into account all sources of inefficiency. More-
over, when comparing the results in Tables 2 and 3, one can notice
that in general the aggregate Luenberger and its components
computed in the traditional way tend to be systematically lower
Table 3
Decomposition of Luenberger indicator (traditional approach) in construction
industries in Spain and Portugal, 2002–2011.
Indicator Spain Portugal S–Z statistics
L  0.004  0.003 36.589*
LTC 0.010 0.008 291.726*
LEC VRS  0.047  0.023 33.151*
LSEC 0.032 0.012 43.846*
n
Denotes significant differences between Spain and Portugal at the critical
1% level.
69
than aggregated measures developed in this paper in both
Portuguese and Spanish samples. Hence, it might indicate that
some inefficiencies related with slacks are presented in the
samples. The differences found between the two indicators may
also reflect the fact that the efficient projection provided by the
model developed in this paper is different from the projection
provided by the traditional Luenberger indicator due to nonradial
versus radial nature of projection.
4. Conclusions
This paper contributes to the literature by developing an input-
specific Luenberger indicator, estimated using a Russell-type
measure in the context of directional distance functions. It
proposed an extended decomposition of input-specific productiv-
ity change taking into account input-specific technological change,
technical efficiency change and scale efficiency change. This
decomposition provides further information concerning the
sources of productivity change.
The paper illustrates the method in the context of the Iberian
construction industry. In particular, a large dataset of Spanish and
Portuguese construction firms was analyzed over the period
2002–2011. The results of the empirical application show that
aggregated productivity change and its components computed
from the approach developed in this paper are different from
those obtained from traditional indicators, which might suggest
the presence of slacks. The results further indicate that aggregate
productivity declined in the period under investigation in both
countries. This decline was mainly driven by the decline of input-
specific productivity of labor and capital. In particular, productivity
decline for capital was the major driving factor behind aggregate
productivity decline in Spain, whereas in Portugal the greatest
decline was associated with labor. This finding can indicate that in
the Spanish construction firms the capacity is highly underutilized
as shown by the decline in capital productivity change. On the
other hand, the labor market reforms in Spain may have con-
tributed to the better performance in labor productivity change in
Spain vis-a-vis Portugal. The productivity decline for labor (Portu-
gal) and capital (Spain) was mainly due to a considerable decrease
in technical efficiency; scale efficiency improved in both countries
during this period. Technological change for labor improved in
Spain and declined in Portugal, whereas the converse occurred for
capital. This means that technological change of capital improved
in Portugal, whereas Spain was more successful in the usage of
human resources.
The results of this study provide insights that may be helpful to
managers of firms, policy makers and industry analysts in Spain
and Portugal in defining strategies and policy measures aimed at
improving the performance of construction firms' specific inputs.
The results imply that Spanish construction firms should reorga-
nize to focus on the improvement of capital utilization, whereas
Portuguese firms need to enhance the utilization of human
resources. Spanish construction firms can improve capital utiliza-
tion by reducing the size of the fixed costs relative to variable
costs. A more flexible labor market in Portugal could help
Portuguese construction firms in improving their utilization of
labor input. In the light of findings of this paper, the policy makers
in Spain could establish policies that decrease the underutilization
of capital of construction firms, while in Portugal they could focus
on labor market reforms.
For future research, from the empirical point of view, it would
be interesting to analyze convergence in productivity levels across
the two countries. It would also be interesting to explore input-
specific productivity change in the construction industry of other
countries or regions in order to depict worldwide patterns of
70 M. Kapelko et al. / Int. J. Production Economics 166 (2015) 64–71

productivity change. Future research could also investigate of residential and nonresidential buildings, one can expect the
whether the results are robust to a longer time horizon, especially results to be different when analyzing the other sectors of
for the period after the start of the crisis. Methodologically, construction such as civil engineering or specialized construction
because one of the main drawbacks of the DEA model applied in activities.
this study is that efficiency and productivity results may be
affected by the sample variation, implying that distances to the
frontier are likely to be underestimated, future study can account
for this by developing the bootstrap approach for the measures
Acknowledgments
estimated in this study following the proposal of Simar et al.
(2012). Finally, a promising avenue for future research would be to
The calculations of adapted Li test have been carried out in the
account for dynamics in the production process by developing
Wroclaw Centre for Networking and Supercomputing (www.wcss.
dynamic input-specific productivity growth measures drawing
wroc.pl), Grant no. 286. The first author acknowledges the visiting
from the work of Silva and Stefanou (2003), Silva and Oude
research grant from Wageningen School of Social Sciences at
Lansink (2013) and Kapelko et al. (2014).
Wageningen University. The funding of this research through a
Given the considerable sample sizes used in this study, which
scholarship SFRH/BPD/86294/2012 from the Portuguese Founda-
even after removing the infeasible observations account for a large
tion of Science and Technology (FCT) is gratefully acknowledged by
fraction of the entire populations of Spanish and Portuguese firms
the second author.
engaged in the construction of residential and nonresidential
buildings, we think that our results can be broadly generalized
to the population of these firms. However, the present empirical
application focuses on small, medium and large construction
firms, excluding microfirms. Hence, our results cannot be directly Appendix A
generalized to microfirms. Also, given that we focus on one
activity within the construction industry related with construction See Table A1 and Table A2.

Table A.1
Evolution of input-specific and aggregate Luenberger indicator in construction industry in Spain, for consecutive years in the period from 2002 to 2011.

Indicator 2002/2003 2003/2004 2004/2005 2005/2006 2006/2007 2007/2008 2008/2009 2009/2010 2010/2011

Materials
Lmaterials 0.001  0.010 0.004  0.005  0.001 0.004 0.003 0.002  0.009
LTC materials 0.113 0.147  0.252 0.088  0.029 0.001 0.048 0.007  0.047
LEC VRS 0.025  0.118 0.195  0.081 0.029  0.206  0.341  0.076  0.010
materials
LSEC materials  0.137  0.039 0.061  0.012  0.001 0.209 0.296 0.071 0.048
Labor
Llabor 0.005  0.006 0.007 0.001  0.005  0.008  0.009  0.003  0.020
LTC labor 0.187 0.179  0.063  0.004 0.035  0.037  0.041 0.004 0.041
LEC VRS  0.074  0.165 0.094  0.006  0.090 0.008  0.067 0.010  0.103
labor
LSEC labor  0.108  0.020  0.024 0.011 0.051 0.022 0.098  0.018 0.043
Capital
Lcapital  0.005  0.005  0.003 0.002  0.002  0.012  0.012  0.007  0.010
LTC capital  0.031  0.062  0.023  0.056  0.060 0.005 0.035  0.008  0.032
LEC VRS 0.033 0.021 0.053 0.025 0.003  0.064  0.095  0.044  0.024
capital
LSEC capital  0.008 0.036  0.033 0.033 0.055 0.047 0.048 0.045 0.046
Aggregate
AL 0.001  0.020 0.008  0.002  0.007  0.016  0.018  0.009  0.039
ALTC 0.269 0.264  0.339 0.028  0.055  0.032 0.042 0.003  0.039
ALEC VRS  0.016  0.262 0.342  0.061  0.058  0.262  0.503  0.110  0.136
ALSEC  0.253  0.023 0.004 0.032 0.106 0.278 0.442 0.098 0.136
No obs. 1815 2079 2213 3052 3560 2958 2204 1825 1382

Table A.2
Evolution of input-specific and aggregate Luenberger indicator in construction industry in Portugal, for consecutive years in the period from 2002 to 2011.

Indicator 2002/2003 2003/2004 2004/2005 2005/2006 2006/2007 2007/2008 2008/2009 2009/2010 2010/2011

Materials
Lmaterials  0.001  0.001 0.009  0.005 0.001 0.001 0.013  0.001 0.007
LTC materials  0.042 0.056  0.102 0.041 0.117 0.033 0.037 0.077  0.085
LEC VRS  0.032  0.191 0.036  0.078  0.136  0.029 0.066  0.099 0.027
materials
LSEC materials 0.074 0.134 0.074 0.032 0.020  0.004  0.090 0.021 0.065
Labor
Llabor  0.010 0.007 0.012  0.008  0.007  0.018  0.005  0.018  0.012
LTC labor  0.010 0.026  0.058 0.069 0.097  0.009  0.191 0.139  0.170
LEC VRS  0.010  0.087 0.012  0.080  0.042  0.040 0.112  0.174 0.055
labor
LSEC labor 0.011 0.067 0.058 0.002  0.062 0.030 0.073 0.017 0.103
Capital
Lcapital  0.005  0.001 0.014 0.004  0.004  0.009  0.001  0.009 0.007

Note: Computing the means over the periods, one will not obtain the same results as means in Table 2 due to the rounding of numbers to three decimals in both tables.
 M. Kapelko et al. / Int. J. Production Economics 166 (2015) 64–71 71

Table A.2 (continued )

Indicator 2002/2003 2003/2004 2004/2005 2005/2006 2006/2007 2007/2008 2008/2009 2009/2010 2010/2011

LTC capital  0.156 0.041 0.032 0.011 0.011  0.073 0.044 0.064 0.101
LEC VRS 0.103 0.024 0.024  0.061  0.046 0.017 0.024  0.100  0.075
capital
LSEC capital 0.048  0.067  0.042 0.055 0.031 0.048  0.070 0.026  0.019
Aggregate
AL  0.016 0.005 0.034  0.009  0.009  0.026 0.007  0.027 0.002
ALTC  0.209 0.123  0.128 0.120 0.226  0.049  0.109 0.281  0.153
ALEC VRS 0.061  0.253 0.072  0.219  0.223  0.052 0.202  0.373 0.007
ALSEC 0.133 0.134 0.090 0.089  0.011 0.074  0.087 0.065 0.148
No obs. 191 226 294 515 508 461 395 371 302

Note: Computing the means over the periods, one will not obtain the same results as means in Table 2 due to the rounding of numbers to three decimals in both tables.

References
Aparicio, J., Pastor, J.T., Zofio, J.L., 2013. On the inconsistency of the Malmquist–
Luenberger index. Eur. J. Oper. Res. 229 (3), 738–742.
Abdel-Wahab, M., Vogl, B., 2011. Trends of productivity growth in the construction
industry across Europe, US and Japan. Constr. Manag. Econ. 29, 635–644.
Briec, W., Kerstens, K., 2009. The Luenberger productivity indicator: an economic
specification leading to infeasibilities. Econ. Model. 26 (3), 597–600.
Caves, D., Christensen, L., Diewert, W., 1982. The economic theory of index numbers
and the measurement of input, output and productivity. Econometrica 50 (6),
1393–1414.
Chambers, R.G., Chung, Y.H., Färe, R., 1996. Benefit and distance functions. J. Econ.
Theory 70, 407–419.
D’Haese, M., Speelman, S., Alary, V., Tillard, E., D’Haese, L., 2009. Efficiency in milk
production on Reunion Island: dealing with land scarcity. J. Dairy Sci. 92 (8),
3676–3683.
European Commission2003. Recommendation concerning the definition of micro,
small and medium-sized enterprises. Official Journal of the European Union.
〈http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=OJ:
L:2003:124:0036:0041:EN:PDF〉.
Eurostat 2014. Structural business statistics – industry and construction.
Ferrier, G.D., Kerstens, K., Eeckaut, Vanden, 1994. Radial and nonradial technical
efficiency measures on a DEA reference technology: a comparison using US
banking data Core Discussion Papers 9423. Louvain Econ. Rev. 60 (4), 449–479.
Färe, R., Grosskopf, S., 2010. Directional distance functions and slacks-based
measures of efficiency. Eur. J. Oper. Res. 200 (1), 320–322.
Färe, R., Grosskopf, S., Lovell, C.A.K., 1994a. Production Frontiers. Cambridge
University Press, Cambridge.
Färe, R., Grosskopf, S., Norris, M., Zhang, Z., 1994b. Productivity growth, technical
progress and efficiency change in industrialized countries. Am. Econ. Rev. 84
(1), 66–83.
Färe, R., Grosskopf, S., Lindgren, B., Roos, P., 1992. Productivity changes in Swedish
pharmacies 1980–1989: a non-parametric Malmquist approach. J. Prod. Anal. 3
(1–2), 85–101.
Fukuyama, H., Weber, W.L., 2009. A directional slacks-based measure of technical
inefficiency. Socio-Econ. Plan. Sci. 43 (4), 274–287.
Horta, I.M., Camanho, A.S., Johnes, J., Johnes, G., 2013. Performance trends in the
construction industry worldwide: an overview of the turn of the century.
J. Prod. Anal. 39 (1), 89–99.
Kapelko, M., Oude Lansink, A., Stefanou, S., 2014. Assessing dynamic inefficiency of
the Spanish construction sector pre- and post-financial crisis. Eur. J. Oper. Res.
237 (1), 349–357.
Kapelko, M., Oude Lansink, A., 2015. Technical efficiency and its determinants in the
Spanish construction sector pre- and post-financial crisis. Int. J. Strateg. Prop.
Manag. 19 (1), 96–109.
Li, Q., 1996. Nonparametric testing of closeness between two unknown distribution
functions. Econom. Rev. 15, 261–274.
Li, Y., Liu, C., 2010. Malmquist indices of total factor productivity changes in the
Australian construction industry. Constr. Manag. Econ. 28 (9), 933–945.
Mahlberg, B., Sahoo, B.K., 2011. Radial and non-radial decomposition of Luenberger
productivity indicator with an illustrative application. Int. J. Prod. Econ. 131,
721–726.
Noblet, C. and Regojo, I.2012. Spain introduces radical labor market reforms. Squiresa
〈http://www.squiresanders.com/files/Event/bddeba0e-a202-49c4-9c58-d557e90589
3d/Presentation/EventAttachment/52ea08cb-d7a2-43c1-8dfe-ad02ac35b882/Spai
n-Introduces-Radical-Labour-Market-Reforms-Newsletter.pdf〉.
Oude Lansink, A., Bezlepkin, I., 2003. The effect of heating technologies on CO2 and
energy efficiency of Dutch greenhouse firms. J. Environ. Manag. 68 (1), 73–82.
Oude Lansink, A., Ondersteijn, Ch., 2006. Energy productivity growth in the Dutch
greenhouse industry. Am. J. Agric. Econ. 88 (1), 124–132.
Oude Lansink, A., Silva, E., 2003. CO2 and energy efficiency of different heating
technologies in the Dutch glasshouse industry. Environ. Resour. Econ. 24 (4),
395–407.
Oude Lansink, A., Pietola, K., Bäckman, S., 2002. Efficiency and productivity of
conventional and organic farms in Finland 1994–1997. Eur. Rev. Agric. Econ. 29
(1), 51–65.
Ruddock, L., Ruddock, S., 2011. Evaluation of trends in the UK construction industry
using growth and productivity accounts. Constr. Manag. Econ. 29, 1229–1239.
Russell, R.R., Schworm, W., 2011. Properties of inefficiency indexes on oinput,
output4 space. J. Prod. Anal. 36, 143–156.
Silva, E., Stefanou, S.E., 2003. Nonparametric dynamic production analysis and the
theory of cost. J. Prod. Anal. 19 (1), 5–32.
Silva, E. and Oude Lansink, A.2013. Dynamic efficiency measurement: a directional
distance function approach. Centro de Economia e Finanças da UPorto, cef.up
Working Paper 2013-07. 〈http://cefup.fep.up.pt/uploads/WorkingPapers/
wp1307.pdf〉.
Simar, L., 2003. Detecting outliers in frontier models: a simple approach. J. Prod.
Anal. 20 (3), 391–424.
Simar, L., Zelenyuk, V., 2006. On testing equality of distributions of technical
efficiency scores. Econom. Rev. 25 (4), 497–522.
Simar, L., Vanhems, A., Wilson, P.W., 2012. Statistical inference for DEA estimators
of directional distances. Eur. J. Oper. Res. 220, 853–864.
Skevas, T., Oude Lansink, A., 2014. Reducing pesticide use and pesticide impact by
productivity growth: the case of Dutch arable farming. Am. J. Agric. Econ. 65
(1), 191–211.
Wilson, P.W., 1993. Detecting outliers in deterministic nonparametric frontier
models with multiple outputs. J. Bus. Econ. Stat. 11 (3), 319–323.
Xue, X., Shen, Q., Wang, Y., Lu, J., 2008. Measuring the productivity of the
construction industry in China by using DEA-based Malmquist productivity
indices. J. Constr. Eng. Manag. 134 (1), 64–71.
You, T., Zi, H., 2007. The economic crisis and efficiency change: evidence from the
Korean construction industry. Appl. Econ. 39, 1833–1842.