Jukuri, open repository of the Natural Resources Institute Finland (Luke) 
   
 
   
All material supplied via Jukuri is protected by copyright and other intellectual property rights. Duplication 
or sale, in electronic or print form, of any part of the repository collections is prohibited. Making electronic 
or print copies of the material is permitted only for your own personal use or for educational purposes.  For 
other purposes, this article may be used in accordance with the publisher’s terms. There may be 
differences between this version and the publisher’s version. You are advised to cite the publisher’s 
version. 
 
This is an electronic reprint of the original article.  
This reprint may differ from the original in pagination and typographic detail. 
 
Author(s): Amer Ait Sidhoum, Philipp Mennig & Johannes Sauer 
Title: Do agri-environment measures help improve environmental and economic 
efficiency? Evidence from Bavarian dairy farmers 
Year: 2023 
Version: Published version 
Copyright:   The Author(s) 2023 
Rights: CC BY 4.0 
Rights url: http://creativecommons.org/licenses/by/4.0/ 
 
Please cite the original version: 
Amer Ait Sidhoum, Philipp Mennig, Johannes Sauer, Do agri-environment measures help improve 
environmental and economic efficiency? Evidence from Bavarian dairy farmers, European Review 
of Agricultural Economics, 2023;, jbad007, https://doi.org/10.1093/erae/jbad007 
Do agri-environment measures help 
improve environmental and economic 
efficiency? Evidence from Bavarian
dairy farmers
Amer Ait Sidhoum *,†, Philipp Mennig ‡ and 
Johannes Sauer‡
†Natural Resources Institute Finland (Luke), Business Economics, 
Finland; ‡Department of Agricultural Production and Resource 
Economics, Technische Universitat München, Germanÿ
*Corresponding author: E-mail: amer.ait-sidhoum@luke.fi
Review coordinated by: Nauges, Céline
Abstract
This study presents an innovative empirical application to the assessment of agri-
environment measures on environmental and economic efficiency. Using a multi-
equation representation with desirable technology and its accompanying undesirable 
by-production technology, we investigate the effects of agri-environment measures 
on farm-level environmental and economic efficiency. A combination of propensity 
score matching and a difference-in-difference approach is used to estimate the policy 
effect. The application focuses on a balanced sample of Bavarian dairy farms surveyed 
between 2013 and 2018. Results suggest that agri-environment schemes do not alter 
farms’ economic efficiency, whereas environmental efficiency does not seem to be 
stimulated by schemes participation.
Keywords: agri-environment schemes, policy evaluation, nitrogen pollution, environ-
mental performance, data envelopment analysis
JEL classification: C23, Q12, Q18, Q57
© The Author(s) 2023. Published by Oxford University Press on behalf of the Foundation for the European Review 
of Agricultural Economics.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://
creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any 
medium, provided the original work is properly cited.
1. Introduction
Nitrogen (N) pollution from agriculture is recognised as one of the most press-
ing environmental problems humanity faces. The N surplus in the environment 
is mainly a result of the intensive use of mineral fertilisers and livestock pro-
duction (Sutton et al., 2011), which has surpassed the planet’s boundaries 
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
2 A. Ait Sidhoum et al.
(Rockstr ̈om et al., 2009; Steffen et al., 2015). The negative environmental 
impacts related to this excess as a result of the human-induced disruption of 
the biochemical nitrogen cycle have only recently begun to overlay the criti-
cal role that nitrogen fertiliser has played in boosting agricultural production. 
Its inefficient application has resulted in losses to the environment in various 
ways. Via leaching and runoff, nitrogen, mainly in the form of nitrate (NO3
−), 
gets into groundwater and surface waters, where it threatens human health 
when entering drinking water (Davidson et al., 2011) and causes unwanted 
eutrophication (Howarth et al., 2000). In gaseous forms (ammonia—NH3, 
nitric oxide—NO, nitrous oxide—N2O and N2), it is emitted to the atmo-
sphere where these types cause considerable harm. NO plays a significant 
role in tropospheric ozone pollution (Hickman et al., 2017), and NH3 and 
NO are precursors to particulate matter air pollution and facilitate the depo-
sition of nitrogen into ecosystems (Galloway et al., 2003). Furthermore, 
high concentrations of ammonia can be harmful for vegetation in case of 
direct air contact (Krupa, 2003). N2O, on the other hand, is a potent green-
house gas that additionally depletes stratospheric ozone (IPCC, 2014; Kanter 
et al., 2017). Finally, high nitrogen loadings threaten biodiversity and ecosys-
tem functioning (Rabalais, Turner and Wiseman, 2002; Vitousek et al.,
1997).
In order to reduce N pollution, numerous measures have been implemented 
by countries, associations of states and private and non-profit players all 
around the globe. The European Union (EU), for example, aims to tackle 
the problem with a number of environmental directives: the Nitrates Direc-
tive, the Habitats Directive, the Water Framework Directive and the National 
Emissions Ceilings Directive. These directives force member states to act 
and are accompanied by regional-, national- and EU-level initiatives, one 
of which are agri-environment schemes (AESs), or agri-environment-climate 
measures (AECM) as they are lately being referred to, as part of the Common 
Agricultural Policy (CAP).
AES in the EU agricultural policy can be traced back to the 1985 Agri-
cultural Structures Regulation. They were conceptualised as tools that com-
pensate farmers for income losses and costs associated with environmentally 
friendly farming practices. After an amendment in 1987, which governed fund-
ing sources and cost flows for environmentally sensitive areas, AES became 
compulsory for all EU member states in 1992 (EU Regulation 2078/92). Since 
then, they are an important component of the second pillar of the CAP. In the 
2014–2020 EU rural development programming period, European, national 
and regional Pillar II spending on environment protection and resource effi-
ciency (the themes to which AES are mainly associated) amounted to 38 billion 
Euros, roughly 25 per cent of the total spending on rural development (Euro-
pean Commission, 2021). In the course of the years, AES spending increased 
and the schemes themselves evolved. Today, top-down action-based schemes 
still dominate; however, more and more spatially targeted (results-based) AESs 
created using a multi-actor approach come into play. They are a response to 
the patchy success of the classical schemes in many environmental categories, 
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
Do agri-environment measures help improve environmental and economic efficiency? 3
including biodiversity (Bisang, Lienhard and Bergamini, 2021; Budka et al., 
2019), water quality (Jones et al., 2017; Kay et al., 2012; Slabe-Erker et al., 
2017) and greenhouse gas emissions (Coderoni and Esposti, 2018; Stetter, 
Menning and Sauer, 2020), where diffuse pollution from nitrogen use causes 
direct harm. The present work attempts to further investigate the impact of 
AES on nitrogen pollution. In contrast to previous studies, we assess the 
programme impact in terms of changes in the (nitrogen) environmental and 
economic efficiency of farms.
Assessing the impact of AES on economic efficiency is critical not only 
for managers who are driven by profit-maximising objectives but also for 
other stakeholders and decision-makers such as policymakers and regulators 
(e.g. environmental agencies). For example, regardless of whether the pro-
ducer decides to join an environmental programme or not, he or she will 
be concerned with improving productivity not only in physical terms but 
also in terms of the economic benefits associated with an optimal alloca-
tion of actual quantities of inputs and outputs. For regulators, minimising the 
level of pollution associated with the production of desirable products is also
critical.
In our case, the production process is modelled in the presence of desir-
able and bad outputs. The desirable output is the value of the dairy products, 
while the bad output is related to nitrogen pollution as measured by the 
farm-level nitrogen balance. Few alternatives exist in the technical litera-
ture on the proper modelling of undesirable outputs in production processes 
(see Dakpo, Jeanneaux and Latruffe, 2016). The most recent type of by-
production approach relies on the concept that a production technology is 
a result of an interaction of two separate sub-technologies: one governs the 
production of good outputs and a second governs the production of bad out-
puts. Introduced by Murty, Robert Russell and Levkoff (2012) and extended 
by Serra, Chambers and Oude Lansink (2014) and Førsund (2021, 2018), 
this approach is one of the most promising methods for modelling pollution-
generating technologies. In this study, we rely on the novel by-production 
approach to estimate our combined economic and environmental1 efficiency
indices.
From a practical point of view, the nitrogen pollution is supposed to be 
influenced by AES that requires the adoption of specific management practices 
or environmentally beneficial actions, e.g. a reduction of nitrogen fertiliser 
use. Clearly, specific agri-environment measures primarily aim at a reduction 
of absolute nitrogen pollution levels with desirable output losses being offset 
1 Although the environmental dimension covers large and complex fields, the environmental 
impacts of nitrogen pollution are well known. High levels of fertiliser application lead to a large 
nitrogen surplus and to correspondingly large flows of nitrogen into the soil and into the air. 
Part of the nitrogen is taken up by crops, but a large portion of these nutrients is emitted to the 
environment. Nitrogen pollution leads to nitrate contamination of the groundwater aquifers, the 
most important source of drinking water. Nitrogen also evaporates as ammonia and causes acid 
rain. Phosphate pollution causes eutrophication of surface water, which endangers plant and fish 
life. For these reasons, ‘nitrogen pollution’ is used in this paper as a proxy of the environmental 
dimension.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
4 A. Ait Sidhoum et al.
by the AES payment. Improving farm efficiency is thus no direct goal of the 
schemes. Furthermore, a reduced nitrogen balance as a result of being enrolled 
in an AES might be linked to output losses that keep the output-pressure ratio 
unaltered. However, studying the effect of AES enrolment on farm efficiency 
is still important for two reasons. First, authors such as van Grinsven et al.
(2019), Grovermann et al. (2019) and Beltrán-Esteve and Picazo-Tadeo (2017) 
claim that innovations have a positive impact on farmers’ efficiency in minimi-
sation of environmental outcomes given economic outcomes. We hypothesise 
that AES participation is likely to stimulate eco-innovation, which in turn has a 
positive impact on economic and environmental efficiency. Second, economic 
and environmental issues need to be looked at jointly in the light of sustain-
able agricultural production. If scheme enrolment leads to less environmental 
pollution in absolute terms and additionally improves economic efficiency, 
further strengthening the role of second pillar AECM and newly introduced 
first pillar eco-schemes would be reasonable. As Kumbhakar and Malikov 
(2018) stated, both economic and environmental efficiencies are crucial from 
a policy standpoint since they are usually taken as the basis for designing 
new or assessing existing policy instruments. Furthermore, policymakers may 
prefer adopting policies aimed at efficiency enhancement rather than imple-
menting restrictive policies (Kuosmanen and Kortelainen, 2005). Economic 
and environmental efficiency improvements are therefore arguably the most 
cost-effective way to reduce environmental pollution while maintaining or 
improving the competitiveness of the agricultural sector.
This paper extends the literature by providing an in-depth analysis of the 
impact of AES on nitrogen pollution. Specifically, we do not purely focus 
on the environmental outcome but link it to classical efficiency analysis, thus 
obtaining measures of environmental and economic efficiency that allow us 
to assess whether AES can be effective in pursuing economic and ecological 
goals simultaneously. Our second contribution relates to the combination of 
the impact evaluation methods, namely, propensity score matching (PSM) and 
difference-in-difference (DiD) with a multi-equation representation using data 
envelopment analysis (DEA).
Just as in the rest of Europe, the agricultural sector has undergone rapid 
structural change in Bavaria in the past decades. While the number of farms 
amounted to around 440,000 in 1960, it decreased to 88,610 in 2017. This 
development was mainly driven by differences in working conditions and 
incomes between economic sectors, technical progress and efficiency gains. 
Especially the use of mineral fertilisers, pesticides and innovative machinery 
resulted in increases in yields and lower producer prices, which in turn forced 
farmers to enlarge their businesses. High fertiliser and pesticide applications, 
however, created negative externalities in the form of environmental pressure. 
Persistently, high loads of nitrogen fertilisers in particular negatively affected 
water quality in Bavaria. Only 15 per cent of all running waters and only 26 out 
of 50 surface water bodies that were evaluated fulfil the criteria set in the Euro-
pean Water Framework Directive (LfU, 2021a). Furthermore, at 6.4 per cent of 
all groundwater quality measuring points, the threshold for nitrate of 50 mg/l is 
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
Do agri-environment measures help improve environmental and economic efficiency? 5
exceeded, and at 30 per cent, values higher than 25 mg/l are reported, indicat-
ing human-induced pollution (LfU, 2021b). In both cases, runoff and leaching 
from the application of nitrogen fertilisers play a key role. This pollution needs 
to be decreased further in order to reach the target of a ‘good state’ of water 
bodies by 2027 as stipulated in the Water Framework Directive.
Measures to improve the state of water bodies and groundwater quality, 
of which AESs are part, need to be effective, while at the same time, they 
should neither affect a farm’s economic performance nor development possi-
bilities. The latter is of particular importance as Bavarian agriculture, which 
compared to the German average is characterised by rather small-scale family 
farms and partly unfavourable natural conditions, competes with national and 
international players. In order to guarantee food supply on the regional level 
with limited natural resources and a low(er) environmental footprint, the sector 
needs to follow a path of sustainable intensification, i.e. farms need to adopt 
production measures that improve either economic or environmental outcomes 
without compromising each other (in the Bavarian context, more emphasis 
might be put on the environmental component). From a farm perspective and 
on the field level, scheme limiting nitrogen fertiliser input, for example, might 
be such a measure if the payment offsets or is higher than income foregone 
and costs incurred. The existence of windfall effects is likely to affect actual 
changes in pollution levels, though.
Given the theoretical promises of such practices with the potential to 
balance the trade-off between economic and environmental efficiency, surpris-
ingly, little empirical studies evaluating them exist. Few to be found mainly 
use field trial data (e.g. Townsend, Ramsden and Wilson, 2016) or simulations 
(e.g. Devkota et al., 2016) and put an emphasis on yields. Holistic farm-level 
assessments, particularly in Central Europe, however, seem underrepresented. 
Contrary to farmers in the global South, European farmers rather aim at 
enhancing their ecological output without sacrificing the economic outcome 
when adopting sustainably intense practices (Charles, Godfray and Garnett, 
2014). The undesirable environmental harm as a by-product of agricultural 
production is thus at the centre of attention. It still interacts with farm-specific 
technical and social factors, which is why it is important to consider the multi-
dimensionality of outcomes (Ait Sidhoum, Serra and Latruffe, 2020). Among 
few studies that use a respective approach for measuring the effect of spe-
cific farming practices on economic and environmental performance at farm 
level are the studies by Gadanakis et al. (2015) and Pérez Urdiales, Lansink 
and Wall (2016). Both studies use a radial eco-efficiency measure based on 
non-parametric DEA. Gadanakis et al. (2015) find that arable farms in the UK 
can improve eco-efficiency by adopting sustainable farming practices, while 
Pérez Urdiales, Lansink and Wall (2016) identify socio-economic character-
istics and attitudes that explain eco-efficiency of Spanish dairy farms. In both 
cases, however, the approaches used lack explicit causal interpretation. This 
shortcoming is addressed by Weltin and Hüttel (2019) who investigate the envi-
ronmental improvement potential of sustainable intensification in the northern 
German Plain using a directional meta-frontier approach and matching. Their 
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
6 A. Ait Sidhoum et al.
results show that adopters of sustainably intense practices have higher mean 
eco-efficiency scores compared to non-adopters but still do not fully exploit 
the potential of ecological improvements.
2. Methodology
Our analytical framework consists of three steps. In the first step, we use PSM 
to control for potential selection bias arising from observable characteristics. 
In step 2, we use DEA to estimate farm-level technical and environmen-
tal efficiency indices. Third, a DiD specification within a single bootstrap 
procedure is applied to derive the average treatment effect of AES on effi-
ciency scores. In the following text, we will explain our framework step
by step.
2.1. PSM
There are several strong theoretical reasons why AESs might improve farm 
performance, but how can we be sure that the better farm performance of 
AES-participating farms compared to non-participating farms is caused by 
AES adoption (or not)? In most instances, experimental data would have been 
of great value to obtain information on the counterfactual state that would 
address the question of causal inference. However, this is not our case, as 
for each farmer only one situation can be observed, we have a missing data 
problem (Rubin, 1976). Therefore, to do this, we have to prevent selec-
tion bias while seeking to separate the AES impact from other determinants 
of farm performance.2 To address the selection bias problem, we use PSM
techniques.
The first step of this procedure consists of estimating the propensity scores 
that are used to match AES-participating with non-participating farms for 
the pre-treatment period 2013. Because agri-environment measures uptake is 
a binary choice, we estimate a logit model for the pre-treatment period as 
follows: 
ln(
𝜋𝑖
1−𝜋𝑖
)= 𝛽0+𝛽𝑋𝑖+𝜀𝑖 (1)
where 𝑖 = 1,…,𝐼 indexes the number of farms and 𝜋𝑖 = 𝑃(𝑇𝑖 = 1|𝑋𝑖) indi-
cates the farmers’ probability3 of treatment assignment, conditional on a vector 
of observed explanatory variables 𝑋𝑖. Successful implementation of the PSM 
technique requires a set of relevant covariates that could potentially affect 
AES participation and/or farm performance (technical and environmental effi-
ciency) but otherwise are not affected by the treatment (scheme participation). 
More specifically, 𝑋 includes a set of variables representing farm and regional 
characteristics (see Table A1).
2 Farms’ performance is affected by both observed factors (e.g. land size, experience, farmers’ 
education, capital, labour and chemical use) and unobserved factors (e.g. managerial ability, 
motivation and environmental awareness).
3 Instead of matching on a set of observable covariates, 𝑋𝑖 showed that matching on the propensity 
score (𝜋𝑖) is sufficient to identify the treatment effect.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
Do agri-environment measures help improve environmental and economic efficiency? 7
Once the propensity scores have been estimated for each of the 𝑖 farms in 
the sample, a matching strategy is used to construct counterfactual outcomes 
for the participating farms. Then, the average treatment effect on the treated 
(ATT) can be estimated as follows: 
ATT =𝐸(𝑌𝑖 (1) |𝑇𝑖 = 1, 𝜋(𝑋𝑖))−𝐸(𝑌𝑖 (0) |𝑇𝑖 = 0, 𝜋(𝑋𝑖)) (2)
where 𝑌𝑖 (1) is the outcome variable (e.g. farm efficiency) when the farm 𝑖
participates in AESs and 𝑌𝑖 (0) is the outcome variable when the farm 𝑖 does 
not participate. 𝑇𝑖 is a dummy variable that takes the value of 1 for the par-
ticipating farms and 0 otherwise, and 𝜋(𝑋𝑖) is the propensity score of farm 
𝑖 conditional on observed covariates 𝑋𝑖. The ATT compares the differences 
in outcomes between the participating farms (first term of the model (2)) 
and matched non-participating farms (second term of the model (2)). The 
validity of PSM further relies on two assumptions: the conditional indepen-
dence assumption and the common support condition (Khandker, Koolwal and 
Sammad, 2009). The conditional independence assumption states that given a 
set of observable covariates 𝑋𝑖, the potential outcomes are independent of 
treatment assignment, which is specified as follows: 
𝑌𝑖 (1) ,𝑌𝑖 (0)⊥𝑇𝑖|𝑋𝑖 (3)
The common support assumption: 0 < 𝑃(𝑇𝑖 = 1|𝑋𝑖) < 1 ensures that 
for each participating farm, there are non-participating farms with similar 
observed characteristics.
Once the matching has been completed and comparable participant and non-
participant farms have been identified, the next step is to derive technical and 
environmental efficiency measures.
2.2. Efficiency measurement
The traditional firm-specific technical efficiency analysis framework has been 
extended to allow for externalities. Within the extended framework, both 
traditional netputs and environmental impacts characterise the production 
technology. However, there is a debate in the production economics literature 
on the appropriate method to use when dealing with environmental impacts 
in production analysis. Two main categories can be distinguished. First, one 
approach represented by a group of studies (e.g. Chambers, Chung and F ̈are, 
1996; F ̈are et al., 2005) relies on the additive directional distance function 
that aims at increasing the good outputs, while reducing pollution at the same 
time. This single-equation approach has been criticised for not being suffi-
ciently suitable to simultaneously capture the trade-off between the desirable 
output and the undesirable output and thus leading to an overstatement of the 
overall efficiency. Second, a multi-equation framework that has its foundations 
by Frisch (1965) that has been followed up in some recent studies (Murty, 
Robert Russell and Levkoff, 2012) relies on a radial measure of a produc-
tion technology that combines two independent sub-technologies reflecting a
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
8 A. Ait Sidhoum et al.
desirable output and undesirable by-production sub-technology. Despite the 
limiting assumption of two independent sub-technologies,4 the multi-equation 
representation remains a very promising approach to tackle the challenges 
associated with by-production technologies.
Similarly to Murty, Robert Russell and Levkoff (2012), we model firm per-
formance as a composition of two sub-technologies: a desirable production 
technology and a pollution-generation technology. In their model, they sep-
arate inputs into non-polluting and polluting inputs in addition to the output 
distinction. In the context of our empirical application, the non-polluting inputs 
(𝑥𝑛)comprise capital, labour, land and so on, while the polluting input (𝑝) con-
sists of nitrogen input, which generates runoff pollution as by-products, such 
as nitrogen pollution. In general terms, the ‘overall’ technology, denoted by 𝑇, 
is represented as the intersection of two sub-technology sets, 𝑇 1 and 𝑇 2. The 
overall technology with its sub-components can be represented as follows: 
𝑇 1 = {(𝑥𝑛,𝑝,𝑦,𝑧) ∶ (𝑥𝑛,𝑝) can produce 𝑦} (4)
𝑇 2 = {(𝑥𝑛,𝑝,𝑦,𝑧) ∶ (𝑝) can produce 𝑧} (5)
where 
𝑇 = 𝑇 1∩𝑇
2
(6)
The desirable sub-technology 𝑇 1 satisfies classic regularity assumptions 
such as the free disposability in desirable outputs and inputs, whereas under 
the undesirable sub-technology 𝑇 2, the bad outputs (nitrogen pollution) and 
the polluting inputs are assumed to be costly disposable. This property implies 
that both polluting inputs and by-products cannot be disposed without addi-
tional cost. In other words and according to Murty, Robert Russell and Levkoff 
(2012)’s perspective, if a given level of nitrogen generates some minimal level 
of runoff, then inefficiency in the use of fertilisers may imply that this level of 
fertilisers application could also generate a higher amount of runoff.
Computing farm efficiency measures that evaluate the farms’ performance 
in each of its sub-technologies requires the estimation of a production fron-
tier. We estimate the production frontier non-parametrically using DEA. First, 
we construct the frontier of the desirable sub-technology 𝑇 1 that only takes 
into consideration the interaction between conventional inputs (𝑥𝑛) and desir-
able output (𝑦) while ignoring the presence of undesirable output. Assuming 
that the observed inputs and outputs combinations are available for 𝐼 farms 
with the superscript 𝑖 = 1,…,𝐼 denoting the individual farms, then the DEA
4 See Dakpo, Jeanneaux and Latruffe (2016) for a good overview.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
Do agri-environment measures help improve environmental and economic efficiency? 9
representation of the desirable sub-technology 𝑇 1 is then expressed as: 
𝑇 1 ={(𝑥𝑛,𝑝,𝑦,𝑧) ∶ 𝑝 ≥∑
𝑖
𝜆𝑖𝑝𝑖,𝑥𝑛 ≥∑
𝑖
𝜆𝑖𝑥𝑖𝑛,𝑦 ≤∑
𝑖
𝜆𝑖𝑦𝑖,
∑
𝑖
𝜆𝑖 = 1,𝜆𝑖 ∈ ℝ𝑁+} (7)
While DEA does not require any a priori assumption on the functional form 
of the production function, some assumptions about how these inputs and out-
puts interact are needed. Our assumption for the desirable sub-technology 
𝑇 1follows the direction in which the inequality constraints are expressed in 
model (7), that is to say the desirable outputs (𝑦) are produced using the 
polluting (𝑝)and non-polluting (𝑥𝑛) inputs. Thus, according to this represen-
tation, additional units of inputs to an existing set of inputs will positively 
contribute to the production of the good outputs. The term 𝜆𝑖 represents 
intensity vector, containing the firms’ weights that are used to define the best-
practice frontier representing the best-practice technology. The assumption 
that 𝑇 1 maintains variable returns to scale (VRS) is captured by the constraint 
∑
𝑖
𝜆𝑖 = 1, 𝜆𝑖 ∈ ℝ𝑁+ ,
whereas the undesirable sub-technology 𝑇 2 under VRS can be given by: 
𝑇 2 ={(𝑥𝑛,𝑝,𝑦,𝑧) ∶ 𝑝 ≤∑
𝑖
𝜇𝑖𝑝𝑖,𝑧 ≥∑
𝑖
𝜇𝑖𝑧𝑖,∑
𝑖
𝜇𝑖 = 1,𝜇𝑖 ∈ ℝ𝑁+} (8)
where 𝜇𝑖 is a vector of intensity variables representing the weight of each 
farm that are used to construct the efficient frontier for 𝑇 2. The undesir-
able sub-technology refers to the performance of farms in controlling the 
generation of by-products. The second process (𝑇 2) responds to the assump-
tion that increased use of nitrogen application (𝑝) promotes increased runoff 
by-products and treats by-products (𝑧) as weakly disposable.
Following Murty, Robert Russell and Levkoff (2012), the global perfor-
mance that recognises the presence of desirable outputs and by-products can 
be captured by the following index: 
𝐸𝑓𝑓𝐺𝑙𝑜𝑏𝑎𝑙 (𝑥𝑛,𝑝,𝑦,𝑧 ∶ 𝑇) =
1
2 min𝜃1
{𝜃1| (𝑥𝑛,𝑝,𝑦 ⊘ 𝜃1,𝑧) ∈ 𝑇
1}
+12 min𝜃2
{𝜃2| (𝑥𝑛,𝑝,𝑦,𝑧 ⊗ 𝜃2) ∈ 𝑇
2}
(9)
where 𝑦 ⊘ 𝜃1 = 𝑦/𝜃1 and 𝑧 ⊗ 𝜃2 = 𝑧𝜃2 and 𝜃1 is the efficiency score for desir-
able output sub-technology, while 𝜃2 is the efficiency score of the undesirable 
sub-technology. Basically, the index in model (9) is simply the average5 
efficiency scores of the two sub-technologies 𝑇 1 and 𝑇 2.
5 See Chambers and Serra (2018) who discussed the possibility to use different weights of the 
different sub-technologies.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
10 A. Ait Sidhoum et al.
2.3. DiD estimation
In the third step of our analysis, the effect of AES participation on farm effi-
ciency is estimated. While PSM allows controlling for possible selection bias 
issues arising from observed characteristics, it has been shown however that 
the unobserved factors could also play a significant role in farmers’ decisions 
to participate in agri-environment programmes (Hynes and Garvey, 2009),
and therefore, by failing to control for unobserved heterogeneity, the estimated 
effect may be biased. One way of dealing with the unobserved heterogeneity 
and being able to control for AES determinants such as environmental aware-
ness or managerial ability is to combine PSM with DiD method (Heckman, 
Ichimura and Todd, 1997; Smith and Todd, 2005). More specifically, we 
employ the DiD6 approach to estimate the difference in change of farm per-
formance indicators (e.g. efficiency scores) over two periods between the 
participants and the matched non-participants. For two time periods, the DiD 
estimator measures the average programme impact as follows (Heckman, 
Ichimura and Todd, 1997):
ATT𝐷𝑖𝐷 =
1
𝑁
𝐼
∑
𝑖=1
{(𝑌𝑖,𝑡2 (1)−𝑌𝑖,𝑡1 (1))−
𝐽
∑
𝑗=1
𝑤𝑖,𝑗(𝑌𝑗,𝑡2 (0)−𝑌𝑗,𝑡1 (0))}
(10)
where 𝑡1represents the pre-treatment period and 𝑡2 represents the post-
treatment period. 𝑖 and 𝑗 are, respectively, the treated and the matched 
non-treated individuals, and 𝑁 is the number of individuals in the treatment 
group. 𝑤𝑖,𝑗 indicates the weights (0 ≤ 𝑤𝑖,𝑗 ≤ 1) given to the 𝑗
𝑡ℎ non-treated 
individual matched to 𝑖𝑡ℎ treated individual.
To derive the ATT, we apply second-stage regression analysis to provide 
empirical evidence of changes in the efficiency of AES-participating farms rel-
ative to changes in the efficiency of non-participating farms. Simar and Wilson 
(2007) demonstrated that traditional second-stage approaches (e.g. censored 
Tobit regression) to inference are unreliable due to serial correlation of DEA 
efficiency measures and show how a truncated regression with bootstrapping 
procedure provides a reliable technique for inference. Following Simar and 
Wilson (2007), we use the single bootstrap approach (Algorithm no. 1) which 
implementation would include the following steps:
1. The use of maximum likelihood (ML) to estimate the parameters of the 
following truncated regression:
𝜃𝑖𝑡 = 𝛼+𝛽𝑇𝑖+𝛾𝑡𝑖+(𝐷𝐷)𝑇𝑖𝑡𝑖+𝛿𝑗𝑍𝑖𝑗+𝜀𝑖𝑡 (11)
where 𝜃𝑖𝑡 is the efficiency score for the farm 𝑖 at time 𝑡. 𝑇𝑖 is a dummy vari-
able representing scheme participation, and 𝑡𝑖 is a time dummy taking the 
value of 0 for the pre-treatment period and 1 for the post-treatment period. 
6 It should be noted that the availability of a panel data structure makes it possible to implement 
the difference–indifference procedure.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
Do agri-environment measures help improve environmental and economic efficiency? 11
The DiD estimator is represented by the coefficient (Coeff.), 𝐷𝐷, which gives 
the estimate for the impact of the AES on the farm-level efficiency scores. 
Finally, 𝑍𝑖 represents a matrix of regressors expected to affect farm techni-
cal and environmental efficiency, and 𝜀𝑖𝑡 is the unobserved time-varying error 
component.
The model (11) generates estimates ( ̄𝛼, ̄𝛽, ̄𝛾, 𝐷𝐷,𝛿𝑗)of (𝛼, 𝛽, 𝛾,
𝐷𝐷,𝛿𝑗)  as well as an estimates of the standard deviation (S.d.) of the error 
component (𝜎𝜀).
2. Based on these estimates, a bootstrap procedure is applied over the 
following three steps 2,000 times:
(a) For each farm, 𝜀𝑖𝑡 is drawn from a 𝑁(0,𝜎2𝜀) distribution with left 
truncation at 1.
(b) Using the drawn error components 𝜀𝑖𝑡, new efficiency scores, ̂𝜃𝑖𝑡 =
̄𝛼+ ̄𝛽𝑇𝑖+ ̄𝛾𝑡𝑖+(𝐷𝐷)𝑇𝑖𝑡𝑖+𝛿𝑗𝑍𝑖𝑗+𝜀𝑖𝑡, are predicted.
(c) The use of the ML method again to estimate the truncated regression 
with the predicted efficiency estimates ̂𝜃𝑖𝑡 on (𝑇𝑖 𝑡𝑖, 𝑇𝑖𝑡𝑖,𝑍𝑖𝑗)  to 
obtain the estimates ( ̂𝛼, ̂𝛽, ̂𝛾, 𝐷𝐷,𝛿𝑗) and (𝜎𝜀)
3. Finally, confidence intervals are obtained for (𝛼, 𝛽, 𝛾, 𝐷𝐷,𝛿𝑗) and (𝜎𝜀)
by using the bootstrap estimates ( ̂𝛼, ̂𝛽, ̂𝛾, 𝐷𝐷,𝛿𝑗) and (𝜎𝜀) and the 
original values ( ̄𝛼, ̄𝛽, ̄𝛾, 𝐷𝐷,𝛿𝑗) and (𝜎𝜀).
3. Data and empirical model
In this study, we use balanced panel data of 1,626 Bavarian dairy farms,7 cov-
ering the period 2013–2018 drawn from Farm Accountancy Data Network in 
Bavaria. This database has been complemented with the official agricultural 
support data (InVeKoS) that provide further details on farm characteristics 
as well as on Pillar I and Pillar II payments received. We also used publicly 
available data from the Bavarian State Statistical Office containing information 
about socio-economic characteristics (e.g. gross domestic product, unemploy-
ment and labour force) at municipality level to improve our matching accuracy. 
The timeframe of concern is from 2014 to 2018 as the last rural development 
programme in the region of Bavaria was initiated in 2014, in accordance with 
the 7-year Multi-annual Financial Framework of the CAP. The Bavarian rural 
development programme offered a large number of AESs targeting various 
environmental categories. Most schemes have a commitment period of 5 years 
and scheme participation is voluntary, which means that farmers self-select 
into treatment based on utility considerations. For this reason, selection bias 
has to be accounted for. A widely used and well-accepted method to address 
selection bias is PSM (see Ali and Abdulai (2010); Mayen, Balagtas and 
Alexander (2010) or Pufahl and Weiss (2009) for examples of empirical studies 
7 The milk revenue of our specialised dairy farms represents at least 70 per cent of their total rev-
enue coming from animal and crop production. Because of their markedly different technology, 
organic farming systems were not taken into account.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
12 A. Ait Sidhoum et al.
in the field of agricultural economics). For the purpose of performing PSM, the 
year 2013 was considered to identify comparable treated farms and untreated 
farms based on observable variables before the scheme’s implementation.
The choice of the period 2013–2018 and the pre-treatment year 2013 was 
motivated by two factors. First, the CAP’s programming periods follow a 7-
year framework, which means that for all 7 years, the first and second pillar 
measures are refined. The period before the 2014–2020 phase ended in 2013, 
so did the 2007–2013 AES, which differed to a certain extent from those of the 
period 2014–2020 and thus represent a ‘new’ treatment. The year 2013 can 
consequently be considered the pre-treatment year. The time range that was 
chosen for the DiD analysis including the pre-treatment year is linked to the 
5-year commitment period of most AESs. Second, nitrate pollution of ground-
water in Bavaria has not improved during the 2007–2013 programming period, 
despite 2007–2013 AES targeting water quality. This resulted in additional 
efforts in the 2014–2020 period.8
In our empirical analysis, we are interested in estimating the impact of 
environmental subsidies on farm-level combined technical and environmental 
efficiency over the period 2014–2018. With the year 2013 being the pre-
treatment phase, the year 2018 is considered to be the post-treatment phase. 
We, therefore, focus on those farms that did not participate in any AESs in the 
initial time period to assess the impact of environmental subsidies. As a result, 
farms that did not receive environmental subsidies in 2013 but did receive them 
during the 2014–2018 period were assigned to the treated group, while farms 
that did not receive any subsidy during the whole period (2013–2018) were 
assigned to the control group.
Then, PSM was performed to balance farm characteristics between farms 
that participated/did not participate in AESs. A central challenge when imple-
menting PSM is which variables to include within the PSM model. While the 
literature has thoroughly discussed the issues associated with which variables 
to include in the propensity score model, yet still there is no scientific consen-
sus regarding the choice of variables in the PSM model (Austin, Grootendorst 
and Anderson, 2007).
Heckman et al. (1997, 1998) demonstrate that data quality is an important 
component of any accurate estimation technique. They showed that estimators 
are only consistent when they are applied on data from a comparison group 
that meets the following requirements: (i) the same data sources are used for 
participants and non-participants, allowing characteristics to be quantified in 
a similar way, (ii) participants and non-participants are from the same area 
and (iii) the data include a diverse set of indicators that influence programme 
participation as well as outcomes. These requirements are obviously met by 
our data.
8 Twenty out of a total of 34 individual schemes (almost 60 per cent) of the most important Bavar-
ian agri-environment programme KULAP can be expected to have either a direct or an indirect 
effect on nitrogen reduction. Among these, 20 schemes are the most popular ones with large 
agricultural areas under programme (e.g. extensive use of grassland, catch crops, organic 
farming).
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
Do agri-environment measures help improve environmental and economic efficiency? 13
While there is no scientific consensus on how many variables to include in 
a propensity score binary model, we are aware of the fact that the inclusion 
of non-significant variables can increase the variance of the propensity score 
estimates; however, they do not bias the estimates nor make them inconsistent. 
Augurzky and Schmidt (2001) discuss the pros and cons of including a lot of 
variables that are not significant. They show that a smaller set of covariates 
can lead to a better estimation of the treatment effect. However, in two of their 
three testing sets, they explicitly include variables that do either not or only 
weakly influence the outcome and variables that are relevant for the outcome 
but irrelevant to the treatment decision. The smallest and best-performing 
set only includes covariates that influence both, which is a necessary condi-
tion for a successful matching procedure. In their conclusion, Augurzky and 
Schmidt (2001, p. 27) emphasise that ‘the main criterion of success for match-
ing remains the balance of the relevant covariates and not the proper estimation 
of the selection equation’. According to Rubin and Thomas (1996), even if the 
variables are not statistically significant, they should be included. Our variable 
selection is based on theoretical economic arguments and empirical evidence 
(Arata and Sckokai, 2016; Chabé-Ferret and Subervie, 2013; Defrancesco 
et al., 2007; Matzdorf and Lorenz, 2010; Mennig and Sauer, 2020). Table A1 
presents descriptive statistics for the variables included in the PSM estimation.
After having defined the treated and untreated farms and the potential rele-
vant covariates for the matching procedure, the propensity score9 is calculated 
using a logit regression as a measure of the probability that a farm will be 
classified as a programme participant. Logit model results for the PSM are 
presented in Table A2. The likelihood ratio test is statistically significant at 
1 per cent level, indicating that all farm characteristics considered are jointly 
significant in explaining programme participation.
Propensity scores were calculated for each observation based on the param-
eter estimates of the logit model, which were then used to match participant 
and non-participant farms. Different matching algorithms10 were tested prior 
to selecting the nearest neighbour estimator (1:1) without replacement. Before 
matching, significant differences are assumed to be found between the treated 
and control groups, and therefore, the resultant balance of the relevant covari-
ates assesses the success of propensity score estimation. Table 1 shows 
9 The propensity score represents the conditional probability of participation for farm given a set 
of i X =xi observed characteristics p (X )=Pr (P =1|X =x i). The propensity score is estimated 
from a logit model in which the binary treatment variable (AES) serves as the dependent variable 
conditional upon the observed variables (covariates).
10 We tested the most common matching algorithms: kernel matching, radius matching and nearest 
neighbour matching without and with replacement from 1 to 10 neighbours. We compared the 
different matching algorithms and found that 1:1 nearest neighbour matching without replace-
ment using a calliper width of 0.3 performed best. Rosenbaum and Rubin (1985) propose the use 
of standardised bias (SB) to compare treated unit means and untreated unit means before and 
after matching as a measure of covariate balance. As noted by Caliendo and Kopeinig (2008), an 
SB below 5 after matching would be seen as sufficient. Our findings indicate that the overall SB 
was reduced from 38.5 to 3.2 per cent by the matching procedure. By comparison, the overall SB 
was reduced to 5.4 per cent, 3.6 per cent and 4.5 per cent with nearest neighbour matching with 
replacement, radius matching and kernel matching, respectively.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
14 A. Ait Sidhoum et al.
T
ab
le
 1
. 
M
ea
n 
an
d 
m
ed
ia
n 
bi
as
 r
ed
uc
tio
n 
of
 r
el
ev
an
t c
ov
ar
ia
te
s 
be
fo
re
 a
nd
 a
ft
er
 m
at
ch
in
g 
th
e 
pr
e-
tr
ea
tm
en
t y
ea
r 
20
13
 B
ef
or
e 
m
at
ch
in
g
 A
ft
er
 m
at
ch
in
g
 S
B
V
ar
ia
bl
es
C
on
tr
ol
 m
ea
n
T
re
at
ed
 m
ea
n
C
on
tr
ol
 m
ea
n
T
re
at
ed
 m
ea
n
B
ef
or
e 
m
at
ch
in
g
A
ft
er
 m
at
ch
in
g
L
iv
es
to
ck
 u
ni
ts
 p
er
 h
a
1.
05
0*
**
0.
84
0
0.
94
9
0.
94
7
−
59
.8
−
0.
7
L
ab
ou
r 
pe
r 
ha
0.
03
6*
*
0.
02
8
0.
03
2
0.
03
2
−
60
.1
6
C
ap
ita
l d
ep
re
ci
at
io
n 
pe
r 
ha
62
8.
81
0*
*
57
2.
21
0
59
2.
27
0
60
2.
59
0
−
16
.5
3
Sa
le
s 
pe
r 
ha
3,
91
3.
70
**
3,
27
5.
60
3,
56
4.
10
0
3,
57
4.
10
0
−
51
.7
0.
8
Fe
rt
ili
se
rs
 p
er
 h
a
17
8.
08
0*
*
15
8.
75
0
16
3.
83
0
16
6.
07
0
−
24
.6
2.
9
Pe
st
ic
id
e 
pe
r 
ha
62
.9
50
63
.4
60
60
.0
68
59
.4
46
1.
3
2.
6
Fe
ed
 p
er
 h
a
60
6.
98
0*
**
50
1.
44
0
56
4.
64
0
55
3.
36
0
−
32
.4
−
3.
5
Sh
ar
e 
of
 a
ra
bl
e 
la
nd
0.
57
1
0.
58
9
0.
57
1
0.
57
3
9
1.
1
Sh
ar
e 
of
 g
ra
ss
la
nd
0.
42
7
0.
41
1
0.
42
5
0.
42
6
−
8.
2
0.
5
Y
ie
ld
 in
de
x 
pe
r 
ha
77
.8
33
**
*
49
.0
25
56
.5
18
58
.7
38
−
51
.4
4
G
D
P
28
,2
40
.0
00
**
*
27
,4
54
.0
00
28
,0
31
.0
00
28
,0
33
.0
00
−
16
.6
0.
1
N
um
be
r 
of
 d
ai
ry
 f
ar
m
s
12
4
14
7
69
69
To
ta
l n
um
be
r 
of
 f
ar
m
s
 2
71
 1
38
**
 a
nd
 *
**
 in
di
ca
te
 s
ta
tis
tic
al
 s
ig
ni
fic
an
ce
 a
t 1
pe
r 
ce
nt
 a
nd
 0
.1
pe
r 
ce
nt
, r
es
pe
ct
iv
el
y,
 o
f 
a 
t-
te
st
 o
n 
th
e 
eq
ua
li
ty
 o
f 
m
ea
n 
di
ff
er
en
ce
s 
be
tw
ee
n 
ob
se
rv
at
io
ns
 f
ro
m
 th
e 
tr
ea
te
d 
an
d 
th
e 
co
nt
ro
l g
ro
up
s.
N
ot
e 
1:
 W
he
n 
us
in
g 
a 
1:
1 
ne
ar
es
t 
ne
ig
hb
ou
r 
m
at
ch
in
g 
al
go
ri
th
m
, 
ob
se
rv
at
io
ns
 t
ha
t 
fa
ll
 o
ut
si
de
 t
he
 r
eg
io
n 
of
 c
om
m
on
 s
up
po
rt
 a
s 
w
el
l 
as
 o
bs
er
va
ti
on
s 
fo
r 
w
hi
ch
 t
he
re
 i
s 
no
 m
at
ch
in
g 
pa
rt
ne
r 
fo
r 
a 
tr
ea
te
d 
in
di
vi
du
al
 th
at
 is
 c
lo
se
st
 in
 te
rm
s 
of
 th
e 
pr
op
en
si
ty
 s
co
re
 h
av
e 
to
 b
e 
di
sc
ar
de
d 
fr
om
 th
e 
an
al
ys
is
. C
on
se
qu
en
tl
y,
 th
e 
sa
m
pl
e 
si
ze
 is
 r
ed
uc
ed
 a
ft
er
 m
at
ch
in
g.
N
ot
e 
2:
 P
SM
 c
an
 e
lim
in
at
e 
th
e 
se
lf
-s
el
ec
tio
n 
bi
as
 is
su
es
 a
ri
si
ng
 f
ro
m
 o
bs
er
ve
d 
ch
ar
ac
te
ri
st
ic
s.
 A
lt
ho
ug
h 
w
e 
ha
ve
 in
cl
ud
ed
 a
s 
m
an
y 
co
va
ri
at
es
 a
s 
po
ss
ib
le
 f
or
 m
at
ch
in
g 
pu
rp
os
es
, s
om
e 
un
ob
se
rv
ab
le
 
fa
ct
or
s 
m
ay
 c
au
se
 e
st
im
at
io
n 
bi
as
es
. T
he
re
fo
re
, w
e 
us
e 
th
e 
R
os
en
ba
um
 s
en
si
tiv
it
y 
an
al
ys
is
 to
 c
he
ck
 w
he
th
er
 th
er
e 
ar
e 
ot
he
r 
fa
ct
or
s 
th
at
 a
ff
ec
t t
he
 q
ua
li
ty
 o
f 
m
at
ch
in
g.
 T
he
 a
na
ly
si
s 
in
di
ca
te
s 
th
at
 th
e 
re
su
lts
 a
re
 q
ui
te
 r
ob
us
t t
o 
un
ob
se
rv
ed
 f
ac
to
rs
. T
he
se
 r
es
ul
ts
 a
re
 a
va
il
ab
le
 f
ro
m
 th
e 
au
th
or
s 
up
on
 r
eq
ue
st
.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
Do agri-environment measures help improve environmental and economic efficiency? 15
covariates’ mean values before and after matching between the two groups. 
These results suggest that no significant differences between participating and 
non-participating farms remain after matching. We can therefore conclude 
that the applied matching algorithm worked well, as the existing observable 
differences have been controlled for.11 The matched treatment and control 
groups are used to estimate the average effect of AES participation on the com-
bined farm-specific technical and environmental efficiency using a DEA-DiD
model.
In our empirical efficiency analysis, we rely on balanced panel data of 
138 Bavarian dairy farms covering the period 2013–2018 (828 observations in 
total). Seven input variables were defined and used in the analysis: livestock 
units expressed in number of cows (𝑥1); total labour (𝑥2 measured in man-
work units); utilised land (𝑥3 in hectares); capital depreciation (𝑥4 in Euros); 
crop-specific inputs, namely expenses for chemicals (pesticides and fertilis-
ers) (𝑥5 in Euros); expenses for feed (𝑥6 in Euros) and quantities of nitrogen 
input (𝑞 in kilograms). The desirable output is total farm sales (𝑦 in Euros). 
Finally, the undesirable output is represented by the nitrogen balance12 (𝑧 in 
kilograms), computed as the difference between the nitrogen contained in the 
inputs and the nitrogen contained in the outputs. Table 2 shows the average val-
ues of the main variables. On average, and for the period under consideration 
(2013–2018), firms in our sample generate total sales of around 223,000 Euros. 
On average, our sample farms manage 58 ha of land, have average livestock 
units of 58 cows, devote 1.75 man-work units of labour per year, have a capital 
depreciation value of 35,575 Euros and spend around 13,000 and 32,200 Euros 
on chemicals (fertilisers and pesticides) and feed, respectively. 
The average nitrogen balance per farm in our sample is slightly less 
than 6,285 kg per year. The nitrogen balance shows substantial heterogeneity 
between farms, with a Coeff. of variation of 0.68. On a per hectare basis, aver-
age nitrogen balances declined from 109 kg/ha in 2014 to 83 kg/ha in 2018. 
Eulenstein et al. (2008) calculate the average nitrogen surplus to be on the 
order of 91 kg/ha, while OECD (2013) estimates the annual average nitro-
gen balance in the period 2007–2009 for the EU 15 agricultural sectors to 
be 65 kg/ha, with substantial variation from 204 kg/ha in the Netherlands to 
25 kg/ha in Greece. OECD (2013) places this value around 85 kg/ha for Ger-
many. The difference in the results reported in the literature usually depends 
on the method used and the type of farms under investigation.
11 The non-participating farms were chosen to be similar to the participating ones across several 
observable characteristics including, among others, labour, land, livestock density and capital 
depreciation. It is, therefore, reasonable to assume that all producers share the same technology.
12 In order to estimate the nitrogen balance, we follow Gamer and Bahrs (2010). We use Wendland 
et al. (2018)’s Coeff. to approximate the quantity of nitrogen contained in milk and meat outputs as 
well as the nitrogen content in feed input and the LFL (2013) Coeff. to approximate the quantities of 
nitrogen that have been fixed by legumes, while for mineral fertilisers, the quantities of nitrogen 
can be computed from the information provided in STATBA (2018).
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
16 A. Ait Sidhoum et al.
T
ab
le
 2
. 
S
um
m
ar
y 
st
at
is
ti
cs
 (
av
er
ag
e 
an
d 
S.
d.
—
in
 p
ar
en
th
es
is
) 
fo
r 
th
e 
m
ai
n 
va
ri
ab
le
s 
in
 th
e 
sa
m
pl
e 
(8
28
 f
ar
m
s)
V
ar
ia
bl
e
Sy
m
bo
l
D
im
en
si
on
20
13
20
14
20
15
20
16
20
17
20
18
Fu
ll
 p
er
io
d
(2
01
3–
20
18
)
To
ta
l s
al
es
𝑦
E
ur
os
21
0,
66
5.
51
 
(1
04
,0
27
.0
2)
23
7,
14
1.
27
 
(1
16
,7
35
.2
5)
21
9,
21
2.
20
 
(1
08
,6
14
.0
0)
20
1,
70
2.
05
 
(1
00
,9
98
.8
9)
21
4,
42
1.
12
 
(1
12
,8
00
.4
5)
25
7,
44
5.
99
 
(1
42
,7
65
.6
9)
22
3,
43
1.
36
 
(1
12
,6
02
.1
1)
L
iv
es
to
ck
 
un
its
𝑥
1
N
um
be
r
55
.4
9 
(2
6.
39
)
57
.6
9 
(2
8.
25
)
58
.2
8 
(2
9.
06
)
59
.0
5 
(3
0.
02
)
60
.1
4 
(3
2.
06
)
60
.6
0 
(3
3.
57
)
58
.5
4 
(2
9.
50
)
L
ab
ou
r
𝑥
2
M
an
-w
or
k 
un
its
1.
70
 (
0.
54
)
1.
73
 (
0.
57
)
1.
75
 (
0.
58
)
1.
76
 (
0.
55
)
1.
79
 (
0.
58
)
1.
79
 (
0.
58
)
1.
75
 (
0.
54
)
L
an
d
𝑥
3
he
ct
ar
es
57
.2
6 
(2
4.
56
)
57
.8
8 
(2
4.
67
)
58
.0
3 
(2
4.
55
)
59
.1
8 
(2
5.
95
)
59
.6
8 
(2
6.
62
)
60
.3
1 
(2
6.
82
)
58
.7
2 
(2
5.
35
)
C
ap
ita
l 
de
pr
ec
ia
tio
n
𝑥
4
E
ur
os
35
,7
37
.0
5 
(2
2,
37
2.
23
)
36
,3
83
.5
1 
(2
3,
24
1.
15
)
36
,2
61
.9
5 
(2
3,
41
4.
60
)
34
,4
97
.5
0 
(2
3,
31
4.
33
)
34
,5
43
.5
0 
(2
5,
11
6.
38
)
36
,0
28
.3
7 
(2
7,
43
7.
59
)
35
,5
75
.3
1 
(2
3,
33
8.
10
)
C
he
m
ic
al
s
𝑥
5
E
ur
os
13
,7
64
.5
2 
(1
0,
30
3.
63
)
14
,3
34
.5
5 
(9
,6
19
.8
2)
13
,8
57
.7
3 
(1
0,
20
8.
55
)
13
,4
06
.2
0 
(1
0,
52
5.
50
)
11
,5
46
.5
4 
(8
,4
01
.3
7)
11
,1
12
.7
6 
(7
,7
92
.3
9)
13
,0
03
.7
2 
(9
,1
83
.2
8)
Fe
ed
𝑥
6
E
ur
os
31
,9
94
.3
6 
(2
0,
02
0.
74
)
33
,2
25
.1
9 
(2
0,
31
5.
27
)
31
,0
12
.2
8 
(1
9,
99
6.
46
)
32
,0
42
.0
8 
(2
1,
40
9.
53
)
31
,7
10
.1
9 
(2
1,
65
4.
38
)
33
,2
38
.6
5 
(2
2,
95
0.
68
)
32
,2
03
.7
9 
(2
0,
33
3.
09
)
N
itr
og
en
 
in
pu
t
𝑞
kg
7,
65
7.
81
 
(4
,7
95
.3
0)
9,
12
1.
79
 
(5
,2
21
.7
9)
8,
47
0.
47
 
(5
,2
71
.8
3)
8,
84
9.
60
 
(5
,8
44
.0
1)
8,
58
4.
21
 
(5
,7
13
.2
3)
8,
02
9.
11
 
(5
,0
51
.0
9)
8,
45
2.
16
(5
,1
55
.1
2)
N
itr
og
en
 
ba
la
nc
e
𝑧
kg
5,
28
5.
20
 
(3
,9
56
.3
7)
6,
46
0.
38
 
(4
,1
48
.0
9)
5,
96
0.
46
 
(4
,3
53
.4
6)
6,
18
8.
52
 
(4
,8
29
.8
2)
5,
94
1.
28
 
(4
,6
81
.5
2)
5,
23
0.
56
 
(3
,9
07
.1
3)
6,
28
4.
89
 
(4
,2
88
.5
9)
N
ot
e:
 M
on
et
ar
y 
va
ri
ab
le
s 
w
er
e 
de
fla
te
d 
to
 2
01
5 
va
lu
es
 u
si
ng
 th
e 
ap
pr
op
ri
at
e 
pr
ic
e 
in
di
ce
s.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
Do agri-environment measures help improve environmental and economic efficiency? 17
4. Results and discussion
4.1. Technical and environmental efficiencies
It is well known that non-parametric efficiency calculations are very sensi-
tive to the presence of outliers. We have used the super-efficiency method 
proposed by Banker and Chang (2006) to detect and remove potential out-
liers in the dataset. Thus, observations with extreme super-efficiency ratings 
were eliminated from the sample. By using the tolerance level of 1.2, 18 
farms were found to be dominating observations (outliers) and thus have 
been removed.13 As a result, we have decided to trim 13 per cent (18 farms 
out of 138) of the observations and work with this trimmed sample going
forward.14
The frequency distributions and the mean values of the overall efficiency 
and its technical and environmental components are summarised in Table 3. 
Non-parametric kernel density distributions of the overall efficiency and both 
technical and environmental efficiencies are presented in Figure 1. The envi-
ronmental efficiency distribution (in blue) shows a highly right-skewed density 
distribution, with around 50 per cent of the sample having an efficiency score 
below 0.65. These results point to a significant difference between calculated 
mean values for technical efficiency (0.920) and nitrogen balance efficiency 
(0.649). This suggests that the sample dairy farms could on average decrease 
their nitrogen pollution levels by around 35 per cent, while maintaining their 
desirable output efficiency unaltered. These differences between technical and 
environmental performance are in line with earlier research that has shown 
that a high level of economic efficiency is associated with a lower level of 
environmental performance (Ait Sidhoum, Hervé Dakpo and Latruffe, 2022; 
Mayen, Balagtas and Alexander, 2010). On average, only one farm was found 
to be fully efficient in controlling nitrogen pollution over the analysed period, 
while the most inefficient farm had a nitrogen efficiency rating of 0.260, imply-
ing that it could maintain its level of desirable output while reducing nitrogen 
pollution by 74 per cent. For instance, using the summary statistics of the 
nitrogen balance (Table 2), the calculated nitrogen pollution efficiency ratings 
indicate that the representative farm could reduce its surplus of nitrogen by 
0.35×6284.89 = 2199.71Kg N.
Our findings indicate a relatively high degree of environmental inefficiency 
in Bavarian dairy farms. This could be due to the fact that fertiliser control reg-
ulations are not implemented effectively (Kirschke et al., 2019). This finding 
is not specific to the region; in fact, various related studies in the literature 
reported low environmental efficiency scores, where some studies explic-
itly focused on dairy farms. Examples of such research include the study by 
13 Note that the removed outliers have been identified from both participants and non-participants 
farms (nine observations in each group), and comparison between treated and untreated units 
has been repeated without these outliers, but results did not change significantly.
14 Despite the fact that such a sample size might appear to be small, it is in line with other studies 
in the efficiency and productivity measurement literatures (Alvarez and Del Corral, 2010; Fraser 
and Cordina, 1999; Pérez Urdiales, Lansink and Wall, 2016; Skevas and Serra, 2016).
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
18 A. Ait Sidhoum et al.
Table 3. Frequency distribution of efficiency scores over the period 2013–2018
Year Technical Environmental Overall
0 < 𝜆 < 0.1 0 0 0
0.1 ≤ 𝜆 < 0.2 0 0 0
0.2 ≤ 𝜆 < 0.3 0 1 0
0.3 ≤ 𝜆 < 0.4 0 3 0
0.4 ≤ 𝜆 < 0.5 0 14 0
0.5 ≤ 𝜆 < 0.6 0 24 0
0.6 ≤ 𝜆 < 0.7 0 39 24
0.7 ≤ 𝜆 < 0.8 17 21 42
0.8 ≤ 𝜆 < 0.9 21 13 41
0.9 ≤ 𝜆 < 1 61 4 12
𝜆 = 1 21 1 1
Average score 0.920 0.649 0.785
Fig. 1. Estimated kernel density distributions of the overall, environmental and technical efficiency 
scores (2013–2018). 
Reinhard and Thijssen (2000), who reported nitrogen pollution efficiency lev-
els of 0.56 for Dutch dairy farms, while Adenuga et al. (2019), who estimated 
environmental efficiency of a sample of Irish dairy farms, found relatively 
high-efficiency ratings of the order of 0.92. The use of different methods 
precludes a direct comparison of the efficiency results; however, Guesmi and 
Serra (2015) and Ait Sidhoum, Serra and Latruffe (2020) have used a compara-
ble approach to estimate the nitrogen efficiency of a sample of Spanish arable 
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
Do agri-environment measures help improve environmental and economic efficiency? 19
Table 4. Average efficiency scores over the period 2013–2018
Year Technical Environmental Overall
2013 0.926 (0.084) 0.617 (0.159) 0.772 (0.096)
2014 0.917 (0.082) 0.723 (0.128) 0.820 (0.086)
2015 0.912 (0.104) 0.672 (0.133) 0.792 (0.095)
2016 0.925 (0.091) 0.633 (0.159) 0.779 (0.095)
2017 0.919 (0.098) 0.631 (0.161) 0.775 (0.104)
2018 0.921 (0.099) 0.619 (0.200) 0.770 (0.117)
Note: S.d. is given in parentheses.
crops and found similar patterns of nitrogen efficiency ratings, 0.69 and 0.57,
respectively.
From Table 4, we observe that the technical efficiency scores show only a 
small difference over the period 2013–2018.15 The absence of fluctuating tech-
nical efficiency levels in the milk sector can be explained by the abolition of 
the milk quotas in 2015, which prevents farmers from adjusting their produc-
tion if the established quotas are exceeded and thus farmers are more likely to 
follow a well-structured and well-designed strategy that aims to expand milk 
production if enough land is available (Samson, Gardebroek and Jongeneel, 
2016). The left-skewed distribution for technical efficiency (in green) shows 
evidence that many farms are performing very well in terms of desirable out-
puts. This implies that most of the farms are highly efficient, while only a few 
of them have an efficiency score below 0.800. 
As mentioned, our results point to a relatively high level of technical effi-
ciency in Bavarian dairy farms, but the fact that no prior studies have been 
carried out to estimate the technical efficiency of Bavarian dairy farms using 
the approach we use precludes a direct comparison of our findings with exist-
ing results reported in the literature. However, an extensive empirical literature 
exists on estimating dairy farm technical efficiency using DEA. A short and 
non-exhaustive list includes the studies by Fraser and Cordina (1999); Hansson 
and Öhlmér (2008); Jaforullah and Whiteman (1999); Latruffe, Fogarasi and 
Desjeux (2012); Reinhard, Knox Lovell and Thijssen (2000); Shortall and 
Barnes (2013) and Stokes, Tozer and Hyde (2007). For Bavarian dairy farms, 
Wimmer and Sauer (2020) applied a stochastic frontier analysis using an 
input distance function to estimate technical efficiency for a sample of dairy 
farms (2000–2014) and found an average technical efficiency of around 0.750, 
whereas in a study of 1,530 dairy farmers covering the period 2007–2011, 
Mennig and Sauer (2020) found an average technical efficiency for dairy farms 
of 0.902.
15 Since the technological level varies from year to year because of technological advances and 
because of different weather conditions, it is common to run efficiency analyses for each year 
separately (Blancard et al., 2006).
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
20 A. Ait Sidhoum et al.
4.2. Effectiveness of agri-environment measures
European citizens are expecting more effective spending of EU budgets. It 
is therefore of interest to investigate the effectiveness of agri-environment 
measures for delivering environmental benefits and improving farms’ eco-
nomic efficiency. This is done by calculating the difference between the farms 
participating in AES and non-participating farms in terms of average efficiency 
levels.
There is a growing body of research reporting on the impact of AES on farm 
performance (Arata and Sckokai, 2016; Baráth, Ferto˝ and Bojnec, 2020; Mary, 
2013; Mennig and Sauer, 2020). However, our reading of the environmental 
economics literature reveals that the impact of AES on the combined environ-
mental and technical efficiency measures has not been investigated so far. Our 
analysis clearly brings a new perspective on the link between AESs and farm 
performance. Our results are shown in Table 5. The average treatment effect 
of AESs on efficiency scores is represented by the DiD estimator (obtained 
from model (11)), which compares the changes in efficiency scores between 
participants and the matched non-participants over the period 2013–2018.16 A 
positive DiD estimator represents an increase in the mean efficiency scores of 
the participants that is greater than the increase of their matched counterparts 
or that the decrease in the mean efficiency scores of the participants is lower 
than the decrease of the non-participants. 
Although agri-environment measures have been launched initially to pre-
vent the negative impacts of intensive agricultural systems on environmental 
aspects, several studies have highlighted the significant role that economic 
factors can play in farmers’ willingness to adopt AESs (Defrancesco et al., 
2007; Mozzato et al., 2018). Thus, the environmental effectiveness of the agri-
environment policies cannot be assessed in isolation but should always involve 
the economic efficiency of the schemes. The corresponding DiD estimator 
that captures the impact of the AES on economic efficiency is not signifi-
cant. This finding is consistent with the notion in which the loss of economic 
resources associated with the adoption of environmentally friendly practices 
has been adequately compensated by the AES. Our results are partially con-
sistent with those recently reported in the literature. The difference between 
AES-participating and non-participating farms in terms of productivity growth 
was found to be positive for arable farms and not significant for dairy farms 
(Mennig and Sauer, 2020). Arata and Sckokai (2016) found that the economic 
impact of AES varies across countries.
16 Since we are relying on an aggregated variable to captures the impact of AES participation on 
efficiency scores and in an attempt to disaggregate the different individual schemes and their 
potential effects at least partly, we formed four agri-environment categories and assigned the 
individual schemes to these categories and included the categories as dummy variables in the 
regression models. This approach was chosen as the number of individual schemes amounts to 
34. The agri-environment categories that were defined followed the main programme goals set by 
the Bavarian State Ministry of Food, Agriculture and Forestry: climate protection, soil and water 
conservation, biodiversity conservation and cultural landscape preservation. Detailed results are 
reported in Tables A5–A7.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
Do agri-environment measures help improve environmental and economic efficiency? 21
Table 5. Average treatment effect of AES on efficiency scores, 2013–2018
 Technical efficiency  Environmental efficiency  Overall efficiency
Treated
mean
Control
mean
Treated
mean
Control
mean
Treated
mean
Control
mean
Pre-treatment 0.925 0.927 0.629 0.606 0.777 0.766
Post-treatment 0.924 0.914 0.666 0.645 0.795 0.779
DiD estimator 0.015
(0.035)
−0.006
(0.028)
0.005
(0.019)
Share of 
grassland
−0.035*
(0.018)
−0.004
(0.013)
−0.003
(0.009)
Insurance −0.031
(0.023)
0.034**
(0.016)
0.004
(0.011)
Capital/labour −0.012
(0.015)
0.067***
(0.01)
0.021***
(0.007)
Age −0.01
(0.039)
−0.105***
(0.029)
−0.082***
(0.021)
Soil type −0.031**
(0.014)
0.017
(0.011)
0.012
(0.008)
Notes: Bootstrapped standard errors are shown in parenthesis.
Significance at the 10 per cent, 5 per cent and 1 per cent level is indicated by *, ** and ***, respectively.
AES participation is projected to affect environmental effectiveness, as they 
offer incentives to participant farmers to adopt environmentally sustainable 
farming practices. However, our findings for the Bavarian dairy farms suggest 
that AES payments do not significantly affect the environmental efficiency, 
indicating that the mean change in the efficiency in controlling nitrogen pollu-
tion from 2013 to 2018 does not significantly differ between the subsidised and 
the non-subsidised farms. The absence of the effect of AES payments on farm-
level environmental efficiency is surprising insofar as participation in these 
schemes17 requires the adoption of sustainable practices that, among others, 
impose restrictions on the use of mineral fertilisers which could be expected 
to reduce nitrogen pollution and thus improves the environmental farm per-
formance, ceteris paribus. In fact, practice-based schemes continue to prevail 
among agri-environment measures because they are relatively easy to comply 
with and do not require a real change of agricultural practices (Burton and 
Schwarz, 2013; Kleijn et al., 2011). This might explain why we do not find a 
significant impact of AES on environmental efficiency. Moreover, the literature 
might provide insights on possible reasons for this non-effectiveness. Batáry 
et al. (2015) perform a meta-analysis that covers a wide swath of literature 
over the last 20 years and conclude that agri-environment measures have been 
17 The most widespread AESs in Bavaria were related to measures for arable land and grassland 
measures. Dairy farms are located mainly in the alpine region and Bavarian forest where grass-
land is one of the most dominating agricultural land uses and most of the adopted AES were 
under the category of grassland measures (mainly aimed at an input reduction and production 
extensification).
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
22 A. Ait Sidhoum et al.
generally successful in improving farmland biodiversity. However, in their 
meta-analysis, Batáry et al. (2015) asserted that the environmental effective-
ness of AES depends on the compatibility of the scheme’s design with respect 
to the specific region in which they are implemented. Kleijn and Sutherland 
(2003) have reached a similar conclusion on the ambiguous pattern of environ-
mental effectiveness of AES aimed at improving farmland biodiversity. Our 
findings are in agreement with the above-mentioned results and found no clear 
support that participating farms differ significantly from non-participants with 
regard to the environmental efficiency. Furthermore, selection bias could not 
be excluded either in this study or in the agri-environment programmes in gen-
eral, as farmers operating in areas experiencing high levels of environmental 
degradation are unlikely to join a programme that limits their decision-making 
options (Brady et al., 2009; Schmit and Rounsevell, 2006), which in turn will 
affect the study findings. When it comes to the effect of AES on the overall 
efficiency scores, the results are similar to those obtained from the above-
discussed AES effects. This result is expected since the overall effectiveness 
(economic and environmental) of the schemes is particularly penalised by the 
lack of the combined economic and environmental effectiveness.
In sum, our findings show that there is no significant connection between 
AES and farm performance, and some possible explanations have been dis-
cussed earlier. Still, it is important to mention that there is a range of studies 
that highlighted the link between subsidies and the deterioration of farm-
ers’ motivation to produce efficiently (Ahovi, Schneider and Lansink, 2021; 
Martinez Cillero et al., 2018; Skevas, Lansink and Stefanou, 2012). This 
could lead to increased use of inputs and poor farming practices (Latruffe and 
Fogarasi, 2009). Therefore, in order to test for the robustness of our results, 
we replicated our estimations considering short post-treatment periods. While 
participating farmers are usually tied to management plans for at least 5 years, 
these robustness checks allow us to examine whether participating farmers 
were able to improve their performance within a short time frame. Results 
of the average treatment effect of AES on efficiency scores over three different 
periods are presented in Table A3. Results remain unchanged.
Table 5 also summarises the results of the regression of the efficiency scores 
against potential influencing variables. Table A4 provides descriptive statistics 
of these variables. The contextual variables we choose are farmers’ age, share 
of grassland, insurance, the ratio of capital to labour which is used as a proxy 
for technology and soil type. Following Mennig and Sauer (2020), the soil 
type dummy variable is measured on the basis of the yield index unit. Our 
farms have been divided into two subsamples, one containing farms with a 
yield index above the median (1= high-quality soil type) and farms below 
that level (0= low-quality soil type). Positive and negative signs of the related 
variables represent the increase and decrease of efficiency scores, respectively. 
Full regression results are available in Tables A5–A7.
The results from the regression of technical efficiency scores against these 
potential influencing variables show that farmers’ age does not appear to have 
a significant impact on technical efficiency. This result is not surprising given 
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
Do agri-environment measures help improve environmental and economic efficiency? 23
the fact that the effect of age on economic performance is widely contested in 
the literature; older farmers who may have more experience are suggested to 
be more efficient (Coelli and Battese, 1996). On the other hand, other stud-
ies suggest that younger farmers are expected to be more efficient since they 
are more knowledgeable about technical developments (Weersink, Turvey and 
Godah, 1990). A higher share of grass on total farm land area is associated with 
efficiency levels. This result is in line with previous analyses of German farms 
within a single output framework (Lakner and Brümmer, 2008; Wimmer and 
Sauer, 2020). Results show that insurance, capital-to-labour ratio and soil type 
do not exert a statistically significant effect on technical efficiency scores. The 
lack of significant effect of the soil type Coeff. is surprising, as the theory and 
common sense suggest that better soils are likely to increase yields, leading to 
increased farmers’ technical efficiency. However, a similar result was reported 
by Mennig and Sauer (2020), who concluded that better soil conditions do not 
lead to productivity improvements.
Turning now to the outcome of the regression of the environmental effi-
ciency scores against the same set of explanatory variables, farms operating 
under better soil conditions do not seem to perform well in terms of envi-
ronmental efficiency. However, the effect of such factor is debatable because 
unfavourable conditions may promote better management practices or may 
have the opposite effect. Indeed, environmental impacts are observed to be 
higher on fertile soils that favour the development of weeds. Farmers with 
higher insurance coverage have higher environmental efficiency. This result is 
in line with Mishra, Wesley Nimon and El-Osta (2005), who explained that 
the coverage provided by the insurance may reduce the use of agrochemicals 
inputs and thus improve nitrogen pollution control. Results show that the share 
of grassland does not exert a statistically significant effect on environmen-
tal efficiency scores. Although a higher capital-to-labour ratio may increase 
economic outcomes through labour-enhancing investments, the impact of 
investments on environmental performance would depend on whether the 
investments are categorised as green technologies or not. In our case, we 
find a positive effect. Finally, we find that younger farmers have higher envi-
ronmental efficiency since they may be more aware of the external effects 
of nitrogen pollution. This finding is consistent with several other studies. 
Concerning the results from the regression of the overall efficiency scores, 
we find similar patterns as those observed for the environmental efficiency
measures.
5. Conclusion
Understanding the effects of AESs on farm-level performance is important for 
policymakers in making sound decisions when designing schemes that meet 
economic and environmental goals. For a sample of Bavarian dairy farms over 
the period 2013–2018, we propose a theoretical framework to analyse the 
impact of agri-environment measures on farm-level technical and environmen-
tal efficiency indices. In our empirical application, environmental efficiency 
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
24 A. Ait Sidhoum et al.
is proxied by nitrogen pollution efficiency as nitrogen surplus from livestock 
farming is recognised as one of the most pressing environmental concerns 
affecting water quality in Bavaria. To control for potential selection bias aris-
ing from observable and unobservable characteristics, we combine matching 
with DiD estimators to identify comparable participating and non-participating 
farms. Our farm performance measures are based on a multi-equation rep-
resentation of the production frontier (Murty, Robert Russell and Levkoff, 
2012).
Our results show that the sample farms have a technical efficiency of 0.920 
on average. Our environmental performance measures focusing on nitrogen 
pollution show an average score of 0.649, implying that there is a considerable 
reduction potential in terms of nitrogen pollution. The lack of effective envi-
ronmental regulation to control dairy farmers’ nitrogen pollution in Bavaria 
in the observation period may explain the low environmental efficiency esti-
mates. It was only in 2017 and 2020 that through major reforms of the 
German fertilisation ordinance, more stringent measures aiming at a reduc-
tion of nitrogen losses to the environment were introduced. AESs do not 
seem to have been successful in reducing nitrogen pollution. Our empirical 
findings show a non-significant impact of AES on environmental efficiency 
scores, as the average change over the period 2013–2018 does not signif-
icantly differ between participants and their matched non-participants. The 
same impact was found on technical efficiency levels. This finding does not 
violate the well-known standards of the World Trade Organization that agri-
environment programmes should not distort trade or production. In general, 
however, much more research is needed on the economic and environmen-
tal effectiveness of AES, especially with respect to how scheme payments 
may influence economic performance (Ansell et al., 2016; Hasler et al.,
2022).
This study illustrates the relevance of accounting for runoff by-products 
when modelling the overall production process of a farm. The inability to 
adequately handle the existence of unintended outputs not only precludes 
researchers from assessing the ceteris paribus reduction in unintended out-
puts but also distorts the reliability of the performance measures that have 
been derived for the desirable production technology. From a policy point of 
view, these aspects are very important, since both economic and environmen-
tal performance, as well as other technological indicators, are frequently taken 
as the basis for the implementation of new or assessment of already exist-
ing policy measures (Kumbhakar and Malikov, 2018). Furthermore, a critical 
point of departure for evaluating the capacity for policies to build synergies 
or trade-offs between environmental and economic performance in agricul-
ture is to consider how these types of performance are related before any 
policy action takes place. Although the widespread idea that the contribu-
tion of agricultural activities to the depletion of natural resources may suggest 
a trade-off relationship between environmental and economic performance, 
complementarities and synergies are likely to occur and create opportunities 
for potential win–win situations where the pursuit of economic objectives 
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
Do agri-environment measures help improve environmental and economic efficiency? 25
might generate corollary benefits in terms of environmental aspects. To address 
this debate, we calculated the Spearman’s rank correlation Coeff. to study 
the association between technical and environmental efficiency scores. Our 
result indicates a significant positive rank correlation (of the order of 0.420) 
between technical efficiency and nitrogen pollution efficiency. This positive 
association does not necessarily invalidate the need for policy intervention. 
In such cases of synergies, it is recommended from a policy perspective 
to support those farmers lying behind in terms of overall efficiency rather 
than trying to enhance environmental performance without considerably com-
promising economic performance (DeBoe, 2020). For instance, policymak-
ers should consider implementing complementary interventions that enforce 
linkages between agri-environment policies and insurance policies that sus-
tain economic growth and enable farmers to adopt innovative environmental
technologies.
Finally, it is worth mentioning that there are other potential avenues for 
future research. First, our empirical application is limited by data availabil-
ity and only nitrogen pollution is considered; thus, a potential extension of 
the current research could be the incorporation of other environmental indi-
cators such as pesticides pollution (Ait Sidhoum, Serra and Latruffe, 2020) 
and biodiversity index (Sipil ̈ainen and Huhtala, 2013). Second, while our 
results, naturally, cannot be extended beyond the specific geographic con-
text of Bavaria, it would be interesting to test the robustness of our findings 
with those derived from other datasets or alternative methods, such as the one 
described by Centorrino et al. (2021), which involve the use of stochastic fron-
tier procedures. Finally, a promising avenue for future research would be more 
context-specific research to investigate the effects of specific AES measures on 
technical and environmental performance.
Acknowledgements
This research is part of the EFFECT project. The authors gratefully acknowl-
edge financial support from the EU Horizon 2020 Programme for Research, 
Technological Development, and Demonstration under grant agreement no. 
817903.
References
Adenuga, A. H., Davis, J., Hutchinson, G., Donnellan, T. and Patton, M. (2019). Envi-
ronmental efficiency and pollution costs of nitrogen surplus in dairy farms: a para-
metric hyperbolic technology distance function approach. Environmental and Resource 
Economics 74: 1273–1298.
Ahovi, E., Schneider, K. and Lansink, A. O. (2021). Technical inefficiency of Dutch 
vegetable farms: specific-input analyses. PLoS One 16: e0250494.
Ait Sidhoum, A., Serra, T. and Latruffe, L. (2020). Measuring sustainability efficiency 
at farm level: a data envelopment analysis approach. European Review of Agricultural 
Economics 47: 200–225.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
26 A. Ait Sidhoum et al.
Ait Sidhoum, A., Hervé Dakpo, K. and Latruffe, L. (2022). Trade-offs between economic, 
environmental and social sustainability on farms using a latent class frontier efficiency 
model: evidence for Spanish crop farms. PLoS One 17: e0261190.
Ali, A. and Abdulai, A. (2010). The adoption of genetically modified cotton and poverty 
reduction in Pakistan. Journal of Agricultural Economics 61: 175–192.
Alvarez, A. and Del Corral, J. (2010). Identifying different technologies using a latent 
class model: extensive versus intensive dairy farms. European Review of Agricultural 
Economics 37: 231–250.
Ansell, D., Freudenberger, D., Munro, N. and Gibbons, P. (2016). The cost-effectiveness of 
agri-environment schemes for biodiversity conservation: a quantitative review. Agricul-
ture, Ecosystems & Environment 225: 184–191.
Arata, L. and Sckokai, P. (2016). The impact of agri-environmental schemes on farm per-
formance in five E.U. member States: a DID-matching approach. Land Economics 92: 
167–186.
Augurzky, B. and Schmidt, C. M. (2001). The propensity score: a means to an end. IZA 
Discussion Papers, No. 271.
Austin, P. C., Grootendorst, P. and Anderson, G. M. (2007). A comparison of the ability 
of different propensity score models to balance measured variables between treated and 
untreated subjects: a Monte Carlo study. Statistics in Medicine 26: 734–753.
Banker, R. D. and Chang, H. (2006). The super-efficiency procedure for outlier identifi-
cation, not for ranking efficient units. European Journal of Operational Research 175: 
1311–1320.
Baráth, L., Ferto˝, I. and Bojnec, Š. (2020). The effect of investment, LFA and agri-
environmental subsidies on the components of total factor productivity: the case of 
Slovenian farms. Journal of Agricultural Economics 71: 853–876.
Batáry, P., Dicks, L. V., Kleijn, D. and Sutherland, W. J. (2015). The role of agri-
environment schemes in conservation and environmental management. Conservation 
Biology 29: 1006–1016.
Beltrán-Esteve, M. and Picazo-Tadeo, A. J. (2017). Assessing environmental perfor-
mance in the European Union: eco-innovation versus catching-up. Energy Policy 104:
240–252.
Bisang, I., Lienhard, L. and Bergamini, A. (2021). Three decades of field surveys reveal a 
decline of arable bryophytes in the Swiss lowlands despite agri-environment schemes. 
Agriculture, Ecosystems & Environment 313: 107325.
Blancard, S., Boussemart, J. P., Briec, W. and Kerstens, K. (2006). Short- and long-run 
credit constraints in French agriculture: a directional distance function framework using 
expenditure-constrained profit functions. American Journal of Agricultural Economics
88: 351–364.
Brady, M., Kellermann, K., Sahrbacher, C. and Jelinek, L. (2009). Impacts of decoupled 
agricultural support on farm structure, biodiversity and landscape mosaic: some EU 
results. Journal of Agricultural Economics 60: 563–585.
Budka, M., Jobda, M., Szałański, P. and Piórkowski, H. (2019). Effect of agri-environment 
measure for the aquatic warbler on bird biodiversity in the extensively managed 
landscape of Biebrza Marshes (Poland). Biological Conservation 239: 108279.
Burton, R. J. F. and Schwarz, G. (2013). Result-oriented agri-environmental schemes in 
Europe and their potential for promoting behavioural change. Land Use Policy 30: 
628–641.
Caliendo, M. and Kopeinig, S. (2008). Some practical guidance for the implementation of 
propensity score matching. Journal of Economic Surveys 22: 31–72.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
Do agri-environment measures help improve environmental and economic efficiency? 27
Centorrino, S., Pérez-Urdiales, M., Bravo-Ureta, B. and Wall, A. (2021). Binary endoge-
nous treatment in stochastic frontier models with an application to soil conservation in 
El Salvador. In 95th Annual Conference, March 29-30, 2021, Warwick, UK (Hybrid) 
312058, Agricultural Economics Society - AES.
Chabé-Ferret, S. and Subervie, J. (2013). How much green for the buck? Estimating addi-
tional and windfall effects of French agro-environmental schemes by DID-matching. 
Journal of Environmental Economics and Management 65: 12–27.
Chambers, R. G., Chung, Y. and F ̈are, R. (1996). Benefit and distance functions. Journal of 
Economic Theory 70: 407–419.
Chambers, R. G. and Serra, T. (2018). The social dimension of firm performance: a data 
envelopment approach. Empirical Economics 54: 189–206.
Charles, H., Godfray, H. and Garnett, T. (2014). Food security and sustainable intensi-
fication. Philosophical Transactions of the Royal Society B: Biological Sciences 369: 
20120273.
Coderoni, S. and Esposti, R. (2018). CAP payments and agricultural GHG emissions in 
Italy. A farm-level assessment. Science of the Total Environment 627: 427–437.
Coelli, T. and Battese, G. (1996). Identification of factors which influence the technical inef-
ficiency of Indian farmers. Australian Journal of Agricultural and Resource Economics
40: 103–128.
Dakpo, K. H., Jeanneaux, P. and Latruffe, L. (2016). Modelling pollution-generating 
technologies in performance benchmarking: recent developments, limits and future 
prospects in the nonparametric framework. European Journal of Operational Research
250: 347–359.
Davidson, E. A., David, M. B., Galloway, J. N., Goodale, C. L., Haeuber, R., Harrison, 
J. A., Howarth, R. W., Jaynes, D. B., Lowrance, R. R., Thomas Nolan, B., Peel, J. L., 
Pinder, R. W., Porter, E., Snyder, C. S., Townsend, A. R. and Ward, M. H. (2011). Excess 
nitrogen in the U.S. environment: trends, risks, and solutions. Issues in Ecology 15.
DeBoe, G. (2020). Economic and environmental sustainability performance of environmen-
tal policies in agriculture. Paris: OECD Publishing.
Defrancesco, E., Gatto, P., Runge, F. and Trestini, S. (2007). Factors affecting farmers? 
Participation in agri-environmental measures: a Northern Italian perspective. Journal of 
Agricultural Economics 59: 114–131.
Devkota, K. P., McDonald, A. J., Khadka, L., Khadka, A., Paudel, G. and Devkota, M. 
(2016). Fertilizers, hybrids, and the sustainable intensification of maize systems in the 
rainfed mid-hills of Nepal. European Journal of Agronomy 80: 154–167.
Eulenstein, F., Werner, A., Willms, M., Juszczak, R., Schlindwein, S. L., Chojnicki, B. 
H. and Olejnik, J. (2008). Model based scenario studies to optimize the regional nitro-
gen balance and reduce leaching of nitrate and sulfate of an agriculturally used water 
catchment. Nutrient Cycling in Agroecosystems 82: 33–49.
European Commission. (2021). European agricultural fund for rural development. https://
cohesiondata.ec.europa.eu/funds/14-20.
F ̈are, R., Grosskopf, S., Noh, D. W. and Weber, W. (2005). Characteristics of a polluting 
technology: theory and practice. Journal of Econometrics 126: 469–492.
Førsund, F. R. (2018). Multi-equation modelling of desirable and undesirable outputs 
satisfying the materials balance. Empirical Economics 54: 67–99.
Førsund, F. R. (2021). Performance measurement and joint production of intended and 
unintended outputs. Journal of Productivity Analysis 55: 157–175.
Fraser, I. and Cordina, D. (1999). An application of data envelopment analysis to irrigated 
dairy farms in Northern Victoria, Australia. Agricultural Systems 59: 267–282.
Frisch, R. (1965). Theory of Production. Dordrecht: D. Reidel Publishing.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
28 A. Ait Sidhoum et al.
Gadanakis, Y., Bennett, R., Park, J. and Areal, F. J. (2015). Evaluating the Sustain-
able Intensification of arable farms. Journal of Environmental Management 150:
288–298.
Galloway, J. N., Aber, J. D., Erisman, J. W., Seitzinger, S. P., Howarth, R. W., Cowling, E. 
B. and Cosby, B. J. (2003). The nitrogen cascade. Bioscience 53: 341–356.
Gamer, W. and Bahrs, E. (2010). Bilanzen von potenziell umweltbelastenden N ̈ahrstof-
fen (N, P, K und S) der Landwirtschaft in Baden-Württemberg. Forschungsauftrag des 
Ministeriums für Ern ̈ahrung und L ̈andlichen Raum Baden-Württemberg, Stuttgart.
Grovermann, C., Wossen, T., Muller, A. and Nichterlein, K. (2019). Eco-efficiency and 
agricultural innovation systems in developing countries: evidence from macro-level 
analysis. PLoS One 14: e0214115.
Guesmi, B. and Serra, T. (2015). Can we improve farm performance? The determinants of 
farm technical and environmental efficiency. Applied Economic Perspectives and Policy
37: 692–717.
Hansson, H. and Öhlmér, B. (2008). The effect of operational managerial practices on eco-
nomic, technical and allocative efficiency at Swedish dairy farms. Livestock Science
118: 34–43.
Hasler, B., Termansen, M., Nielsen, H. Ø., Daugbjerg, C., Wunder, S. and Latacz-
Lohmann, U. (2022). European agri-environmental policy: evolution, effectiveness, and 
challenges. Review of Environmental Economics and Policy 16: 105–125.
Heckman, J., Ichimura, H., Smith, J. and Todd, P. (1998). Characterizing selection bias 
using experimental data. Econometrica 66: 1017–1098.
Heckman, J. J., Ichimura, H. and Todd, P. E. (1997). Matching as an econometric evalua-
tion estimator: evidence from evaluating a job training programme. Review of Economic 
Studies 64: 605–654.
Hickman, J. E., Huang, Y., Wu, S., Diru, W., Groffman, P. M., Tully, K. L. and Palm, C. 
A. (2017). Nonlinear response of nitric oxide fluxes to fertilizer inputs and the impacts 
of agricultural intensification on tropospheric ozone pollution in Kenya. Global Change 
Biology 23: 3193–3204.
Howarth, R. W., Swaney, D. P., Butler, T. J. and Marino, R. (2000). Climatic control on 
eutrophication of the Hudson River estuary. Ecosystems 3: 210–215.
Hynes, S. and Garvey, E. (2009). Modelling farmers’ participation in an agri-environmental 
scheme using panel data: an application to the rural environment protection scheme in 
Ireland. Journal of Agricultural Economics 60: 546–562.
IPCC. (2014). Climate change 2014: synthesis report. Contribution of Working Groups I, 
II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate 
Change. Geneva.
Jaforullah, M. and Whiteman, J. (1999). Scale efficiency in the New Zealand dairy 
industry: A non-parametric approach. Australian Journal of Agricultural and Resource 
Economics 43: 523–541.
Jones, J. I., Murphy, J. F., Anthony, S. G., Arnold, A., Blackburn, J. H., Duerdoth, C. P., 
Hawczak, A., Hughes, G. O., Pretty, J. L., Scarlett, P. M., Gooday, R. D., Zhang, Y. S., 
Fawcett, L. E., Simpson, D., Turner, A. W. B., Naden, P. S. and Skates, J. (2017). Do 
agri-environment schemes result in improved water quality? Journal of Applied Ecology
54: 537–546.
Kanter, D. R., Wentz, J. A., Galloway, J. N., Moomaw, W. R. and Winiwarter, W. (2017). 
Managing a forgotten greenhouse gas under existing U.S. law: an interdisciplinary 
analysis. Environmental Science & Policy 67: 44–51.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
Do agri-environment measures help improve environmental and economic efficiency? 29
Kay, P., Grayson, R., Phillips, M., Stanley, K., Dodsworth, A., Hanson, A., Walker, A., 
Foulger, M., McDonnell, I. and Taylor, S. (2012). The effectiveness of agricultural stew-
ardship for improving water quality at the catchment scale: experiences from an NVZ 
and ECSFDI watershed. Journal of Hydrology 422–423: 10–16.
Khandker, S. R., Koolwal, G. B. and Sammad, H. A. (2009). Handbook on Impact, 
Quantitative Methods and Practices. Washington, DC: The World Bank.
Kirschke, S., H ̈ager, A., Kirschke, D. and V ̈olker, J. (2019). Agricultural nitrogen pollu-
tion of freshwater in Germany. The governance of sustaining a complex problem. Water 
(Switzerland) 11: 2450.
Kleijn, D., Rundl ̈of, M., Scheper, J., Smith, H. G. and Tscharntke, T. (2011). Does conser-
vation on farmland contribute to halting the biodiversity decline? Trends in Ecology & 
Evolution 26: 474–481.
Kleijn, D. and Sutherland, W. J. (2003). How effective are European agri-environment 
schemes in conserving and promoting biodiversity? Journal of Applied Ecology 40: 
947–969.
Krupa, S. V. (2003). Effects of atmospheric ammonia (NH3) on terrestrial vegetation: a 
review. Environmental Pollution 124: 179–221.
Kumbhakar, S. C. and Malikov, E. (2018). Good modeling of bad outputs: editors’ 
introduction. Empirical Economics 54: 1–6.
Kuosmanen, T. and Kortelainen, M. (2005). Measuring eco-efficiency of production with 
data envelopment analysis. Journal of Industrial Ecology 9: 59–72.
Lakner, S. and Brümmer, B. (2008). Determinants of technical efficiency of organic 
grassland farming in Germany. Konferenzbeitrag Auf Dem IAMO-Forum 25: 27.
Latruffe, L., Fogarasi, J. and Desjeux, Y. (2012). Efficiency, productivity and technology 
comparison for farms in Central and Western Europe: the case of field crop and dairy 
farming in Hungary and France. Economic Systems 36: 264–278.
Latruffe, L. and Fogarasi, J. (2009). Farm performance and support in central and western 
Europe: a comparison of Hungary and France. Working Paper SMART-LERECO.
LfL. (2013). Basisdaten (Düngeberatung/Düngeverordnung). Bayerische Landesanstalt für 
Landwirtschaft.
LfU. (2021a). Ökologischer Zustand der Oberfl ̈achengew ̈asser - LfU Bayern. 
https://www.lfu.bayern.de/umweltdaten/indikatoren/natur_landschaft/oekozustand_
oberflaechengewaesser/index.htm.
LfU. (2021b). Nitrat im Grundwasser - LfU Bayern. https://www.lfu.bayern.de/
umweltdaten/indikatoren/umwelt_gesundheit/nitrat_im_grundwasser/index.htm.
Martinez Cillero, M., Thorne, F., Wallace, M., Breen, J. and Hennessy, T. (2018). The 
effects of direct payments on technical efficiency of Irish beef farms: a stochastic frontier 
analysis. Journal of Agricultural Economics 69: 669–687.
Mary, S. (2013). Assessing the impacts of pillar 1 and 2 subsidies on TFP in French crop 
farms. Journal of Agricultural Economics 64: 133–144.
Matzdorf, B. and Lorenz, J. (2010). How cost-effective are result-oriented agri-
environmental measures?-An empirical analysis in Germany. Land Use Policy 27: 
535–544.
Mayen, C. D., Balagtas, J. V. and Alexander, C. E. (2010). Technology adoption and tech-
nical efficiency: organic and conventional dairy farms in the United States. American 
Journal of Agricultural Economics 92: 181–195.
Mennig, P. and Sauer, J. (2020). The impact of agri-environment schemes on farm pro-
ductivity: A DID-matching approach. European Review of Agricultural Economics 47: 
1045–1053.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
30 A. Ait Sidhoum et al.
Mishra, A. K., Wesley Nimon, R. and El-Osta, H. S. (2005). Is moral hazard good for 
the environment? Revenue insurance and chemical input use. Journal of Environmental 
Management 74: 11–20.
Mozzato, D., Gatto, P., Defrancesco, E., Bortolini, L., Pirotti, F., Pisani, E. and Sartori, L. 
(2018). The role of factors affecting the adoption of environmentally friendly farming 
practices: Can geographical context and time explain the differences emerging from 
literature? Sustainability (Switzerland) 10: 3101.
Murty, S., Robert Russell, R. and Levkoff, S. B. (2012). On modeling pollution-
generating technologies. Journal of Environmental Economics and Management 64:
117–135.
OECD. (2013). OECD Compendium of Agri-environmental Indicators. Paris: OECD.
Pérez Urdiales, M., Lansink, A. O. and Wall, A. (2016). Eco-efficiency among dairy 
farmers: the importance of socio-economic characteristics and farmer attitudes. Envi-
ronmental and Resource Economics 64: 559–574.
Pufahl, A. and Weiss, C. R. (2009). Evaluating the effects of farm programmes: results from 
propensity score matching. European Review of Agricultural Economics 36: 79–101.
Rabalais, N. N., Turner, R. E. and Wiseman, W. J. (2002). Gulf of Mexico hypoxia, a.k.a. 
“The dead zone. Annual Review of Ecology and Systematics 33: 235–263.
Reinhard, S., Knox Lovell, C. A. and Thijssen, G. J. (2000). Environmental efficiency 
with multiple environmentally detrimental variables; estimated with SFA and DEA. 
European Journal of Operational Research 121: 287–303.
Reinhard, S. and Thijssen, G. (2000). Nitrogen efficiency of Dutch dairy farms: A shadow 
cost system approach. European Review of Agricultural Economics 27: 167–186.
Rockstr ̈om, J., Steffen, W., Noone, K., Persson, Å., Chapin, F. S., Lambin, E. F., Lenton, 
T. M., Scheffer, M., Folke, C., Schellnhuber, H. J., Nykvist, B., De Wit, C. A., Hughes, 
T., Van Der Leeuw, S., Rodhe, H., S ̈orlin, S., Snyder, P. K., Costanza, R., Svedin, 
U., Falkenmark, M., Karlberg, L., Corell, R. W., Fabry, V. J., Hansen, J., Walker, B., 
Liverman, D., Richardson, K., Crutzen, P. and Foley, J. A. (2009). A safe operating 
space for humanity. Nature 461: 472–475.
Rosenbaum, P. R. and Rubin, D. B. (1985). Constructing a control group using multivariate 
matched sampling methods that incorporate the propensity score. American Statistician
39: 33–38.
Rubin, D. B. (1976). Inference and missing data. Biometrika 63: 581–592.
Rubin, D. B. and Thomas, N. (1996). Matching using estimated propensity scores: relating 
theory to practice. Biometrics 52: 249–264.
Samson, G. S., Gardebroek, C. and Jongeneel, R. A. (2016). Explaining production expan-
sion decisions of Dutch dairy farmers. NJAS - Wageningen Journal of Life Sciences 76: 
87–98.
Schmit, C. and Rounsevell, M. D. A. (2006). Are agricultural land use patterns influenced 
by farmer imitation? Agriculture, Ecosystems & Environment 115: 113–127.
Serra, T., Chambers, R. G. and Oude Lansink, A. (2014). Measuring technical and envi-
ronmental efficiency in a state-contingent technology. European Journal of Operational 
Research 236: 706–717.
Shortall, O. K. and Barnes, A. P. (2013). Greenhouse gas emissions and the technical 
efficiency of dairy farmers. Ecological Indicators 29: 478–488.
Simar, L. and Wilson, P. W. (2007). Estimation and inference in two-stage, semi-parametric 
models of production processes. Journal of Econometrics 136: 31–64.
Sipil ̈ainen, T. and Huhtala, A. (2013). Opportunity costs of providing crop diversity 
in organic and conventional farming: would targeted environmental policies make 
economic sense? European Review of Agricultural Economics 40: 441–462.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
Do agri-environment measures help improve environmental and economic efficiency? 31
Skevas, T., Lansink, A. O. and Stefanou, S. E. (2012). Measuring technical efficiency in 
the presence of pesticide spillovers and production uncertainty: the case of Dutch arable 
farms. European Journal of Operational Research 223: 550–559.
Skevas, T. and Serra, T. (2016). The role of pest pressure in technical and environmen-
tal inefficiency analysis of Dutch arable farms: an event-specific data envelopment 
approach. Journal of Productivity Analysis 46: 139–153.
Slabe-Erker, R., Bartolj, T., Ogorevc, M., Kavaš, D. and Koman, K. (2017). The impacts of 
agricultural payments on groundwater quality: spatial analysis on the case of Slovenia. 
Ecological Indicators 73: 338–344.
Smith, J. A. and Todd, P. E. (2005). Does matching overcome LaLonde’s critique of 
nonexperimental estimators? Journal of Econometrics 125: 305–353.
STATBA. (2018). Statistisches Bundesamt (annual reports). Fachserie 4 Reihe 8.2 
Düngemittelversorgung. Statistisches Bundesamt, Wiesbaden.
Steffen, W., Richardson, K., Rockstr ̈om, J., Cornell, S. E., Fetzer, I., Bennett, E. M., Biggs, 
R., Carpenter, S. R., De Vries, W., De Wit, C. A., Folke, C., Gerten, D., Heinke, J., Mace, 
G. M., Persson, L. M., Ramanathan, V., Reyers, B. and S ̈orlin, S. (2015). Planetary 
boundaries: Guiding human development on a changing planet. Science 347: 1259855.
Stetter, C., Menning, P. and Sauer, J. (2020). Going beyond average—using machine learn-
ing to evaluate the effectiveness of environmental subsidies at micro-level. Working 
Paper to be presented at the 94th Annual Conference of the Agricultural Economics 
Society.
Stokes, J. R., Tozer, P. R. and Hyde, J. (2007). Identifying efficient dairy producers using 
data envelopment analysis. Journal of Dairy Science 90: 2555–2562.
Sutton, M. A., Oenema, O., Erisman, J. W., Leip, A., Van Grinsven, H. and Winiwarter, W. 
(2011). Too much of a good thing. Nature 472: 159–161.
Townsend, T. J., Ramsden, S. J. and Wilson, P. (2016). How do we cultivate in England? 
Tillage practices in crop production systems. Soil Use and Management 32: 106–117.
van Grinsven, H. J. M., van Eerdt, M. M., Westhoek, H. and Kruitwagen, S. (2019). Bench-
marking eco-efficiency and footprints of Dutch agriculture in European context and 
implications for policies for climate and environment. Frontiers in Sustainable Food 
Systems 3: 13.
Vitousek, P. M., Mooney, H. A., Lubchenco, J. and Melillo, J. M. (1997). Human 
domination of Earth’s ecosystems. Science (1979) 277: 494–499.
Weersink, A., Turvey, C. G. and Godah, A. (1990). Decomposition measures of technical 
efficiency for Ontario dairy farms. Canadian Journal of Agricultural Economics/Revue 
Canadienne D’agroeconomie 38: 439–456.
Weltin, M. and Hüttel, S. (2019). Farm eco-efficiency: can sustainable intensification make 
the difference? FORLand-Working Paper 10. Berlin: Humboldt-Universit ̈at zu Berlin.
Wendland, M., Diepolder, M., Offenberger, K. and Raschbacher, S. (2018). Leitfaden für 
die Düngung von Acker-und Grünland. Bayerische Landesanst. für Bodenkultur und 
Pflanzenbau.
Wimmer, S. and Sauer, J. (2020). Diversification economies in dairy farming - Empir-
ical evidence from Germany. European Review of Agricultural Economics 47:
1338–1365.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
32 A. Ait Sidhoum et al.
Appendix
Table A1. Summary statistics for the covariates used in the PSM in the pre-treatment year 
2013
Dimension Average S.d. Min Max
AES 1 if yes, 0 
if no
0.54 0.50 0 1.00
Livestock unit number 57.11 27.43 8.00 162.00
Labour Man-work 
units
1.79 0.62 0.40 5.00
Land Hectares 66.48 35.55 12.75 290.05
Capital depreciation Euros/ha 597.55 339.56 29.81 3,145.65
Total sales Euros/ha 3,567.60 1,256.20 1,390.59 10,294.10
Fertilisers Euros/ha 167.49 78.84 0 487.14
Pesticides Euros/ha 63.22 39.48 0 231.47
Feed Euros/ha 6.13 0.64 2.69 7.73
Farmer’s age Number 56.96 9.98 33.00 91.00
Share of arable land % 0.58 0.19 0 0.97
Share of grassland % 0.42 0.19 0.03 1.00
Share of rented land % 0.61 0.36 0.02 2.89
Yield index Number/ha 62.07 56.58 5.31 317.65
Agricultural income Euros/ha 1,091.56 602.35 −572.70 4,059.40
Dummy variable 
‘Swabia’
1 if yes, 0 
if no
0.15 0.36 0 1.00
Dummy variable 
‘Lower Franconia’
1 if yes, 0 
if no
0.10 0.30 0 1.00
Dummy variable 
‘Middle Franconia’
1 if yes, 0 
if no
0.23 0.42 0 1.00
Dummy variable 
‘Upper Franconia’
1 if yes, 0 
if no
0.20 0.40 0 1.00
Dummy variable 
‘Upper Palatinate’
1 if yes, 0 
if no
0.17 0.38 0 1.00
Dummy variable 
‘Lower Bavaria’
1 if yes, 0 
if no
0.03 0.17 0 1.00
Dummy variable 
‘Upper Bavaria’
1 if yes, 0 
if no
0.12 0.33 0 1.00
Dummy variable ‘no 
agric. education’
1 if yes, 0 
if no
0.04 0.21 0 1.00
Dummy variable 
‘skilled worker’
1 if yes, 0 
if no
0.54 0.50 0 1.00
Dummy vari-
able ‘University 
education’
1 if yes, 0 
if no
0.41 0.49 0 1.00
Gross value added in 
agriculture, forestry, 
fishing
Euros 
(million)
72.46 32.34 6.00 144.00
(continued)
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
Do agri-environment measures help improve environmental and economic efficiency? 33
Table A1. (Continued)
Dimension Average S.d. Min Max
Gross domestic 
product per capita
Euros 27,818.60 4,782.03 18,470.00 55,265.00
Unemployment rate % 0.03 0.01 0.01 0.07
Workforce Number 36,464.44 12,961.64 21,672.00 76,017.00
Farmland rental price Euros/ha 227.78 74.47 108.00 412.00
Number of 
observations
 271
Table A2. Estimation of the propensity score
Logistic regression
LR chi2(27)= 101.26
Prob > chi2 = 0.0000
Log likelihood=−135.624
Pseudo R2 = 0.272
Number of observations= 271
Dependent variable: AES
 Regressors Coef. z-stat P-value
 Livestock per ha −0.984 −1.13 0.26
Labour per ha −2.441 −0.14 0.889
Land 0.040 3.88 0.000
Capital depreciation per ha −0.040 −0.11 0.909
Total sales per ha 0.829 0.62 0.538
Fertilisers per ha 0.001 0.22 0.83
Pesticides per ha −0.286 −1.36 0.174
Feed per ha −0.326 −0.89 0.376
Ln farmers’ age −0.917 −0.97 0.334
Share arable land −7.580 −2.69 0.007
Share grassland −2.576 −2.5 0.013
Share rented land 0.155 0.62 0.537
Ln yield index per ha 0.674 1.71 0.087
Agricultural income per ha 0.475 0.63 0.528
Dummy variable ‘master’s certificate or university degree’ 1.012 1.31 0.19
Dummy variable ‘in education or skilled worker’ 0.621 0.81 0.418
Dummy variable ‘Swabia’ −17.295 −0.01 0.993
Dummy variable ‘Lower Franconia’ −16.946 −0.01 0.993
Dummy variable ‘Middle Franconia’ −15.657 −0.01 0.993
Dummy variable ‘Upper Franconia’ −14.582 −0.01 0.994
Dummy variable ‘Upper Palatinate’ −15.675 −0.01 0.993
Dummy variable ‘Upper Bavaria’ −17.383 −0.01 0.993
Ln gross domestic product per capita 1.752 1.51 0.132
Unemployment rate 0.446 0.42 0.674
Gross value added in agriculture, forestry, fishing 0.818 2.25 0.024
Intercept − 9.502 −0.01 0.996
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
34 A. Ait Sidhoum et al.
Table A3. Average treatment effect of AES on efficiency scores with varying length of 
post-treatment period
Technical 
efficiency
Environmental 
efficiency
Overall 
efficiency
2014 as post-treatment
DiD estimator −0.002 0.019 0,008
Bootstrapped Std. Err. 0.029 0.030 0.020
z-value −0.14 0.63 0.40
P>|z| 0.889 0.528 0.692
2015 as post-treatment
DiD estimator 0.018 0.012 0.013
Bootstrapped Std. Err. 0.036 0.025 0.019
z-value 0.52 0.47 0.70
P>|z| 0.602 0.636 0.481
2016 as post-treatment
DiD estimator 0.018 0.006 0.011
Bootstrapped Std. Err. 0.035 0.024 0.018
z-value 0.54 0.24 0.62
P>|z| 0.589 0.807 0.533
The results for the other explanatory variables are not shown to conserve space but are available from the authors.
Table A4. Descriptive statistics of the contextual variables (720 observations)
Variable Dimension Average S.D.
Share of grassland (%) 41 0.17
Insurance Euros 6018.76 2294.10
Capital/labour Ratio 19,689.07 10,400.47
Age Years 54 9.45
Category (%)
Soil type 1 50
0 50
Climate AES 1 23
0 77
Soil water AES 1 31
0 69
Biodiversity AES 1 25
0 75
Cultural landscape AES 1 8
0 92
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
Do agri-environment measures help improve environmental and economic efficiency? 35
Table A5. Detailed results of the truncated regression analysis with bootstrapped confi-
dence intervals (technical efficiency)
95% Confidence 
interval
Variables Coef.
Bootstrap
Std. Err. z P>|z| Lower Upper
Technical efficiency
Constant 1.334 0.255 5.220 0.000 0.844 1.828
AES −0.013 0.035 −0.360 0.717 −0.081 0.058
Time −0.035 0.030 −1.150 0.250 −0.095 0.023
DID 0.015 0.036 0.410 0.684 −0.058 0.084
Share of grassland −0.035 0.018 −1.930 0.054 −0.071 −0.001
Insurance −0.031 0.023 −1.340 0.179 −0.077 0.013
Capital/labour −0.012 0.015 −0.790 0.428 −0.040 0.017
Age −0.010 0.039 −0.260 0.792 −0.085 0.070
Soil type −0.031 0.014 −2.280 0.023 −0.058 −0.003
D2014 −0.006 0.029 −0.210 0.831 −0.066 0.051
D2015 −0.034 0.029 −1.160 0.247 −0.092 0.024
D2016 −0.013 0.030 −0.430 0.667 −0.072 0.047
D2017 −0.041 0.029 −1.430 0.154 −0.099 0.014
D2018 – – – – – –
Climate AES 0.002 0.018 0.130 0.896 −0.033 0.038
Soil water AES −0.034 0.017 −1.980 0.048 −0.068 −0.001
Biodiversity AES −0.016 0.015 −1.100 0.273 −0.044 0.014
Cultural landscape 
AES
0.078 0.033 2.390 0.017 0.021 0.148
Note: The Coeff. of D2018 has been omitted because of collinearity.
Table A6. Detailed results of the truncated regression analysis with bootstrapped confi-
dence intervals (environmental efficiency)
95% Confidence 
interval
Variables Coef.
Bootstrap
Std. Err. z P>|z| Lower Upper
Environmental efficiency
Constant 0.066 0.186 0.350 0.724 −0.307 0.431
AES 0.049 0.027 1.820 0.068 −0.004 0.103
Time −0.006 0.023 −0.270 0.790 −0.050 0.042
DID −0.006 0.028 −0.230 0.820 −0.062 0.049
Share of grassland −0.004 0.013 −0.320 0.750 −0.030 0.023
Insurance 0.034 0.016 2.080 0.038 0.003 0.066
Capital/labour 0.067 0.010 6.890 0.000 0.048 0.086
Age −0.105 0.029 −3.610 0.000 −0.162 −0.047
(continued)
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023
36 A. Ait Sidhoum et al.
Table A6. (Continued)
95% Confidence 
interval
Variables Coef.
Bootstrap
Std. Err. z P>|z| Lower Upper
Soil type 0.017 0.011 1.570 0.116 −0.004 0.039
D2014 0.112 0.022 5.020 0.000 0.068 0.156
D2015 0.058 0.022 2.590 0.010 0.015 0.103
D2016 0.021 0.022 0.960 0.339 −0.020 0.066
D2017 0.010 0.022 0.470 0.638 −0.032 0.055
D2018 – – – – – –
Climate AES −0.019 0.013 −1.370 0.169 −0.045 0.006
Soil water AES −0.026 0.013 −2.010 0.044 −0.051 0.000
Biodiversity AES −0.056 0.012 −4.590 0.000 −0.079 −0.032
Cultural landscape 
AES
0.047 0.021 2.250 0.024 0.004 0.088
Note: The Coeff. of D2018 has been omitted because of collinearity.
Table A7. Detailed results of the truncated regression analysis with bootstrapped confi-
dence intervals (overall efficiency)
95% Confidence 
interval
Variables Coef.
Bootstrap 
Std. Err. z P>|z| Lower Upper
Overall efficiency
Constant 0.849 0.130 6.520 0.000 0.601 1.101
AES 0.022 0.018 1.180 0.237 −0.014 0.058
Time −0.011 0.016 −0.690 0.491 −0.042 0.020
DID 0.005 0.019 0.290 0.776 −0.032 0.042
Share of grassland −0.003 0.009 −0.370 0.710 −0.022 0.015
Insurance 0.004 0.011 0.370 0.708 −0.018 0.027
Capital/labour 0.021 0.007 3.100 0.002 0.007 0.034
Age −0.082 0.021 −3.970 0.000 −0.123 −0.042
Soil type 0.012 0.008 1.620 0.105 −0.002 0.027
D2014 0.049 0.016 3.140 0.002 0.018 0.079
D2015 0.020 0.015 1.350 0.177 −0.009 0.049
D2016 0.003 0.016 0.210 0.837 −0.028 0.034
D2017 −0.004 0.015 −0.270 0.787 −0.034 0.025
D2018 – – – – – –
Climate AES −0.009 0.009 −0.940 0.347 −0.028 0.009
Soil water AES −0.017 0.009 −1.900 0.058 −0.035 0.000
Biodiversity AES −0.036 0.009 −4.210 0.000 −0.053 −0.019
Cultural landscape 
AES
0.035 0.014 2.460 0.014 0.007 0.063
Note: The Coeff. of D2018 has been omitted because of collinearity.
D
ow
nloaded from
 https://academ
ic.oup.com
/erae/advance-article/doi/10.1093/erae/jbad007/7085979 by N
atural R
esources Institute Finland (Luke) user on 31 M
arch 2023