Universitat Politècnica de Catalunya

Institut de Sostenibilitat
at
Politècni
ca de
Catalun Doctoral Thesis:
ya
“The productive efficiency in agriculture:
recent methodological advances”

By

BOUALI GUESMI

Under the supervision of

Thesis director: Dr. Teresa Serra Devesa

Tutor: Professor José María Gil

PhD Program: Sustainability

Main Subject: Agricultural Economics

Barcelona, November 2013

 Table of contents

Table of contents ........................................................................................................................ ii

List of Tables & Figures............................................................................................................ iii
Abstract ..................................................................................................................................... iv
Resumen ..................................................................................................................................... v
Dedication ................................................................................................................................. vi
Acknowledgments .................................................................................................................... vii
Chapter 1: Introduction .............................................................................................................. 1
References .................................................................................................................................. 7
Chapter 2: The productive effciency of organic farming: the case of grape sector in Catalonia9
2.1. Introduction ..................................................................................................................... 11
2.2. Material and methods...................................................................................................... 14
2.3. Results ............................................................................................................................. 16
2.3.1. Characteristics of farms and farmers ......................................................................... 16
2.3.2. Model specification and research results ................................................................... 19
2.4. Discussion and concluding remarks ............................................................................... 23
References ................................................................................................................................ 27
Chapter 3: Technical effciency of Kansas arable crop farms: a local maximum likelihood
approach ................................................................................................................................... 41
3.1. Introduction ..................................................................................................................... 43
3.2. Methodology ................................................................................................................... 45
3.3. Data and results ............................................................................................................... 51
3.3.1. Data ............................................................................................................................ 51
3.3.2. Empirical results ........................................................................................................ 52
3.4. Concluding remarks ........................................................................................................ 57
References ................................................................................................................................ 59
Chapter 4: Technical and environmental effciency of Catalan arable crop farms ................... 73
4.1. Introduction ..................................................................................................................... 75
4.2. Literature review ............................................................................................................. 77
4.3. Methodology ................................................................................................................... 80
4.4. Data and Results ............................................................................................................. 85
4.4.1. Data ............................................................................................................................ 85
4.4.2. Empirical results ........................................................................................................ 87
4.5. Concluding remarks ........................................................................................................ 89
References ................................................................................................................................ 91
Chapter 5: Conclusions ............................................................................................................ 99
References .............................................................................................................................. 105

ii
 List of Tables & Figures

Table 2.‎1.‎Sample‎farms’‎agronomic,‎economic‎and‎demographic‎characteristics ................. 31

Table 2.2. Summary statistics for the variables used in the analysis ....................................... 34
Table 2.3. Model specification tests ......................................................................................... 35
Table 2.4. Maximum likelihood estimates for the stochastic production frontier model ....... 36
Table 2.5. Production and scale elasticities.............................................................................. 38
Table 2.6. Frequency distribution of technical efficiency (TE) for the conventional and
organic farms ............................................................................................................................ 39
Figure 2.‎1. Histogram, normal and nonparametric densities of technical efficiency............... 40
Table ‎3.1. Summary statistics for the variables of interest ...................................................... 64
Table ‎3.2. Summary statistics for the local estimates of  u2 ,  v2 and  ................................ 65
Table ‎3.3. Estimated LML production elasticities and observed input share for Kansas ........ 66
Table ‎3.4. Frequency distribution of technical efficiency scores ............................................. 67
Table ‎3.5. Kolmogorov-Smirnov test ....................................................................................... 68
Table ‎3.6. Cluster analysis results ............................................................................................ 69
Figure ‎3.1. Distribution of the number of observations at each data point. ............................. 70
Figure ‎3.2. Distribution of localized estimates of input elasticities and returns to scale ......... 71
Table 4.1. Summary statistics for the variables used in the analysis ....................................... 95
Table 4.‎2. Summary statistics for DEA results ........................................................................ 96
Table 4.3. Frequency distribution of DEA efficiency scores ................................................... 97

iii
 Abstract

Firm-level productivity and efficiency analyses have important implications for the evaluation
of their economic viability and sustainability. The assessment‎ of‎ a‎ firm’s‎ performance‎
requires the use of an adequate methodological approach to derive sound efficiency estimates.
By targeting economic sectors not previously investigated and using new methodological
approaches, this thesis contributes to the literature both from a methodological and empirical
point of view.
Three specific objectives have been pursued in three papers that constitute the main
body of the present thesis. The main purpose of the first paper is to compare the efficiency
ratings of organic and conventional grape farms in Catalonia. To do so, we fit a stochastic
production frontier to cross sectional, farm-level data collected from a sample of 141 Catalan
farms that specialize in grape growing. Results show that organic farmers, on average, are
more efficient than their conventional counterparts (efficiency ratings are on the order of 0.80
and 0.64, respectively). Apart from adoption of organic practices, experience is also found to
improve technical efficiency. Conversely, technical efficiency tends to decrease with the
relevance‎of‎unpaid‎family‎labor,‎farm‎location‎in‎less‎favored‎areas,‎and‎farmers’‎ concerns
for environmental preservation.
In the second paper, local maximum likelihood (LML) methods, recently proposed by
Kumbhakar et al. (2007), are applied to assess the technical efficiency of a sample of arable
crop Kansas farms. LML techniques overcome the most relevant limitations associated to
mainstream parametric and nonparametric frontier models. Results suggest that Kansas farms
reach technical efficiency levels on the order of 91%. These results are compared with data
envelopment analysis and stochastic frontier analysis efficiency estimates.
The last paper focuses on the assessment of technical and environmental efficiency of
Catalan arable crop farms. Specifically, we apply the methodology recently developed by
Coelli et al. (2007) and extend it to a consideration of the stochastic conditions under which
production takes place, as proposed by Chambers and Quiggin (1998 and 2000). Results
suggest that sample farms reach technical and environmental efficiency levels on the order of
93% and 74%, respectively.

iv
 Resumen

El análisis de la productividad y la eficiencia tiene importantes implicaciones para la evaluación

de la viabilidad económica y la sostenibilidad de las empresas. Dicha evaluación requiere el uso
de un enfoque metodológico adecuado que produzca estimaciones de eficiencia no sesgadas.
Mediante el estudio de sectores económicos no analizados con anterioridad y la adopción de
nuevos enfoques metodológicos, esta tesis contribuye a la literatura, tanto desde el punto de vista
metodológico como empírico.
La tesis estudia tres cuestiones principales que se reflejan en tres artículos científicos
independientes, que constituyen el elemento central de la misma. El principal objetivo del primer
artículo es el de comparar la eficiencia técnica de las explotaciones de uva ecológicas y
convencionales en Cataluña. Para ello utilizamos el modelo de la frontera de producción
estocástica. El análisis se basa en datos de corte transversal de una muestra de 141 explotaciones
catalanas especializadas en la producción de uva. Los resultados sugieren que los agricultores
ecológicos son, de promedio, técnicamente más eficientes que los convencionales (los ratios de
eficiencia son 0,80 y 0,64, respectivamente). Además de la adopción de técnicas ecológicas, la
experiencia también incrementa la eficiencia técnica. En cambio, las explotaciones con una mayor
proporción de trabajo no remunerado, que se encuentran en una zona desfavorecida y/o que
tienden a tener fuertes preferencias por preservar el medio ambiente, son generalmente menos
eficientes.
En el segundo artículo, se utilizan los métodos de máxima verosimilitud local (LML)
propuestos recientemente por Kumbhakar et al. (2007) para estimar la eficiencia técnica de una
muestra de explotaciones agrícolas especializadas en la producción de cereales, oleaginosas y
proteaginosas en Kansas. Las técnicas LML permiten superar muchas de las limitaciones
asociadas a los modelos de frontera paramétricos y no paramétricos. Los resultados sugieren que
las explotaciones de Kansas son técnicamente eficientes, con niveles de eficiencia del orden del
91%. Estos resultados se comparan con los ratios de eficiencia obtenidos a través del análisis de la
envolvente de datos y de la frontera de producción estocástica.
El último artículo se centra en la estimación de la eficiencia técnica y medioambiental de
una muestra de explotaciones Catalanas especializadas en la producción de cereales, oleaginosas y
proteaginosas. Para ello se aplica una versión ampliada de la metodología desarrollada
recientemente por Coelli et al. (2007), la cual se extiende para considerar explícitamente las
condiciones estocásticas bajo las cuales tiene lugar la producción. Para ello se utilizan los
métodos estado-contingente propuestos por Chambers y Quiggin (1998 y 2000). Los resultados
sugieren que las explotaciones presentan niveles de eficiencia técnica y medioambiental del orden
del 93% y 74%, respectivamente.

v
 Dedication

I dedicate this doctoral thesis to my mother “Mbarka”

and my father “Mohamed “for their patience and sacrifice
in making me what I am today.

 BOUALI

vi
 Acknowledgments

First of all, without the guidance of ALLAH, this work would not have been done successfully.
My deepest and sincere gratitude to Allah: who gives me all the valuable things that I have and guides
my way, not only during the realization of this thesis but during my whole life.

I would like to express my sincere thanks to many people and institutions that have helped me
to achieve my academic goals and who deserve my acknowledgment.

First among those I want to thank is my dissertation director Dr. Teresa Serra, “Teia”. I
would like to express my deepest thanks and gratitude for her detailed supervision, insightful
guidance, encouragement, constructive criticism and friendly treatment throughout my PhD study.
Without her guidance, this dissertation would not have been accomplished. I am honored to have
learnt from her innumerable lessons and insights in the academic research. I owe you so much.

I am very thankful to my dissertation tutor, Prof. José Maria Gil Roig, for his valuable advice,
support, encouragement and friendly treatment throughout this thesis. I am grateful for his time and
effort spent in reviewing this research.

I would also like to thank Prof. Allen Featherstone and Dr. Zein Kallas for their valuable
suggestions and contributions to this dissertation.

I gratefully acknowledge the Centre de Recerca en Economia i Desenvolupment

Agroalimentari (CREDA-UPC-IRTA) for providing a dynamic scientific atmosphere and financial
support during the realization of my PhD study.

I am deeply and forever indebted to all my family for their tremendous love, continued
support, encouragement and their valuable prayers throughout my entire life. Especially, my deepest
thanks go to my wife “Ibtissem” for her patience and support in both the easy and hard times that I
have faced.

I am thankful to my colleagues at the CREDA center and my friends in Spain and in Tunisia
whom I consider part of my family, for providing a friendly working atmosphere and always being
ready to lend a hand when I need them.

Finally, I am thankful to all my professors at the “Ecole Supérieur d’Agriculture de Mograne

(ESAM)”, “Universitat Politècnica de Catalunya (UPC)”, “Universitat de Barcelona (UB)” and “the
Mediterranean Agronomic Institute of Zaragoza (IAMZ)”, for providing me financial support and
scientific formation during all my studies.

vii
Chapter 1

Introduction
2
Technical efficiency is a prerequisite for economic efficiency, which in turn ensures the
economic‎ viability‎ and‎ sustainability‎ of‎ a‎ firm.‎ In‎ being‎ a‎ useful‎ tool‎ to‎ diagnose‎ a‎ firm’s‎
economic problems, assessment of technical efficiency has drawn broad research interest.
Efficiency requires rational input allocation to achieve the desired output levels, which is
important for producers who try to optimize their production decisions, and strengthens the
firms’ capacity to face changing market conditions, increasing input costs, economic
hardships and rapid technological progress. It is also relevant for policy makers interested in
enhancing‎ firms’‎ economic‎ performance‎ and‎ competitiveness,‎ promoting‎ economic‎
development and sustainable economic practices.
The analysis of technical efficiency assesses to what extent firms are able to maximize
their output levels with minimum use of inputs. Since the pioneering work by Farrell (1957),
the scientific community has proposed a wide array of techniques to derive firm-level
efficiency measures. Two main approaches namely, parametric (Stochastic Frontier Analysis -
SFA) and nonparametric approaches (Data Envelopment Analysis - DEA), have emerged as
alternatives and have been extensively used in the efficiency literature (see, for a few
examples, Tzouvelekas et al., 2001, 2002; Oude Lansink et al., 2002; Sipiläinen and Oude
Lansink, 2005; Lohr and Park, 2006).
The‎ assessment‎ of‎ a‎ firm’s‎ performance‎ requires the use of an adequate
methodological approach to derive unbiased efficiency estimates. Several studies have shown
that technical efficiency estimates are sensitive to estimation methods and functional form
specifications (Ferrier and Lovell, 1990; Coelli and Perelman, 1999; Ruggiero and Vitaliano,
1999; Chakraborty et al., 2001). Inadequate representations of the production frontier and
error distributions can lead to misleading results (Kumbhakar et al., 2007; Martins-Filho and
Yao, 2007; Serra and Goodwin, 2009). The efficiency measurement literature has
progressively evolved to incorporate new advances, refinements and extensions. This thesis
focuses on a few of the most recent methodological developments.
Both parametric (SFA) and nonparametric (DEA) techniques have been shown to
suffer from different shortcomings. The stochastic parametric approach addresses the main
shortcomings of the deterministic approach and permits to distinguish between inefficiency
and‎exogenous‎shocks‎that‎are‎outside‎the‎firm’s‎control‎(Aigner et al., 1977; Meeusen and
Van den Broeck, 1977). However, it requires specification of a parametric frontier function to
capture production characteristics (e.g: Aigner et al., 1977; Meeusen and Van den Broeck,
1977). In this regard, SFA relies on two strong assumptions: the specification chosen to
represent the deterministic frontier and the distributional assumption of the composite error

3
term. In contrast, the nonparametric approach (DEA), through which all firms are compared
with the “best‎practice”‎or‎“benchmark‎performance”‎frontier,‎does not rely on the definition
of a functional form characterizing the underlying technology and therefore avoids
misspecification problems. A disadvantage of this technique is the ignorance of the stochastic
error term which implies that all deviations from the frontier are attributed to inefficiency. As
a result, TE ratings obtained from the nonparametric approach are generally lower than those
obtained under the parametric alternative (SFA) (Sharma et al., 1999; Puig-Junoy and Argiles,
2000; Wadud and White, 2000).
To overcome the limitations of both aforementioned approaches without foregoing
their advantages, a new methodological approach based on local modeling methods has been
recently proposed by Kumbhakar et al. (2007). In contrast to parametric models, this method
does not require strong assumptions regarding the deterministic and stochastic components of
the frontier: the parameters characterizing both production and error distribution are localized
with respect to the covariates. As opposed to nonparametric approaches, local modeling
methods allow for stochastic variables and variable measurement errors when estimating
efficiency scores. The local modeling approach by Kumbhakar et al. (2007) is based on local
maximum likelihood (LML) principles (Fan and Gijbels, 1996).
Recent advances in the efficiency literature not only refer to methodological but also
conceptual issues. Traditional performance measures focus almost exclusively on the efficient
use of conventional inputs and outputs. However, as the environmental sustainability of
economic activities has become of increasing interest, firm-performance studies have evolved
to include environmental concerns and conventional efficiency measures have been extended
to include both technical and environmental dimensions. Late developments within the
literature on environmental efficiency have stressed the necessity to consider the materials
balance condition in order to provide sound measures of firms’‎environmental performance.
Based on this principle, Coelli et al. (2007) suggest a new approach which, in contrast to
previous modeling approaches, does not require the introduction of an extra pollution variable
in the production model.
This thesis employs both local production frontier estimation techniques to derive
production frontier parameters, as well as environmental efficiency measures to extend
technical performance measures with environmental considerations. Its scientific contribution
is both methodological and of an empirical nature, and is organized in three main core
chapters that constitute three independent scientific articles. The analysis of the efficiency
with which agricultural holdings operate is the guideline of this thesis. The first article uses

4
well-known SFA techniques to conduct a comparative study of technical efficiency ratings for
organic and conventional grape farms in Catalonia, Spain. The assessment of organic Spanish
farms’‎ economic‎ viability‎ has‎ received‎ scant‎ attention‎ by‎ the‎ literature and this work
contributes to fill this gap. This first article is also pioneering in that it measures the
contribution of farmers’‎ preferences‎ regarding‎ environmental‎ preservation‎ and‎ economic‎
performance to efficiency. The analysis is conducted on cross sectional, farm-level data
collected from a sample of organic and conventional Catalan farms that specialize in grape
growing, and based upon a stochastic production frontier in which inefficiency effects are
assumed to be a function of firm-specific characteristics.
The second article implements local estimation techniques to study efficiency of
Kansas farms that specialize in the production of cereals, oilseeds and protein (COP) crops.
The relevant role of Kansas in US arable crop production justifies the decision to study
technical efficiency of Kansas arable crop farms. The analysis is based on farm-level data
obtained from farm account records from the Kansas Farm Management Association (KFMA)
dataset. In spite of the interesting features of local estimation methods, its use has been
limited to a few empirical studies due to implementation complexities. Further, while the
existing literature on technical efficiency has broadly compared parametric (SFA) and
nonparametric approaches (DEA), no study had previously compared semiparametric
techniques with mainstream methods. The second article in this dissertation sheds light on this
issue.
The third research article focuses on environmental efficiency measurement of a
sample of Spanish arable crop farms. Since very few proposals to measure environmental
efficiency based on the materials balance principle have been proposed and empirically
implemented, there is scope for significant literature contributions. The third paper builds on
the proposal by Coelli et al. (2007), expands it to allow for production risk and makes a
twofold contribution to the literature. To date, there are no published studies that have focused
on the assessment of technical and environmental efficiency of Spanish agriculture using this
methodology. Furthermore, environmental efficiency studies have failed to explicitly consider
the stochastic conditions under which production takes place. We do so by implementing the
recently developed state-contingent methods as proposed by Chambers and Quiggin (1998
and‎ 2000),‎ which‎ represents‎ a‎ relevant‎ extension‎ to‎ Coelli‎ et‎ al’s‎ (2007)‎ proposal.‎ To‎ our‎
knowledge, no previous published work has studied environmental efficiency using state-
contingent methods. To the extent that the use of environmentally damaging inputs affects
production risk, a model that ignores risk will produce biased efficiency estimates. The

5
analysis is based on farm-level data collected using a questionnaire distributed among 190
Catalan arable crop agricultural holdings.
In addition to this general introduction and the concluding remarks section, this
present thesis is organized into three chapters containing the three research articles
summarized above.‎The‎first‎paper‎(chapter‎2),‎entitled‎“The‎productive‎efficiency‎of‎organic
farming:‎the‎case‎of‎grape‎sector‎in‎Catalonia”‎has‎been‎published‎in‎the‎Spanish Journal of
Agricultural Research.‎The‎second‎paper‎(chapter‎3),‎entitled‎“Technical‎efficiency‎of‎Kansas‎
arable crop farms: a local maximum likelihood approach”,‎is‎under review in the Agricultural
Economics journal.‎ The‎ third‎ paper‎ (chapter‎ 4),‎ entitled‎ “Technical‎ and‎ environmental‎
efficiency‎ of‎ Catalan‎ arable‎ crop‎ farms”,‎ is‎ under‎ review‎ in‎ the‎ Applied Economic
Perspectives & Policy journal. The final chapter synthesizes the main findings achieved in the
three previous chapters. Based on these results, some policy implications as well as
recommendations for future studies are derived.

6
References

Aigner, D., Lovell, C.A.K., Schmidt, P., 1977. Formulation and estimation of stochastic
frontier production functions models. Journal of Econometrics 6: 21-37.
Farrell, M.J., 1957. The measurement of productive efficiency. Journal of the Royal
Statistical Society, Series A, CXX, Part 3: 253-290.
Chakraborty, K., Biswas, B., Lewis, W.C., 2001. Measurment of technical efficiency in
public education: A stochastic and nonstochastic production function approach. Southern
Economic Journal 67: 889-905.
Chambers, R.G., Quiggin, J., 1998. Cost functions and duality for stochastic technologies.
American Journal of Agricultural Economics 80: 288-295.
Chambers, R.G., Quiggin, J., 2000. Uncertainty, Production, Choice, and Agency.
Cambridge, UK: Cambridge University Press.
Coelli, T.J., Perelman, S., 1999. A comparison of parametric and non-parametric distance
functions: With application to European railways. European Journal of Operational
Research 117: 326-339.
Coelli, T., Lauwers, L., Van Huylenbroeck, G.V., 2007. Environmental efficiency
measurement and the materials balance condition. Journal of Productivity Analysis 28: 3-
12.
Fan, J., Gijbels, I., 1996. Local polynomial modeling and its applications. London, UK:
Chapman & Hall.
Ferrier, R.S., Lovell, C.A.K., 1990. Measuring cost efficiency in banking: An econometric
and linear programming evidence. Journal of Econometrics 46: 229-245.
Kumbhakar, S.C., Byeong, U.P., Simar, L., Tsionas E.G., 2007. Nonparametric stochastic
frontiers: A local maximum likelihood approach. Journal of Econometrics 137: 1-27.
Lohr, L., Park, T.A., 2006. Technical efficiency of US organic farmers: The complementary
roles of soil management techniques and farm experience. Agricultural and Resource
Economics Review 35: 327-338.
Martins-Filho, C., Yao, F., 2007. Nonparametric frontier estimation via local linear
regression. Journals of Econometrics 141: 283-319.
Meeusen, W., Van Den Broek, J., 1977. Efficiency estimation from Cobb-Douglas production
functions with composed error. International Economic Review 18: 435-444.

7
Oude Lansink, A., Pietola, K., Bäckman, S., 2002. Efficiency and productivity of
conventional and organic farms in Finland 1994–1997. European Review of Agricultural
Economics 29: 51-65.
Puig-Junoy, J., Argiles, J.M., 2000. Measuring and explaining farm inefficiency in a panel
data set of mixed Farms. Pompeu Fabra University, Working Paper, Barcelona, Spain.
Ruggiero, J., Vitaliano, D., 1999. Assessing the efficiency of public schools using data
envelopment analysis and frontier regression. Contemporary Economic Policy 17: 321-
331.
Serra, T., Goodwin, B.K., 2009. The efficiency of Spanish arable crop organic farms, a local
maximum likelihood approach. Journal of Productivity Analysis 3: 113-124.
Sharma, K.R., Leung, P., Zaleski, H.M., 1999. Technical, allocative and economic
efficiencies in swine production in Hawaii: A comparison of parametric and nonparametric
approaches. Agricultural Economics 20: 23-35.
Sipiläinen, T., Oude Lansink A., 2005. Learning in organic farming-an application on Finnish
dairy farms. Manuscript presented at the XIth Congress of the EAAE. Copenhagen,
Denmark, August 24–17.
Tzouvelekas, V., Pantzios, C.J., Fotopoulos, C., 2001. Technical efficiency of alternative
farming systems: The case of Greek organic and conventional olive-growing farms. Food
Policy 26: 549–569.
Tzouvelekas, V., Pantzios, C.J., Fotopoulos, C., 2002. Empirical evidence of technical
efficiency levels in Greek organic and conventional farms. Agricultural Economics Review
3: 49-60.
Wadud, A., White, B., 2000. Farm household efficiency in Bangladesh: A comparison of
stochastic frontier and DEA methods. Applied Economics 32: 165-73.

8
Chapter 2

The productive efficiency of organic farming:

the case of grape sector in Catalonia1

1
Publication Information: Guesmi, B, Serra, T., Kallas, Z., Gil, J.M., 2012. The productive efficiency of organic
farming: the case of grape growing in Catalonia. Spanish Journal of Agricultural Research 10: 552-566.
10
2.1. Introduction

Intensive agricultural systems have caused several negative externalities on humans, animals
and the environment. Impacts on human health, pollution of underground and surface water,
loss of biodiversity, or overutilization of natural resources are just a few examples of these
externalities. Social concerns regarding the negative externalities derived from conventional
agriculture have been growing. Over the last few years, there has also been an increase in
consumer awareness pertaining to the consequences of food choices on their health and the
environment. These concerns have led to changes in European Union (EU) agricultural
policies that have progressively incorporated environmental considerations. Interest in
alternative agricultural practices that are more environmentally friendly has also been
growing. Organic farming, which replaces chemical inputs with organic fertilizers and non-
chemical crop protection inputs, has received substantial attention within the EU.
Since the beginning of the 1990s, the EU has made a significant effort to enhance and
develop organic agriculture. In order to increase the supply of organic products, EU countries
have provided financial assistance for organic producers. Conversion subsidies have been
introduced to compensate for the lower incomes obtained during the early stages of
conversion. As a result, organic farming has quickly grown within the EU-27 countries from
0.70 million hectares in 1993 to 7.20 million hectares in 2007 (Eurostat, 2007; Willer and
Kilcher, 2009). The organic area share over the total utilized agricultural area is around 4% in
the EU-27, which is among the highest in the world. Organic farming in Spain has grown
faster than in other EU-27 member states. While Spain ranked 10th in‎the‎EU’s‎organic area
distribution with 4,235 ha in 1991, it currently ranks second with almost one million ha
(Lampkin, 1996; MARM, 2008; FiBL, 2009). Spain was the first contributor to the increase in
the‎ EU’s‎ organic‎ area‎ in‎ 2006‎ (FiBL, 2009). The rapid and substantial increase is mainly
explained by economic strategies adopted by farmers who consider organic production to be
profit maximizing when accounting for subsidies received and price premiums for their
produce (Armesto-López, 2008).
Despite the prominent position of Spain in the EU, the share of organic farming in the
Spanish utilized agricultural area (UAA) (3.70%) is still below the EU-27 average (4%). As in
Europe, more than 60% of the Spanish organic area is devoted to grassland, while arable
crops are the most important organic crop with almost 275,823 ha, representing more than one
third of the organic crop area. Olive groves are the second most common organic crop (22%),
followed by dried fruits (15%) and grapes (7%) (MARM, 2008).

11
 Grape is a perennial crop that, compared to other crops, has relatively low nutritional
needs and adapts well to marginal soils (Winkler et al., 1974; Pongracz, 1978). This feature is
considered very relevant to produce organically and makes conversion easier than for other
crops. While other crops suffer many problems over the period of transition from
conventional to organic, grape cultivation does not as long as the minimum level of nutrient
needs is guaranteed to avoid productivity loss. These features make organic grape production
a technically feasible, economically attractive and sustainable activity. Selection of resistant
varieties in organic viticulture plays a vital role in ensuring high immunity against pests and
diseases, high adaptation to the environmental conditions (rainfall, temperature, frost,
humidity and soil quality), high productivity and profitability. Other operations are considered
important to guarantee an excellent growing season for organic grape. Organic vineyard
requires correct training operations to facilitate pruning (a critical practice), spraying and
harvesting.
By the end of the 2000s decade, 70% of the worldwide organic grape production area
was located in the EU-27, where Italy, France and Spain were the main producers. Within the
EU-27, Spain represented 33% of the total (organic and conventional) vineyard area
(Eurostat, 2008) and 15% of the organic vineyard area, behind Italy (32%) and France (17%).
The Spanish organic grape area represented 1.70% of total grape area. In terms of production,
Spain generated 23% of total grapes produced in the EU-27 and 9% of worldwide production.
Catalonia plays a significant role in organic farming in Spain, recording an average
annual growth rate of 37% since 1995 (CCPAE, 2009). While the major organic producer in
Spain is Andalucía (with around 60% of total area), Catalonia ranked fourth with 62,331 ha
farmed by 909 producers in 2008. Further, 19% of the total Spanish organic food industry was
concentrated in Catalonia. The Catalan organic vineyard area represented around 7% of the
total organic grape area in Spain (being the fourth most relevant share). Since 1995, this area
rapidly grew with an average annual growth of about 21%. The area increased from 207 ha in
1995 to 2,241 ha in 2008 (CCPAE, 2009). In terms of production, Catalonia contributes 7% to
total Spanish vineyard production. We aim to study the technical efficiency (TE) with which
Catalan grapes operate.
While conversion subsidies are useful in promoting organic conversion, they do not
guarantee that converting farms will be economically viable in the future. An important first
step towards economic viability is to ensure that organic production processes are technically
efficient. TE is a prerequisite for economic efficiency, which is also a necessary condition for
economic sustainability (Tzouvelekas et al., 2001). Knowledge about productivity and

12
efficiency differences between conventional and organic farms is important for policy makers
who are interested in promoting sustainable farming practices, farmers who try to optimize
their production decisions, as well as other economic agents such as food processors and
retailers who process and sell organic food. In the following lines a literature review on
organic farming is presented.
The relevance of the organic farming movement has led many authors to evaluate the
current situation and expectations on the future development of organic farms. Among these
studies, the analyses on the adoption of organic farming practices have gained special
relevance. Different methodologies, ranging from descriptive qualitative analyses to highly
sophisticated econometric exercises, have served this purpose. Within the adoption literature,
a first group of studies has been interested in understanding the determinants that motivate
farmers to adopt the organic technology (Fairweather, 1999; Lohr and Salomonson, 2000;
Pietola and Oude Lansink, 2001; Acs et al., 2007). A second group has focused on the amount
of time it takes a farmer to adopt organic practices (Padel, 2001; Parra et al., 2007).
Despite the development of organic farming worldwide and especially in Europe, the
literature on the TE performance of organic farming is sparse, which is mainly due to data
scarcity on organic farms (Oude Lansink et al., 2002). In recent years there have been a few
attempts to study this issue. Different approaches have been used to estimate the differences
in TE between conventional and organic farms and different results have been derived. While
some authors have utilized a parametric approach, specifically a Stochastic Frontier Analysis
(SFA), others have relied on non-parametric methods, specially the Data Envelopment
Analysis (DEA).
Oude Lansink et al. (2002) used DEA to compare organic and conventional crop and
livestock farms in Finland and found that organic producers have higher efficiency than
conventional farms (efficiency ratings for organic and conventional producers were 0.96 and
0.72, respectively), but use a less productive technology. In another recent DEA-based study,
Bayramoglu and Gundogmus (2008) suggested that conventional raisin-producing households
in Turkey are superior to organic producers in terms of TE (0.90 vs. 0.86). Both studies
assumed variable returns to scale in order to compute TE.
Tzouvelekas et al. (2001; 2002a,b) used the Stochastic Production Frontier (SPF)
approach to evaluate the TE ratings achieved by Greek organic and conventional farms. They
found organic producers to be more efficient than conventional ones. In contrast with this
finding, Madau (2007) applied a SPF model and found that Italian conventional cereal farms
were significantly more efficient than organic farms (0.90 vs. 0.83). Serra and Goodwin

13
(2009) is the only study that compares the efficiency ratings of organic and conventional
arable crop farming in Spain. In this analysis, the SPF model was estimated by a local
maximum likelihood approach. Results showed that organic farms have efficiency levels that
are slightly below conventional farms (0.94 vs. 0.97). The output-oriented measure of
efficiency is the most widely used method to determine TE levels.
In spite of the recent relevant growth of organic farming in Spain, the literature on the
TE of organic farming in this country is very thin. Our work contributes to the scarce
literature on organic farming in Spain by carrying out a comparative study of TE ratings for
organic and conventional grape farms in Catalonia. Additionally, we attempt to identify the
factors that affect TE levels. SPF methodology is used for this purpose. By measuring
efficiency we can assess whether economic agents use their resources optimally to reach their
production objectives. Productivity differences between the two agricultural practices are also
assessed by means of computing the output elasticity of different inputs and the productivity
measure proposed by Kumbhakar et al. (2009).

2.2. Material and methods

The assessment of farm TE and the factors that explain TE provides valuable information to
improve farm management and economic performance. In the presence of technical
inefficiencies, farmers can increase their production levels without the need to increase the
use of inputs that are usually scarce, or to adopt new technologies or practices. Avoiding
sources of inefficiency and waste of resources is a requisite for economic sustainability.
Generally, a farmer who operates with a high TE level obtains better economic results than a
farmer who does not. In this regard, productive efficiency studies have important implications
for economic performance, technological innovation and the overall input use in the
agricultural sector.
There are two main approaches widely used in the literature to estimate TE:
parametric (SFA or deterministic frontier analyses) and non-parametric methods (data
envelopment analysis, DEA). Non-parametric techniques are more flexible than parametric
approaches in that they can be implemented without knowing the true specification of the
functional form characterizing the production technology. However, they do not allow the
researcher to isolate inefficiency effects from random noise or random shocks.
To overcome the identification problem posed by non-parametric models, an
alternative method can be used: SFA. This approach, that was introduced simultaneously by

14
Aigner et al. (1977) and Meeusen and Van den Broeck (1977), distinguishes between
exogenous‎ shocks‎ outside‎ the‎ firm’s‎ control‎ and‎ inefficiency.‎ Contrary to DEA and
deterministic frontier analyses, SFA accounts for random noise and can be used to conduct
conventional tests of hypotheses. On the other hand, SFA requires the specification of a
distributional form for the inefficiency term and a functional form for the production function.
Results of SFA are sensitive to these assumptions. Since agricultural production outcomes are
stochastically determined due to random climatic influences, and since agricultural production
studies are likely to be affected by measurement and variable omission errors (Coelli, 1995;
Chakraborty et al., 2002; Oude Lansink et al., 2002), it is necessary to choose a robust model
that reflects and accounts for these issues. In this regard, we select SFA as a method to
correctly and consistently estimate TE.
The SPF proposed by Aigner et al. (1977) and Meeusen and Van den Broeck (1977)
can be specified as:

yi  f (Xi ; ) exp  ei  ; ei  vi  ui , i =1,2,...,N (1)

where yi denotes the level of output for the i-th observation (firm); Xi is the vector of input

quantities used by the i-th firm in the production process;  is the vector of parameters to be
estimated; and f ( Xi ; ) is a suitable functional form for the frontier. The error term ei in

equation (1) can be decomposed into two components, ui and vi ; it is assumed that ui and vi

are independently distributed from each other. The first component, vi , is a standard random
variable capturing the random variation in output due to statistical noise that arises from (a)
the unintended omission of relevant variables from vector Xi ; (b) from measurement errors
and approximation errors associated with the choice of the functional form; (c) unexpected
stochastic changes in production (weather influences, for example); and (d) other factors that
are not under the control of the farm. Component vi is usually assumed to be symmetric,

independent and identically distributed as N(0,  v2 ). The second component ui N+(μ,  u2 ), is

a one-sided, non-negative random variable representing the stochastic shortfall of the i-th
farm output from its production frontier, as a result of the existence of technical inefficiency.
The output oriented measure of TE can be expressed as the ratio of observed output to
the corresponding stochastic frontier output, a measure that takes a value between 0 and 1:

15
 yi
TEi =  exp  ui  (2)
f  Xi ;   exp  vi 

Reifschneider and Stevenson (1991), Huang and Liu (1994) and Battese and Coelli
(1995) proposed stochastic frontier models in which the inefficiency effects ( ui ) are
expressed as a linear function of explanatory variables reflecting farm socio-economic and
demographic characteristics and a random error. Following Battese and Coelli (1995) we used
the following TE effects model:

M
ui   0    m Zmi   i (3)
m 1

where Zmi are farm-specific variables associated with technical inefficiencies;  0 and  m are

parameters to be estimated; and  i is a random variable with zero mean and finite variance

 M

 2 , defined by the truncation of the normal distribution such that  i    0    m Zmi  . The
 m 1 
M
mean of ui ,    0    m Zmi , is farm-specific and the variance components are assumed to
m 1

be equal (  u2   2 ).
Following Battese and Coelli (1995), we estimate the parameters of the model defined
in equations (1) and (3) by maximum likelihood procedures. The log likelihood function and
the derivation of TE estimates followed the approach used in Battese and Coelli (1995). The
estimation was carried out using the parameterization by Battese and Corra (1977) who
replace  v2 and  u2 with  2   u2   v2 and    u2 /( u2   v2 ) . The next section is devoted to
present research results.

2.3. Results

2.3.1. Characteristics of farms and farmers

Our analysis uses cross sectional, farm-level data collected from a sample of Catalan farms
that specialize in grape growing. This research focuses on Catalonia because of the important
role played by the Catalan vineyard sector within the Spanish organic agriculture and the

16
exponential growth that this sector has experienced since 1995. It is thus relevant to
investigate the characteristics of this type of farming and compare them with the
characteristics of the conventional sector. Data were collected by face-to-face questionnaires
during the period from March to June 2008 in the major Catalan organic grape-growing areas.
These areas were identified based on organic farming systems certification by the Official
Certification Organism in Catalonia (Consell Català de la Producció Agrària Ecològica,
CCPAE).
Geographically, our sample farms are concentrated in three different Catalan
provinces (Barcelona, Tarragona and Lleida). For each organic farm, at least three
neighboring conventional farms were also selected. This neighboring criterion allows for two
subsamples (organic and conventional) with an analogous composition (Tzouvelekas et al.,
2001; Madau, 2007). Our final sample consists of 26 organic farmers and 115 neighboring
conventional farms. The following lines provide a description of sample farms both from an
agronomic and economic perspective, as well as the demographic characteristics of sample
farmers. Summary statistics for sample farm and farmer characteristics are presented in Table
2.1.
Based on a scale from 0 to 10, farmers were asked to grade soil quality and erosion.
Although both groups have similar perceived soil quality and erosion, a large number of
organic farms (53%) are located in a less favored area or in an area with specific difficulties
that limit agricultural productivity (Council Regulation EC 1257/1999). In contrast, only a
quarter of conventional farms are located in these areas. On average, organically farmed soil
is steeper (9%) than conventionally farmed soil (3%). The difference in slope is statistically
significant. Although both farm types strongly rely on rainfed agriculture, irrigation practices
are relatively more important within the organic group (16% vs. 7%).
Land use patterns do not differ greatly between organic and conventional farms. On
average 64% of conventionally cultivated land is devoted to produce grapes. Arable crops are
the second most common conventional crop (19%), followed by fruits (10%) and olive groves
(9%). Organic farms devote, on average, 69% of their land to grape production, mainly at the
expense of arable crops that now represent 11% of cultivated land. Many different cultivars,
with different abilities to withstand climatic conditions and diseases, are used within organic
and conventional farms. However, both farm types use a similar range of grape varieties. The
most common varieties spread among all farmers are ‘Macabeu’ (69.50%), ‘Parellada’
(58.87%), ‘Ull de llebre’, (42.55%), ‘Xarello’ (37.59%), ‘Merlot’ (30.50%), ‘Cabernet’
(22.70%) and ‘Garnatxa’ (18.44%).

17
 Contrary to conventional farms that have, on average, 45 ha of agricultural land,
organic farms are mainly small holdings with only 19 ha. The land tenure status is similar
between farm types, with owned land representing 46% (45%) of total organic (conventional)
land. Farm output is defined as the quantity of grapes produced and expressed in physical
units (kg). Conventional‎ farms’‎ total‎ output‎ averages‎ 120,364‎ kg,‎ which‎ is‎ twice‎ organic‎
farms’‎ total‎ output‎ (59,969‎ kg).‎ However,‎ organic‎ farms’‎ yields‎ are‎ only‎ 16%‎ lower‎ than‎
conventional‎farms’‎yields.‎The‎difference‎in‎total‎output‎and‎yields‎is‎statistically‎significant.
The average price received by organic farms more than doubles the average
conventional price, suggesting statistically significant organic price premiums. The proportion
that‎agricultural‎revenue‎represents‎within‎total‎farmers’‎revenue,‎which‎measures‎the degree
of diversification in income sources, is 68% (77%) for organic (conventional) farms. Hence,
organic farmers have more diversified income sources. Subsidies (almost 70% of organic
farms receive public subsidies) and price premiums compensate for the low yields and high
costs‎ in‎ organic‎ farming,‎ leading‎ to‎ substantially‎ higher‎ incomes‎ on‎ a‎ per‎ hectare‎ basis:‎ €‎
4,004 vs. €‎2,670.
Consistent with previous research, statistically significant differences regarding input
use are found between the two groups: our organic sample farms are more labor intensive
than conventional farms. Both types of farms strongly rely on unpaid family labor which
represents 69% (73%) of total labor in organic (conventional) farms. On a per hectare basis,
expenses in fertilizers and crop protection products are much higher in organic farms (381‎€‎
ha-1 vs. 294‎€‎ha-1).‎Total‎costs‎per‎hectare‎are‎€‎1,814‎(€‎1,509)‎for‎organic‎(conventional)‎
farms. Consistently with higher per hectare input costs borne by organic farms, this group
uses 0.66 agricultural machines per hectare (machines include any farm equipment: tractors,
manure spreaders, pre-pruning, cultivators, shredders, etc.), while conventional farms use
0.50 machines. Organic farmers appear to have less access to bank loans than conventional
counterparts. A 50% of the latter are able to get credit, while less than 30% of the former have
access to bank loans. Farmers mainly use the loans for operation and investment.
The‎difference‎between‎income‎and‎costs‎per‎hectare‎leads‎to‎profits‎per‎hectare‎of‎€‎
2,435‎for‎organic‎farms‎and‎€‎1,283‎for‎conventional‎ones.‎Hence,‎organic‎profits‎per‎hectare‎
almost double conventional profits. Regarding the marketing of agricultural output, both
organic and conventional farms strongly rely on sales to processing companies and
cooperatives. These sales represent around 71% and 73% of conventional and organic
production sales, respectively. Conventional and organic farmers are members of different
agricultural‎ associations‎ such‎ as‎ cooperatives,‎ farmers’‎ associations‎ and‎ syndicates,‎ organic‎

18
farming associations and protected designations of origin (PDOs). PDOs constitute the most
attractive form of association: 68% and 60% of conventional and organic farmers
respectively, engage with these organizations which increase the market outlets for their
production.
There is a predominance of 45 years old male farmers. While organic farmers have an
average of 15 years of experience managing the farm, conventional farmers have typically
been managing the farm for about 18 years. Primary and unfinished secondary education is
the most common educational profile characterizing both organic and conventional farmers.
The family size for both groups is similar and between 3 and 4 members. Organic and
conventional farmers differ in terms of their preferences2, which helps to better understand
production and adoption decisions. When it comes to production decisions, conventional
farmers are more worried about farm economic performance (profit), whereas the organic
group is more concerned about protecting the environment.

2.3.2. Model specification and research results

In order to study productivity and efficiency of our sample of organic and conventional
Catalan grape farms, we specify our SFA as follows:

     
4 4 4
ln yi   0   0o    j   jo .Dc / o ln X ji  12   jk   jko .Dc / o ln X ji .ln Xki   vi  ui  (4)
j 1 j 1 k 1

where the subscript i  1, 2,........, N denotes the firm number and j , k  1, 2,........, J agricultural
inputs. The dependent variable ( yi ) represents grape production (in kg) by the i-th farm.

Inputs included are: ( X 1 ) total land devoted to grape, measured in hectares; ( X 2 ) total labor

(both hired and family labor), expressed in hours; ( X 3 ) total amount of capital, measured as

the number of machines used in the farm; and ( X 4 ) the expenditure in fertilizers and crop

protection‎ products‎ (in‎ €).3 Dc / o is a dummy variable that reflects the agronomic technique

2
Farmers were asked to rate their preferences for economic profit and environmental preservation from 1 to 10
(1 = not important, 10 = very important). The median of the responses is used to define a dummy variable for
each type of preferences. The first dummy takes the value of 1 if the farmer rated the relevance of economic
profit with the highest punctuation, i.e., 10 and zero otherwise, while the second dummy is one if the punctuation
was above 8.
3
To keep the model size manageable and due to the limited number of observations available, most of the inputs
considered are aggregate inputs.

19
(1=organic; 0=conventional). Summary statistics for the variables used in the analysis are
presented in Table 2.2.
M
The inefficiency model is specified as ui   0    m Zmi   i , with M  10 . The
m 1

selection of Zmi variables is based on previous literature, data available and our knowledge of
the sector studied. Since previous research has widely shown that organic practices differ
from conventional ones regarding efficiency ratings, ( Z1 ) is defined as a dummy variable that

reflects the agronomic technique ( Z1  Dc / o ).‎ Farmers’‎ experience,‎ usually‎ included‎ in‎ TE‎
studies (either as age or years of experience), is considered as the number of years dedicated
to agriculture ( Z 2 ). In line with Karagiannias et al. (2006) who shows that TE of both organic
and conventional milk farms depends on specialization, the degree of specialization measured
as the proportion of vineyard revenue to total agricultural revenue is reflected in ( Z 3 ). Madau
(2007) advocates that farms located in less favored areas or in mountain areas have lower TE
scores than the rest. A dummy variable that indicates whether the farm is located in a less
favored area or not ( Z 4 ) is used. In line with Karagiannias et al. (2006) findings, debt is also

considered through ( Z 5 ), defined as a dummy variable equal to 1 if the farmer has financial
debt and zero otherwise. Tzouvelekas et al. (2002b) conclude that organic farming subsidies
tend to negatively affect efficiency levels. A dummy variable equal to 1 if the farm receives
subsidies and zero otherwise is thus included ( Z 6 ). Tzouvelekas et al. (2001) show that
family-operated organic and conventional olive-growing farms tend to be less efficient than
farms with strong dependence on hired labor. We thus define ( Z 7 ) as the proportion of family
labor to total labor.‎ The‎ two‎ dummy‎ variables‎ described‎ above‎ that‎ reflect‎ farmers’‎
preferences for economic profit and for environmental preservation ( Z 8 and Z 9 ) are also
considered.‎Farmers’‎preferences‎have‎not‎been‎used‎by previous literature when explaining
efficiency, which represents a contribution of our analysis. The proportion of owned land to
total land ( Z10 ) is also included as previous research has shown that the share of rented land is
related to TE (Larsen and Foster, 2005). The model is estimated using Frontier 4.1 software
(Coelli, 1996).
A series of specification tests were carried out to ensure that the model specification
correctly represents our sample farms (see Table 2.3). In being a parametric approach, SFA
requires specification of the functional form representing the production technology. Since
this form is unknown, we have selected a flexible functional form (a translog - see equation

20
(4)) and compared it against another more restrictive and parsimonious specification: the
Cobb-Douglas. At the 5% level of significance, we reject the null hypothesis ( H 0 : ij  0 ),

which suggests that the translog form is the suitable specification for our data. This implies
that output elasticities and substitution elasticities depend on input levels. Further it also
involves the relevance of input interactions when explaining production. The second test
( H 0 :  *jk  0 ) indicates that the neutral stochastic frontier model (Huang and Liu, 1994) is the

adequate representation, i.e., that input use does not interfere with the variables found to
explain inefficiency. Concerning the nature of the inefficiency effects, we test whether these
are stochastic or not. We reject the null hypothesis ( H 0 :   0 ) implying that the technical
inefficiency effects are stochastic and farmers are not fully technically efficient. The fourth
test ( H 0 :    m  0 ) that aims to assess whether inefficiency effects are absent from the

model or not, is also rejected. In addition, through the fifth test ( H 0 :  m  0 ), we study the
influence of firm characteristics on TE levels. The null hypothesis is rejected, indicating that
the variables included in the inefficiency‎ effects‎ equation‎ significantly‎ influence‎ farms’‎
efficiency.
Another specification test carried out concerns geographically induced differences
among farms. Differences among areas not only refer to rainfall but also to winter freeze and
spring frost patterns, diseases brought during hot seasons, sunlight exposure, land quality and
slope, crop varieties used in different regions, etc. In order to capture these geographical
differences, a set dummies representing provinces is included. Since our sample farms are
concentrated in three different provinces of Catalonia (namely, Barcelona, Tarragona and
Lleida), two dummies, one representing Barcelona and the other for Tarragona are included
and a likelihood-ratio test is used to determine whether the two dummies are statistically
different from zero. Results show that we cannot reject the null hypothesis
( H 0 : DBarcelona ; DTarragona  0 ), which involves that the model without regional dummies in the

production equation adequately fits our data. Results of the estimation of the stochastic
frontier are reported in Table 2.4.
Production function results are best interpreted by means of input elasticities. Contrary
to the Cobb-Douglas functional form in which coefficients have a direct interpretation as
input elasticities, deriving the marginal influence of inputs on output in a translog form is not
straightforward. Input elasticities are computed for our translog model as follows:
 ln Y  /  ln  X k   k   kk ln X ki   kj ln X ji . Elasticities are computed at
j k

21
the data means and their standard deviations derived using the delta method (Snedecor and
Cochran, 1989) (Table 2.5).
In conventional farming, land has the highest elasticity estimate. Land is followed by
fertilizers and crop protection products, capital and labor. In organic farming, the highest
elasticity is achieved by fertilizer and crop protection inputs. Land area and capital display
similar contribution to output increases, while labor presents the lowest contribution to
organic grape output.
The high elasticity of the expenditures in fertilizers and crop protection products in
organic farming contrasts with the relatively low elasticity of the equivalent inputs in
conventional production methods (0.69 vs. 0.22). This implies that grape output is more
responsive to fertilizers and crop protection inputs in organic production than in conventional
technology. Land area elasticity is higher in conventional farming, which is compatible with
conventional yields being above organic ones. The elasticity of conventional grape output
with respect to land is above one half, indicating that a 1% increase in agricultural land would
lead to approximately a 0.56% increase in output. Given the restrictions faced by organic
farmers to use chemical inputs, mechanical methods are likely to become relevant, which is
reflected in the higher elasticity of capital in organic farms relative to conventional ones.4
Regarding the average scale elasticity, organic farms exhibit increasing returns to scale while
conventional farms operate under decreasing returns to scale. The small size of organic farms
relative to conventional ones makes it especially beneficial to increase organic farm size and
take advantage of economies of scale. The global productivity index proposed by Kumbhakar
et al. (2009) suggests that conventional farms are, on average, 12% more productive than their
organic counterparts. However, as will be seen below, the latter group of farms operates
closer to their production frontier than the former.
In Table 2.4, we observe that the estimate of  is close to one and highly significant,
indicating that inefficiency effects explain most of residual variation. As noted above, ten
explanatory variables are used as determinants of the inefficiency effects. Parameter estimates
of the inefficiency effects model are shown in Table 2.4. Apart from adoption of organic
practices, our results identify experience, family labor share in total labor, farm location and
farmer environmental preferences as the variables that are more relevant in explaining
technical inefficiencies. Our analysis reveals that holdings located in less-favored areas are

4
While variable input use was collected distinguishing between grape and non-grape activities, capital was not.
As a result, capital is not grape-specific. An alternative model weighting capital by the proportion of grape land
on total land was estimated and results, available upon request, changed very little.

22
less efficient compared to the other farms. As expected, farmers with more experience tend to
reach higher efficiency scores. This implies that TE increases‎ with‎ farmer’s‎ skills‎ and‎
practice. Farms that rely on a higher proportion of unpaid labor are found to be less efficient.
Farms, whose manager has strong environmental preservation preferences, tend to be less
efficient. Our results also show that the level of farm debt, subsidies, degree of farm
specialization, tenure regimes of land and the preferences regarding economic profit do not
have a significant impact on efficiency ratings. The dummy variable that reflects the
agronomic technique by identifying organic farms has a negative and statistically significant
sign, indicating that inefficiency decreases with the organic technology.
Technical efficiency scores for both farming methods are calculated as an output-
oriented measure and results are presented in Table 2.6 with decile ranges from the computed
frequency distribution. The histogram and kernel distributions of efficiency are plotted in
figure 2.1. The average TE score is 80% for organic farms and 64% for conventional ones. In
other words, organic (conventional) farmers reach 80% (64%) of their maximum potential
output. Moreover, these TEs range from a minimum of 17% (10%) to a maximum of 100%
(100%) for organic (conventional) farmers, indicating a lower dispersion in organic farming.
Almost 54% of organic farmers have efficiency ratings above 90%, whereas only 16% of
conventional farmers show these high performance levels. Therefore, our results indicate that
if organic (conventional) farms effectively used available resources and maintained current
technology, they would be able to increase their output by 20% (36%) on average. Improving
TE levels can reduce production costs and improve the economic viability of farms.

2.4. Discussion and concluding remarks

The present study aims to compare technical efficiency of organic and conventional grape
farms in Catalonia. Consistent with previous studies looking at the performance of organic
farming (Offerman and Nieberg, 2000; Oude Lansink et al., 2002; Oude Lansink and Jensma,
2003), we find that organic farming is on average 90% more profitable, on a per hectare basis,
than conventional farming. However, organic farms face higher production costs per hectare
and require more labor than conventional farms. This finding is compatible with previous
research on organic farming in Spain (Serra et al., 2008). Organic farms also exhibit
increasing returns to scale, while conventional farms operate under decreasing returns to
scale, meaning that organic farms could become more profitable with larger operations.

23
However, in line with previous literature (Tzouvelekas et al., 2001), conventional farmers are
found to be more worried about farm economic performance (profit), whereas the organic
group is more concerned about protecting the environment.
Our empirical findings suggest that organic farmers, on average, reach higher TE
ratings than their conventional counterparts (80% and 64%, respectively). Our results differ
from the findings by Bayramoglu and Gundogmus (2008), who assessed the efficiency of the
Turkish‎ grape‎ sector,‎ and‎ are‎ consistent‎ with‎ Tzouvelekas‎ et‎ al.’s‎ results‎ (2002a),‎ who‎
focused on the Greek grape sector. Higher efficiency scores attained by organic farms should
warrant their economic viability in the agricultural sector. Several reasons may explain the
higher average level of TE observed in organic farming. The higher costs per hectare
supported by organic farming are likely to motivate farmers to effectively use their inputs and
improve their agricultural performance. As noted by Tzouvelekas et al. (2001), information
on how to adequately apply organic farming techniques may be expected to improve
production performance. In this regard, the EU and national regulations concerning organic
farming may help organic farmers to be more efficient relative to their conventional
counterparts. Moreover, attractive organic price premiums can also explain the higher efforts
by organic farms to increase TE, given the high marginal income derived from production.
An interesting finding is the high elasticity of the expenditures in fertilizers and crop
protection products found in organic farming. Since organic farms cannot use non-authorized
chemical fertilizers and pesticides, organic fertilizers and biological controls are important
factors in organic grape production. Organic farms usually make a more rational and less
arbitrary use of these inputs relative to conventional farms. A more restricted and well-
managed use of these inputs contributes to explain the higher elasticity that they display in
organic farming.5
The low elasticity of labor in both types of farms can be explained by the high share of
family labor and the usual lack of qualified labor in this sector. Tzouvelekas et al. (2001,
2002a) found that family-operated farms are less efficient than farms with stronger
dependence on hired labor. Larsen and Foster (2005) also suggested that the share of hired
labor has a positive effect on TE for both organic and conventional farms. Another study
conducted by Lambarraa et al. (2007) concluded that a higher level of inefficiency may be
associated to a higher proportion of unpaid labor. More recently, Serra and Goodwin (2009),

5
The main difference between organic and conventional farms relies on the use of chemical inputs (mainly
fertilizers and pesticides), which is controlled by different regulations. The legal framework of organic farming
contributes to a rational use of these inputs.

24
found a negative labor elasticity characterizing the conventional technology indicating an
overuse of this input. The authors associated this result to the relevance of unpaid family labor
in their sample of farms. As we have seen in the descriptive analysis of sample farms, organic
farms are much more labor-intensive (on a per unit of land) than conventional farms. In spite
of this intensive use, labor elasticity is higher in organic than in conventional farming, which
is compatible with organic methods being more labor demanding than conventional practices.
Both types of farms (organic and conventional) suffer from relevant technical
inefficiencies. As suggested by previous findings (Tzouvelekas et al., 2001; Madau, 2007),
farms that are located in a less favored area tend to be less efficient. The finding is not
surprising given the environmental and production constraints faced by the first group. A
farmer who holds additional experience is more likely to have higher efficiency levels. This
implies‎ that‎ TE‎ increases‎ with‎ farmer’s‎ skills‎ and‎ practice.‎ In‎ line‎ with‎ the‎ findings‎ of‎
Tzouvelekas et al. (2001), farms with a higher proportion of unpaid labor are found to be less
efficient than farms with a stronger dependence on hired labor. Another interesting finding is
adoption of organic practices can improve technical efficiency under which farmers are
operating. However, organic farms show a lower productivity than conventional ones which is
compatible with Oude Lansink et al. (2002) results. TE‎ can‎ be‎ affected‎ by‎ farmers’‎
preferences regarding the need to preserve the environment. Producers that place a higher
value on preserving the environment through their production tend to be more inefficient.
Organic subsidies usually compensate organic farmers for reduced yields and adoption
costs. Though subsidies have been often criticized for making economic agents less
responsive to changing market conditions and increasing inefficiencies, our results show no
statistically significant effect of subsidies on efficiency. There are, however, a number of
things that policy makers can do to improve efficiency in grape farming. First, better
promoting extension services providing detailed information to farmers may be expected to
improve production performance. Second, since family labor is found to generate
inefficiencies, promoting a more professionalized management of agricultural holdings by
decreasing non-specialized family labor in favor of a more specialized labor force may
enhance the performance of organic farming.
Improving TE allows for a reduction in production costs and increases
competitiveness, which can help farmers face changing market conditions and economic
hardships. Farm margins can be squeezed when market conditions change, consumers become
more and more demanding and unwilling to pay higher price premiums, or middlemen in the
marketing chain and retailers increase their marketing power. In this regard, improving TE

25
can help farmers endure times of economic distress. Increasing profit levels can be achieved
by means of increased organic price premiums and subsidies, or alternatively, by means of
reduced production costs. A strategy based on cost reduction is especially relevant in the
organic sector.

26
References

Acs, S., Berentsen, P.B.M., Huirine, R.B.M., 2007. Conversion to organic arable farming in
the Netherlands: a dynamic linear programming analysis. Agricultural Systems 94: 405-
415.
Aigner, D., Lovell, C.A.K., Schmidt, P., 1977. Formulation and estimation of stochastic
frontier production functions models. Journal of Econometrics 6: 21-37.
Armesto-López, X.A., 2008. Organic farming in Spain: two case studies. Journal of
Sustainable Agriculture 31: 29-55.
Battese, G.E., Corra, G.S., 1977. Estimation of a production frontier model: with application
to the pastoral zone of Eastern Australia. Australian Journal of Agricultural Economics 2:
169-179.
Battese, G.E., Coelli, T.J., 1995. A model for technical inefficiency effects in a stochastic
frontier production function for panel data. Empirical Economics 20: 325-332.
Bayramoglu, Z., Gundogmus, E., 2008. Cost efficiency on organic farming: a comparison
between organic and conventional raisin-producing households in Turkey. Spanish Journal
of Agricultural Research 6: 3-11.
CCPAE, 2009. Statistics of the organic agricultural sector in Catalonia. Consell Català de la
Producció Agrària Ecològica.
http://www.ccpae.org/docs/estadistiques/2009/00_2009_ccpae_estadistiques.pdf. Accessed
20 February 2009.
Chakraborty, K., Misra, S., Johnson, P., 2002. Cotton farmers’‎technical‎efficiency:‎stochastic
and non-stochastic production function approaches. Agricultural and Resource Economics
Review 31: 211-220.
Coelli, T.J., 1995. Recent developments in frontier modeling and efficiency measurement.
Australian Journal of Agricultural and Resource Economics 39: 219-245.
Coelli, T.J., 1996. A guide to frontier version 4.1: a computer program for stochastic frontier
production and cost function estimation. CEPA Working Papers 7-96, Dept. Econometrics,
University of New England, Armidale, Australia.
EC, 2009. Rural development policy 2007-2013: aid to farmers in less favoured areas (LFA).
Council Regulation (EC) No 1257/99, European Commission.
http://ec.europa.eu/agriculture/rurdev/lfa/index_en.htm. Accessed 15 December 2011.
Eurostat, 2007. Organic farming and vineyard survey: database.
http://epp.eurostat.ec.europa.eu. Accessed 25 Janury 2009.

27
Eurostat, 2008. Agriculture statistics. Main results 2006-2007.
http://epp.eurostat.ec.europa.eu. Accessed 25 Janury 2009.
Fairweather, J.R., 1999. Understanding how farmers choose between organic and
conventional production: Results from New Zealand and policy implications. Agriculture
and Human Values 16: 51-63.
FiBL, 2009. Fibl statistics on organic farming in Europe. Research Institute of Organic
Agriculture, Frick, Switzerland. http://www.organic-europe.net. Accessed 12 August 2009.
Huang, C.J., Liu, J.T., 1994. Estimation of a non-neutral stochastic frontier production
function. Journal of Productivity Analysis 7: 171-180.
Karagiannias, G., Salhofer, K., Sinabell, F., 2006. Technical efficiency of conventional and
organic farms: some evidence for milk production. OGA Tagungsband.
Kumbhakar, S.C., Tsionas, E.G., Sipiläinen, T., 2009. Joint estimation of technology choice
and technical efficiency: an application to organic and conventional dairy farming. Journal
of Productivity Analysis 31: 151-161.
Lambarraa, F., Serra, T., Gil, J.M., 2007. Technical efficiency analysis and decomposition of
productivity growth of Spanish olive farms. Spanish Journal of Agricultural Research 5:
259-279.
Lampkin, N., 1996. European organic farming statistics 1996. Aberystwyth, UK: Welsh
Institute of Rural Studies.
Larsen, K., Foster, K., 2005. Technical efficiency among organic and conventional farms in
Sweden 2000-2002: a counterfactual and self selection analysis. Paper presented at the
American Agricultural Economics Association Annual Meeting, Providence, Rhode Island,
July 24-27.
Lohr, L., Salomonson, L., 2000. Conversion subsidies for organic production: results from
Sweden and lessons for the United States. Agricultural Economics 22: 133-146.
Madau, F.A., 2007. Technical efficiency in organic and conventional farming: Evidence form
Italian cereal farms. Agricultural Economics Review 8: 5-21.
MARM, 2008. Estadísticas 2008 Agricultura Ecológica España. Spanish Ministry of
Environment and Rural and Marine Affairs, Madrid.
Meeusen, W., Van Den Broek, J., 1977. Efficiency estimation from Cobb-Douglas production
functions with composed error. International Economic Review 18: 435-444.
Offerman, F., Nieberg, H., 2000. Economic performance of organic farms in Europe: organic
farming in Europe. Economics and Policy 5: 1-198.

28
Oude Lansink, A., Jensma, K., 2003. Analysing profits and economic behaviour of organic
and conventional Dutch arable farms. Agricultural Economics Review 4: 19-31.
Oude Lansink, A., Pietola, K.S., Backman, S., 2002. Efficiency and productivity of
conventional and organic farms in Finland 1994-1997. European Review of Agricultural
Economics 29: 51-65.
Padel, S., 2001. Conversion to organic farming: a typical example of the diffusion of an
innovation. Socialogia Ruralis 4: 40-61.
Parra López, C., De Haro Gimenez, T., Calatrava Requena, J., 2007. Diffusion and adoption
of organic farming in the Southern Spanish olive groves. Journal of Sustainable
Agriculture 30: 105-152.
Pietola, K.S., Oude Lansink, A., 2001. Farmer response to policies promoting organic farming
technologies in Finland. European Review Agricultural Economics 28: 1-15.
Pongracz, D.P., 1978. Practical Viticulture. Cape Town, South Africa: Iowa State University
Press.
Reifschneider, D., Stevenson, R., 1991. Systematic departures from the frontier: a framework
for the analysis of firm inefficiency. International Economic Review 32: 715-723.
Serra, T., Goodwin, B.K., 2009. The efficiency of Spanish arable crop organic farms, a local
maximum likelihood approach. Journal of Productivity Analysis 3:113-124.
Serra, T., Zilberman, D., Gil, J.M., 2008. Differential uncertainties and risk attitudes between
conventional and organic producers: the case of Spanish arable crop farmers. Agricultural
Economics 39: 219–229.
Snedecor, G.W., Cochran, W.G., 1989. Statistical methods. Ames, IA, USA: Iowa State
University Press.
Tzouvelekas, V., Pantzios, C.J., Fotopoulos, C., 2001. Technical efficiency of alternative
farming systems: the case of Greek organic and conventional olive-growing farms. Food
Policy 26: 549-569.
Tzouvelekas, V., Pantzios, C.J., Fotopoulos, C., 2002a. Empirical evidence of technical
efficiency levels in Greek organic and conventional farms. Agricultural Economics Review
3: 49-60.
Tzouvelekas, V., Pantzios, C.J., Fotopoulos, C., 2002b. Measuring multiple and single factor
technical efficiency in organic farming: the case of Greek wheat farms. British Food
Journal 104: 591-609.
Willer, H., Kilcher, L., 2009. The world of organic agriculture: statistics and emerging trends
2009. IFOAM, Bonn; Fibl, Frick; ITC, Geneva.

29
Winkler, A.J., Cook, J.A., Kliewer, W.M., Lider, L.A., 1974. General Viticulture. Berkeley,
CA, USA: University of California Press.

30
Table 2.1. Sample‎farms’‎agronomic, economic and demographic characteristics
Organic Conventional T-test of mean
Unit of
Variable name difference
measure Average SD1 Average SD1 Significance level2
Agronomic characteristics
Total land ha 18.90 12.82 45.33 75.68 0.00*
Proportion of land devoted to grape % 68.48 28.49 63.48 30.43 0.43
Proportion of land devoted to arable
% 10.58 16.94 19.11 29.03 0.05*
crops
Proportion of land devoted to fruits % 10.76 18.78 10.12 14.75 0.87
Proportion of land devoted to olive
% 10.18 15.33 8.61 11.93 0.63
groves
Total output kg 59,969.04 45,217.33 120,364.27 8,2454.49 0.00*
(0 low, 10
Soil quality 6.71 1.47 6.38 1.44 0.31
high)
(0 low, 10
Erosion 3.23 1.96 3.67 1.94 0.31
high)
Soil slope % 8.93 9.41 3.13 3.05 0.03*
Proportion of irrigated land % 15.62 30.26 7.15 19.78 0.18
(1 yes, 0
Farms in LFA3 0.53 0.25 0.00**
no)
Economic, structural and other characteristics
Output € 33,933.52 28,062.50 36,613.27 24,474.25 0.67
Price €‎kg-1 0.75 0.58 0.33 0.19 0.00*

31
Table 2.1. Sample‎farms’‎agronomic,‎economic‎and‎demographic‎characteristics (continued)
Organic Conventional T-test of mean
Unit of
Variable name difference
measure Average SD1 Average SD1 Significance level
Share of agricultural income in total
% 68.65 27.66 77.32 24.10 0.15
income
Share of output sold to processing
% 73.08 42.51 70.70 42.82 0.80
companies and cooperatives
Proportion of owned land % 46.31 45.81 44.56 37.94 0.86
Family labor share % 68.85 29.71 73.02 25.48 0.51
(1 yes, 0
Subsidy 0.69 0.58 0.30
no)
(1 yes, 0
Credit 0.27 0.50 0.03**
no)
(1 yes, 0
PDO4 association 0.60 0.68 0.47
no)
(1=10, 0
Economic profit preferences 0.46 0.54 0.47
otherwise)
Environmental preservation (1‎if‎≥8,‎0‎
0.92 0.70 0.02**
preferences otherwise)
Demographic characteristics
Age year 43.31 13.78 44.56 10.66 0.67
Years of experience year 15.42 9.90 18.16 11.63 0.23
number of
Family size 3.35 1.35 3.85 1.36 0.09
person
Statistics on a per hectare basis
Yield kg ha-1 6,848.31 3,261.73 8,173.19 3,177.51 0.02*
Revenue €‎ha -1
4,004.43 2,478.48 2,670.09 1,971.22 0.00*

32
Table 2.1. Sample‎farms’‎agronomic,‎economic‎and‎demographic‎characteristics (continued)
Organic Conventional T-test of mean
Unit of
Variable name difference
measure Average SD1 Average SD1 Significance level
Total revenue (revenue from grape
€‎ha-1 4,232.73 2,314.79 2,791.54 1,985.31 0.00*
and other farm activities)
Labor hours ha-1 458.93 240.41 285.76 303.34 0.00*
-1
Machinery N ha 0.66 0.53 0.49 0.71 0.18
Other variable inputs (farming
overheads and young vine plant €‎ha-1 860.51 751.98 834.94 1822.24 0.91
expenditures)
Fertilizers and crop protection €‎ha-1 380.92 579.56 294.12 399.34 0.48
Total cost (specific grape production
costs, farming overheads, labor €‎ha-1 1,813.93 1,421.67 1,508.55 1,922.69 0.38
costs)
Profit (total revenue minus total
€‎ha-1 2,435.07 2,293.04 1,282.99 2,805.43 0.04*
cost)
1 2 3
SD: standard deviation. *,** indicate statistical significance at the 5%, and chi-square statistical significance at the 5%, respectively. LFA: less favored areas.
4
PDO: protected designations of Origin.

33
Table 2.2. Summary statistics for the variables used in the analysis
Organic Conventional T-test of mean
Unit of
Variable name difference
measure Average SD1 Average SD1 Significance level2
Output y kg 59,969.04 45,217.33 120,364.27 8,2454.49 0.00*
Grape land X1 ha 8.44 4.94 14.22 7.55 0.00*
Labor X2 hours 3,084.52 1,109.65 2,891.92 1,461.70 0.46
Capital X3 machines 4.38 2.45 4.77 2.21 0.47
Fertilizers and crop protection X4 € 3,520.42 6,638.52 3,776.70 3,930.05 0.85
(1 organic,
Agronomic technique Z1 0.18 0.82
0 non-org.)
Experience Z2 year 15.42 9.90 18.16 11.63 0.23
Specialization Z3 % 72.19 29.76 74.41 26.43 0.73
(1 yes, 0
Farms not in LFA3 Z4 0.46 0.75 0.00**
no)
(1 yes, 0
Credit Z5 0.27 0.50 0.03**
no)
(1 yes, 0
Subsidy Z6 0.69 0.58 0.30
no)
Family labor share Z7 % 68.85 29.71 73.02 25.48 0.51
(1=10, 0
Economic profit preferences Z8 0.46 0.54 0.47
otherwise)
Environmental preservation (1‎if‎≥8,‎0‎
Z9 0.92 0.70 0.02**
preferences otherwise)
Owned land share Z10 % 46.31 45.81 44.56 37.94 0.86
1 2 3
SD: standard deviation. *,** indicate statistical significance at the 5%, and chi-square statistical significance at the 5%, respectively. LFA: less favored areas.

34
Table 2.3. Model specification tests
Restrictions Model λ  0.95
2
Decision
H 0 : ij  0 Cobb-Douglas 79.76 31.41 Reject
H 0 :  jk  0
* Neutral Stochastic frontier 50.56 55.76 Accept
H 0 :   m  0 No inefficiency effects 31.44 20.41 Reject
H0 :  0 No stochastic factor 90.27 5.14 Reject
H 0 : m  0 No firm- specific factors 39.90 19.67 Reject
H 0 : DBarcelona ; DTarragona  0 No regional dummies 2.00 5.99 Accept

35
Table 2.4. Maximum likelihood estimates for the stochastic production frontier model
Variable1 Parameter Estimate SE2
Frontier production function
Constant β0 0.524 0.021***
Constanto β0o -0.060 0.082
Land area β1 1.199 0.033***
Labor β2 0.080 0.043*
Capital β3 0.054 0.010***
Fertilizer and crop protection β4 -0.205 0.009***
Land areao β1o 0.335 0.186*
Laboro β2o -0.369 0.238
Capitalo β3o 0.018 0.084
Fertilizer and crop protection.o β4o -0.338 0.245
(Land area) × (Land area) β11 -0.264 0.074***
(Labor) × (Labor) β22 -0.210 0.067***
(Capital) × (Capital) β33 -0.389 0.088***
(Fertilizer and crop protection) ×
β44 -0.133 0.022***
(Fertilizer and crop protection)
(Land area) × (Labor) β12 -0.236 0.084***
(Land area) × (Capital) β13 0.535 0.104***
(Land area) × (Fertilizer and crop
β14 0.191 0.031***
protection)
(Labor) × (Capital) β23 -0.066 0.042
(Labor) × (Fertilizer and crop protection) β24 -0.018 0.099
(Capital) × (Fertilizer and crop protection) β34 -0.224 0.031***
(Land area) × (Land area)o β11o 0.628 0.189***
(Labor) × (Labor)o β22o 2.027 1.776
(Capital) × (Capital)o β33o -0.437 0.326
(Fertilizer and crop protection) × (Fertilizer
β44o -0.474 0.211**
and crop protection) o
(Land area) × (Labor)o β12o -1.334 0.685*
(Land area) × (Capital) o β13o 0.563 0.231**
(Land area) × (Fertilizer and crop
β14o -0.043 0.099
protection) o
(Labor) × (Capital) o β23o -0.126 0.525
(Labor) × (Fertilizer and crop protection) o β24o 0.684 0.304**
(Capital) × (Fertilizer and crop protection) o β34o -0.586 0.251**

36
Table 2.4. Maximum likelihood estimates for the stochastic production frontier model
(continued)
Variable Parameter Estimate SE2
Inefficiency effects model
Constant δ0 -0.523 0.941
Dc/o δ1 -1.180 0.315***
Experience δ2 -0.020 0.010*
Specialization δ3 -0.450 0.560
Farm is not located in a less favored area δ4 -0.534 0.286*
Credit δ5 0.035 0.248
Subsidy δ6 0.402 0.297
Family labor share δ7 0.961 0.513*
Economic profit preferences δ8 0.007 0.260
Environmental preservation preferences δ9 0.712 0.285***
Owned land δ10 -0.648 0.394
 2   v 2   u2 2 0.590 0.107***
   u2  2  0.999 4E-08***
log likelihood function -24.607
1
Superindex o represents the interaction of the variable with the organic farming dummy variable. 2 SE: standard error. ***,
** and * indicate statistical significance at the 1%, 5% and 10% respectively.

37
Table 2.5. Production and scale elasticities
Conventional Organic
Elasticities with respect to 1
Estimate SE Estimate SE1
Land area 0.558 0.024*** 0.323 0.138**

Labor 0.041 0.026 0.075 0.003***

Capital 0.165 0.017*** 0.323 0.024***

Fertilizer and crop protection 0.219 0.028*** 0.686 0.083***

Returns to scale 0.983 1.407

Productivity differential 0.12

(Kumbhakar et al., 2009)
1
SE: standard error. *** indicates that the parameter is significant at the 1%.

38
Table 2.6. Frequency distribution of technical efficiency (TE) for the conventional and
organic farms
TE: Range(%) Conventional (%) Organic (%)
<20 2 1.74 1 3.85
20-30 8 6.96 2 7.69
30-40 11 9.57 0 0.00
40-50 10 8.70 0 0.00
50-60 18 15.65 2 7.69
60-70 16 13.91 1 3.85
70-80 21 18.26 5 19.23
80-90 11 9.56 1 3.85
90-100 18 15.65 14 53.84
Sample size 115 100 26 100
Mean 64.25 79.63
SE1 22.64 25.68
Minimum 9.69 17.36
Maximum 99.99 99.98
1
SE: standard error

39
 60 Conventional … normal kernel
__ nonparametric kernel
50

40

30
P
e 20
r
10
c
e 0
60
Organic
n
50
t
40

30

20

10

0
0 0.2 0.4 0.6 0.8 1.0
technical efficiency

Figure 2.1. Histogram, normal and nonparametric densities of technical efficiency

40
Chapter 3

Technical efficiency of Kansas arable crop farms:

a local maximum likelihood approach6

6
Publication information: Guesmi, B., Serra, T., Featherstone, A.M., 2012. Technical efficiency of Kansas
arable crop farms: a local maximum likelihood approach. Agricultural Economics (first-round review).
42
3.1. Introduction

Technical efficiency is a prerequisite for economic efficiency, which in turn ensures the
economic viability and sustainability of a firm. Assessment of firms’‎ technical‎ efficiency‎
levels has drawn broad research interest. Such study is important for producers, as it assists
rational‎input‎allocation‎to‎achieve‎desired‎output‎levels,‎which‎strengthens‎a‎firms’‎capacity‎
to face changing market conditions, increasing input costs and economic hardships. It is also
relevant‎ for‎ policy‎ makers‎ interested‎ in‎ enhancing‎ firms’‎ economic‎ performance‎ and‎
competitiveness, and promoting economic development.
As is well known, the analysis of technical efficiency assesses to what extent firms are
able to maximize their output levels with minimum use of inputs. Two main approaches have
been widely used in the efficiency literature namely, parametric (Stochastic Frontier Analysis
- SFA) and nonparametric approaches (Data Envelopment Analysis - DEA) (Tzouvelekas et
al., 2001, 2002; Oude Lansink et al., 2002; Sipiläinen and Oude Lansink, 2005; Lohr and
Park, 2006). While both encompass several advantages, they are also characterized by a
number shortcomings. An important difference between these two approaches lies on the fact
that the stochastic production frontier (SPF) allows for the stochastic component of
production. This makes SFA suited to assess performance of production processes involving
random variables. Most agricultural technologies are stochastic in nature, due to unexpected
production changes resulting from weather influences and other factors that are not under the
farm control. Also, agricultural production studies may be affected by measurement and
variable omission errors, further emphasizing the relevance of stochastic approaches (Coelli,
1995; Chakraborty et al., 2002; Oude Lansink et al., 2002). The SFA further facilitates
inference, as it permits to conduct conventional statistical tests of hypotheses. However, this
approach presents important drawbacks: it relies on the assumption of a parametric functional
form representing the production frontier, as well as on a distributional assumption for the
random noise and inefficiency error components. Several studies show that technical
efficiency results are sensitive to estimation methods and functional form specifications
(Ferrier and Lovell, 1990; Coelli and Perelman, 1999; Ruggiero and Vitaliano, 1999;
Chakraborty et al., 2001). Inadequate parametric representations of the frontier and the error
distributions can lead to biased efficiency estimates (Kumbhakar et al., 2007; Martins-Filho
and Yao 2007; Serra and Goodwin, 2009).
Nonparametric DEA techniques overcome the most relevant limitations of SFA: they
do not rely on specific functional forms. However, nonparametric approaches do not allow for

43
stochastic variables and measurement errors, which precludes separating inefficiency effects
from random noise or random shocks, i.e., all production shortfalls are attributed to the
inefficiency term. As a result, technical efficiency ratings obtained from the nonparametric
approach (DEA) are generally lower than those obtained under the parametric alternative
(SFA) (Sharma et al., 1999; Puig-Junoy and Argiles, 2000; Wadud and White, 2000). Both
methods however have been found to lead to similar rankings of technical performance of
decision making units (DMUs).
Recently, a new methodological approach based on local modeling methods has been
developed (Kumbhakar et al., 2007) to overcome the limitations of parametric and non-
parametric approaches, without foregoing their advantages. In contrast to parametric models,
this method does not require strong assumptions regarding the deterministic and stochastic
components of the frontier: the parameters characterizing both production and error
distribution are allowed to depend on the covariates through a process of localization. As
opposed to nonparametric approaches, local modeling methods allow for stochastic variables
and variable measurement errors when estimating technical efficiency scores. Furthermore,
these techniques accommodate the heterogeneity in the data by deriving observation-specific
variances of the inefficiency and noise components of the error term (Serra and Goodwin,
2009). The local modeling approach by Kumbhakar et al. (2007) is based on local maximum
likelihood (LML) principles (Fan and Gijbels, 1996).
In spite of the interesting features of this approach, the complexity of implementing
the method has limited its use to a few empirical studies.7 The work by Serra and Goodwin
(2009) constitutes a notable exception. The present study focuses on estimating technical
efficiency ratings of a sample of cereals, oilseeds and protein crop (COP) farms in Kansas
using flexible LML methods that are compared with the results of DEA and SFA techniques.
Our article contributes to the scarce literature on the use of local modeling techniques to
assess technical efficiency. While the existing literature on technical efficiency has broadly
compared parametric (SFA) and nonparametric (DEA) approaches, to date, there is no study
that compares technical efficiency scores under DEA, SFA and LML. In addition, ours
constitutes the first study that assesses the efficiency of Kansas arable crop farms using local
modeling approaches (Rowland et al., 1998; Cotton et al., 1999; Serra et al., 2008). The
relevance of Kansas as a leading US producer of arable crops makes the analysis especially
interesting. In 2010, Kansas generated almost 20% and 50% of total wheat and sorghum

7
The software code to estimate the model is available upon request.

44
produced in the US, respectively. Kansas is also a leading corn and soybean producer, with
around 5% of the global US production. The relevant role of Kansas in US arable crop
production justifies our decision to study technical efficiency of Kansas arable crop farms.
The paper is organized as follows. In the next section we describe the methodology
used in our empirical analysis. The third section presents the data and results from the
empirical implementation. We finish the paper with concluding remarks.

3.2. Methodology

Several approaches have been used in the literature to assess firm-level performance. SFA and
the DEA constitute two mainstream techniques that have their intrinsic drawbacks and
advantages. One attractive advantage of the nonparametric DEA is that it does not require an
aprioristic specification of the frontier functional form and error distribution. However, DEA
methods ignore the stochastic component of production that may arise from unobserved
heterogeneity and measurement errors. This may lead to biased and misleading technical
efficiency measures. While this problem is addressed by the SFA, the major drawback of the
latter is that it relies on strong assumptions regarding specification of the production frontier
and the error distribution. The parametric approach is thus likely to be influenced by
misspecification issues and yield biased efficiency estimates. These shortcomings have been
widely discussed in the technical efficiency literature and several methodological
improvements have been proposed.
The nonparametric techniques by Cazals et al. (2002), Aragon et al. (2005), or Daouia
and Simar (2007)‎ are‎ robust‎ to‎ outliers,‎ but‎ still‎ rely‎ on‎ the‎ so‎ called‎ “deterministic”‎
assumption intrinsic of DEA (Kuosmanen et al., 2009). Kuosmanen (2006) proposed another
approach namely Stochastic Nonparametric Envelopment of Data (StoNED) which allows
combining stochastic frontier and a deterministic, nonparametric approach. However, this
method still requires a priori assumptions on inefficiency and noise distribution. The same
limitation can be attributed to Fan et al. (1996), who proposed a semiparametric method based
on a two-step pseudo-likelihood estimator. An alternative technique recently proposed by
Kumbhakar et al. (2007) overcomes the limitations of SFA and DEA, without foregoing their
advantages. Based on the LML principle proposed by Fan and Gijbels (1996), this new
approach localizes the parameters of the stochastic and the deterministic components of the
frontier model (flexibilized) with respect to the covariates.

45
 Since our analysis is based on a large number of Kansas farms over a broad
geographic region with different climatic conditions, heterogeneity is likely to characterize
the sample (different farm sizes, uneven skills, etc). The LML approach is suited to deal with
heteroscedasticity in both noise and inefficiency, as it localizes the standard errors
characterizing the distribution of efficiency and noise components of the error term. Based on
this approach, we seek to assess the technical efficiency with which COP Kansas farms
operate and compare efficiency ratings with scores derived from the DEA and SFA
alternatives. In the following lines, a description of the theoretical framework of the LML
stochastic frontier is presented.
The stochastic frontier models proposed by Aigner et al. (1977) and Meeusen and Van
den Broeck (1977) can be specified as follows Yi  0   T X i  ui  vi , where Yi and X i are

independent and identically distributed random variables. Yi represents output produced by

firm i  1,..., N and the vector representing input use is X i  d

. The betas are unknown

parameters. As usual, the stochastic frontier has a composite error term, ui  vi , where

ui  0 is the inefficiency term and vi is the random noise term. The parametric estimation of
stochastic frontier models requires definition of the joint probability density function (pdf) of

Y , X  , which is decomposed into a marginal pdf for X , pdf  x  =p(x) and a conditional pdf

for Y given x , pdf  y | x   g  y,  x   , where   x  

k
is the localized vector of parameters

to be estimated, and g is a function assumed to be known.

The LML is built upon the anchorage parametric model proposed by Aigner et al.

(1977). The conditional pdf for Y given X  x is defined as: Y  r  X   u  v . Following

Kumbhakar et al. (2007), the inefficiency term u is assumed to follow a half normal

distribution u | X  x 
N  0,  u2  x   , the error term v is assumed to have a normal

distribution ( v | X  x N  0,  v2  x   ) and u and v are assumed to be independently

distributed, conditional on X . The local polynomial approximation is used to estimate the

three dimensional local parameter vector   x    r  x  ,  u2  x  , v2  x   . The conditional log-

T

likelihood function is written as L    i 1 log g Yi ,  X i   and can be locally approximated

N

using an mth order local polynomial fit:

46
  
N
LN 0 ,1 ,...,m    log g Yi ,0  1  X i  x   ...  m  X i  x  K H  X i  x 
m
(1)
i 1

where x represents a fixed interior point in the support of the pdf p  x  ,  j   j1 ,..., jk  for
T

j  0,1,..., m , and K H  u   H K  H 1u  , being K a multivariate kernel function and H a

1

positive definite and symmetric bandwidth matrix. The local polynomial estimator ˆ  x  is

given by ˆ  x   ˆ0  x  where

ˆ  x  ,...,ˆ
0 m  x    arg max
0,...,
m
LN 0 ,1 ,..., m  (2)

The LML estimator can be derived using a local linear fit (Kumbhakar et al., 2007).
To do so, the random noise and inefficiency components are assumed to follow a local normal
and a half normal distribution, respectively. The conditional pdf of   v  u is specified as:

2       x 
f  | X  x         , (3)
  x     x      x  

where 
2
 x    u2  x    v2  x  ,   x    u  x   v  x  and  . and  . represent the

probability and the cumulative distribution functions of a standard normal variable,

respectively. The local linear parameter is given by   x    r  x  ,  2  x  ,   x   and the

T

conditional pdf of Y given X is expressed as:

 y  r  x     x 
g  y;  x         y  r  x  
2
   (4)
  x    x     x  

Therefore, the approximation of the conditional local log-likelihood function is specified as:

1 Yi  r  X i      Xi  
2
N
1
L      log   X i  
2

2  2  Xi 
 log     Y  r  X    (5)
   
i i
i 1 2
  2
X i 

47
In the present study, a local linear model for the frontier r  xi  and a local constant model for

the parameters of the error term is used that allows rewriting expression (5) as:

  Yi  r0  r1T  X i  x    
0  
2

 log    Yi  r0  r1  X i  x  
N
1 1
LN 0 , 1     log  0 
 2 T
 K H  X i  x  (6)
i 1  2 2  02   2 
  0 

where 0   r0 ,  02 , 0  and 1  r1T . The local linear estimator of the model is given by ˆ0 :
T

ˆ  x  ,..., ˆ  x   arg max L

0 1
0 ,1
N 0 , 1  . (7)

The local likelihood function (6) does not differ substantially from the conventional
likelihood function used in SFA (8). Observations in the former are weighted using the
multivariate kernel function ( K H ).

1 1 N N
  
Ln L   log  02 
2 02
 2
  log    i 20  (8)
 0 
i
2 i 1 i 1

The choice of the order of the local polynomial being fit may affect the quality and robustness
of the estimation. There exists a trade-off between bias and variance. While higher order fits
may be used with the purpose of reducing bias, variance in estimates may increase which may
lead to numerical instability. This is not optimal in the sense of minimizing the kernel
function, promoting the efficiency of the estimates and selecting the true bandwidth
(Cleveland and Loader, 1996; Hengartner et al., 2002; De Brabanter et al., 2013). Further, as
explained by Fan and Gijbels (1996), since the modeling bias is primarily controlled by the
bandwidth, the order of the local polynomial is less crucial. As a result, Fan and Gijbels
(1996) recommend the use of the lowest odd order polynomial determined as p    1,
where  represents the order of the derivative required, or occasionally p    3 . Hall and
Racine (2013) and Fan and Gijbels (1996) consider local linear regressions as one of the best
bias correction methods, specially in the boundary areas. Furthermore, several authors prefer

48
to choose a polynomial of order p  1 for computational ease considerations (Cleveland,
1979; Heij et al., 2004; Wu and Zhang, 2006; Hassouneh et al., 2012).
As noted above, LML allows deriving observation-specific estimates taking into
account the heterogeneity in inefficiency and noise terms. Following Jondrow et al. (1982),
the efficiency measure for a particular point can be obtained from the following expression:

uˆi 
 
ˆ 0  X i  ˆ0  X i    ˆ  X i  ˆ0  X i  ˆ 0  X i  ˆ  X i  ˆ0  X i  

 
(9)
1  ˆ 02  X i    ˆ  X i  ˆ0  X i  ˆ 0  X i  ˆ 0  X i  
 

where ˆ  X i   Yi  rˆ0  X i  . In the case of variables measured in logs, the efficiency score is

ˆ  exp  uˆ   0,1 . Since LML allows deriving local parameter estimates based
given by eff i i

on‎ kernel‎ regression,‎ each‎ farm’s‎ reference‎ set‎ is‎ more‎ homogeneous‎ relative‎ to‎ other‎
alternative efficiency-measurement techniques, which is likely to lead to higher efficiency
levels.
Finding a solution to the maximization problem in (7) requires specifying starting
values. To do so, we follow Kumbhakar et al. (2007) and start with the local linear least

squares estimator of r̂0  x  and r̂1  x  and the SFA estimators of ˆ 2 and  . The local

intercept r̂0  x  is corrected for the moment condition along the lines of the parametric
Modified Ordinary Least Squares (MOLS) estimator. Kumbhakar et al. (2007) recommend

using the following expression for such purpose rˆ0MOLS  x   rˆ0  x   2ˆ u2  ,

 
where ˆ u2  ˆ 2ˆ 2 1  ˆ 2 . Hence, initial values for solving (7) are obtained from

 
T
0  rˆ0MOLS , ˆ 2 , ˆ and 1  rˆ1  x  .
T

 
The product kernel chosen is h d  j 1  K h1  x j  , where K . represents the
d

Epanechnikov Kernel and d represents the number of covariates. Fan (1993) suggested that
using the Epanechnikov Kernel maximizes estimated efficiency. The bandwidth is adjusted
for different variable scales and sample sizes and is defined as: h  hbase sx N 1 5 ; where s x

represents the vector of empirical standard deviations of the covariates and N represents the
number of observations. The choice of the optimal value for hbase is based on the cross

49
validation criterion (CV) proposed by Kumbhakar et al. (2007). The CV, for a given value
of hbase , is computed by minimizing the following expression:

  
2
1 N 
CV  hbase    Yi  rˆ0   x   ui   ,
i i
 (10)
N i 1  

i  i 
where rˆ0 and ui are the leave-one-out versions of the local linear estimators defined above.
The cross-validation procedure, which involves estimating the model several times, leaving
one unit out at a time, allows controlling for the unobserved characteristics of observations
which in turn ensures the efficiency of estimates (Beck, 2001).
As‎ noted‎ above,‎ apart‎ from‎ Kumbakar‎ et‎ al.’s‎ (2007)‎ LML‎ proposal,‎ efficiency‎ of‎
Kansas farms is also assessed by DEA and SFA approaches. The random parameter approach
is used to derive SFA estimates. Following Greene (2002), the simulated log-likelihood
function assuming normal and half normal distributions can be defined as:

  /  ]  [(Y   X ) ]    1    (Y it  ir X it ) 
2

Ln LS   i 1
N 1

R
log
1
 log   i it ir it
  log   i   log  2  i (11)
R r 1
2   u2  v2   u  2 2 2
 

where r represents the number of replications and t indicates time period. The DEA linear
programming model can be expressed as (Färe et al., 1994):

max 
 ,

s.t.
 yi  Y   0 (12)
xi  X   0
N 1  1
0

where 1     , N is the number of farms, X is a d  N matrix of inputs, Y is a 1 N

matrix of outputs. Technical efficiency scores are given by 1  . The constraint N1  1 is
included to allow for variable returns to scale (VRS). As is well known, without such
constraint, constant returns to scale (CRS) are assumed (Charnes et al., 1994). To test for

50
divergence between the efficiency distributions obtained from LML, SFA and DEA methods,
the standard Kolmogorov-Smirnov (KS) two-sample (two-tail) test statistic is conducted:

D  max F a ( x, N )  F b ( x, N ) (13)

where F a ( x, N ) represents the empirical distribution function for a sample a with total
observations N .

3.3. Data and results

3.3.1. Data

The empirical application focuses on a sample of Kansas farms that specialize in the
production of COP crops. Farm-level data are obtained from farm account records from the
Kansas Farm Management Association (KFMA) dataset and cover the period 2000-2010.
Data available include farm production and input use, financial and socio-economic
characteristics, as well as farm structural characteristics. To ensure that COP is the main farm
output, farms whose COP sales represent at least 90% of total farm income were selected.
This criterion allows obtaining a relatively homogeneous sample of farms. The dataset is an
unbalanced panel that contains 1,258 observations.
We define farm output ( yi ) as an implicit quantity index that is computed as the ratio
of production in currency units to the output price index. Since information on market prices
is unavailable at the farm-level, a Paasche price index is built on the basis of state-level cash
unit prices and production data. Output yi includes the predominant crops in Kansas
(Albright, 2002): wheat, corn, soybean and sorghum. The inputs considered as explanatory
variables are COP land ( x1 ) measured in acres, total labor input ( x2 ), mainly composed of
family labor, and expressed in annual working units (AWUs), as a fraction of 10-hours per
day, chemical inputs ( x3 ), other inputs ( x4 ) and capital ( x5 ). Chemical inputs are defined as
a quantity index that includes the use of fertilizers and pesticides, and is obtained by dividing
input expenditures by its corresponding price index. Other inputs, also defined as a quantity
index, include fuel and seed expenses. Capital input ( x5 ) aggregates the value of machinery,
other equipment and buildings used in the production process, and is determined by dividing

51
capital value by its corresponding price index. Input prices are measured using national input
price indices. Monetary values are measured at constant 2000 prices. Data unavailable from
the Kansas database include country-level input price indices and state-level output prices and
quantities and are obtained from the United States Department of Agriculture (USDA) and the
National Agricultural Statistics Service (NASS).
Table 3.1 provides summary statistics for the variables used in the analysis. Sample
farms use, on average, 293 AWUs, of which 82% represents unpaid family labor. In contrast
to the European Union (EU) arable crop farms that are mainly small holdings with around 116
acres (Farm Accountancy Data Network, FADN 2012), Kansas farms devote 1,278 acres on
average to COP production. More than 80% of the COP area is allocated to wheat, soybeans,
sorghum and corn production. The average value of farm production (around 154 thousand
dollars) almost doubles the EU value (about 84 thousand dollars). However, per acre statistics
suggest that EU farms are much more intensive than Kansas farms: while EU farms have an
average‎income‎of‎441‎dollars‎per‎acre,‎Kansas‎income‎is‎122‎dollars‎per‎acre.‎Sample‎farms’‎
investments in machinery and buildings are on the order of 163 thousand dollars. On per acre
basis, Kansas farms are less intensive in capital use (150 dollars per acre) relative to the EU
farms with investment ratios on the order of 1,666 dollars per acre (FADN, 2012). To ensure
immunity against pests and diseases and to avoid productivity loss due to pest infestations,
Kansas farmers spend around 38 thousand dollars annually on chemical inputs. On a per acre
basis, expenses in fertilizers and crop protection products are much higher in EU farms (178
dollars per acre versus 29 dollars per acre). Expenses in other inputs, seeds and energy is
rather low compared to chemical input costs, and on the order of 24 thousand dollars.

3.3.2. Empirical results

Using the aforementioned variables and following Kumbhakar et al. (2007), we specify the
anchorage parametric model as a Cobb-Douglas function:

5
log Y  0    j log X j  u  v (14)
j 1

It is relevant to note that rigidities associated to this production frontier are overcome by
estimating the frontier for each observation in the sample, i.e., flexibility is achieved through
varying parameter estimates. To select the bandwidth required to derive the LML estimator of

52
(14), we use the CV procedure described above. It is worth noting that with multiplicative
multivariate kernels, an observation i will only be considered in the LML estimation if all

covariates xi fall into the interval  xi  hi , xi  hi ) ; where hi  hbase sxi N . If even one of the
1 5

components fails to fall into this interval, the observation will not be considered for the
estimation. Such procedure requires relatively large values for hbase in order to have a
sufficiently large subsample of observations to locally estimate the stochastic production
frontier.8 The more important the sample heterogeneity is, the bigger the required bandwidth.
We start with a crude grid of values to then focus on a finer grid for the selection of the
optimal hbase according to (10). Final results show that at the optimal hbase  11 , the

bandwidths h1 , h2 , h3 , h4 and h5 take values of 2.38, 2.34, 3.37, 3.47 and 2.86, respectively.
The number of observations at each data point is, on average, is 1,041. The distribution of the
number of observations at each data point is presented in figure 3.1 below. Once we select the
adequate bandwidth for our data, we then derive local parameter estimates. 9
Descriptive statistics for the variation of local estimates of  u2 and  v2 are shown in
table 3.2. These statistics confirm the presence of heteroscedasticity and indicate an important
degree of variation among observations regarding the shares of the inefficiency term to the
noise term (    u2 /  v2 ). Figure 3.2 illustrates the variation of the parameters of the
deterministic component of the frontier. Since we use a Cobb–Douglas functional form,
coefficients represent input elasticities. Variation of the localized estimates suggests that
assuming the same input elasticities for all observations may not be reliable. Variation is
specially relevant for land, with an elasticity that ranges from 19% to 43%, followed by
chemical inputs, labor and capital, that have an elasticity fluctuating from 26% to 42%, 2% to
14% and 21% to 33%, respectively. Input elasticities indicate that the average farm operates
under constant returns to scale with a mean scale elasticity equal to 1.005 and a standard
deviation of 0.089.

8
A minimum of 9 observations is required and this was imposed.
9
The monotonicity condition of production functions implies that production should monotonically increase in
all inputs, and is certainly an important concept in efficiency analyses (Henningsen and Henning, 2009). While
DEA implicitly imposes monotoncity, we impose it in SFA and LML techniques. Technical efficiency
measurement generally assumes that producers maximize output given input quantities, but not that producers
maximize their profit. Thus, in contrast to monotonicity, there is not necessarily a technical motivation for a
production function to be quasi-concave (Henningsen and Henning, 2009).

53
 Production elasticity estimates show, on average, that chemical input use has the
highest potential to increase output, followed by land, capital, other inputs and labor (table
3.3). The low contribution of labor to farm production increases can be attributed to the high
share of family labor. Since this labor type usually involves an opportunity cost but not an
actual cost, incentives to use it efficiently may be weaker than for other inputs. The fact that
capital, land and other inputs have lower elasticities than chemicals suggests that the latter are
used less intensively. Input cost shares can be used as a reference for estimated elasticities.
Under perfect competition and constant returns to scale assumptions, output elasticity with
respect to input should equal the input cost share (Shumway and Talpaz, 1980; Krishnapillai
and Thompson, 2012). Table 3.3 shows that there is no substantial difference between shares
and elasticities, which provides supporting evidence of the reasonability of our findings, i.e.,
our empirical findings suggest that Kansas farms are likely to be operating under perfect
competition and constant returns to scale.
Small differences between estimated output production elasticities of inputs and
observed factor shares can be attributed to production spillovers, excess returns, omitted
variables, or measurement errors (Stiroh, 2002). Average estimated elasticities of chemical
inputs (0.293), other inputs (0.199) and labor (0.048) are slightly below average factor shares
(0.334, 0.241, 0.053, respectively). In contrast, land area and capital input elasticities are
higher than factor shares (0.248 and 0.217 vs. 0.173 and 0.197, respectively). This suggests
that land and capital productivity outweighs marginal costs and that these two inputs are
under-used.
Table 3.4 illustrates the distribution of LML efficiency estimates. The same table also
presents the distribution under the alternative DEA and SFA approaches. A translog
production function10 is defined as the anchorage model for SFA (equation 15), which is
estimated using the random parameter technique.

5 5 5
log Y  0    j log X j  12   jk log X j log X k  u  v (15)
j 1 j 1 k 1

Localized technical efficiency estimates show a high average score, on the order of 0.905,
indicating that farmers reach about 91% of their maximum potential output. Therefore, our

10
To economize space, parameter estimates of the translog are not presented. However, results are available
upon request.

54
sample farms could increase their output by a further 9% by simply using their inputs more
efficiently, without incurring extra input costs or adopting new technologies. In the presence
of inefficiencies, the use of existing technologies is more cost effective as a means to improve
output than adopting new technologies (Shapiro, 1977; Belbase and Grabowski, 1985).
Average DEA-CRS efficiency scores (0.808) are lower than LML ratings. Under VRS,
however, efficiency evaluations are much closer to LML results (0.917).11 Average SFA
technical efficiency scores (0.804) are also below LML scores. At the 5% level of
significance,‎the‎KS‎test‎indicates‎that‎DEA‎and‎SFA‎score‎distributions‎differ‎from‎LML’s‎
(table 3.5). Given the fact that LML overcomes the most relevant limitations of DEA and
SFA methods, its robustness should be higher.12 Reliable information about farm efficiency
performance is relevant to identify inefficient farms and define adequate policy and
management strategies. Defining these strategies based on DEA or SFA estimates, may lead
to targeting the wrong farms, while overlooking inefficient farms that need to improve their
performance levels. As shown in table 3.4, more than a half and 40% of Kansas farms are
assigned efficiency scores below 80% when using DEA-CRS and SFA estimates,
respectively, while only 6% of total observations exhibit this performance under LML.
Hence, both DEA-CRS and SFA approaches are likely to overestimate inefficiency.
In the following lines, we compare the results derived in our analysis with findings by
previous research. Following Balcombe et al. (2006), the purpose of this comparison is to
check the confidence and robustness of our findings, i.e., whether they concur or not with
other results derived from different methods and whether they are or not within the range of
existing estimates in the literature. Our results differ from those in Serra et al. (2008) who
used the same database, but focused on the period 1998-2001.‎ Through‎ Kumbhakar’s‎
stochastic frontier model (2002), Serra et al. (2008) obtained mean technical inefficiency
levels‎ of‎ 0.30,‎ versus‎ 0.09‎ in‎ our‎ analysis.‎ The‎ use‎ of‎ different‎ methodologies‎ or‎ farmers’‎
performance improvement over time can explain differences in efficiency scores across
studies. However, our results are closer to other findings by Rowland et al. (1998) for a
sample of Kansas swine operations from 1992 through 1994, or Cotton et al. (1999), for a
sample of multi-output Kansas farms during the period 1985 to 1994. Both authors used

11
DEA results suggest that Kansas farms do not operate at optimal scale.
12
Robustness assessment requires simulation exercises. Kumbhakar et al. (2007) show how LML outperforms
parametric frontier methods in a number of situations and conclude that LML should always be preferred to
traditional MLE techniques with anchorage models. Formal comparison (robustness assessment) between DEA
and LML has not been conducted, which offers scope for future research.

55
nonparametric DEA techniques to derive efficiency estimates and obtained mean efficiency
scores of 0.89 and 0.91, respectively.
Under the LML approach, technical efficiencies range from a minimum of 0.095 to a
maximum of one, indicating important dispersion and heterogeneity within Kansas farms.
Using LML technical efficiency levels to identify cluster membership, cluster analysis
classifies Kansas farms into three distinct groups. In order to characterize the groups,
differences in farm and farmer characteristics across clusters are assessed using anova and
cross tabulation analyses, depending on the quantitative or categorical nature of the data.
These characteristics include farm size (categorically defined through three dummy variables
representing farms that cultivate less than 500 acres, between 500 and 1000 acres, and more
than 1000 acres); proportion of irrigated land (share of irrigated land to total land); rented
land share (proportion of rented land to total land); unpaid labor share (proportion of family
labor to total labor); farm output (defined as a quantity index as explained above) and yields;
farm manager age; education expenses; public subsidies and non-agricultural income
measured in dollars. Results of this analysis are presented in table 3.6. On average, the first,
second and third groups have an efficiency level of 0.78, 0.88 and 0.96, respectively.
Anova and cross tabulation analysis results (table 3.6) show that increased farm size,
measured as the extension of cultivated land, brings higher efficiency, which is compatible
with the presence of increasing returns to scale in small farms. Farm size differences across
efficiency groups are statistically significant and lead to different output indices and yields.
Efficiency is further positively related to output and yields, being differences across groups
statistically significant.
Although all farm types strongly rely on rainfed agriculture, irrigation practices are
relatively more important among medium and high efficiency groups. While all three farm
types mainly use unpaid family labor, a higher relevance of family labor is associated to
poorer performance. Hence, it seems that actual costs of paid labor exert a positive influence
on farm performance, relative to the opportunity costs of unpaid workforce. Along the same
lines, those farms with better efficiency levels are the ones that can afford higher land rental
costs.
Farmers with higher education expenses tend to be more efficient. However, the
difference across groups is not statistically significant. An interesting finding is that younger
farmers are likely to be more technically efficient. Subsidies received from government differ
across clusters and have a positive relationship with efficiency, which may be due to the fact
that subsidies are paid based on farm size. Non-agricultural activities show a significant

56
negative relationship with efficiency, suggesting that farmers who diversify their income
sources by conducting off-farm activities tend to be less efficient.
Improving efficiency performance may require actions at the farm, policy and
academic levels. Academically, further research could be conducted so as to identify
additional inefficiency causes, as well as their marginal impact on efficiency. Those elements
with higher marginal impacts on efficiency should be the ones receiving further attention by
farm managers and policy makers. Refined methods including risk and risk attitudes may
allow more accurate efficiency estimates. We find evidence that higher education and
professionalized labor force may lead to higher efficiency. As a result, any action aimed at
promoting high-quality‎and‎skilled‎work‎is‎likely‎to‎increase‎farms’‎economic‎sustainability.‎
At the policy level this could be achieved through information campaigns and extension
services. Similar results should be obtained through skilled labor force development
initiatives adopted by the farm manager (education and training actions). Efficiency is also
positively‎ related‎ to‎ a‎ farm’s‎ size.‎ Increased‎ farm‎ size‎ may‎ be‎ pursued‎ through land
acquisitions and rental. To the extent that land market rigidities are relevant, land rental may
offer a flexible alternative that may even result in higher profits than purchasing own land
(actual rental costs induce higher efficiency relative to the opportunity costs of own land).
Adopting irrigation practices or promoting farm specialization constitute other efficiency
improving alternatives. Generational change in farming may also prove to be very useful.

3.4. Concluding remarks

The relevance of deriving‎ reliable‎ technical‎ efficiency‎ scores‎ to‎ assist‎ firms’‎ management‎
decisions as well as policy design, makes it essential to use methodologies that produce farm-
level non-biased efficiency ratings. The parametric SFA and the nonparametric DEA
approaches have focused the attention of mainstream efficiency literature. Both approaches
have been widely criticized for their shortcomings that may lead to biased efficiency
estimates.
Recently, Kumbhakar et al. (2007) proposed a new approach, namely the LML
method. The method estimates the parameters of the deterministic and stochastic components
of the frontier locally. LML methods overcome the shortcomings of SFA and DEA without
foregoing their advantages. However, some of the major drawbacks of this approach are the
“curse‎of‎dimensionality”‎and‎the‎estimation‎convergence‎issues‎that‎are‎likely‎to‎arise.‎These‎
drawbacks do not allow estimating too many parameters, thus restricting the alternative

57
functional forms that can be considered. In any case, rigidities associated to simple production
function specifications are overcome through varying parameter estimates. Another LML
limitation is the selection of the order of the local polynomial, as there is a trade-off between
bias and variance of the estimation. High order polynomials will reduce the bias, but at the
cost of the variance of fitting. In spite of the complexity of this approach, it is highly
recommended to derive reliable and unbiased estimates. LML techniques are used in this
article to assess the efficiency levels achieved by Kansas farms specialized in cereals, oilseeds
and protein crop production and compares them with those obtained from DEA and SFA
models. Farm-level data obtained from farm account records from the KFMA dataset
covering the period 2000-2010 are used.
Empirical results support the relevance of using the LML approach through the
variation in the localized parameter estimates, representing the variance of the composite error
term and input elasticities. Results show high mean efficiency scores (0.905) indicating that
farmers could increase their output by 9% keeping their input bundle constant. Technical
efficiency scores derived from the LML approach are higher than those of the DEA-CRS and
SFA models, but close to DEA-VRS ratings. According to the KS test, the efficiency score
distributions obtained from DEA and SFA differ from LML distribution ratings. Since LML
allows both for stochastic error terms, as well as for flexibility in the functional form
representing the frontier function, efficiency scores derived under LML should be more
reliable and less biased than efficiency ratings under nonparametric DEA and SFA
alternatives.
One limitation of our analysis is the use of a single output instead of a multi-output
technology (in which sorghum, wheat, soybean and corn output would be considered
separately). However, Nauges et al. (2011) suggested that using multiple outputs can conduct
to biased estimates due to the endogeneity problem. Our research can be extended in many
different ways. Different methodological innovations to assess efficiency have been recently
introduced in the literature. Noteworthy are the refinements regarding the measurement of
technical efficiency in the presence of uncertainty through state-contingent techniques
(Chambers and Quiggin, 2000). Failure to properly allow for risk can lead to biased efficiency
estimates‎ (O’Donnell‎ et‎ al.,‎ 2010).‎ Extension of LML methods to a consideration dynamic
issues constitutes another area that merits further attention. This is left for the near future
research as a means to improve the specification of the frontier technology.

58
References

Aigner, D., Lovell, C.A.K., Schmidt, P., 1977. Formulation and estimation of stochastic
frontier production functions models. Journal of Econometrics 6: 21-37.
Albright, M.L., 2002. Kansas Agriculture: An economic overview for 2001, Research Papers
and Presentations, Kansas State University (KSU) Research and Extension.
http://www.agmanager.info/farmmgt/income/papers/. Accessed 20 November 2011.
Aragon, Y., Daouia, A., Thomas-Agnan, C., 2005. Nonparametric frontier estimation: a
conditional quantile-based approach. Econometric theory 21: 358-389.
Balcombe, K., Fraser, I., Kim, J.H., 2006. Estimating technical efficiency of Australian dairy
farms: a comparison of alternative frontier methodologies. Applied Economics 38: 2221-
2236.
Beck, N., 2001. Time-series-cross-section data. Statistica Neerlandica 55: 111-133.
Belbase, K., Grabowski, R., 1985. Technical efficiency in Nepalese agriculture. Journal of
Development Areas 19: 515-525.
Brümmer, B., 2001. Estimating confidence intervals for technical efficiency: the case of
private farms in Slovenia. European Review of Agricultural Economics 28: 285-306.
Cazals, C., Florens, J.P., Simar, L., 2002. Nonparametric frontier estimation: a robust
approach. Journal of Econometrics 106: 1-25.
Chakraborty, K., Biswas, B., Lewis, W.C., 2001. Measurment of technical efficiency in
public education: A stochastic and nonstochastic production function approach. Southern
Economic Journal 67: 889-905.
Chakraborty, k., Misra, S., Johnson P., 2002.‎Cotton‎farmers’‎technical‎efficiency:‎Stochastic‎
and non-stochastic production function approaches. Agricultural and Resource Economics
Review 31: 211-220.
Chambers, R., Quiggin, J., 2000. Uncertainty, production choice and agency: The state-
contingent approach. Cambridge, UK: Cambridge University Press.
Charnes, A., Cooper, W., Lewin, A., Seiford, L., 1994. Data envelopment analysis, theory,
methodology and applications. Norwell, MA: Kluwer Academic Publishers.
Cleveland, W.S., 1979. Robust locally-weighted regression and smoothing scatterplots.
Journal of the American Statistical Association 74: 829–836.
Cleveland, W.S., Loader, C., 1996. Smoothing by local regression: principles and methods.
New York: Springer.

59
Coelli, T.J., 1995. Recent developments in frontier modeling and efficiency measurement.
Australian Journal of Agricultural and Resource Economics 39: 219-245.
Coelli, T.J., Perelman, S., 1999. A comparison of parametric and non-parametric distance
functions: With application to European railways. European Journal of Operational
Research 117: 326-339.
Cotton, M.K., Langemeier, M.R., Featherstone, A.M., 1999. Effects of weather on Multi-
Output: Efficiency of Kansas Farms. Journal of the ASFMRA 62: 85-91.
Daouia, A., Simar, L., 2007. Nonparametric efficiency analysis: a multivariate conditional
quantile approach. Journal of Econometrics 140: 375-400.
De Brabanter, K., De Brabanter, J., De Moor B., Gijbels I., 2013. Derivative Estimation with
Local Polynomial Fitting. Journal of Machine Learning Research 14: 281-301.
FADN, 2012. FADN public database.
http://ec.europa.eu/agriculture/rica/database/database_en.cfm. Accessed 16 April 2012.
Fan,‎J.,‎1993.‎Local‎linear‎regression‎smoothers‎and‎their‎minimax‎efficiencies.‎The Annals of
Statistics 21: 196-216.
Fan, J., Gijbels, I., 1996. Local polynomial modeling and its applications. London, UK:
Chapman & Hall.
Fan, Y., Li, Q., Weersink, A., 1996. Semiparametric estimation of stochastic production
frontier models. Journal of Business and Economic Statistics 14: 460-468.
Färe, R., Grosskopf, S., Lovell, C.A.K., 1994. Production frontiers. Cambridge, UK:
Cambridge University Press.
Ferrier, R.S., Lovell, C.A.K., 1990. Measuring cost efficiency in banking: An econometric
and linear programming evidence. Journal of Econometrics 46: 229-245.
Greene, W., 2002. Alternative panel data estimators for stochastic frontier models. Paper
presented at the conference on current developments in productivity and efficiency
measurement. University of Georgia, USA, October 25-26.
Hall, P.G., Racine, J.S., 2013. Infinite order cross-validated local polynomial regression.
http://aysps.gsu.edu/sites/default/files/documents/publications/February%2028%20Semina
r.pdf. Accessed 15 April 2013.
Hassouneh, I., Serra, T., Goodwin, B.K., Gil, J.M., 2012. Non-parametric and parametric
modeling‎of‎biodiesel,‎sunflower‎oil,‎and‎crude‎oil‎price‎relationships.‎Energy Economics
34: 1507-1513.
Heij, C., De Boer, P., Franses, P.H., Kloek,T.,VanDijk, H.K., 2004. Econometric methods
with applications in Business and Economics. Oxford, UK: Oxford University Press.

60
Hengartner, N.W., Wegkamp, M.H., Matzner-Løber, E., 2002. Bandwidth selection for local
linear regression Smoothers. Journal of the Royal Statistical Society: Series B (Statistical
Methodology) 64: 791-804.
Henningsen, A., Henning, C.H.C.A., 2009. Imposing regional monotonicity on translog
stochastic production frontiers with a simple three-step procedure. Journal of Productivity
Analysis 32: 217-229.
Jondrow, J., Lovell, C.A.K., Materov, I.S., Schmidt, P., 1982. On the estimation of technical
inefficiency in stochastic frontier production models. Journal of Econometrics 19: 233-
238.
Krishnapillai, S., Thompson, H., 2012. Cross section translog production and elasticity of
substitution in U.S. manufacturing industry. International Journal of Energy Economics
and Policy 2: 50-54.
Kumbhakar, S.C., 2002. Specification and estimation of production risk, risk preferences and
technical efficiency. American Journal of Agricultural Economics 84: 8-22.
Kumbhakar, S.C., Byeong, U.P., Simar, L., Tsionas E.G., 2007. Nonparametric stochastic
frontiers: A local maximum likelihood approach. Journal of Econometrics 137: 1-27.
Kuosmanen, T., 2006. Stochastic nonparametric envelopment of data: combining virtues of
sfa and dea in a unified framework. MTT Discussion Paper No. 3/2006.
Kuosmanen, T., Johnson, A., Kortelainen, M., 2009. Stochastic DEA: myths and
misconceptions. Paper presented at the XIth EWEPA Conference. Pisa, Italy, June 23-26.
Li, Q., 1999. Nonparametric testing the similarity of two unknown density functions: Local
power and bootstrap analysis. Journal of Nonparametric Statistics 11: 189-213.
Lohr, L., Park, T.A., 2006. Technical efficiency of US organic farmers: The complementary
roles of soil management techniques and farm experience. Agricultural and Resource
Economics Review 35: 327-338.
Martins-Filho, C., Yao, F., 2007. Nonparametric frontier estimation via local linear
regression. Journals of Econometrics 141: 283-319.
Meeusen, W., Van Den Broek, J., 1977. Efficiency estimation from Cobb-Douglas production
functions with composed error. International Economic Review 18: 435-444.
Nauges,‎ C.,‎ O’Donnell,‎ C.J.,‎ Quiggin,‎ J.,‎ 2011.‎ Uncertainty‎ and‎ technical‎ efficiency in
Finnish agriculture: a state-contingent approach. European Review of Agricultural
Economics 38: 449-467.
O’Donnell,‎ C.‎ J.,‎ Chambers,‎ R.G., Quiggin, J., 2010. Efficiency analysis in the presence of
uncertainty. Journal of Productivity Analysis 33: 1-17.

61
Oude Lansink, A., Pietola, K., Bäckman, S., 2002. Efficiency and productivity of
conventional and organic farms in Finland 1994–1997. European Review of Agricultural
Economics 29: 51-65.
Puig-Junoy, J., Argiles, J.M., 2000. Measuring and explaining farm inefficiency in a panel
data set of mixed Farms. Pompeu Fabra University, Working Paper, Barcelona, Spain.
Rowland, W.W., Langemeier, M.R., Schurle, B.W., Featherstone, A.M., 1998. A
nonparametric efficiency analysis for a sample of Kansas swine operations. Journal of
Agricultural and Applied Economics 30: 189-199.
Ruggiero, J., Vitaliano, D., 1999. Assessing the efficiency of public schools using data
envelopment analysis and frontier regression. Contemporary Economic Policy 17: 321-
331.
Serra, T., Goodwin, B.K., 2009. The efficiency of Spanish arable crop organic farms, a local
maximum likelihood approach. Journal of Productivity Analysis 3: 113-124.
Serra, T., Zilberman, D., Gil J.M., 2008. Farms’‎ technical‎ inefficiencies‎ in‎ the‎ presence‎ of‎
government programs. The Australian Journal of Agricultural and Resource Economics
52: 57-76.
Shapiro, K.H. 1977. Sources of technical efficiency: the roles of modernization and
information. Economic Development and Cultural Change 25: 235-243.
Sharma, K.R., Leung, P., Zaleski, H.M., 1999. Technical, allocative and economic
efficiencies in swine production in Hawaii: A comparison of parametric and nonparametric
approaches. Agricultural Economics 20: 23-35.
Shumway, C.R., Talpaz, H., 1980. U.S aggregate agricultural production elasticities estimated
by an ARIMA factor share adjustment model. Southern Journal of Agricultural Economics
12: 119-125.
Simar, L., Zelenyuk, V., 2006. On testing equality of distributions of technical efficiency
scores. Econometric Reviews 25: 497-522.
Sipiläinen, T., Oude Lansink A., 2005. Learning in organic farming-an application on Finnish
dairy farms. Manuscript presented at the XIth Congress of the EAAE. Copenhagen,
Denmark, August 24–17.
Stiroh, K. J., 2002. Are ICT spillovers driving the new economy? Review of Income and
Wealth 48: 33-57.
Tzouvelekas, V., Pantzios, C.J., Fotopoulos, C., 2001. Technical efficiency of alternative
farming systems: The case of Greek organic and conventional olive-growing farms. Food
Policy 26: 549–569.

62
Tzouvelekas, V., Pantzios, C.J., Fotopoulos, C., 2002. Empirical evidence of technical
efficiency levels in Greek organic and conventional farms. Agricultural Economics Review
3: 49-60.
USDA, 2011. Agricultural prices summary, crop values annual summary and crop production
annual summary. http://usda.mannlib.cornell.edu/MannUsda. Accessed 13 August 2011.
Wadud, A., White, B., 2000. Farm household efficiency in Bangladesh: A comparison of
stochastic frontier and DEA methods. Applied Economics 32: 165-173.
Wu, H., Zhang, J., 2006. Nonparametric regression methods for longitudinal data Analysis:
mixed-effects modeling approaches. A John Wiley & Sons, Inc., Hoboken, NJ.

63
Table 3.1. Summary statistics for the variables of interest

N=1258
Variable
Mean Standard deviation
Total output (index) 154,193.14 164,521.51
Capital (index) 162,547.25 158,754.89
Land (acres) 1,277.89 1,103.34
Labor (AWU) 292.68 252.84
Chemicals inputs (index) 38,296.45 41,985.78
Other inputs (index) 24,398.16 25,388.22
Statistics on a per acre basis
Total output (dollars/acre) 122.50 66.52
Capital (dollars/acre) 150.36 131.49
Labor (AWU/acre) 0.24 0.14
Chemicals inputs
29.11 16.98
(dollars/acre)
Other inputs (dollars/acre) 19.48 12.68

64
Table 3.2. Summary statistics for the local estimates of  u2 ,  v2 and 

 u2  v2 
Maximum
1.14 0.26 22.20
(100%)
Third quartile
0.03 0.10 0.59
(75%)
Median (50%) 0.02 0.09 0.48
First quartile
1.93E-5 0.09 0.01
(25%)
Minimum (0%) 6.93E-7 7.59E-4 0.30E-2

65
Table ‎3.3. Estimated LML production elasticities and observed input share for Kansas
farms

Variable Estimated elasticities Observed input share

Land area 0.248 (0.070) 0.173 (0.129)

Labor 0.048 (0.066) 0.053 (0.084)
Capital 0.217 (0.050) 0.197 (0.118)
Chemical inputs 0.293 (0.047) 0.334 (0.132)
Other inputs 0.199 (0.041) 0.241(0.106)
Note: standard deviation in parenthesis

66
Table 3.4. Frequency distribution of technical efficiency scores
Observations (%)
TE: Range (%)
LML1 DEAVRS2 DEACRS3 SFA4 LML DEAVRS DEACRS SFA
<80 VRS
71 2 645 508 5.64 0.16 51.27 40.38
80-85 132 31 449 419 6
10.49 2.46 35.69 33.07
85-90 431 363 103 297 34.26 28.86 8.19 23.61
90-95 244 657 31 37 19.40 52.23 2.46 2.94
95-100 380 205 30 0 30.21 16.30 2.38 0
Mean 0.905 0.917 0.808 0.804
Standard deviation 0.084 0.035 0.047 0.073
Minimum 0.095 0.779 0.678 0.046
Maximum 1.000 1.000 1.000 0.941
1
LML: local maximum likelihood. 2VRS: variable returns to scale. 3CRS: constant return to scale. 4SFA: Stochastic Frontier
Analysis.

67
Table 3.5. Kolmogorov-Smirnov test
Test Value p-value
LML vs. DEA VRS 0.264 0.000
LML vs. DEA CRS 0.713 0.000
LML vs. SFA 0.607 0.000

68
Table 3.6. Cluster analysis results
Cluster Test of
Farm TE Low Medium High difference
0.78 0.88 0.96 between means
(0.10) (0.01) (0.04) (significance
Small <500 acres 64% 22.3% 12.7% level)
Medium (500- 1000 (0.10)
Farm size 18.7% 24.8% 28.7% 0.000**
acres)
Big >1000 acres 17.2% 52.9% 58.7%
Irrigated land share (ratio) 0.05 0.09 0.09 0.007*
(0.15) (0.19) (0.19)
Rented land share (ratio) 0.49 0.62 0.57 0.000*
(0.41) (0.33) (0.33)
Unpaid labor share (ratio) 0.92 0.79 0.81 0.000*
(0.23) (0.29) (0.29)
Output (index) 55,404.96 160,623 181,890 0.000*
(78,526.86) (140,915.8) (186,897.4)
Yields (dollars/acre) 91.81 119 134.89 0.000*
(54.18) (47.14) (77.29)
Age (year) 63.28 56.88 58.23 0.000*
(15.12) (13.37) (13.52)
Education (dollars) 194.05 573.11 641.53 0.112
(1,203.36) (2,462.07) (3,109.10)
Subsidies (dollars) 10,450.91 24,986.47 27,443.01 0.000*
(15,610.36) (34,302.09) (29,764.41)
Non agricultural income 17,535.24 11,938.25 9,653.33 0.000*
(dollars) (41,322.00) (19,712.15) (19,518.52)
Number of observations 203 431 624
Standard deviation in parenthesis. *,** indicate F-statistical and chi-square statistical significance at the 1%, respectively.

69
Figure 3.1. Distribution of the number of observations at each data point.

70
Figure 3.2. Distribution of localized estimates of input elasticities and returns to scale

71
72
Chapter 4

Technical and environmental efficiency

of Catalan arable crop farms13

13
Publication information: Guesmi, B., Serra, T., 2013. Technical and environmental efficiency of Catalan
arable crop farms. Applied Economic Perspectives & Policy journal (first-round review).
74
4.1. Introduction

Intensive agricultural systems have several harmful impacts on humans, animals and the
environment. This has increased social and political concerns regarding agriculture-related
negative‎externalities.‎At‎the‎political‎level,‎these‎concerns‎have‎led‎European‎Union’s‎(EU)‎
agricultural policies to increasingly focus on environmental considerations. Interest in
promoting agricultural practices that minimize pollution has been growing. Consistently,
different policies have been devised to encourage farmers to use less chemical inputs and to
adopt environmentally friendly practices. These alternative practices however, can affect the
productivity and efficiency with which farms are operating, which in turn can influence their
economic viability.
Since‎ its‎ inception,‎ the‎ EU’s‎ Common‎ Agricultural‎ Policy‎ (CAP)‎ has‎ been‎
continuously reshaping itself. While initial objectives focused on farm income support, policy
scopes have been widened to embrace environmental preservation. Late CAP reform
proposals made by the European Commission aim at aligning the CAP with the targets of the
“Europe‎ 2020‎ framework”‎ and‎ call‎ for‎ environmental‎ sustainability,‎ higher‎ efficiency,‎
effectiveness and equitability. Noteworthy among the reform proposals is the aim to
redistribute CAP direct payments on the basis of both economic and environmental criteria. In
light of current CAP reform debates, it is important to develop tools to support monitoring the
impacts of policy and to assist in better targeting policy measures.
Derivation of farm-level technical and environmental efficiency (TE and EE,
respectively) indices should be a relevant tool for improved CAP payment redistribution.
While high TE measures are a pre-requisite for economic sustainability, high environmental
performance indicators should contribute to environmental sustainability of agricultural
practices. Recent literature on efficiency has been debating on the adequate methods to derive
these measures.
Farm-level transition to environmentally sustainable practices can be regarded as a
three-stage process involving different degrees of environmental impact reduction. The three
stages are efficiency, substitution and redesign (Wossink and Denaux, 2007). Our analysis
focuses on the first phase, which aims at minimizing the use of polluting inputs and
optimizing input allocation to achieve the desired output levels. A farm can be considered as
environmentally inefficient, if pollution per unit of input is above an ideal minimum. On the
other hand, technical inefficiencies arise when firms are unable to maximize their output
levels with minimum use of inputs (Farrell, 1957). While privately-run farms are likely to

75
achieve a high level of TE regarding conventional input/output use, they are less likely to be
environmentally efficient due to a lack of economic incentives and information, bounded
rationality, or lack of external competitive pressure regarding environmental performance
(Wossink and Denaux, 2007).
During the last decades, the scientific community has produced several research
studies that attempt to evaluate the aggregate external costs of modern agriculture (Pimentel
et al., 1992 and 1995; Evans, 1995 and 1996; Bailey et al., 1999; Tiezzi, 1999; Pretty et al.,
2000; Le Goffe, 2000). Tegtmeier and Duffy (2004) place the value of the negative impact of
agriculture on water, land, air and human health around 29.44 - 95.68 dollars per hectare in
the USA. In another study, Pretty et al. (2000) obtained a greater value of 32514 dollars for the
United Kingdom, being the most relevant sources of environmental damage: contamination
by pesticides, greenhouse gas emissions, damage to wildlife and habitats, as well as food
poisoning by bacteria and viruses and other disease agents.
In contrast to most sustainability indicators that have been defined at an aggregate
level, farm-level efficiency measures are directly linked to firm management decision making.
While aggregate-level studies can be very useful for politicians and society at large, and can
help designing suitable agricultural and environmental policies, they do not provide useful
information for decision making units (DMU) who are more concerned about the economic
and environmental performance of their holdings. Thus, unlike many environmental
performance measures that have been defined at the aggregate level, our study focuses on
estimating combined measures of TE and EE at the microeconomic level. To achieve this
objective‎ Coelli‎ et‎ al.’s‎ (2007)‎ efficiency‎ measures‎ are‎ extended‎ to‎ a‎ consideration‎ of‎ the‎
stochastic environment in which production takes place.
The paper is organized as follows. In the next section, a literature review and the
contribution of this work to previous literature is presented. Then, we describe the
methodology used in our empirical analysis. The fourth section presents the data and results
from the empirical implementation. We finish the paper with concluding remarks.

14
Pretty et al. (2000) express this amount in GBP. The exchange rate used is 1GPB=1.56 US, which was
obtained from http://www.measuringworth.com/.

76
4.2. Literature review

The first attempts to measure firm-level‎EE‎considered‎a‎firm’s‎environmental‎impacts‎either‎

as an input or an output. Färe et al. (1989), using Data Envelopment Analysis (DEA)
techniques, incorporated pollution into productive efficiency analyses as a weakly disposable
bad output, which implies the assumption that the reduction of pollution is expensive. The
latter implies that, for a given technology, reducing pollution comes at the cost of reducing
the amount of the good output. The authors developed a hyperbolic productive efficiency
model that treats desirable and undesirable outputs asymmetrically, i.e., while producers get
credit for providing desirable outputs, they are being penalized for generating undesirable
outputs.‎This‎“enhanced”‎efficiency‎measure‎was‎compared‎with‎the‎conventional‎efficiency‎
index where, by holding inputs fixed, one expands desirable outputs and ignores pollution.
The comparison of these two efficiency measures shows the extent to which ignoring
undesirable outputs distorts the magnitude of efficiency. Färe et al. (1996) decomposed the
overall factor productivity for US fossil-fuel-fired electric utilities into an environmental
index (under the weak disposability assumption for the undesirable outputs) and a productive
efficiency index. They demonstrated that ignoring pollution leads to significant divergence in
the rankings of the electric utilities. In another study, Färe et al. (2001) used Malmquist-
Luenberger productivity index as a measure to weight the relative importance of bad outputs.
They found that ignoring bad outputs leads to underestimate the annual productivity growth
of US manufacturing sectors for the 1974-1986 period (on average, 1.7% vs. 3.6%).
Piot-Lepetit and Vermersch (1998) studied the TE and EE of a sample of French
farms specialized in pig production. They considered pollution caused by organic nitrogen as
an undesirable output. Using DEA techniques and assuming weak disposability of organic
nitrogen, the authors found a limited ability to reduce nitrogen pollution for given output
levels. Nitrogen surplus has also been treated as an environmentally detrimental input rather
than as undesirable output (Reinhard et al., 1999). The authors provided separate estimates of
output-oriented TE and input-oriented EE of Dutch dairy farms. They found that the latter
achieve high levels of TE (0.89) and low levels of EE (0.44).
The work conducted by Arandia and Aldanondo (2007) constitutes an exception to
published literature on TE and EE of Spanish farms. They focused on a sample of wine farms
and studied their TE by means of a directional distance function that is fit to data by using
DEA methods. Two bad outputs were considered: nitrogen and pesticide pollution. Under the
strong disposability assumption, the average inefficiency was around 25%. If the weak

77
disposability was imposed, inefficiency levels were reduced to 4%, the difference being
caused by the opportunity costs of pollution reduction.
Another EE measure different from other approaches proposed in the literature was
developed by Reinhard et al (2002). This methodology is based on two stages. While the first
stage estimates both TE and EE using Stochastic Frontier Analysis (SFA), the second stage
uses SFA to regress EE obtained from the first stage on explanatory variables that may
explain EE. Conditional EE is then derived from the one sided error component of this second
stage. Relatively low conditional environmental (0.57) and environmental (0.43) efficiency
levels of Dutch dairy farms were found.
Asmild and Hougaard (2006) used a DEA model to estimate TE and EE of a sample of
Danish pig farms. Unlike other studies that consider pollution as an undesirable output or
input, Asmild and Hougaard (2006) disaggregated the nutrient surpluses into two flows: input
and output. They suggested that, assuming variable returns to scale (VRS), EE levels are
around 34-56%, which shows an important margin to improve the environmental performance
of these farms.
While previous literature has paid considerable attention to adjusting efficiency and
productivity measures by considering negative externalities associated to production, a few
recent analyses have incorporated the provision of environmental goods in the vector of
farms’‎ good‎ outputs‎ (Omer‎ et‎ al.,‎ 2007;‎ Areal‎ et‎ al.,‎ 2012).‎ Areal‎ et‎ al.‎ (2012)‎ used‎ SFA‎
based on Bayesian procedure to assess TE of dairy farms in England and Wales and found
efficiency scores and ranking according to these scores to change with incorporation of
environmental outputs in the output vector. As a proxy for the provision of environmental
goods, they used the share of permanent and rough grassland to total agricultural land area.
Average efficiency scores changed from 0.91 to 0.83.
Førsund et al. (2008) and Murty et al. (2011) have shown that reduced-form
technologies that consider pollution either as an input or as a weakly disposable output, have
serious weaknesses. The materials balance principle, that represents the key role of inputs in
residual generation, is proposed as an appropriate method to model pollution. Reinhard and
Thijssen (2000) used the shadow cost approach to assess environmental performance of Dutch
dairy farms, based on the materials balance condition. They found that mean technical and
nitrogen efficiency are on the order of 0.84 and 0.56, respectively.
Based on materials balance principle, Coelli et al. (2007) suggested a new approach
which, in contrast to previous research, does not require the introduction of an extra pollution
variable in the production model. Coelli et al. (2007) illustrate their proposal by studying the

78
environmental performance of a sample of farms specializing in pig-finishing in Belgium
using DEA techniques under constant returns to scale (CRS) assumption. Results suggest that
farms can produce their current output level with 15.7% fewer nutrient pollution. Thus, by
improving‎ farm’s‎ efficiency‎ performance,‎ and‎ based‎ on‎ a‎ cost‎ reducing‎ strategy,‎ farmers‎
avoid adopting expensive pollution abatement technologies.
While the production economics literature has heavily debated on the proper
specification of technologies that involve both intended and unintended outputs such as
pollution, less attention has been paid to the incorporation of the fundamentally stochastic
nature of production into efficiency and productivity analyses. Productivity and efficiency
analyses have often ignored‎the‎stochastic‎nature‎of‎production.‎Even‎the‎“stochastic‎frontier”‎
model is typically grounded on the assumption that the underlying technology is non-
stochastic. In industries such as agriculture where uncertainty is more the norm than the
exception, this can lead to biased efficiency and productivity estimates, because effects due to
uncertainty can either be attributed to productivity or efficiency differences. For example, a
bad production outcome due to a stochastic factor beyond the control of farmers may be
misconstrued as an inefficient production choice. An important challenge in production
economics is thus to appropriately model the stochastic environment under which production
takes place.
In line with mainstream efficiency analysis, Coelli et al. (2007) rely on the assumption
that the underlying production technology is deterministic. The state-contingent approach
proposed by Chambers and Quiggin (1998 and 2000) and built upon the theory developed by
Debreu (1959), is based on the assumption that production under uncertainty can be
represented by differentiating outputs according to the state of nature in which they are
realized. This leads to a stochastic technology based on a state-contingent input
correspondence. Under conventional representations of stochastic technologies, input-output
relationships are studied conditional on the realized state of nature. Chambers and Quiggin
(2000) have shown that while these representations are an extension of conventional non-
stochastic representations of technology and can be easily empirically implemented, they
impose relevant restrictions on the interaction between stochastic outputs and variable inputs.
For example, non-substitutability between state-contingent outputs is imposed, i.e, it is
assumed that producers can only respond to random shocks by modifying their input bundle,
but not by re-allocating state-contingent outputs. This representation is known as the output-
cubical technology (Chambers and Quiggin, 2000), and has been shown to potentially lead to
important‎biases‎in‎efficiency‎estimates‎(O’Donnell‎et‎al.,‎2010).‎

79
 Ex-ante measures of the random variables are required to overcome the output-
cubicality assumption. Since historically, efforts have been focused on collecting ex-post
production data, ex-ante measures are usually unavailable. Our work is innovative in that it
relies on data collected by means of a survey that elicited information on ex-ante state-
contingent outputs. When such measures are on hand, the methods used to study deterministic
technologies can be easily applied to evaluate state-contingent technologies. By using the
Arrow-Debrew-Savage framework, Chambers and Quiggin (2000) show that, in the presence
of‎risk,‎the‎firm’s‎cost‎and‎input‎demand‎functions‎depend‎on‎the‎outputs‎in‎all‎possible‎states‎
of nature. Further, recognition that individuals with monotonic preferences minimize cost
permits evaluation of production decisions independently on any specific assumption on risk
attitudes.
Despite the relevance of state-contingent techniques, there are very few empirical applications
based on this methodology.‎ O’Donnell‎ and‎ Griffiths‎ (2006)‎ use‎ Bayesian‎ techniques‎ to‎
estimate a state-contingent production frontier. Chavas (2008) develops a state-contingent
cost function and an econometric method to recover the ex-ante technology from the ex-post
production data. Following Chavas (2008), Serra et al. (2010) apply state-contingent
techniques to assess production decisions in US agriculture over the last century. Previous
empirical approaches have not relied on survey-elicited ex-ante production data. This is a
relevant contribution of this article to previous research.

4.3. Methodology

Recently, traditional measures of TE have been extended to integrate pollution considerations.

Late developments within this literature have stressed the necessity to consider the materials
balance‎condition‎in‎order‎to‎provide‎sound‎measures‎of‎farms’‎environmental‎performance.‎
Coelli‎et‎al.’s‎(2007)‎proposal,‎based‎on‎this‎principle,‎is‎extended‎to‎allow‎for‎the‎stochastic‎
conditions of production.
Consider a firm that uses a vector of k  1, 2,..., K inputs, x  K
 to produce a state-

contingent output, y   y1 ,..., yS   S

 . The set of states of nature is represented

by   1, 2,...,S , and ys represents the output realized under state of nature s . The feasible

production set under the state-contingent approach, T , can be derived as follows:

80
T  y, x x can produce y; x  K
 ,y S
 , (1)

where the production technology is defined to be convex, non-increasing in inputs, non-

decreasing in outputs, and strongly disposable in inputs and outputs. Let z  S
 represent a
surplus measure calculated using a materials balance equation, which is specified as a linear
function of input and output vectors:

z  ax  by , (2)

where a and b represent vectors of known non-negative constants. The optimization problem
seeks to determine the optimal combination of inputs for a given amount of output that
minimizes the amount of the surplus (pollution caused in the production process).
Under the assumption that the output vector y is fixed or that the output vector is not
capable of converting the polluting input into a usable form (i.e., b equals the zero vector),
the first component of the surplus in equation (2) will be minimized when the aggregate
pollutant content of inputs  M  ax  is minimized. For a given vector of k  1, 2,..., K

pollution contents, a  K
 , the minimum pollution associated with producing a specified
amount of output, can be expressed as:


M  y, a   min ax y, x  T ,
x
 (3)

Denote x e the solution to the minimization problem in (3). axe and ax represent the
minimum and the observed environmental damage, respectively. Following Farrell (1957),
the technically efficient input vector x t can be determined by solving the following
optimization problem:


TE  y, x   min   x, y  T ,

 (4)

where  is a scalar that takes a value between zero and one. x t is determined by xt   x and

the corresponding amount of pollution can be approximated by axt .

81
 Under the state-contingent approach to modeling risk, the production function is
transformed from an ex-post to an ex-ante representation of technology (Chambers and
Quiggin, 2000). Since ex-ante production is conditional upon the state of nature, each ex-post

output quantity has a multivariate ex-ante representation ( y   y1 ,..., yS   S

 ). The increase

in the output vector size under the state-contingent approach may bring dimensionality issues
that are specially relevant under SFA, as the number of parameters to be estimated grows
substantially. Further, ex-ante state-contingent outputs tend to be strongly correlated with
each other (Chavas, 2008), which is associated to potential multicollinearity issues in
econometric model estimation. Overcoming multicollinearity problems in parametric
approaches usually involves working with a reduced version of the actual state space. As is
well known and in contrast to SFA, nonparametric DEA techniques neither rely on specific
functional forms, nor on the covariance among input and output variables, which reduces the
risk of misspecification issues that may lead to biased efficiency estimates. Gong and Sickles
(1992) suggest that DEA techniques become more attractive as potential SFA
misspecification issues grow. Following Färe et al. (1994), the DEA linear programming
model to assess input-oriented TE levels can be expressed as:

min 
 ,

s.t.
 y i  Y  0 (5)
 x i  X  0
N1  1
0

where N represents the number of farms. The constraint N1  1 is included to allow for
VRS. Efficiency scores derived under VRS are compared with those obtained under CRS.
The EE measure proposed by Coelli et al. (2007) is expressed as a ratio of minimum pollution
over observed pollution:

EE  axe ax (6)

82
where EE takes a value between zero and one, the latter indicating that the firm is fully
environmentally efficient. The EE scores for observation i , using the DEA method (Coelli et
al., 2007), are obtained from the following minimization problem:

min ai x ie
 , xie

s.t.
(7)
 y i  Y  0
x ie  X  0
N1  1
0

According‎ to‎ Coelli‎ et‎ al.’s‎ (2007)‎ model,‎ environmental‎ inefficiencies‎ are‎ caused‎ both‎ by‎
technical inefficiencies that imply an excessive use of polluting inputs, and by allocative
inefficiencies involving an inappropriate input mix given the observed a vector. Hence, EE is
decomposed into two components: TE and environmental allocative efficiency (EAE):

TE  axt ax  a  x  ax  , (8)

and

EAE  axe axt , (9)

Allocative efficiency is thus defined as the ratio of minimum pollution to the amount of
pollution generated by the technically efficient input vector. All these efficiency measures
(TE, EE, and EAE)15 take a value ranging from zero to one and can be related through the
following expression:

EE  TE  EAE . (10)

Though a generalization of research results is difficult to make and data and

methodologies used by different analyses are rather heterogeneous, damages derived from
pesticide use are found to be one of the most relevant agriculture-related externalities. This
15
See Coelli et al. (2007) for a graphical representation of the approach to measuring EE, TE and EAE.

83
study focuses on pollution derived from pesticide use. In line with Morse et al. (2006),
Wossink and Denaux (2007) and de Koeijer et al. (2002), we build an index of pesticide
contamination that accounts for the amount of pesticide applied and its toxicity. Through a
farm-level survey, we collected detailed information on the quantities of active ingredients
applied through herbicides, fungicides and insecticides. The total quantity of active
ingredients, expressed in liters, was considered as a polluting input. As well known, different
active ingredients have different environmental and health effects. To derive a single measure
of pesticide pollution, we used a weighting procedure.
While different indices have been elaborated to measure the impacts of different active
ingredients on the environment, animal and human health, they usually focus on a limited list
of active ingredients. The Acceptable Daily Intake (  i ), obtained from the Footprint (2012)
dataset, is the only index covering the full range of active ingredients used by our sample
farms.  i measures the quantity of active ingredients that can be daily ingested over a
lifetime, without implying a significant health risk for humans. 16 This index is usually
measured in mg per kilos of body weight per day (mg/kgbw/day). The vector of  i ’s can be
represented as:

α = (1 ,  2 , ...,  n' ) , (11)

the vector of weights applied to each active ingredient is expressed as:

a   a1 , a2 ,..., an '  . (12)

min(α)
where ai  . The vector of weights was scaled so that their sum be equal to one.
i
n
Pesticide, insecticide and herbicide pollution is approximated by:  a AI
i =1
i i , being AI i the

quantity of active ingredient i applied.

16
While the use of Acceptable Daily Intakes offers a first approximation to toxicity, a more relevant measure of
aggregate environmental impact should include the number of people affected by pollution.

84
4.4. Data and Results

4.4.1. Data

Our analysis uses cross sectional, farm-level data collected using a questionnaire that was
distributed among 190 agricultural holdings specialized in the production of cereals, oilseeds
and protein (COP) crops. The survey was conducted during the planting season in 2011 and
includes detailed information on planned input use. We also obtained data on ex-ante outputs
for three alternative states of nature: bad, normal and ideal growing conditions
y   y1 , y2 , y3  .

Since‎farmers’‎responses‎regarding‎crop‎yield‎distribution‎are‎likely‎to‎be‎subjective,‎
eliciting ex-ante output information is a complex process. Subjective opinions about what
characterizes a bad, normal or ideal crop yield can lead to biased responses. To reduce
subjectivity, one could provide farmers with detailed information on each crop-growing
scenario (rainfall, temperature, frost, etc.). This, however, would increase survey complexity
and cost, as a full characterization of crop-growing scenarios would require provision of a
relevant amount of data during the interview. For example, providing the number of frost days
during the growing season would render an incomplete picture if not accompanied by the
distribution of these days over time. Incomplete scenarios would lead to highly inaccurate
responses.
Final survey design was a trade-off between complexity and subjectivity, and was
conditional upon feedback with technicians from Unió de Pagesos, which constitutes the
largest farmer association in Catalonia. Unió de Pagesos was in charge of administering the
survey. Highly qualified technicians from this institution recommended to obtain point
estimates of yields under bad ( y1 ), normal ( y2 ) and ideal ( y3 ) growing conditions, without
projecting specific scenarios. Yields under normal conditions usually represent average yields
realized during a sufficiently long period of time. Once normal yields are identified, it is
relatively easy for‎farmers‎to‎provide‎yields‎under‎bad‎and‎ideal‎conditions.‎Unió‎de‎Pagesos’‎
technicians based their recommendation to obtain these point estimates on the argument that
the more complex scenario-based alternative, would not provide substantially different results
if designed correctly.
Chambers and Quiggin (2000) and Rasmussen (2003) defined state-allocable inputs as
inputs that, ex-ante, can be allocated to different states of nature. To include these inputs in

85
the production representation, the authors expanded the input vector to allow identifying the
different impacts that inputs have on different states of nature. Budget restrictions, as well as
the need to keep the survey short, precluded obtaining information on state-allocable inputs.
This is left for further research as a means to improve the representation of the stochastic
technology.
Output  ys  represents farm COP production in Euros‎(€)‎under‎state s  1, 2,3 . Inputs

included in the analysis are:  X 1  total hectares (ha) of land planted to COP;  X 2  hired and

family labor, expressed in hours;  X 3  capital input that aggregates the replacement value (in

€)‎ of‎ machinery,‎ other‎ equipment‎ and‎ buildings‎ used‎ in‎ the‎ production‎ process;‎  X 4  the

expenditure‎in‎fertilizers‎expressed‎in‎€;‎  X 5  pesticide, herbicide and insecticide use (liters

of active ingredients);  X 6  seed‎expenses,‎expressed‎in‎€;‎  X 7  energy input including fuels

and‎lubricants,‎expressed‎in‎€;‎  X 8  contract‎work,‎expressed‎in‎€.‎

Table 4.1 provides summary statistics for the variables used in the analysis. Sample
farms cultivate, on average, 75 hectares, a farm size above the national and the EU arable crop
farms’‎average,‎of‎around‎52‎and‎46‎hectares,‎respectively‎(Farm‎Accountancy‎Data Network,
FADN 2012). More than 95% of the COP area is devoted to wheat (36%) and barley (60%)
production. Sample farms devote, on average, 552 hours to COP production during the
growing season, of which more than 90% represents unpaid family labor.
Depending on crop growing conditions, farmers expect to obtain different output
levels. While under bad conditions the average value of COP production is around 31
thousand‎€‎per‎farm,‎under‎ideal‎conditions‎average‎output‎is‎on‎the‎order‎of‎70‎thousand‎€.‎
Under normal conditions, the value of COP production (51‎ thousand‎ €) generated by our
sample‎farms‎almost‎doubles‎the‎EU‎farms’‎average‎output‎(28‎thousand‎€). Per ha statistics
suggest our sample farms are more intensive than both national and EU farms: while EU and
national‎farms‎have‎respectively,‎an‎average‎income‎of‎627‎and‎428‎€‎per‎ha,‎670‎€‎per‎ha‎is‎
the average income of our sample farms under normal growing conditions.
Machinery‎and‎buildings‎used‎by‎sample‎farms‎amount‎to‎134‎thousand‎€,‎or‎2,304‎€‎
per‎ha,‎above‎EU’s‎average‎investment‎ratios‎on‎the‎order‎of‎1,497‎€‎per‎ha‎(FADN,‎2012)‎
and‎it‎is‎much‎higher‎than‎the‎national‎average‎(536‎€‎per‎ha).‎While‎sample‎farms‎cultivate‎
more land than EU arable crop farms, they spend less money in fertilizer than the latter (5,315
vs.‎ 9,279‎ €‎ annually).‎ Total‎ quantity‎ of‎ active‎ ingredients‎ used‎ by‎ our‎ sample‎ farms‎ is,‎ on‎
average,‎ on‎ the‎ order‎ of‎ 85‎ liters.‎ Expenses‎ in‎ pesticides‎ amount‎ to‎ 2,975€,‎ below‎ EU’s‎

86
average‎crop‎farms‎(5,120€).‎Annual‎expenses‎in‎seeds and energy are on the order of 3,866
and‎4,913‎€,‎respectively.

4.4.2. Empirical results

The DEA results are presented in table 4.2. Results show mean TE scores on the order of 0.93
and 0.87 under VRS and CRS, respectively, suggesting that our sample farms could use on
average 7% (13%) fewer inputs to produce the same level of their current output. Returns to
scale were studied to find that farms operate under increasing returns to scale. More than 70%
of the observations have efficiency ratings greater than 0.90, showing relatively high
performance levels. However, under CRS, only about one half of farms exhibit this
performance. Under VRS (CRS) assumption, technical efficiencies range from a minimum of
0.57 (0.18) to a maximum of one, suggesting that our sample farmers present different skills
to manage their holdings.
Our TE findings are consistent with previous studies looking at the performance of
crop farms (Mathijs et al., 2001; Oude Lansink et al., 2002). Mathijs et al. (2001) compared
the efficiency of family farms and partnerships with large-scale successor organizations of the
collective and state farms (LSOs) using DEA methods. For crop farms and for the periods
1991-1992 and 1994-1995, they showed that, under CRS, partnerships display higher average
TE (1.00 and 0.97, respectively) than do family farms (0.82 and 0.81) and LSOs (0.93 and
0.93) in 1991-1992 and 1994-1995, respectively. When VRS were assumed, partnerships and
family farms were found fully technically efficient, while LSOs’s‎TE‎was‎on‎the order of 0.97
in 1991-1992. However, in 1994-1995 LSOs became more efficient than family farms and
partnerships (1.00, 0.98 and 0.97, respectively). In another study, Oude Lansink et al. (2002)
used DEA to compare organic and conventional crop and livestock farms in Finland and
found that organic crop producers have higher efficiency than conventional farms under CRS
(VRS) : 0.91 (0.96) and 0.67 (0.72), respectively. In contrast, our results are far from the
findings by Latruffe et al. (2005), who used DEA to assess the technical and scale efficiency
of crop and livestock farms in Poland for two periods 1996 and 2000. They found scores
under CRS (VRS) for crop farms on the order of 0.66 (0.70) and 0.57 (0.67) in 1996 and
2000, respectively.
Comparison with other studies that use different methodologies does not aim at
recommending one particular methodology over another. Instead, we aim at checking the

87
confidence and robustness of our findings, whether they concur or not with other results
derived from different methods and whether they are or not within the range of existing
estimates (Balcombe et al, 2006). Serra and Goodwin (2009) used a local maximum
likelihood approach to assess technical efficiency of Spanish conventional and organic arable
crop farms. Mean efficiency scores of 0.97 and 0.94 for conventional and organic crop farms
were derived, respectively. These values are relatively close to our findings for VRS. In
contrast, our findings are far from those found by Hadley (2006) and Zhu and Oude Lansink
(2010). Hadley (2006) used SFA to estimate farm-level TE in England and Wales for the
period 1982-2002 and found a mean TE of 0.75 for crop farms. Zhu and Oude Lansink (2010)
used an output distance SFA function to analyze the impacts of CAP reforms on TE of crop
farms in Netherlands, Germany and Sweden for the period 1995-2004. They found an average
TE level of 0.64 in Germany, 0.76 in the Netherlands and 0.71 in Sweden.
The average EE scores are 0.74 (0.58) under VRS (CRS) assumption, indicating that
farmers should be able to produce their current output with 26% (42%) fewer pesticide,
herbicide and insecticide use. Results suggest that farmers who are environmentally efficient
tend to be more technically efficient (high positive correlation of around 0.90 between the two
measures has been found), supporting that an efficient use of chemical inputs improves both
environmental and technical performance. As opposed to previous studies that found an
adverse effect of environmental regulations on productivity (Färe et al., 2001), the high
correlation between TE and EE for our sample farms implies complementarity between
economic and environmental sustainability. Environmental efficiencies range from a
minimum of 0.02 to a maximum of 1, suggesting important variability within sample farms.17
Environmental efficiency ratings imply that there is substantial scope to reduce
chemical input use leaving current output levels unaltered. This reduction is likely to alleviate
the negative environmental impacts of chemical inputs (contamination of surface and
groundwater,‎loss‎biodiversity,‎etc…).‎If‎sample‎farms‎were‎environmentally‎efficient,‎around‎
26 % of current use of pesticides, insecticides and herbicides would be avoided (around 4,200
liters of active ingredients for our sample farms).18 This reduction in pollution levels would
take place at no cost and would not require the adoption of pollution abatement technologies.
A more rational input use would suffice. Since environmental allocative inefficiency (EAE) is

17
While previous analyses have modeled environmental efficiency, to our knowledge, none of them has focused
on arable crop farms. Since comparing environmental efficiency across different types of farming (pig farming
versus arable crop farming, for example) is not adequate, comparison with previous research results is not made
here.
18
Pesticide use savings would be above 40% under constant returns to scale assumption.

88
found to be the main source of environmental inefficiency, it looks like farmers are not using
the correct input mix, given the observed level of riskiness associated to each active
ingredient. Improvements in technical efficiency performance will also lead to higher
environmental efficiency levels of our sample farms.

4.5. Concluding remarks

The productive efficiency literature has paid very little attention to environmental
performance issues. Growing social and political concerns for the environmental impacts of
agriculture make it necessary to study environmental and technical performance using robust
methodologies that enable to derive reliable indicators. Recently, Coelli et al. (2007) proposed
a new approach based on the materials balance concept that represents the relevant role of
inputs in generating residuals.
This‎study‎contributes‎to‎the‎literature‎by‎extending‎Coelli‎et‎al.’s‎(2007)‎proposal‎to‎
allow for the stochastic environment in which production takes place. The extension is based
on the state-contingent methods (Chambers and Quiggin, 2000). The model is applied to
derive combined technical and environmental efficiency levels achieved by 190 Catalan farms
specialized in cereals, oilseeds and protein crop production. To our knowledge, this is the first
empirical application that derives TE and EE measures using both the materials balance and
state-contingent frameworks.
Our empirical findings suggest that our sample farms, on average, reach technical
efficiency scores of 93% and thus that they can reduce input use by 7% while leaving output
levels unaltered. The average environmental efficiency score, on the order of 74%, indicates
ample scope to improve environmental performance and reduce pesticide use and pollution.
These inefficiencies are, to a large extent, caused by allocative inefficiencies that involve an
inappropriate input mix.
Some policy recommendations to increase the relatively low EE levels are as follows.
First, since chemical input is partly applied out of habit (farmers tend to do what they have
done in the past), information and training courses on how to adequately apply chemical
inputs‎ may‎ improve‎ the‎ agricultural‎ sector’s‎ environmental‎ performance.‎ Second,‎ CAP‎
subsidy redistribution on the basis of environmental criteria, may act as an effective tool to
motivate farmers to adopt environmentally friendly practices. Finally, since environmental
inefficiencies are mainly due to allocative issues, providing farmers with better information
on the environmental impacts of different pesticides, herbicides and insecticides, should

89
improve environmental performance. Besides economic incentives penalizing the use of those
chemicals with stronger harmful effects, encouraging farmers to produce environmental
goods should further ensure better environmental efficiency levels.
One limitation of this analysis is that budget restrictions, as well as the need to keep
the survey short, precluded obtaining information on state-allocable inputs, as well as on
possible sources of inefficiency. This is left for future research as a means to improve the
representation of the stochastic technology.

90
References

Arandia, A., Aldanondo, A., 2007. Eficiencia técnica y medioambiental de las explotaciones
vinícolas ecológicas versus convencionales. Revista Española de Estudios Agrosociales y
Pesqueros 215-216: 155-184.
Areal, F.J., Tiffin, R., Balcombe, K., 2012. Provision of environmental output within a multi-
output distance Function Approach. Ecological Economics 78: 47-54.
Asmild, M., Hougaard, J.L., 2006. Economic versus environmental improvement potentials of
Danish pig farms. Agricultural Economics 35: 171-181.
Bailey, A.P., Rehman, T., Park, J., Keatunge, J.D.H., Trainter, R.B., 1999. Towards a method
for the economic evaluation of environmental indicators for UK integrated arable farming
systems. Agriculture, Ecosystems and Environment 72: 145-158.
Balcombe, K., Fraser, I., Kim, J.H., 2006. Estimating technical efficiency of Australian dairy
farms: a comparison of alternative frontier methodologies. Applied Economics 38: 2221-
2236.
Chambers, R.G., Quiggin, J., 1998. Cost functions and duality for stochastic technologies.
American Journal of Agricultural Economics 80: 288-295.
Chambers, R.G., Quiggin, J., 2000. Uncertainty, Production, Choice, and Agency.
Cambridge: Cambridge University Press.
Chavas, J.P., 2008. A cost approach to economic analysis under state-contingent production
uncertainty. American Journal of Agricultural Economics 90: 435-446.
Coelli, T., Lauwers, L., Van Huylenbroeck, G.V., 2007. Environmental efficiency
measurement and the materials balance condition. Journal of Productivity Analysis 28: 3-
12.
Debreu, G., 1959. Theory of Value. Cowles Foundation Monograph 17. New Haven, CT: Yale
University Press.
De Koeijer, T.J., Wossink, G.A.A., Struik, P.C., Renkema, J.A., 2002. Measuring agricultural
sustainability in terms of efficiency: the case of Dutch sugar beet growers. Journal of
Environmental Management 66: 9-17.
Evans, R., 1995. Soil Erosion and Land Use: Towards a Sustainable Policy. Cambridge
Environmental Initiative. Cambridge: University of Cambridge.
Evans, R., 1996. Soil Erosion and its Impact in England and Wales. London: Friends of the
Earth Trust.

91
FADN, 2012. FADN public database.
http://ec.europa.eu/agriculture/rica/database/database_en.cfm. Accessed 20 September
2012.
Färe, R., Grosskopf, S., Lovell, C.A.K., Pasurka, C., 1989. Multilateral productivity
comparisons when some outputs are undesirable: a non-parametric approach. The Review
of Economics and Statistics 71: 90-98.
Färe, R., Grosskopf, S. and Lovell, C.A.K., 1994. Production frontiers. New York, NY:
Cambridge University Press.
Färe, R., Grosskopf, S., Tyteca, D., 1996. An activity analysis model of the environmental
performance of firms application to fossil-fuel-fired electric utilities. Ecological
Economics 18: 161-175.
Färe, R., Grosskopf, S., Pasurka, C.A., 2001. Accounting for air pollution emissions in
measures of state manufacturing productivity growth. Journal of Regional Science 41: 381-
409.
Farrell, M.J., 1957. The measurement of productive efficiency. Journal of the Royal
Statistical Society, Series A, CXX, Part 3: 253-290.
Førsund, F.R., 2008. Good modeling of bad outputs: pollution and multiple-output
production. Working Paper 0809-8786, Department of Economics, University of Oslo.
Footprint dataset, 2012.
http://sitem.herts.ac.uk/aeru/footprint/en/index.htm. Accessed 17 June 2012.
Gong, B.H., Sickles, R.C., 1992. Finite sample evidence on the performance of stochastic
frontiers and data envelopment analysis using panel data. Journal of Econometrics 51:
259-284.
Hadley, D., 2006. Patterns in technical efficiency and technical change at the farm-level in
England and Wales, 1982–2002. Journal of Agricultural Economics 57:81–100.
Latruffe, L., Balcombe, K., Davidova, S., Zawalinska, K., 2005. Technical and scale
efficiency of crop and livestock farms in Poland: does specialization matter? Agricultural
Economics 32: 281-296.
Le Goffe, P., 2000. Hedonic pricing of agriculture and forestry externalities. Environmental
and Resource Economics 15: 397-401.
Mathijs, E., Swinnen, J.F. M., 2001. Production organization and efficiency during transition:
an empirical analysis of East German agriculture. The Review of Economics and Statistics
83: 100-107.

92
Morse, S., Bennett, R., Ismael, Y., 2006. Environmental impact of genetically modified cotton
in South Africa. Agriculture, Ecosystems and Environment 117: 277-289.
Murty, S., Russell, R.R., Levkoff, S., 2011. On modeling pollution-generating technologies.
Working paper, Department of Economics, University of Exeter Business School.
O’Donnell,‎ C.J.,‎ Griffiths,‎ W.E., 2006. Estimating state-contingent production frontiers.
American Journal of Agricultural Economics 88: 249-266.
O’Donnell,‎ C.J.,‎ Chambers,‎ R.G.,‎ Quiggin,‎ J., 2010. Efficiency analysis in the presence of
uncertainty. Journal of Productivity Analysis 33:1-17.
Omer, A., Pascual, U., Russell, N.P., 2007. Biodiversity conservation and productivity in in-
tensive agricultural systems. Journal of Agricultural Economics 58: 308-329.
Oude Lansink, A., Pietola, K.S., Backman, S., 2002. Efficiency and productivity of
conventional and organic farms in Finland 1994-1997. European Review of Agricultural
Economics 29: 51-65.
Pimentel, D., Acguay, H., Biltonen, M., Rice, P., Silva, M., Nelson, J., Lipner, V., Giordano,
S.,‎Harowitz,‎A.,‎D’Amore,‎M.,‎1992. Environmental and economic cost of pesticide use.
Bioscience 42: 750-760.
Pimentel, D., Harvey, C., Resosudarmo, P., Sinclair, K., Kunz, D., McNair, M., Crist, S.,
Shpritz, L., Fitton, L., Saffouri, R., Blair, R., 1995. Environmental and economic costs of
soil erosion and conservation benefits. Science 267: 1117-1123.
Piot-Lepetit, I., Vermersch, D., 1998. Pricing organic nitrogen under the weak disposability
assumption: an application to the French pig sector. Journal of Agricultural Economics
49:85-99.
Pretty, J.N., Brett, C., Gee, D., Hine, R.E., Mason, C.F., Morison, J.I.L., Raven, H., Rayment,
M.D., van der Bijl, G., 2000. An assessment of the total external costs of UK agriculture.
Agricultural Systems 65: 113-136.
Rasmussen, S., 2003. Criteria for optimal production under uncertainty. The state-contingent
approach. Australian Journal of Agricultural and Resource Economics 47: 447-476.
Reinhard, S., Lovell, C.A.K, Thijssen, G., 1999. Econometric estimation of technical and
environmental efficiency: an application to Dutch dairy farms. American Journal of
Agricultural Economics 81: 44-60.
Reinhard, S., Thijssen, G., 2000. Nitrogen efficiency of Dutch dairy farms: a shadow cost
system approach. European Review of Agricultural Economics 27: 167-186.
Reinhard, S., Lovell, C.A.K., Thijssen, G., 2002. Analysis of environmental efficiency
variation. American Journal of Agricultural Economics 84: 1054-1065.

93
Serra, T., Goodwin, B.K., 2009. The efficiency of Spanish arable crop organic farms, a local
maximum likelihood approach. Journal of Productivity Analysis 3:113-124.
Serra, T., Stefanou, S. Oude Lansink, A., 2010. A dynamic dual model under state-contingent
production uncertainty. European Review of Agricultural Economics 37: 293-312.
Tegtmeier, E., Duffy, M., 2004. External costs of agricultural production in the United States.
International Journal of Agricultural Sustainability 2:1-20.
Tiezzi, S., 1999. External effects of agricultural production in Italy and environmental
accounting. Environmental and Resource Economics 13: 459-472.
Wossink, G.A.A., Denaux, Z.S., 2007. Efficiency and innovation offsets in non-point source
pollution control and the role of education. International Journal of Agricultural
Resources, Governance and Ecology 6: 79-95.
Zhu, X., Oude Lansink, A., 2010. Impact of CAP subsidies on technical efficiency of crop
farms in Germany, the Netherlands and Sweden. Journal of Agricultural Economics 61:
545–564.

94
Table 4.1. Summary statistics for the variables used in the analysis
Variable Mean Standard deviation
( y1 )1 30,576.97 33,155.16
2
Output‎(€)‎ ( y2 ) 50,958.65 51,672.67
( y3 )3 70,431.86 74,793.60
Land (ha) ( X1 ) 74.81 72.68
Labor (hours) ( X2 ) 552.10 656.46
Capital‎(€)‎‎‎‎‎‎‎‎‎‎‎‎‎‎‎‎‎ ( X3 ) 133,721.63 126,557.95
Fertilizers‎(€) ( X4 ) 5,314.76 7,352.18
Pesticides (liters of
( X5 ) 85.34 117.19
active ingredients)
Seeds‎(€) ( X6 ) 3,866.07 3,750.11
Energy‎(€) ( X7 ) 4,912.87 5,334.41
Contract‎work‎(€) ( X8 ) 2,916.54 4,018.97
Statistics on a per hectare basis
( y1 ) 391.31 131.76
Output‎(€/ha) ( y2 ) 669.85 144.29
( y3 ) 912.83 214.04
Capital‎(€/ha) 2,303.74 2,785.71
Labor (hours/ha) 6.88 4.86
Fertilizers‎‎(€/ha) 71.75 63.61
Pesticides (liters of
1.04 0.72
active ingredients /ha)
Seeds‎(€/ha) 52.46 16.98
Energy‎(€/ha) 66.97 43.68
Contract‎work‎(€/ha) 63.17 68.42
1
y1 : bad growing conditions. 2 y2 : normal growing conditions. 3 y3 : ideal growing conditions.

95
Table 4. 2. Summary statistics for DEA results
Efficiency TE EE EAE
measures VRS CRS VRS CRS VRS CRS
Mean 0.93 0.87 0.74 0.58 0.76 0.61
Standard deviation 0.11 0.16 0.37 0.41 0.35 0.40
Minimum 0.57 0.18 0.02 0.00 0.04 0.00
Maximum 1.00 1.00 1.00 1.00 1.00 1.00

96
Table 4.3. Frequency distribution of DEA efficiency scores
Efficiency scores: VRS CRS
Range (%) TE EE EAE TE EE EAE
<60 2
VRS 63 57 17
VRS 96 88
60-70 9 2 5 14 4 8
70-80 19 3 3 23 3 4
80-90 25 4 6 34 3 4
90-100 134 117 118 101 83 85
Mean 0.93 0.87 0.74 0.58 0.76 0.61
Standard deviation 0.11 0.16 0.37 0.41 0.35 0.40
Minimum 0.57 0.18 0.02 0.00 0.04 0.00
Maximum 1.00 1.00 1.00 1.00 1.00 1.00

97
98
Chapter 5

Conclusions
100
Farm-level productivity and efficiency analyses have important implications both for firm
management decisions and policy design. By targeting firms not previously investigated and
using new methodological approaches, this thesis contributes to the literature both from a
methodological and an empirical point of view. The thesis is integrated by three independent
research articles. The first article’s‎scientific‎contribution‎is‎mainly‎of‎an‎empirical‎nature,‎as‎
it targets organic farms in Spain that have received little attention by the efficiency and
productivity literature. More specifically, it focuses on assessing productivity and technical
efficiency differences between organic and conventional grape farms in Catalonia, as well as
the factors that affect technical efficiency levels. The second article, with a strong
methodological orientation, uses recent innovative techniques to study technical efficiency of
Kansas farms specialized in arable crop production. Finally, the third research paper extends
recent proposals to derive combined environmental and technical efficiency measures, and
applies them to study performance of Catalan arable crop agricultural holdings.
In spite of the significant recent growth in organic farming in Spain, the literature on
the performance of Spanish organic farming is still insignificant. The first research paper
contributes to fill this gap. The methodological approach adopted in the first paper consists of
a stochastic frontier model in which inefficiency effects are assumed to be a function of firm-
specific characteristics. Our research is pioneer in that it measures the contribution of farmers’‎
preferences regarding environmental preservation and economic performance to efficiency.
The analysis is based upon a sample of 141 organic and conventional Catalan farms that
specialize in grape growing.
Our empirical findings suggest that organic farmers, on average, display higher
technical efficiency scores than their conventional counterparts (80% and 64%, respectively).
However, organic farms show lower productivity than conventional ones. Our results identify
adoption of organic practices, experience, family labor share in total labor, farm location and
farmer environmental preferences as the variables that are more relevant in explaining
technical inefficiencies. Holding more experience and/or using organic practices leads to
higher efficiency levels. Conversely, farms that rely on a higher proportion of unpaid labor,
are located in a less favored area, or whose manager has strong environmental preservation
preferences, tend to be less efficient.
In the second research article, local maximum likelihood methods, recently proposed
by Kumbhakar et al. (2007), are used to assess Kansas farms efficiency levels. The analysis is
based on farm-level data obtained from farm account records from the Kansas Farm
Management Association dataset covering the period 2000-2010. In spite of the interesting

101
features of local estimation methods, its use has been limited to a few empirical studies due to
implementation complexities.
Empirical results support the relevance of using the LML approach through the
variation in the localized parameter estimates representing the variance of the composite error
term and input elasticities. Results show that Kansas farms reach 91 % of their maximum
potential output indicating that farmers could increase their output by 9% without the need to
increase input use and without changing current technology. Technical efficiency scores
derived from the LML approach [0.905] are higher than those of the DEA model under CRS
[0.808] and SFA [0.804] and close to DEA-VRS [0.917] ratings. According to the KS test, the
efficiency score distributions obtained from DEA and SFA differ from LML distribution
ratings. Since LML allow for both stochastic error terms, as well as for flexibility in the
functional form representing the frontier function, LML efficiency scores may be more
reliable and less biased than those derived under nonparametric DEA and SFA alternatives.
In the third research paper, we use the recent Coelli‎et‎al.’s‎(2007)‎method‎based‎on‎
the materials balance concept to assess both technical and environmental efficiencies. We
expand this method to allow for the stochastic environment in which production takes place.
The extension is based on the state-contingent methods proposed by Chambers and Quiggin
(1998 and 2000). To our knowledge, no previous published work has studied environmental
efficiency using state-contingent methods. This constitutes our major contribution to the
literature. On the other hand, to date, no studies have previously focused on the assessment of
technical and environmental efficiency of Spanish agriculture using this methodology.
The analysis is based on farm-level data collected using a questionnaire distributed
among 190 Catalan arable crop agricultural holdings. Our empirical findings suggest that
sample farms present high average technical efficiency scores of 93%, indicating that they can
reduce input use by 7% while leaving output levels unaltered. The average environmental
efficiency score, on the order of 74%, indicates ample scope to improve environmental
performance and reduce pesticide use and pollution by 26%. These inefficiencies are, to a
large extent, caused by allocative inefficiencies.
As well known and also shown in this thesis, efficiency estimates are very sensitive to
the method used to estimate the frontier (parametric or non-parametric), to the functional form
representing the production frontier and the distribution of the error term. The use of
improved techniques is thus key for meaningful efficiency analyses. This thesis implements
different methodologies in the analysis of farm performance, from well-known methods to

102
very innovative approaches, providing credible case studies in applying different approaches
and providing evidence for their value.
Improving technical efficiency allows for a reduction in production costs and increases
competitiveness. This is specially useful to agricultural sectors where consumers are generally
unwilling to pay higher prices and the marketing power of middlemen and retailers becomes
more and more relevant. Firm management and policy implications are proposed along the
thesis that are based on the obtained results. According to these, promotion of extension
services that transfer knowledge to farmers is expected to improve production performance
through added education and experience. Enhanced efficiency levels may also be pursued
through more professionalized management of agricultural holdings by using more
specialized labor force. Information and training courses on how to adequately apply chemical
inputs could enhance the‎ agricultural‎ sector’s‎ environmental‎ performance.‎ Similar results
might be achieved by redistributing CAP subsidies on the basis of environmental criteria, or
by promoting economic incentives that penalize the use of harmful chemicals. Providing
better information on the environmental impacts of different pesticides, herbicides and
insecticides and how to adequately use them may also lead to improved environmental
performance.
Several shortcomings affecting our analysis as well as proposals for future research
can be pointed out. One limitation is the small number of organic farms used in the first
research paper. Collecting additional farm-level organic farming data would increase the
reliability and the number of farms represented by our results. The curse of dimensionality
affecting local maximum likelihood techniques used in the second article makes it difficult to
use more sophisticated representations of production technology. Further, panel data
techniques are not taken into account when estimating the model. Developing local maximum
likelihood methods applicable to panel data techniques is another pending research issue.
Different methodological innovations to assess efficiency have been recently introduced in
the literature that could be applied to our data. Noteworthy are the innovations regarding
dynamic efficiency measurement that do not‎rely‎on‎the‎assumption‎of‎firm’s‎ability‎to‎adjust
instantaneously and that allow for the dynamic linkages of production decisions (Tsionas,
2006; Silva and Stefanou, 2007; Rungsuriyawiboon and Stefanou, 2007; Serra et al., 2010;
Emvalomatis et al., 2011;Serra et al., 2011). Extension of local maximum likelihood methods
to a consideration dynamic issues constitutes another area that merits further attention. In the
third article, budget restrictions, as well as the need to keep the survey short, precluded
obtaining information on state-allocable inputs, as well as on possible sources of inefficiency.

103
This is left for future research as a means to improve the representation of the stochastic
technology. Allowing for the impacts of other pollution sources such as fertilizers, will allow
deriving more reliable environmental performance estimates.

104
References

Chambers, R.G., Quiggin, J., 1998. Cost functions and duality for stochastic technologies.
American Journal of Agricultural Economics 80: 288-295.
Chambers, R.G., Quiggin, J., 2000. Uncertainty, Production, Choice, and Agency.
Cambridge: Cambridge University Press.
Coelli, T., Lauwers, L., Van Huylenbroeck, G.V., 2007. Environmental efficiency
measurement and the materials balance condition. Journal of Productivity Analysis 28: 3-
12.
Emvalomatis, G., Stefanou, S., Oude Lansink, A., 2011. A reduced-form model for dynamic
efficiency measurement: application to dairy farms in Germany and The Netherlands.
American Journal of Agricultural Economics 93: 161-174.
Kumbhakar, S.C., Byeong, U.P., Simar, L., Tsionas E.G., 2007. Nonparametric stochastic
frontiers: A local maximum likelihood approach. Journal of Econometrics 137: 1-27.
Oude Lansink, A., Pietola, K., Bäckman, S., 2002. Efficiency and productivity of
conventional and organic farms in Finland 1994–1997. European Review of Agricultural
Economics 29: 51-65.
Rungsuriyawiboon, S., Stefanou, S., 2007. Dynamic efficiency estimation: an application to
U.S. electric utilities. Journal of Business and Economic Statistics 25: 226-238.
Silva, E., Stefanou, S., 2007. Dynamic efficiency measurement: theory and application.
American Journal of Agricultural Economics 89: 398-419.
Serra, T., Stefanou, S., Oude Lansink, A., 2010. A dynamic dual model under state-contingent
production uncertainty. European Review of Agricultural Economics 37: 293-312.
Serra, T., Oude Lansink, A., Stefanou, S., 2011. Measurement of dynamic efficiency: A
directional distance function parametric approach. American Journal of Agricultural
Economics 93: 756-767.
Tsionas, M., 2006. Inference in dynamic stochastic frontier models. Journal of Applied
Economics 21: 669-676.

105