Travisi und Nijkamp - 2008 - Valuing environmental and health risk in agricultu

ECOLOGICAL ECONOMICS 67 (2008) 598– 607
available at www.sciencedirect.com
www.elsevier.com/locate/ecolecon
ANALYSIS
Valuing environmental and health risk in agriculture:
A choice experiment approach to pesticides in Italy
Chiara Maria Travisi a, ⁎, Peter Nijkamp b
a Fondazione Eni Enrico Mattei, Milano, Italy and Department of Management Economics and Industrial Engineering,
Politecnico di Milano, Milano, Italy
b Department of Spatial Economics, Free University, and Tinbergen Institute, Amsterdam, The Netherlands
ARTICLE INFO
ABSTRACT
Article history:
Received 5 February 2007
Received in revised form
10 December 2007
Accepted 14 January 2008
Available online 12 February 2008
Keywords:
Pesticide risks
Food safety
Willingness-to-pay
Choice experiment
Stated choice
JEL classification:
H23; I12; Q25; Q18; R0
The widespread use of pesticides in agriculture shows a complex ramification of multiple
negative externalities, ranging from food safety-related effects to the deterioration of
farmland ecosystems. Recent research has demonstrated that the assessment of the
economic implications of such negative processes is fraught with many uncertainties. This
paper presents the results of an empirical study recently conducted in Northern Italy aimed
at estimating the economic value of reducing the wide-ranging impacts of pesticide use, by
deploying a Choice Experiment approach. The experimental design provides a meaningful
tool to assign monetary values to the negative environmental effects associated with
agrochemicals use. In this connection, the paper addresses in particular the reduction of
farmland biodiversity, groundwater contamination and harm to human health. The
resulting estimates confirm that, on average, respondents demonstrate a substantial
willingness-to-pay a premium for agricultural goods (in particular, foodstuffs) produced in
environmentally-benign ways.
© 2008 Elsevier B.V. All rights reserved.
1. Introduction
Modern intensive agriculture produces significant negative
externalities that have been broadly documented in the
scientific literature (Pimentel et al., 1992; Pimentel and
Greiner, 1997). The order of magnitude of these externalities
has been addressed in several studies in the scientific,
political and economic literature on recent agro-environmental
regulations, on pesticide and fertilizer-reduction
strategies, and on the assessment of the associated economic
costs. Challenging questions and new opportunities to
provide policy makers with relevant insights on the best
option to be developed against pesticide risks are broadly
discussed. Relevant issues here concern, inter alia: how to
accelerate the implementation of pesticide risk reduction and
management strategies; and how to choose, from amongst
the range of possible pesticide reduction measures, those
actions that are able to provide the highest level of risk
abatement at the lowest collective cost. In this context, the
present paper examines the use of a Choice Experiment (CE)
⁎ Corresponding author. Fondazione Eni Enrico Mattei (FEEM). Research programme: Sustainability Indicators and Environmental
Valuation (SIEV). Corso Magenta 63, 20123 - Milano, Italy. Tel.: +39 02 52036801; fax: +39 02 52036946.
E-mail address: chiara.travisi@feem.it (C.M. Travisi).
0921-8009/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.ecolecon.2008.01.011

ECOLOGICAL ECONOMICS 67 (2008) 598– 607
599
methodology to assess the economic value of pesticide risk
reductions 1 . The CE survey took place in Milan, Italy, with the
aim of providing estimates of the willingness-to-pay (WTP) of
consumers to achieve improvements in the environmental
and health safety of agriculture. This allows us to study in
detail the preferences of consumers for alternative food
bundles that differ in production practices which are more
or less dependent on the use of pesticides. The remainder of
this paper is organized as follows. Section 2 presents the
current situation concerning pesticide risk management in
the European Union (EU) political context, and explores the
potential to use economic valuation methods in general.
Section 3 presents a CE strategy for assessing the benefits of
pesticide risk reductions that relies on the Random Utility
Model (RUM) formulation in order to study the respondents'
behaviour. Section 4 presents the survey instruments and
describes the interviews conducted with a sample of 484
consumers intercepted at three shopping malls in Milan.
Then, Section 5 links the selected theoretical model to the
empirical exercise, using the CE questionnaire and the
respective economic valuation exercise, while Section 6
discusses the range of the economic estimates and interprets
the results. Section 7 offers concluding remarks.
2. Background to pesticide risk management
2.1. Policy challenges
Pesticides are chemicals that require particular attention
because most of them have inherent properties that make
them dangerous to health and the environment. The European
Thematic Strategy on the sustainable use of pesticides
(currently being developed) identifies a set of policy objectives
that will have to be reached in the coming years to achieve a
higher level of sustainability in chemical-based agricultural
production. Minimizing the hazards and risks to health and
the environment from the use of pesticides is a key point in
this strategy that will need to be supported by several policy
actions. Amongst other things, the EU strategy includes:
encouraging the use of low input or pesticide-free crop
farming, particularly by raising users' awareness; promoting
the use of codes of good practice; and consideration of the
possible application of financial instruments. The strategy
assumes: i) the imposition of penalties on users by reducing or
cancelling benefits provided by support schemes; ii) the
introduction of special levies on pesticides to raise awareness
of the detrimental effects of over-intensive pesticide use and
further reduce reliance on chemical inputs in modern
agriculture; and iii) the harmonization of the value-added
tax rates for pesticides (which vary between 3 and 25% in the
various Member States). In this context, Italian agricultural
policy aims to decrease the risks attached to the use of
pesticides by providing economic incentives for organic
1 The survey also included a Contingent Valuation question in
which respondents were asked to report a maximum WTP for
eliminating the negative pesticide impacts (for further details, see
Travisi and Nijkamp, 2004).
farming and Integrated Pest Management (IPM) 2 . Moreover,
the design of eco-labelling for fresh food that is produced with
more benign agricultural practices is a major concern for both
agribusiness and policy makers in the Italian agricultural
sector. Agribusinesses, such as supermarkets and food
producers, appear to be interested in estimating consumer
demand for a product with additional environmental attributes,
while other agribusinesses, such as seed and chemical
companies, and technology and equipment dealers, are
interested in farmers' WTP for a new eco-product or service
(for a discussion, see Lusk and Hudson, 2004). In relation to
pesticide policy purposes, economic theory suggests that an
efficient incentive or tax should be set equal to the marginal
damage associated with pesticide use. Similarly, estimates of
individuals' WTP for pesticide risk reduction would provide
key information for policy makers in order to introduce price
differentials in products, according to the type and severity of
pesticide risks related to their production processes. In this
perspective, a proper incentive programme for Italian farmers,
or the design of eco-labelling, would require an estimation of
individuals' WTP for pesticide risk reduction. In the current
panorama, therefore, the availability of an economic estimate
of the social benefits of reduced pesticide risk could be pivotal,
allowing us to identify the optimal value-added tax rates for
pesticides or incentives to use less risky chemicals.
2.2. Economic valuation
Over the last two decades, an extensive empirical economic
literature on pesticide risk valuation has emerged (for two
meta-analyses, see Florax et al., 2005 and Travisi et al., 2006).
The WTP estimates available mainly refer to the negative
externalities on human health (e.g. Wilson, 2002) and, to a
lesser extent, to damage to environmental agro-ecosystems
(e.g. Brethour and Weersink, 2001; Cuyno et al., 2001). Despite
the relative abundance of surveys that have provided estimations
of WTP for the reduction of several pesticide risks, to our
knowledge, there are still only a few Conjoint Analysis (CA)
approaches to the valuation of pesticide risks in this
literature 3 . Foster and Mourato (2000) and Schou et al. (2002)
employed contingent ranking techniques to determine the
WTP for the reduction of human health effects, and loss of
farmland biodiversity. In their survey, Foster and Mourato
(2000) estimated the marginal value of reducing risks for bird
biodiversity and human health, whereas Schou et al. (2002)
valued the benefits of the reduced use of pesticides in field
margins with a focus on the biodiversity of partridges. The use
2 Italy has the third highest level of pesticide consumption in
the OECD countries at 13% of total purchases, and a rate of
consumption of about 7.7 kg of pesticide per hectare of
agricultural land treated.
3 Conversely, a growing number of CA studies have been
implemented to analyse consumers' demand for a wide array of
food safety and quality issues, including genetically modified
feed, mad-cow disease, growth hormones, health risk (in general),
distance to the producer, taste and other quality attributes, as
well as food certification/labelling for country/region of origin,
organic or free range production as well as for certain other
quality attributes, for a variety of food products; see, e.g.: Loureiro
and Umberger, 2003, 2005; Goldberg and Roosen, 2007.

600 ECOLOGICAL ECONOMICS 67 (2008) 598– 607
of pesticides was not mentioned to the respondents, who were
asked to express their preferences for a generic change in
biodiversity. More recently, Wikström (2003) used CE to
estimate the WTP for sustainable coffee produced with low
pesticide input, while Hasler and Birr Pederson (2004) applied
CE for valuing increased groundwater quality provided by a
generic change in groundwater protection. Contributing to
this literature, the present paper provides an empirical
application of the CE technique to the valuation of pesticide
risks in Italy, by considering the main areas of actual risk for
the Italian context.
3. A CE strategy to value alternative
agricultural production scenarios
3.1. Statement of the valuation problem
This study aims to assess people's preferences for alternative
scenarios of agricultural production methods which lead to a
healthier environment (e.g. integrated pest management,
organic agriculture), by focusing on the environmental and
economic effects they generate. However, the elicitation of the
citizens' preferences for, and economic valuation of, alternative
agricultural scenarios is complicated for two reasons.
First, the negative environmental externalities of pesticide use
− such as pollution of soil, surface and groundwater, higher
mortality of sensitive animal and insect species, effects on
human health, etc. − are not bought and sold on regular
markets with proper prices. This implies that we need to apply
non-market valuation techniques. Second, even though lowinput
agricultural practices have been applied in Europe and
Italy, they have not been monitored with regard to their
environmental and economic effects, so that we have to resort
to Stated Preferences (SP) non-market valuation techniques,
i.e. relying on what people say they would do under hypothetical
experimental circumstances, rather than studying their
actual behaviour. Although the analysis concerns hypothetical
circumstances, we are interested in estimating the value of
reducing the main actual pesticide risks in the Italian context:
biodiversity, soil and groundwater contamination, acute illness
in humans. We deploy here the Choice Experiment (CE)
as our main methodology.
3.2. A CE approach
Choice Experiment (CE) is a non-market valuation method
that makes it possible to infer people's preferences for a set of
alternatives, described by a set of relevant attributes (see, e.g.,
Louviere et al., 2000). Respondents are asked to view the
various environmental risks of pesticide use in agricultural
production as foodstuff attributes to be taken into account in
the grocery purchase decision. Moreover, we can calculate the
WTP for a pesticide risk decrease based on the preferences of a
selected sample of individuals, whereas, for instance, it is the
householders' preferences that are usually elicited in the
hedonic price approaches (Söderqvist, 1998). In addition, CE
has an attractive advantage over Contingent Valuation (CV). A
typical pesticide abatement policy scenario involves various
aspects that can have a significant impact on people's wellbeing.
Moreover, pesticide risks are intrinsically multiple and
heterogeneous, since the range of targets, exposure vehicles,
and toxicological and eco-toxicological end points is very
wide. A number of questions, therefore, need to be addressed:
what type of pesticide is targeted by the policy? What level of
pesticide reduction does the policy provide? When and at
what cost will the policy be implemented? CE can separately
estimate the preferences of individuals for all these aspects.
Our analysis of the responses to the CE questions uses
Random Utility Modelling (RUM) (McFadden, 1986). The model
is then estimated with a nested 4 logit (for more details, see
Louviere et al., 2000; Green, 2002). As a plausible nesting for the
status quo/agricultural scenarios model, we assume that a
respondent decides whether to keep the status quo or to
purchase an alternative agricultural production scenario, and
then, conditional on not keeping the status quo, chooses
between the two single agricultural alternatives. Valuation
results are presented and discussed in the fifth section.
4. The CE survey on pesticide risks
4.1. Survey design
The questionnaire was developed by using the results from
two focus groups and one pre-test 5 , and in collaboration with a
team of eco-toxicologists. Focus groups and the pre-test were
necessary in order to: test the appropriateness of the
attributes (and their levels) included in the CE questions;
select a proper payment vehicle for the WTP experiment and
test bids; refine the initial draft questionnaire. On the basis of
the results provided by this fieldwork, some modifications in
the draft questionnaire were included. The pre-test was
conducted on two university campuses 6 and the final survey
was carried out in Milan, Italy, between May and June 2003.
The survey questionnaire was self-administered by the
respondents who were approached at three shopping malls
in Milan by a trained team of three interviewers. Respondents
are those persons in the household who are usually responsible
for grocery shopping. The enumerators were instructed
to stop potential respondents and ask them to take the
4 For discrete choice problems where the independence assumption
is suspect, Hausman and McFadden (1984) have
proposed the nested logit model. Rather than regarding all the
alternatives as if they were elements of a single choice set, the
nested logit model assumes that choice proceeds through a set of
‘nested’ choice sets. It thus allows a variety of response patterns
to a change in the characteristics of one alternative, resulting in a
relaxation of the IIA assumption.
5 A pre-test on 40 respondents was undertaken in April 2003 in
Milan.
6 University campuses and shopping centres were considered to
be suitable locations to maximize the visibility of our questionnaire
and the sampling size, thus curbing the generally high costs
of surveys. On the university campuses interviewers asked people
to take the questionnaire, bring it home and ask the member of
the family responsible for the daily food shopping to complete it.
In shopping centres a stand was located at the entrance. People
were asked to take the questionnaire before shopping, complete
it, sitting at the desk and then drop it off to the interviewer after
shopping.

ECOLOGICAL ECONOMICS 67 (2008) 598– 607
601
questionnaire, complete it, and then drop it off after shopping.
Overall, 484 questionnaires were distributed, 302 of which
were returned in a completed form. The return rate was about
62%.
The questionnaire consisted of three sections. The first
section introduced the subject of the environmental externalities
of pesticide use in modern agriculture, by using a costbenefit
perspective, which emphasized existing trade-offs
between the positive and negative externalities associated
with agricultural production based on the use of synthetic
inputs. First, we referred to pesticide risks in general, and
asked respondents their opinion on the current environmental
situation and the detrimental effects of modern agriculture.
Other questions included: i) How serious are
environmental problems compared with other problems in
our society, and which of these problems deserve higher
public investment?; ii) How serious are pesticide problems
compared with other environmental problems, and which of
these problems should be the priority for public investment?;
iii) Which type of pesticide impact is more severe, and why?;
iv) Had the respondents ever personally experienced any of
these impacts?; and v) How much were they informed about
pesticide impacts? We then focused on some specific dimensions
of pesticide risk. The questionnaire described the actual
current Italian situation concerning pesticides, providing
information on, and graphical aids to indicate, both their
benefits and risks, and giving the reasons for these positive
and negative effects. In particular, the questionnaire focused
on three environmental dimensions currently affected by
pesticides: farmland biodiversity; soil and groundwater contamination;
and the health effects on an exposed population.
To facilitate the questionnaire's comprehension, respondents
were provided with a card summarizing all these
information in a simple graphical manner. This card was
available to the respondent during the whole CE exercise. The
second section of the questionnaire contained the CE exercise.
Preliminary to the CE questions, we informed the respondents
that a reduction of pesticide risk exposure is possible by
implementing some pesticide management policies, and that
the Italian government was about to do this. Policy options
consisted of a change in agricultural production practices that
are designed to reduce the rate of pesticide application on
fields, without any change in the products' quality. However,
this would increase production costs, leading to an increase in
retail costs too. We clearly explained to the respondents that
the implementation of the pesticide risk reduction policies
would be costly to the agricultural sector, and that some of the
increased production costs would fall on consumers, through
an increase in retail prices. We explained how a reduction in
risks would be possible; what range of reduction would be
achievable; who would provide this reduction; how it would be
provided; and the economic effect of such risk reduction to the
respondent. Respondents were therefore asked to view the
various externalities of pesticide use due to conventional
agricultural practices as food attributes to be taken into
account in daily purchase decisions. The third section of the
questionnaire gathered additional information in order to
obtain a clearer image of the respondents' profile, attitudes,
socio-economic conditions, and exposure to pesticides. Questionnaire
debriefs closed the survey, in order to explore
whether the respondents had a reasonably good comprehension
of the survey material and choice tasks.
4.2. CE questions
Following the above explanation, the respondents focused on
the CE questions. These questions were formalized by two
elements: i) a decrease in pesticide impacts (provided by a
change in agricultural practices); ii) and an increase in the
foodstuff shopping expenditure. Relevant risk attributes were
selected with the assistance of a team of Italian ecotoxicologists
specialized in pesticide risk assessment. We
therefore identified the main environmental and health
dimensions affected by pesticides in Italy, selected indicator
variables that best describe each environmental effect, and set
a realistic range of variation that is reachable thanks to the
proposed change of agricultural practices. All indicators were
selected to describe, as accurately as possible, the main areas
of well-documented environmental damage in Italy. Attributes
and attributes' levels were also tested during focus
groups to allow the fine-tuning of this part of the survey
instrument. We focused on the following dimension of
damage due to pesticide use: farmland bird biodiversity, soil
and groundwater pesticide contamination, and threats to
human health.
To avoid misinterpretation of attributes' levels, we formalized
such pesticide effects in terms of damage, rather that in terms of
risk (i.e. probability of occurrence). The attributes and the
attributes' levels are described in Table 1. Status quo levels are
national-specific, and refer to the current Italian situation. The
impact on biodiversity was formalized in terms of the number of
endangered farmland bird species. Respondents were provided
Table 1 – List of the attributes used in the CE value application and corresponding percentage values
Attributes Current level Level-1 Level-2 Level-3
Expenditure for foodstuff
[€/household per month]
Biodiversity [no. endangered
farmland bird species]
Soil and groundwater
[% contaminated agricultural land]
Human health
[no. cases illness per year]
Current expenditure E (⁎) E+50€ [increase by 10%] E+100€ [increase by 20%] E+200€ [increase by 40%]
15 9 [decrease by 40%] 6 [decrease by 60%] 3 [decrease by 80%]
65 45 [decrease by 30%] 25 [decrease by 60%] 15 [decrease by 80%]
250 150 [decrease by 40%] 100 [decrease by 60%] 50 [decrease by 80%]
Note: (⁎) As indicated by each respondent.

602 ECOLOGICAL ECONOMICS 67 (2008) 598– 607
Fig. 1 – Example of card with current and future pesticide risk levels provided to facilitate the respondents' comprehension of
trade-off between attributes.
with some examples of endangered birds, from amongst those
most commonly known. The impact on soil and groundwater
was formalized by using the percentage of farmland areas
affected by soil and aquifer pesticide contamination. Respondents
were informed that about 65% of farmland soil and
aquifers (i.e. groundwater) contain a non-negligible concentration
of pesticide compounds: ‘non-negligible’ meaningthatitis
able to produce short-term and long-term negative effects on
non-target terrestrial and aquatic ecosystems. This attribute
therefore captures diffuse pesticide impacts on the ecosystems'
life- support function. The impact on human health was
expressed in terms of cases per year of acute illness (i.e. leading
to hospitalization), both as a result of work and domestic
exposure. To get a point of reference, respondents were informed
about the number of yearly cases of hospitalization due to food
poisoning and poisoning from domestic cleansing. In addition,
during the CE questions, respondents were provided with a card
summarizing current national pesticide risk levels and their
possible future range of variation reachable by reducing pesticide
use in agricultural practices. These were presented both in
absolute values and percentage terms in order to facilitate the
comprehension of trade-offs between attributes (see Fig. 1).
Special attention was given to the selection of the payment
vehicle. A common trend amongst previous studies estimating
WTP for reducing pesticide risk is to use the price premium for a
single food product with particular additional environmental
characteristics compared with the price of a pre-existing substitute
(for a discussion, see Florax et al., 2005). Since
respondents have to focus on a good which is private and
deliverable, this approach reduces the problem of hypothetical
bias. Nevertheless, the results from our fieldwork showed that
respondents have difficulties in understanding the overall cost
to them of the proposed policies and tend to overestimate their
WTP 7 . Thus we referred to an overall increase in the respondent's
agricultural foodstuff expenditure (what we call a green
shopping payment vehicle) 8 . Bids were selected on the basis of
the average national foodstuff expenditure, which is set at
about €600 per month per household (i.e. about €185 per
person per month), and next tested during focus groups. The
proposed bids corresponded to an increase of the household's
monthly grocery expenditure by approximately 10, 20 and 40%,
respectively (see Table 1).
Using a cyclical design based on an orthogonal fractional
factorial 9 , we generated 9 choice sets, each consisting of three
7 In a pilot version of the questionnaire, the payment vehicle
was a price premium on a single packet of spaghetti. Respondents
were, however, disturbed by a “single green product” perspective.
Respondents were not able to clearly understand the monthly/
yearly cost to them of the pesticide policy, and tended to
overestimate their WTP.
8 An additional advantage of this payment vehicle is that the
estimation of the overall benefits from pesticide risk reduction is
very much simplified and less biased by approximations.
9 The design of the choice sets is consistent with principles of
experimental design (Bunch et al., 1993; Lazari and Anderson,
1994). In particular, we used a shifted or cyclical design, which
generally has a superior efficiency compared with other strategies
for generating main effects designs (Bunch et al., 1993). These
shifted designs use an orthogonal fractional factorial to provide
the basic alternatives for each choice set. Thus, starting from a set
of 243 possible permutations of the hypothetical agricultural
scenario (3 levels 4 attributes ), we generated the fractional factorial
using a simple routine in the software package SPSS®. Subsequently,
alternative profiles were cyclically combined to generate
9 choice sets. These satisfy the principle of orthogonality, level
balance, and minimal overlap (see Huber and Zwerina, 1996).

ECOLOGICAL ECONOMICS 67 (2008) 598– 607
603
Fig. 2 – Example of choice set.
alternative profiles. The first one was fixed and corresponded to
the status quo scenario, i.e. the conventional scenario of
agricultural practices, priced at the household's monthly grocery
expenditure level, as reported by each respondent, and characterized
by the current national pesticide damage levels. The
other two profiles varied from card to card and corresponded to
agricultural scenarios that provide lower pesticide risk levels.
Internal coherence was verified. All combinations were asked in
roughly equal frequencies. Respondents were instructed to
select the most preferred one (see Fig. 2).
5. Empirical findings
5.1. Some basic statistics of the questionnaire
Table 2 shows the survey statistics and the socio-demographics
of the sample. The average respondent is 34 years old, has a
household income of roughly €2100 a month, and has completed
secondary school. The average household size is 3.5, with
15% of the sample having at least one person in the household
who is younger than 15. The household monthly food expenditure
is about €612, corresponding to about 30% of the monthly
income. Moreover, 26% of the respondents had a strong
environmental attitude and 12.2% were very concerned about
pesticide risks. Comparing the environment with other problems
in Italy, respondents ranked environment as the third
important area for public investment after public health and
education. 68.9% of the sample population considered public
investments for environmental safety very important, compared
with the 77.3 and 71.5% who found investment for,
respectively, public health and education very important. 56.3%
of the population believe that the media should inform the
public about pesticide risks, while only 10.2% indicated that they
were ‘not at all’ or ‘slightly’ informed. The main differences
between the socio-demographics of our sample and those of the
population of Milan concern age and income level. The average
age of our sample is rather low – 34 as opposed to 44 years old –
and the household income is 25% lower than the Milan average.
This suggests that we should control for these individual
characteristics in our statistical model of the choice responses.
On the basis of the responses to the choice questions and to
the control questions, we believe that the respondents had a
reasonably good comprehension of the survey material and
choice tasks. Only 4.4% complained that they had insufficient
Table 2 – Survey statistics and socio-demographics of the
sample
Mean or
percentage
Individual characteristics
Age 33.9
Female 61.6
Household size 3.5
Households with one or more persons under 15 15.1
Years of schooling 13.0
Monthly household income (in Euros) 2,098.1
Monthly household grocery expenditure (in Euros) 611.9
Attitudinal characteristics
‘Very much sensitive’ to sensitive to
26.1
environmental and health issues (⁎)
Believe that public investment for environmental 68.9
safety is very important
Believe that public investment for public health is 77.3
very important
Believe that public investment for education is very 71.5
important
‘Very well’-informed about pesticide risks (⁎) 12.2
‘Not at all’ or ‘slightly’-informed about
10.2
pesticide risks (⁎)
Believe that media should inform public opinion 56.3
about pesticide risks
Respondents debriefs
Found some questions hard to understand 8.5
Found not enough information provided 4.4
Note: (⁎) Based on a 5-point Likert scale.

604 ECOLOGICAL ECONOMICS 67 (2008) 598– 607
Table 3 – List of explanatory variables
Variable
PRICE Continuous; monthly household grocery expenditure
BIODIV Categorical variable; takes on value 15, 9, 6, 3
GRWATER Categorical variable; takes on value 65, 45, 25, 15
HEALTH Categorical variable; takes on value 250, 150, 100, 50
AGE Continuous variable providing the respondent's age
CONCERN Categorical variable based on a 5-point Likert-scale:
from 1 ‘not at all’ to 5 ‘very much’ informed on
environmental and health risks due to pesticides
information, and the relatively low proportion of 8.5% reported
that they had found some of the questions difficult to understand.
5.2. Model specifications and estimation results
Data from our unlabelled experiment are examined with the
following additive RUM model specifications 10 . Damage attributes
are formalized as categorical variables (see Table 3).
Model 1 is the simplest model that we consider in order to
investigate the effect that each attribute can have on the
respondents' preferences for alternative agricultural scenarios.
Formally, we have:
5.2.1. Model 1
V ¼ b 1 PRICE þ b 2 BIODIV þ b 3 GRWATER þ b 4 HEALTH:
In this model formulation, PRICE refers to the cost of the
policy to the respondents. BIODIV denotes the variable for the
level of biodiversity damage. GRWATER denotes the variable for
the level of contamination of soil and groundwater. Similarly,
HEALTH denotes the variable for the level of human health
impacts. Ceteris paribus, β 1 can be interpreted as the coefficient
of the cost of the pesticide policy to the respondent. β 2 provides
the effect of a unit decrement of the biodiversity damage (for
instance: from 15 to 14 endangered bird species), on the probability
of choosing an agricultural scenario. β 3 provides the effect of
a unit decrement of the soil and groundwater contamination
(i.e. by 1%) on the probability of choosing an agricultural scenario.
Finally, β 47 provides the effect of a unit decrement of human
health impact (for instance from 250 to 249 cases of acute intoxication)
on the probability of choosing an agricultural scenario.
The estimation results are shown in Table 4. Model (1)
shows a Pseudo-R 2 of 0.562, which indicates a very high fit
(Hensher et al., 2005; Hensher and Johnson, 1981). All variables
are highly statistically significant and have the expected sign:
the sign of PRICE and of all the various pesticide damages being
negative. This signals that any risk level reduction from the
status quo is welcomed by the respondents. The results from
Model (1), therefore, confirm the expectation that the pesticide
risk reduction elasticity is positive (Florax et al., 2005).
We also want to control for differences in the respondent's
profile with respect to the consumer's food choice and the economic
valuation of alternative agricultural scenarios. Model (2) 11
10 Note that all the indexes for the respondents and alternatives
have been omitted.
11 We also explored other model specifications. Those presented
show the highest goodness of fit.
ð1Þ
integrates information on the respondents' profile, and
assess the respective effect on the valuation of the agricultural
option.
5.2.2. Model 2
V ¼ b 1 PRICE þ b 2 BIODIV þ b 3 GRWATER þ b 4 HEALTH
þb 5 HEALTH AGE þ b 6 GRWATER CONCERN
Model (2) incorporates in the utility function the respondents'
level of age and concern towards environmental and
health issues (see Table 3). It involves the cross terms of
HEALTH and AGE, GRWATER and CONCERN. AGE is a continuous
variable providing the respondent's age; CONCERN is
a categorical variable taking values from 1 ‘not at all’ to 5 ‘very
much’ informed on environmental and health risks due to
pesticides (see Table 3). The results are presented in Table 4.
The results of log–likelihood ratio tests show that adding
demographic and attitudinal variables adds significantly to
Model (1). The pseudo-R 2 is again very high (0.565). All the
effects of interaction of the choice attributes with AGE and
CONCERN are statistically significant, meaning that individual
utility is sensitive to the respondent's level of age and concern
to environmental and health issues. The interaction between
AGE and HEALTH is negative and highly statistically significant,
illustrating a positive relationship between respondents'
age level and WTP estimates for biodiversity. Similarly, the
coefficient of the interaction between CONCERN and
GRWATER is significant and negative, meaning that people
concerned about pesticide risks are more prone to pay to
reduce soil and groundwater contamination.
In order to capture the empirical magnitude of the effect of
these three socio-demographic factors, we provide a sensitive
analysis to the valuation results in Table 5. Results refer to
BIODIV, GRWATER and HEALTH, whose values are estimated
for a set of different age and concern profiles. These figures are
Table 4 – Estimated coefficients of nested logit models
Model (1) Model (2)
PRICE −0.014⁎⁎⁎ −0.015⁎⁎⁎
(0.794 − 03 ) (0.828 −03 )
BIODIV −0.068⁎⁎⁎ −0.075⁎⁎⁎
(0.016) (0.016)
GRWATER −0.0263⁎⁎⁎ −0.044⁎⁎⁎
(0.004) (0.010)
HEALTH −0.006⁎⁎⁎ −0.003⁎
(0.001) (0.002)
HEALTH ×AGE
−0.100 −03⁎⁎⁎
(0.433 −04 )
GRWATER ×CONCERN
−0.004⁎
(0.003)
SAMPLE 1358 1337
Log–likelihood −811.159 −792.572
Pseudo-R 2 0.562 0.565
LR test of significance of all coefficients 37.174
(pb0.001)
Note: Significance is indicated by ⁎⁎⁎, ⁎⁎ and ⁎ for the 1, 5, and 10%
level, respectively, with standard error in parentheses. Calculations
are performed with nested logit procedures in Nlogit 4.0.
ð2Þ

ECOLOGICAL ECONOMICS 67 (2008) 598– 607
605
Table 5 – Sensitivity analysis to the valuation results
Average respondent (1) Age − (2) Age + (3) Concern − (2) Concern + (3)
BIODIV 3.87 3.49 5.23 2.68 35.71
GRWATER 2.27 2.05 3.07 1.57 20.95
HEALTH 0.15 0.14 0.21 0.11 1.43
Based on Model (2).
Note: (1) Average respondent with respect to age and concern profile. (2) ‘Age −’ and ‘Concern −’ are measured at the first quartile of the
distribution. (3) ‘Age +’ and ‘Concern +’ are measured at the third quartile of the distribution.
computed by multiplying the related interaction estimate by
individual profiles, measured at the sample mean, and at the
first and third quartile of the distribution. An average respondent
(with respect to age and concern), is associated to a
marginal WTP for BIODIV, GRWATER and HEALTH equal to,
respectively, €3.8, €.32and €0.15 per month.
In a similar way, we can run the same valuation exercise,
and thus estimate the marginal WTP for BIODIV, GRWATER
and HEALTH, when considering changes in the individual
characteristics namely her/his age profile and concern level.
6. Welfare analysis
We have developed now a framework for the economic
valuation of several relevant pesticide impacts on ecosystems
and human health using choice experiments. This approach
allows us to estimate the effect that different environmental
attributes of a foodstuff can have on daily household grocery
shopping decisions. Our results indicate that the choice between
alternative scenarios depends in predictable ways on
their environmental and economic shopping attributes. Thus,
respondents consider food purchased in shops to be less
attractive if the pesticide pollution generated by the underlying
agricultural production process is higher.
From an analytical point of view, the results of the choice
modelling experiment perform well in terms of fit and theoretical
validity. The Pseudo-R 2 values of the estimated models
are very high (N0.56) (Hensher et al., 2005), and the signs of
major estimated coefficients are statistically significant
and consistent with the theoretical predictions. Respondents
evaluate a price increase negatively, but evaluate a risk reduction
positively. This is true for any type of pesticide risk
considered (biodiversity, farmland contamination and human
health).
Marginal utilities vary depending on the type of pesticide
risk and provide mixed results (see Table 6). For a unit level of
Table 6 – Willingness-to-pay estimates
Mean Lower-bound Upper-bound
BIODIV 4.86 2.54 7.39
GRWATER 1.88 1.34 2.48
HEALTH 0.43 0.01 0.87
Based on Model (1).
Note: Willingness-to-pay is expressed in Euros per household per
month. Upper and lower bounds are calculated using the delta
method (Goldberger, 1991, pp.110) at 95% confidence levels.
risk decrease, reductions are priced higher if they refer to
biodiversity. However, the interpretation of results is complicated
by the use of different unit of measure for different risks.
To facilitate the interpretation of these results, we can consider
unit trade-offs between attributes shown by their rate of
substitution (see Table 7).
In Table 7 we can see that, on average, respondents are
willing to tolerate (i.e. to trade) six additional cases of human
illness to save one entire species of farmland birds, and two
cases of human illness to reduce soil and groundwater contamination
by 1%. Similarly, Foster and Mourato (2000) find
that respondents are willing to tolerate six to eight additional
cases of human illness to save an entire farmland bird species.
However, for bird biodiversity and human illness, our estimates
are higher than those by Foster and Mourato (2000).
They calculate a willingness-to- pay (WTP) of about €20 per
household per year to save one farmland bird species, and a
WTP of about €3 per household per year to avoid one case of
human illness. Our average estimates (based on Model (1)) are
approximately: €58.3 per household per year to save one
farmland bird species; €5.1 per household per year to avoid
one case of human illness; and €22.5 per household per year
to reduce soil and groundwater contamination by 1% 12 .
According to our estimations, therefore, the annual WTP of
an Italian household ranges from €874 to protect all the 15
endangered bird species; €1,465 to eliminate soil and groundwater
contamination in farmland areas (currently set at 65%);
to €1,286 to eliminate all the cases of acute pesticide intoxication
(250 cases a year).
Finally, our results confirm that WTPs are affected by the
respondents' socio-economic profiles, in terms of age and
level of concern about pesticide risks. The coefficient
estimates for HEALTH⁎ AGE, GRWATER ⁎ CONCERN in Model
(2) suggest that the individual age and concern profiles are
likely to influence the WTP for pesticide risk abatement in a
predictable way. In particular, respondent more aged, are
more prone to pay to purchase risk abatements, and so do
respondents with a higher concern profile (see Table 5).
These results signal the importance of knowing as accurately
as possible the respondents' socio-economic and attitudinal
12 Differences from the estimations by Foster and Mourato (2000)
might derive both from differences in modelling and elicitation
features. In fact, whereas they employ contingent ranking and
use a price premium on a single food product, a loaf of bread, we
use choice experiment and employ a green shopping payment
vehicle.

606 ECOLOGICAL ECONOMICS 67 (2008) 598– 607
Table 7 – Unit trade-offs across choice attributes
HEALTH GRWATER BIODIV
HEALTH [no. cases of illness] 1 0.4 0.1
GRWATER [% contaminated 2 1 0.4
farmland]
BIODIV [no. endangered bird
species]
6 3 1
Note: Trade-offs are calculated on the basis of a basic nested logit
model (Model (1)) including the choice attributes PRICE, BIODIV,
GRWATER, HEALTH. See footnote 11.
features, and of improving the methods for gathering such
information.
7. Conclusions
This article has provided an economic assessment of the nonmarket
benefits of safety improvements in the environmental
and health safety of agricultural production that can be
achieved by non-conventional agricultural practices based on
lower pesticide use. The valuation is based on a questionnaire
survey undertaken in Milan, one of the biggest metropolitan
areas located in the North of Italy. The valuation exercise
employs the Choice Experiment (CE) technique, which is an
innovation in the pesticide risk valuation literature (for a
discussion, see Travisi et al., 2006). The biggest advantage of CE
compared with Contingent Valuation (CV) is that respondents
are forced to make trade-offs − not only between environmental
issues and money − but also between different aspects
of environmental safety. These are important and typical
features of environmental decision making and are central to
the debate on the most preferred type of pesticide policy in
Italy and Europe. In this concern, our study suggests that the
welfare gain of a policy aimed at reducing pesticide risks would
strictly depend on the policy target in terms of: i) level of
provision of the risk reduction; and ii) dimension of pesticide
risk to be reduced. At present, the literature addresses in
particular the relation between level of pesticide risk decline
and WTP estimates (e.g. Florax et al., 2005). However, more
efforts are still required to improve the knowledge on the more
appropriate rate of substitution between different pesticide
risk targets. It is indeed crucial to set efficient pesticide policy
priorities and targets. In this connection, the transfer of values
developed in a non- market setting is appealing but difficult in
itself. The great variety of pesticide risks concerned and
methods applied in the literature suggests that more primary
research is needed. In particular, it is important that future
valuation effort carefully specifies both the baseline level of
risk and the change in the risk level. More attention is also
necessary for the income-specific and potentially locationspecific
(i.e. respondent-specific) nature of the valuation of
reductions in pesticide risk exposure.
Acknowledgements
The authors are grateful to two anonymous referees for helpful
comments. The authors wish to thank the research team on
ecotoxicology led by Marco Vighi (University of Milano-Bicocca),
for providing scientific support during the survey design. A
special word of thanks goes to Anna Alberini (Maryland University)
and Jeroen van den Bergh (Free University of Amsterdam)
for helpful suggestions on an earlier draft.
REFERENCES
Brethour, C.A., Weersink, A., 2001. An economic evaluation of the
environmental benefits from pesticide reduction. Agricultural
Economics 25, 219–226.
Bunch, D.S., Louviere, J.J., Anderson, D., 1993. A Comparison of
Experimental Design Strategies for Multinomial Logit Models:
The Case of Generic Attributes. Working Paper.
Cuyno, L.C.M., Norton, G.W., Rola, A., 2001. Economic analysis of
environmental benefits of integrated pest management: a
Philippine case study. Agricultural Economics 25, 227–233.
Florax, R.J., Travisi, C.M., Nijkamp, P., 2005. A meta-analysis of the
value of reducing pesticide risk exposure. European Review of
Agricultural Economics 32, 141–167.
Foster, V., Mourato, S., 2000. Valuing the multiple impacts of
pesticides use in the UK: a contingent ranking approach.
Journal of Agricultural Economics 51, 1–21.
Goldberger, A.S., 1991. A course in econometrics. Harvard
University Press, Cambridge.
Goldberg, I., Roosen, J., 2007. Scope insensitivity in health risk
reduction studies: a comparison of choice experiments and the
contingent valuation method for valuing safer food. Journal of
Risk and Uncertainty 34 (2), 123–144.
Green, W.H., 2002. Econometric analysis, 4th edition. Prentice Hall,
Upper Saddle River, NY.
Hasler, B., Birr Pederson, K., 2004. The multiple values of
groundwater protection in Denmark: designing a choice
experiment study, Conference proceedings of the 90th EAAE
Seminar, 28- 29 October 2004, Rennes, France.
Hausman, Jerry A., McFadden, Daniel, 1984. Specification tests for
the multinomial logit model. Econornetrica 52 (5), 1219–1240.
Hensher, D.A., Johnson, L.W., 1981. Applied Discrete Choice
Modelling. John Wiley and Sons, New York.
Hensher, D.A., Rose, J.M., Greene, W.H., 2005. Applied Choice
Analysis: A Primer. Cambridge University Press, Cambridge, UK.
Huber, J., Zwerina, K., 1996. The importance of utility balance in
efficient choice set designs. Journal of Marketing Research 33,
307–317.
Lazari, A.G., Anderson, D.A., 1994. Design of discrete choice set
experiments for estimating both attribute and availability
cross effects. Journal of Marketing Research 31, 375–383.
Loureiro, M.L., Umberger, W.J., 2003. Estimating consumer
willingness to pay for country-of-origin labeling. Journal of
Agricultural Resource Economics 28, 287–301.
Loureiro, M.L., Umberger, W.J., 2005. Assessing consumer
preferences for country-of-origin labeled products. Journal of
Agricultural Applied Economics 37, 49–63.
Louviere, J.J., Hensher, D.A., Swait, J.D., 2000. Stated choice
methods; analysis and applications. Cambridge University
Press, Cambridge.
Lusk, J.L., Hudson, D., 2004. Willingness-to-pay estimates and their
relevance to agribusiness decision making. Review of
Agricultural Economics 26 (2), 152–169.
McFadden, D., 1986. The choice theory approach to market
research. Marketing Science 5, 275–297.
Pimentel, D., Greiner, A., 1997. Environmental and socio-economic
costs of pesticide use. In: Pimentel, D. (Ed.), Techniques for
reducing pesticide use: economics and environmental benefits.
John Wiley and Sons, Chichester, pp. 51–78.
Pimentel, D., Acquay, H., Biltonen, M., Rice, P., Silve, M., Nelson, J.,
Lipner, V., Giordano, S., Horowitz, A., D'amore, M., 1992.

ECOLOGICAL ECONOMICS 67 (2008) 598– 607
607
Environmental and human costs of pesticide use. Bioscience
42, 750–760.
Schou, J.S., Hasler, B., Kaltoft, P., 2002. Valuing biodiversity effects
of pesticide use. What does perception of uncertainty mean for
survey design? Conference Proceedings. Risk and Uncertainty
in Environmental and Resource Economics, Wageningen, The
Netherlands.
Söderqvist, T., 1998. Valuing chemical characteristics: a hedonic
approach. In: Swanson, T., Vighi, M. (Eds.), Regulating Chemical
Accumulation in the Environment. Cambridge University
Press, Cambridge, UK.
Travisi, C.M., Nijkamp, P. (2004). Willingness to Pay for Agricultural
Environmental Safety: Evidence from a Survey of Milan, Italy,
Residents, Fondazione Eni Enrico Mattei SIEV Working Paper
Series, 100.
Travisi, C.M., Nijkamp, P., Vindigni, G., 2006. Pesticide risk
valuation in empirical economics: a comparative analysis.
Ecological Economics 56 (4), 455–474.
Wikström, D. (2003). Willingness-to-Pay for Sustainable Coffee. A
Choice Experiment Approach. Master's Thesis, Department of
Business Administration and Social Science, Luleå University
of Technology, Luleå, Sweden.
Wilson, C., 2002. Pesticide avoidance: a result from a Sri-Lankan
study with health policy implications. In: Hall, D.C., Moffitt, L.J.
(Eds.), Economics of Pesticides, Sustainable Food Production,
and Organic Food Markets. Elsevier Science, Amsterdam.