www.elsevier.com/locate/ecolecon

Ecological Economics 5

ANALYSIS

Analysing the social benefits of soil conservation measures using

stated preference methods

Sergio Colombo a,*, Javier Calatrava-Requena a, Nick Hanley b

a Department of Agricultural Economics, Instituto Andaluz de Investigacion Agraria (IFAPA), Andalucıa Government,

Camino de Purchil s/n, 18004 Granada, Spainb Department of Economics, University of Stirling, Stirling FK94LA, Scotland, UK

Received 29 March 2005; received in revised form 13 September 2005; accepted 18 September 2005

Available online 2 November 2005

Abstract

The paper estimates the benefits of programmes to mitigate the off-site impacts of soil erosion for a watershed in Andalusia,

Spain. Two stated preference methods are used, namely choice experiments and contingent valuation, to obtain estimates of the

social benefit from soil erosion reductions under two different methodologies. We emphasise the relative merits of the choice

experiment method to provide useful inputs to policy design. However, employing both methods allows us to undertake a

convergent validity test and thus to provide more defensible social benefit estimates. The attributes used in the choice

experiment include water quality impacts (which we find to have the highest marginal values), impacts on wildlife and the

area subject to a control programme. The contingent valuation design includes an attempt to reduce bias by reminding

respondents about substitutes. Results are used to suggest upper limits on per hectare payments for soil conservation

programmes.

D 2005 Elsevier B.V. All rights reserved.

Keywords: Convergent validity test; Choice experiments; Contingent valuation; Soil erosion

1. Introduction

Soil is a non-renewable resource which is indis-

pensable for life. In the south of Spain, soil erosion

rates are extremely high due to climatic conditions,

soil characteristics and the nature of tillage systems

0921-8009/$ - see front matter D 2005 Elsevier B.V. All rights reserved.

doi:10.1016/j.ecolecon.2005.09.010

* Corresponding author.

E-mail address: scolombo@ugr.es (S. Colombo).

employed in many olive groves, a cultivation style

that cause severe land degradation yet which occupies

most of the cultivated area of the eastern part of

Andalusia region (MAPA, 2000). One reason for

high soil erosion rates is the fact that all olive groves

are privately owned, and farmers are not rewarded by

the market for reducing soil erosion rates to the level

which society demands. This is for several reasons.

First, since the effects of soil erosion occur with a time

lag, farmers may not perceive these effects, or may

8 (2006) 850–861

S. Colombo et al. / Ecological Economics 58 (2006) 850–861 851

discount future losses more highly than society would

choose to. Third, many of the off-site impacts of soil

erosion have environmental effects which go un-

priced by markets. These off-site impacts include

deteriorations in water quality, desertification of the

landscape, and loss of wildlife habitat.

Rational responses on the part of government

require, from the economist’s viewpoint, information

on the costs of these environmental consequences, and

of peoples’ preferences for soil conservation pro-

grammes which reduce these costs. However, there

are relatively few studies worldwide on the economic

costs of the off-farm effects of soil erosion (Clark et al.,

1985; Feather et al., 1999, Hansen et al., 2002). In

Spain, some valuations of the off-farm effects of soil

erosion have been undertaken, for example ICONA

(1982), Almansa and Calatrava-Requena, 2002 and

Colombo et al., 2003. Almansa and Calatrava-Requena

(2002) studied a case specific to a reforestation project

in a small municipality of Almeria. The study of

Colombo et al. (2003) investigated the off-site effects

of soil erosion in the same watershed1 considered in

this study, but was based on a pilot study sample, and

did not include a choice experiment exercise.

This paper adds to this literature by comparing

estimates of the value of soil erosion control pro-

grammes obtained using contingent valuation (CV)

and choice experiments (CE), and by estimating

values for different attributes of soil conservation

plans using the latter technique. This is felt to be

particularly valuable in helping policy makers re-

design soil erosion control programmes which can

increase social benefits. Finally, we use our results

to suggest upper bounds on the per hectare pay-

ments that the government should offer to farmers

in the watershed. The rest of the paper is structured

as follows: a brief description of CV and CE under-

pinnings and a summary of the studies that have

carried out convergent validity tests using these

techniques is presented in Section 2. The question-

naire design is then presented in Section 3 and

results are shown in Section 4. Some conclusions

are offered in Section 5.

1 For a description of the watershed and the erosion rates in it, see

Colombo et al. (2003).

2. Methodological framework

Stated preference approaches rely on direct ques-

tioning of the affected population to estimate the

economic value, measured as either Willingness to

Pay or Willingness to Accept Compensation, for a

change in the level of supply of some environmental

good. The main difference between the two methods

used here is how they approach this problem. Con-

tingent Valuation takes a bwhole goodQ approach, andasks WTP/WTAC for a discrete change in some

environmental good—such as water quality. Whilst

the nature of this change may be explained with

reference to the attributes of this good (such as eco-

logical quality, or pesticide levels), the method does

not estimate values for these individual attributes. In

contrast, the Choice Experiment method is based on

Lancaster’s characteristics theory of value (Lancaster,

1966). An environmental good is characterised as a

collection of attributes, and the levels these take.

Experiment design methods are then used to construct

choice tasks through which respondents reveal the

marginal values they place on each attribute. To

value a programme which changes the levels of

many of these attributes simultaneously–such as

with a soil erosion control programme–the compen-

sating surplus (WTP or WTAC) must be evaluated by

effectively summing over attribute values.

Since the underlying methodology of both CV and

CE is now well-known and widely discussed, we do

not present it here (see, for example, Bateman et al.,

2002). Rather, we briefly summarise how the existing

literature has compared the application of CV and CE

to the same environmental change, before explaining

how the CV and CE exercises were defined in our

study.

Previous authors have sought to compare CV and

CE estimates of the same underlying utility change as

a test of bconvergent validityQ (Mitchell and Carson,

1989). This test asks whether two valuation methods,

applied to the environmental change, produce signifi-

cantly different estimates of value. bPassingQ the con-vergent validity test would mean a failure to reject the

null hypothesis of no significant difference between

the two estimates, and is seen by many as a useful

measure of the validity of stated preference estimates

(although, as Hanley and Spash pointed out many

years ago, both answers could be equally wrong:

S. Colombo et al. / Ecological Economics 58 (2006) 850–861852

Hanley and Spash, 1994). As Mitchell and Carson

themselves noted, b. . .neither measure (that is, neither

estimate) is assumed to be a truer measure of the

construct than the otherQ (p. 204).In the literature there are actually rather few studies

that compare Willingness to Pay estimates derived

from CE and CV methods. Examples can be found

in Boxall et al. (1996), Hanley et al. (1998), Adamo-

wicz et al. (1998), Christie and Azevedo (2002) and

Lehtonen et al. (2003). Differences emerge in terms of

the relative magnitude of CE and CV estimates. For

instance, Boxall et al. (1996) and Adamowicz et al.

(1998) found that WTP estimated from the CV

method was higher than that obtained from CE. Han-

ley et al. (1998) and Christie and Azevedo (2002)

found a reverse relation, with CE welfare estimates

higher than the CV ones. Only Christie and Azevedo

(2002) and Lehtonen et al. (2003) report formal tests

for the statistical significance of these differences: in

the former paper, the authors concluded that they were

not able to support the equality hypothesis between

CV and CE value estimates, although a significant

difference was found in the latter study.

2 The description of the attributes and attribute levels were pro-

vided with in much greater detail to respondents in a colour booklet3 108 profiles is the smallest orthogonal and balanced design tha

allows the estimation of the 15 main effects plus the 90 two-way

attribute interactions (Louviere et al., 2000).

3. Study design

The main objective of the study was to identify

people’s preferences towards reducing the off-farm

effects of soil erosion in the Alto Genil watershed in

Andalusia. The questionnaire consisted of three parts.

The first part contained two questions aimed at elicit-

ing the relative importance that respondents attributed

to the environment with respect to other four areas of

public interest (education, health, crime, and cultural

heritage), and the relative importance that respon-

dents’ gave to the reduction of soil erosion among

three other areas of environmental concern (water, air

and biota). The information gathered in these ques-

tions was used in the CV exercise to reduce the effect

of budgetary substitutes bias, as noted below.

The second part of the survey contained the CV

and CE exercises (it is important to note that each

respondent was asked to complete both the CV and

CE tasks). The design of the CE exercise followed

several steps. At the beginning, through a wide bib-

liographic review of the agronomic, geological and

environmental literature, the main off-site effects of

soil erosion were identified. Subsequently, by means

of focus groups and informal interviews, we identified

the best subset of soil erosion effects to be used as CE

attributes. The attribute levels described the likely

future conditions with and without the implementation

of soil erosion reduction projects, and were selected

using the geographic information systems, experts’

advice and existing empirical studies. By means of

the geographic information system, we obtained a

detailed description of the areas in which the project

had to be implemented, and used this to predict

changes in attribute values. For instance, providing

experts with information regarding the number of

hectares that would be reforested and with which

trees or bush species, we predicted the expected den-

sity of flora and fauna with and without project execu-

tion. The number of jobs created was obtained by

relating the expected rise in agricultural production

due to soil erosion reduction (Pastor et al., 1999) to

the increase in farm work of olive collection, using the

formula of Lopez (1992). A particular effort was

dedicated in the definition of the monetary attribute

level, due to its central role in the welfare analysis. A

contingent valuation pilot study, details of which can

be seen in Colombo and Calatrava-Requena (2002),

allowed the definition of the mean value of the mone-

tary level and its range. The attributes and attribute

levels finally chosen are shown in Table 12. By the

inclusion of the barea of the project execution (km2)Qattribute, we tested the hypothesis of sensitivity to

spatial scope in the CE estimates. A positive and

significant regression coefficient of this attribute

would indicate that respondents prefer a larger area

to be covered by the soil conservation plan, other

things being equal. Other attributes included land-

scape impacts, wildlife impacts, effects on water qua-

lity, and effects on rural employment.

This set of attributes and levels forms a universe of

1,062,153 possible combinations. By means of the

experimental design techniques (Louviere, 1988) an

orthogonal fraction of the complete factorial3 was

drawn, yielding 108 combinations to be presented to

.

t

Table 1

Attributes and attribute levels used in the study

Attributes Levels

Landscape change: desertification

of the semiarid areas

Degradation

Small improvement

Improvement

Surface and ground water quality Low

Medium

High

Flora and fauna quality Poor

Medium

Good

Rise of agricultural productivity:

jobs created (number)

0

100

200

Area of project execution (km2) 330

660

990

Extra tax (euros) 6.01

12.02

18.03

24.04

30.05

36.06

Table 2

The status quo attributes and levels

Attributes Levels

Landscape change: desertification Degradation

S. Colombo et al. / Ecological Economics 58 (2006) 850–861 853

respondents. Since it is unrealistic that a respondent

can complete 108 choice tasks and considering that

soil erosion might be an unfamiliar issue to respon-

dents, we decided to present only four choice sets to

each respondent. The 108 profiles were therefore split

into 27 blocks, taking care to minimise both the

number of blocks and intra block correlation. The

status quo option described the expected environmen-

tal situation in the watershed in 50 years if no soil

conservation measures are implemented (Table 2).

Alternatives A and B represented the expected envir-

onmental situation in the watershed in 50 years with

different soil conservation measures. An example of

choice set is shown in Table 34.

In the CV exercise, respondents are asked to

state their maximum WTP to have, in 50 years

time, an environmental situation characterised by a

reduction of landscape desertification, a bmedium

qualityQ of the surface and ground waters, a

bmedium quality of flora and faunaQ, the creation

4 To reduce complexity at the choice stage and to avoid fatigue

effects, the description of the status quo option was dropped from

the choice set and provided to respondents on a separate sheet.

of 100 jobs and the area of project implementation

of 330 km2. This is one of the scenarios which we

can simulate in the choice experiment. The refer-

ence situation, which would result if no hypotheti-

cal payment was forthcoming, was the status quo

option as in the CE. The two stated preference

methods were thus designed to be directly compar-

able with each other in terms of valuing a given

policy scenario. The CV design followed the

scheme used in Colombo et al. (2003). A bdoublestepQ process used to account for possible budgetary

substitute bias was used. Respondents who stated a

positive WTP for the erosion process (WTP 1)

were then asked to express a WTP for three pos-

sible environmental bsubstitute projectsQ in the

watershed. These were (I) reducing sewage pollu-

tion in their town, (II) reducing local urban air

pollution and (III) conserving local biodiversity. In

a second step, respondents were then informed

about their combined WTP (i.e. the sum of the

four WTP values, considered as an benvironmental

WTPQ). They were then reminded about their

responses earlier in the questionnaire to questions

concerning the importance that they assigned to

environment issues in general relative to other

demands on public funds, and to soil erosion rela-

tive to other environmental problems faced in the

region. They were then asked to reconsider/restate

the WTP for the four environmental projects (their

environmental WTP), and finally to reconsider/

restate their WTP value for the soil erosion project

(WTP 2).

The third part of the questionnaire gathered socio-

economic information. The survey was administered

to 345 citizens selected so as to be representative of

the watershed population in terms of their residence

of the semiarid areas

Surface and ground water quality Low

Flora and fauna quality Poor

Agricultural productivity: job created (number) 0

Area of project execution (km2) 0

Extra tax (euros) 0

Table 3

Example of choice set card presented to respondents

Situation A Situation B Status quo

Landscape change: desertification of

the semiarid areas

Worsening Improvement Neither situation A nor situation B

is worth the extra tax payment.

Surface and ground water quality Low High

Flora and fauna quality Medium High

Agricultural productivity: job created (number) 200 100

Area of project execution (km2) Two third All I choose the Status Quo option 5

Extra tax (euros) 18.03 24.04

I choose situation A 5 I choose situation B 5

S. Colombo et al. / Ecological Economics 58 (2006) 850–861854

municipality. The survey format was face to face

interviews.

4. Results

Of the 345 citizens who were interviewed, 19 did

not complete the questionnaire and were excluded from

the analysis, 74 expressed a protest answer and did not

respond to the CE cards; 51 displayed a zero WTP in

their CE response by choosing always the status quo

option. Some 201 persons thus fully completed the

survey, providing 1008 (252*4) valid observations

for choice model estimation, and (201+51) responses

for the CV analysis. Sample characteristics were com-

pared to the general Andalusian population character-

istics on the basis of the distribution of age, gender and

urban–rural area of residence, and were found to be

representative of the regional population.

4.1. Analysis of choice experiment data

In a random utility context, the probability that

individual n will choose option i over any other option

j belonging to the complete choice set C is given by:

Probin ¼ Prob Vin þ einNVjn þ ejn� �

8jaC ð1Þ

where V is the deterministic component of utility, and eis a random component. A typical assumption is that

this random component is independently and identi-

cally distributed (IID) with and extreme-value distribu-

tion5. Under this hypothesis an explicit form of the

5 This assumption leads to the IIA axiom (Louviere, 2000). If the

IIA property does not hold the analyst can use other models (nested

logit model, random parameters logit model, multinomial probit

model) that, despite a higher level of analytical complexity, relax

the IID assumption.

probability of choice of Eq. (3) is satisfied by the

conditional logit model (McFadden, 1973):

Probin ¼ exp kVinð Þ=X

j

exp kVjn

� �8jaC ð2Þ

where k is a scale parameter which is inversely pro-

portional to the standard deviation of the error terms

and Vin and Vjn are conditional indirect utility func-

tions assumed to be linear in parameters,

Vjn ¼ Cj þX

bjkXjk þX

cjn SnTCj

� �ð3Þ

where: Cj is an alternative specific constant, Xjk is the

k attribute value of the alternative j; bjk is the coeffi-

cient associated to the k attribute, Sn is the socio-

economic characteristics vector of individual n and cjnis the vector of the coefficients associated to the

individual socioeconomic characteristics6. Having

estimated these parameters, the marginal rate of sub-

stitution (MRS) between any pair of attributes a and b

can be calculated by the formula (4)

MRS ¼ � battribute a=battribute bð Þ ð4Þ

In addition, willingness to pay measures relative to

different environmental scenarios can be obtained by

applying (5) where V0 is the utility associated to the

binitial stateQ, V1 is the utility associated to an

improved state encompassed in the study and bm is

the monetary attribute coefficient, interpreted as the

marginal utility of income.

Economic surplus ¼ � 1=bm V0 � V1ð Þ ð5Þ

We began by estimating a conditional logit

model: results are shown in Table 4. Model I

6 Since respondents’ characteristics do not vary across alterna-

tives, singularities arise in model estimation unless the socioeco-

nomic characteristics are introduced as interactions with either the

attributes or the alternative specific constants.

Table 4

Choice model results

Model I Model II Model III

Coefficients Std.

error

Coefficients Std.

error

Coefficients Std.

error

Coefficients

Constant � 2.899* 0.287 � 7.590* 0.681 � 9.821* 1.459

Landscape desertification: small improvement 0.939* 0.143 1.017* 0.148 1.527* 0.245

Landscape desertification: improvement 1.367* 0.142 1.516* 0.149 2.520* 0.288

Surface and ground water quality: medium 0.910* 0.154 1.052* 0.162 1.704* 0.281

Surface and ground water quality: high 1.348* 0.148 1.502* 0.155 2.349* 0.299

Flora and fauna quality: medium 0.656* 0.145 0.774* 0.152 1.179* 0.237

Flora and fauna quality: good 0.882* 0.141 1.049* 0.148 1.426* 0.250

Jobs created 0.006* 0.001 0.007* 0.001 0.010* 0.001

Degraded area treated 0.001* 0.0002 0.001* 0.0002 0.001 0.0004

Tax � 0.046* 0.006 � 0.057* 0.007 � 0.108* 0.014

Constant*solidaritya 0.014* 0.005 0.025* 0.007

Constant*enjoymentb 0.016* 0.008 0.016 0.012

Constant*erosionc 0.003* 0.002 0.004** 0.002

Constant*genderd 0.039* 0.023 0.052 0.032

Constant*agee � 0.072* 0.027 � 0.109* 0.039

Constant*marital statusf 0.063* 0.026 0.071** 0.037

Constant*occupationg 0.109* 0.026 0.199* 0.041

Constant* incomeh 0.00013* 0.00004 0.00018* 0.0001

Standard deviation

Landscape desertification: small improvement 0.832** 0.416

Landscape desertification: improvement 0.676 0.522

Surface and ground water quality: medium 0.830** 0.420

Surface and ground water quality: high 1.193* 0.375

Flora and fauna quality: medium 0.133 0.566

Flora and fauna quality: good 1.033** 0.431

Jobs created 0.009* 0.002

Degraded area treated 0.003* 0.0004

Number of observations 1008 1008 1008

Log likelihood at constant � 1049.50 � 1049.50 � 1049.50

Log likelihood at convergence � 912.70 � 821.945 � 755.397

LR 273.60 455.11 588.206

Pseudo R2 130 258 318

a Importance that respondents assigned to solidarity (likert scale 1–10).b Importance that respondents assigned to leisure activities in a natural environment (likert scale 1–10).c Percentage share of 1 public pound that respondents gave to the funding of soil erosion reduction projects among other natural resource care

projects (improve air, water and biodiversity quality).d Respondents’ gender (female: 0; male: 1).e Respondents’ age (less than 50: 0; more than 50: 1).f Respondents’ marital status (not married: 0; married: 1).g Respondents’ occupation (not active worker: 0; active worker: 1).h Respondents’ income.

* Significant at 1% confidence level.

** Significant at 5% confidence level.

S. Colombo et al. / Ecological Economics 58 (2006) 850–861 855

S. Colombo et al. / Ecological Economics 58 (2006) 850–861856

represents the regression coefficients of an attri-

butes-only7 conditional logit model; whilst model II

includes socioeconomic and attitudinal characteristics

of respondents8. The significance of the regression

coefficients in model II implies that it is the preferred

model to be used for welfare estimation. Model II is

highly significant (LRTest=455.11, significant at the

0.000 level), and when tested for the IIA assumption

passed the Hausmann and McFadden (1984) test with

a v2 value of 10.60 ( p =0.91).

All regression coefficients are significant at the 1%

level. The signs of all the utility function parameters

are theoretically consistent. In all cases, the highest

attribute level has associated with it the highest utility.

The three qualitative environmental attribute coeffi-

cients (landscape desertification, surface and ground

water quality and flora and fauna quality) suggest

increasing utility as respondents are offered better

environmental conditions. The coefficients of these

qualitative environmental attributes are calculated

from dummy variables: they reflect the relative utility

with respect to the status quo situation. So, in the case

of the landscape desertification, the coefficient of the

bmediumQ improvement level represents the average

change in the utility that respondents experience in

going from the worse level (reference) to the medium

improvement level. The differences between the med-

ium and the high levels of the qualitative attributes

were also checked fitting the model with the medium

level of each attribute used as reference. Again, the

difference between the attribute’ levels were found to

be significant at a 1% confidence level, except for

flora and fauna quality where the difference between

the medium and the highest level were found to be

significant at the 5% confidence level. The signifi-

cance of the social attribute (jobs created) reveals that

respondents care about employment: this is not sur-

prising considering that there are a lot of seasonal

workers in the Andalusia region whose incomes

depend on the collection of olives, almonds and vege-

7 The inclusion of the second order interactions, although permit-

ting a better fit, did not provide significant changes in the regression

coefficients of attributes or in the welfare measures.8 Initially all the socioeconomic characteristics gathered in the

survey were considered in the model. Subsequently, the non-sig-

nificant ones at 10% level were dropped from the model.

tables. It is interesting that the coefficient representing

the area included in the project (the number of square

kilometres of the watershed in which will be imple-

mented in the soil erosion reduction project) is sig-

nificant, showing that a scope effect is present in the

data.

The negative sign of the constant reveals the pre-

sence of a status quo bias, i.e. the utility associated

with moving away from the current situation is nega-

tive and significant. As Adamowicz et al. (1998)

pointed out status quo bias is a common economic

phenomenon that may happen because respondents

(1) find the task of selecting the preferred option too

complex; (2) they are uncertain about the trade-off

they would be willing to make; and (3) do not trust the

government to actually implement any of the soil

erosion control projects. Soil erosion is a rather tech-

nical issue about which people are often unfamiliar.

The rather small amount of information given in the

survey may have been insufficient for some respon-

dents to feel comfortable to make a positive choice

about the design of soil erosion control policy, whilst

the status quo offered an easy opt-out.

The socioeconomic interactions with the constant

show that people who assigned higher importance to

community solidarity, or who enjoy activities closely

related with the natural environment, have a greater

probability of choosing either alternative A or B.

People who allocated a higher share of the public

budget to reducing soil erosion among the other

three areas of environmental care (water, air and

biota) were also more likely to chose options A or

B rather than the status quo. Residents older than 50

years were more likely to choose the no change

option relative to younger people. Marital status

and occupation influences choice of the preferred

alternative as well, in the sense that married and

active worker people are the groups that have a

greater probability of choosing either alternative A

or B. The income coefficient shows that the higher

the income the higher the probability of opting for

soil erosion control.

Including socioeconomic characteristics as in

model II is also one way of incorporating preference

heterogeneity in the model. A further step to include

respondents’ heterogeneity is to use the random para-

meters logit model (Train, 1998). In this approach, the

utility function described in (3) is augmented with a

Table 5

Implicit prices and confidence intervals

Model IIa Model IIIa

Implicit price (o) Implicit price (o)

Attribute coefficients

Landscape desertification: small improvement 17.78 (12.02; 25.21) 14.09 (9.67; 19.87)

Landscape desertification: improvement 26.51 (20.05; 35.76) 23.25 (18.53; 29.79)

Surface and ground water quality: medium 18.39 (12.67; 25.96) 15.72 (11.03; 21.47)

Surface and ground water quality: high 26.27 (20.10; 34.67) 21.67 (17.02; 27.26)

Flora and fauna quality: medium 13.53 (7.96; 19.54) 10.87 (6.87; 15.56)

Flora and fauna quality: good 18.34 (13.11; 24.57) 13.16 (8.86; 18.26)

Jobs created 0.119 (0.088; 0.160) 0.089 (0.065; 0.123)

Degraded area treated 0.014 (0.007; 0.023) 0.005 (� 0.004; 0.013)

a 95% Confidence interval in parenthesis.

S. Colombo et al. / Ecological Economics 58 (2006) 850–861 857

vector of parameters g that incorporate individual’s

preference deviations with respect to the mean pre-

ference values expressed by the vector b:

Vjn ¼ Cj þX

bjkXjk þX

gnkXjk þX

cjn SnTCj

� �

ð6Þ

where g is a vector of deviation parameters which

represents the individual’s tastes relative to the

average (b) and the other terms are as previously

defined. The researcher can estimate b and g; the

gnk terms, as they represent personal tastes, are

assumed constant for a given individual across all

the choices they make, but not constant across

people. This implies that IIA is not a property of

the RPL model. In order to estimate the model it is

necessary to make an assumption over how the bcoefficients are distributed over the population:

Train (1998) assumes them to be distributed either

log-normally or normally.

Model III (Table 4) gives results from this random

parameter set-up. By incorporating heterogeneity the

model fitting rises significantly. The attributes-only

model has a pseudo-rho square (q2) of 0.13; by add-

ing the socioeconomic and taste characteristics the q2

rises to 0.258, and, in the random parameter logit

model it assumes a value of 0.3189. In the RPL

9 Simulations conducted by Domenich and McFadden (1975)

compare values of q2 between 0.2 and 0.4 to values between 0.7

and 0.9 of the R2 in the case of the ordinary linear regression.

model, mean effects are very similar to those of

model II, with the exception that the parameter esti-

mate for the degraded area of project execution is no

longer significant. Among the socioeconomic and

attitudinal variables, enjoyment and gender are no

longer significant determinants of choice. Coefficient

values of model III are consistently bigger than those

of model II. This reflects the fact that in the RPL

model the variance in parameters is treated explicitly

as a separate component of the error (gnk, Xjk) such

that the remaining error (enj) is free of this variance.

This does not happen in the conditional logit model,

where the error term encompasses all the variance not

explained by the model itself.

The standard deviation terms in the random

parameter model are significant for six attributes,

suggesting that preferences do indeed vary among

respondents. Their magnitudes are reasonable relative

to the mean coefficient values, the exception being

for the job attribute and especially for the degraded

area attribute, where the standard deviation is three

times the mean attribute value. This high heterogene-

ity could explain the big difference between the impli-

cit price of the degraded area treated attribute in model

II and model III.

Table 5 shows the implicit prices for all the attri-

butes considered. Overall, the confidence intervals of

the implicit prices overlap considerably, so that, at

least in this case, allowing for heterogeneous prefer-

ences makes little difference to welfare estimates.

These implicit prices for all the attributes are positive,

implying that respondents have a positive WTP for

increases in the quality or quantity of each attribute. In

Table 6

Compensating surplus for three possible scenarios

Scenarios Model IIa Model IIIa

Compensating

surplus (o)

Compensating

surplus (o)

Scenario 1 �10.77(�16.87; �4.37)

�14.56(�19.40; �9.67)

Scenario 2 �36.11(�43.75; �29.75)

�34.20(�40.69; �28.57)

Scenario 3 �53.48(�62.33; �45.68)

�44.03(�51.53; �36.81)

a 95% Confidence interval in parenthesis.

S. Colombo et al. / Ecological Economics 58 (2006) 850–861858

case of quantitative attributes, the implicit prices

represent the WTP to achieve one unit more (one

more job, one more square kilometre) of the attribute

considered. If the attributes are qualitative, the impli-

cit prices reflect the WTP for a discrete change in the

attribute level, for example to change from the blowQto the bmediumQ level of surface and ground water

quality. The implicit prices afford some understanding

of the relative importance of the attributes and can be

used by policy makers to assign more resources in

favour of the attributes which have higher implicit

prices. For models II and III, the reduction of land-

scape desertification is the effect that produces the

highest marginal impact on WTP, followed by the

improvement of water quality to a bhighQ level. Note-worthy is that, according to the economic principle of

decreasing marginal utility, the marginal WTP for the

environmental attributes decreases as one moves

toward higher attribute levels. In Table 5 it is possible

to observe, by subtracting the marginal WTP asso-

ciated with the highest level from the marginal WTP

Table 7

Mean and median WTP before and after the iterative process

Statistics WTP 1

WTP for the erosion project WTP for all pr

Mean 26.58 76.46

Median 12.02 30.05

Standard deviation 38.24 124.93

95% Confidence intervals 21.86–31.30 61.03–91.89

Null hypothesis for medians WTP 1=medians WTP 2

Erosion control project

All environmental projects

of the medium level, that the implicit prices of moving

from a medium to a high improvement are around 1/2

of those expressed to move from the low to the

medium level.

Using the choice model parameters, it is also pos-

sible to obtain compensating surplus estimates for a

wide range of policy scenarios. As an example, the

following three scenarios have been used to illustrate

the overall willingness to pay for improvements with

respect to the status quo conditions in the environ-

mental and social quality of the watershed.

Scenario 1: Landscape desertification is charac-

terised by a small improvement; water quality is

improved to a medium level; flora and fauna qual-

ity is improved to medium; jobs created are 100

and the watershed degraded area treated is equal to

330 km2. This scenario corresponds to that used in

the CV exercise.

Scenario 2: Landscape desertification is charac-

terised by a larger improvement; water quality is

high; flora and fauna quality is improved to a

medium level; jobs created are 200 and the

watershed degraded area treated is equal to 660

km2.

Scenario 3: Landscape desertification is charac-

terised by a larger improvement; water quality is

high; flora and fauna quality is good; jobs created

are 200 and the watershed degraded area treated is

equal to 990 km2.

Estimated compensating surplus (willingness to pay)

and 95% confidence intervals for the three scenarios

described above are calculated by means of Eq. (5)

WTP 2

ojects WTP for the erosion project WTP for all projects

16.18 41.06

6.01 24.04

25.07 52.12

13.08–19.28 34.62–47.50

v2 Prob. value of accepting null

7.773 0.000

8.152 0.000

Table 8

Comparison between CV and CE mean estimates

Statistics CV estimate CE (model II) CE (model III)

WTP erosion project WTP Scenario 1 WTP Scenario 1

Mean 16.18 10.77 14.56

95% Confidence interval (13.08; 19.28) (4.37; 16.87) (9.67; 19.40)

95% Confidence interval of the difference (o) Sign

Poe et al. (1997) testa CV-CE(model II) �1.37, 12.77 0.065

CV-CE(model III) �18.13, 6.71 0.177

a H0: WTP CV=WTP CE; H1: WTP CVNWTP CE.

S. Colombo et al. / Ecological Economics 58 (2006) 850–861 859

and are shown in Table 610. Estimated compensating

surplus for the change from the status quo to the

scenarios are consistent across the range of policy op-

tions used in this study and as expected, WTP in-

creases as we move towards better environmental and

social conditions in the watershed. For example, sce-

nario 2 compared to scenario 1 presents a lower land-

scape desertification, a higher water quality, 100 more

jobs and 330 km2 of degraded area treated more. All

these changes bring an increase of 25.34 o in the

case of model II and 19.64 o in the case of model III

in the average respondent’s WTP. Allowing for pre-

ference heterogeneity does not appear to influence

significantly the compensating surplus estimates, and

there is no clear pattern of differences in magnitude

between model II and model III estimates. What

is observable is that the RPL estimates are more sta-

tistically precise than the conditional logit model

estimates.

4.2. The contingent valuation exercise

Basic statistics for positive CVWTP bids are shown

in the second column of Table 7. Some 11% of the

sample stated a genuine zero answer; whilst the per-

centage of protest bidders was 23%. Mean and median

WTP for the erosion project before considering the

substitute projects were 26.58 o and 12.02 o,

whereas after considering these substitutes, WTP fell

to 16.18 o and 6.01 o respectively. The statistics for

benvironmental WTPQ, i.e. referring toWTP for all four

environmental projects (air, water, soil and biodiver-

sity), show the same trend, decreasing from 76.46 o

10 In the estimation, the socioeconomic characteristics were set at

the population mean level.

and 30.05 o to 41.06 o and 24.04 o. There exists a

significant difference (az0.01) between these two

measures as indicated by the Wilcoxon matched-pair

signed-rank test. This reveals the presence of an em-

bedding effect in the WTP estimates before we asked

respondents to consider the substitute projects. This

result underlines the importance of considering in CV

questionnaires a specific ddexerciseTT to deal with the

embedding problem to reduce the risk of biased esti-

mates. The omission of the double step procedure in

our case would have led to an over-estimation of WTP

for the soil erosion project of 64%.

4.3. A convergent validity test

In this study, the significance of the difference

between the two sets of stated preference estimates

were tested following the convolution test proposed

by Poe et al. (1997) in which we took a series of 1000

random draws from the asymptotic distribution of the

parameters. The results of the comparison of CV and

CE estimates in terms of estimated WTP are shown in

Table 811. In the case of model II there is rather weak

evidence to support the convergent validity of the CV

and CE exercises, since the mean WTP from the CE

exercise lies outside the 95% confidence interval of

the CV WTP mean. The Poe et al. (1997) approach

suggests that the CV and CE mean do not differ at

95% confidence levels, but only permits a weak,

borderline acceptance of mean equality. This is not

the case of model III, where the null hypothesis of

mean equality test can be more strongly accepted. So,

11 For comparability, 1000 draws of the open-ended WTP welfare

measure were also simulated from the mean and standard error o

the sample distribution, assuming normality.

f

Table 9

CV and CE aggregate WTP

CV CEscenario 1 CEscenario 2 CEscenario 3

Aggregate

WTP (o)

4,691,275

Model II 3,123,766 10,466,305 15,502,583

Model III 4,222,522 9,915,053 12,765,068

Per hectare

value

(o ha�1)

142.15

Model II 94.66 158.58 156.59

Model III 127.95 150.22 128.94

S. Colombo et al. / Ecological Economics 58 (2006) 850–861860

allowing for taste heterogeneity using the random

parameter approach seems to aid convergent validity.

These individual welfare estimates can be aggre-

gated to determine the WTP of the population to deter-

mine the social value of the soil erosion off-farm effects

in the watershed12. Table 9 shows the aggregate WTP

for the CV method and for the CE for the three

scenarios considered. Depending on the magnitude

of the improvement of the environmental and social

conditions in the watershed that can be achieved by

the implementation of the soil erosion reduction pro-

ject, the social benefits of reducing soil erosion are

between 3.1 and 15.5 million euros per year.

5. Conclusions

Soil erosion produces both on-site and off-site

impacts, which particularly in the Mediterranean

regions of Europe are serious enough to warrant policy

action. However, to ensure costs are not bexcessiveQrelative to benefits, and to aid the optimal design of soil

conservation policy, policy makers need information

on public preferences for this policy intervention.

12 The watershed population is 375,002 people. In the aggregation

we added the people who state a bprotestQ answer (the 22.7% of the

population) considering them as they would have expressed a zero

WTP, as recommended by the NOAA panel (Arrow, 1993). By

including protest answers as zero WTP and by considering only the

people living in the watershed, the aggregate WTP values shown are

under-estimates of total economic value for the reductions in soil

erosion impacts, because (1) people that expressed a bprotestQanswer might have a positive value on reducing soil erosion and

(2) people living outside the catchments (especially downstream)

might well derive utility from these improvements, i.e. have a

positive WTP.

We found that both CE and CV were suitable to

evaluate the off-site effects of soil erosion. We also

found that the welfare estimates did not differ markedly

between the two approaches. In the CV method, we

employed a means of reducing bias by reminding

respondents of substitute environmental goods: this

proved to have a large effect on stated WTP. The CV

method gave estimates of the benefits of a pre-defined

package of potential improvements from a soil conser-

vation strategy. In contrast, the CE approach allowed us

to assign economic values to different attributes of

strategy that policy-makers can fine-tune.Water quality

and landscape benefits were valued highly relative to

wildlife and employment benefits. However, a status

quo bias was detected that may suggest that the choice

task proved difficult for some respondents.

Individual WTP for the reduction of the external

effects of soil erosion extend over the 11–53 o inter-

val, depending on the magnitude of the environmental

and social improvement achieved with project imple-

mentation, and the area of project execution. By aggre-

gating this value to the population we estimate that the

social value of reducing the off-farm effects of soil

erosion in the catchment lie in the 3.1–15.5 million ointerval, i.e. about 95–160 o per hectare. The per-

hectare welfare estimates obtained in this study span

the current subsidy that the Andalusia’s Government

gives to farmers that adopt soil conservation measures

(132.22 o ha� 1 year� 1). Thus, public administration

is currently paying a subsidy to farmers that is in

consonance with what society is willing to pay.

Another way of thinking about this is to say that

payments aboveo160 per hectare might lead to nega-

tive net social benefits—although recall that we have

not included benefits to visitors and non-residents in

our study. However, it is not clear that even very high

payments would lead benoughQ farmers to adopt soil

conservation strategies to achieve the kinds of benefits

included in Scenario 3. This is not only because the

opportunity costs per hectare might exceed this value

in some parts of the region, but also because the lack of

up-take of soil conservation adoption measures is not

only due to financial issues13 but also to social, cul-

tural and institutional matters. Further research is

13 The average costs of implementing and maintaining the soi

erosion reduction systems hypothesised in this paper are actually

not that much greater than the costs of traditional tillage system

l

.

S. Colombo et al. / Ecological Economics 58 (2006) 850–861 861

needed in these fields to fully understand the reasons

that bring farmers to the decisions on adopting soil

conservation measures.

Acknowledgements

The authors would like to thank the Spanish Minis-

terio de Agricultura, Pesca y Alimentacion (Project

INIA RTA01-128), and the Consejerıa de Agricultura

y Pesca of the Andalucian Government (Project, PIA

8.01-01) for financial support. We are grateful to Pro-

fessor Jordan Louviere for his assistance in the devel-

opment of the experimental design used in this study.

References

Adamowicz, W., Boxall, P., Williams, M., Louviere, J., 1998. Stated

preference approaches for measuring passive use values. Choice

experiments and contingent valuation. Am. J. Agri. Econ. 80,

64–75.

Almansa, M.C., Calatrava-Requena, J., 2002. Valuing erosion con-

trol actions in south-eastern Spain: contingent valuation and

local awareness analysis. 7th Biennial Conference of the Inter-

national Society for Ecological Economics, Tunes.

Bateman, I., Carson, R., Day, B., Hanemann, M., Hanley, N., Hett,

T., Jones-Lee, M., Loomes, G., Mourato, S., Ozdemiroglu, E.,

Pearce, D., Sugden, R., Swanson, J., 2002. Economic Valuation

with Stated Preference Techniques, A Manual. Edward Elgar.

Cheltenhan, UK/Northampton, Ma, USA.

Boxall, P.C., Adamowicz, W., Williams, M., Swait, J., Louviere, J.,

1996. A comparison of stated preference methods for environ-

mental valuation. Ecol. Econ. 18, 243–253.

Christie, M., Azevedo, C., 2002. Testing the consistency in benefit

estimates across contingent valuation and choice experiments: a

multiple policy option application. Paper presented at the 2nd

World Congress of Environmental and Resource Economists,

Monterrey, California, June.

Clark, E.H., Haverkamp, J.A., Chapman, W., 1985. Eroding Soils:

The off-farm Impacts. The Conservation Foundation, Washing-

ton, DC.

Colombo, S., Calatrava-Requena, J., 2002. Towards the economic

evaluation of the intangible effects of soil erosion. Paper pre-

sented to the Xth congress of the European Association of

Agricultural Economists, EAAE, Zaragoza, Spain.

Colombo, S., Calatrava-Requena, J., Hanley, N., 2003. The eco-

nomic benefits of soil erosion control: an application of the

contingent valuation method in the Alto Genil basin of southern

Spain. J. Soil Water Conserv. 58, 367–371.

Feather, P., Hellerstein, D., Hansen L., 1999. Economic Valuation of

Environmental Benefits and the Targeting of Conservation Pro-

gram: The case of the CRP. AER-778. United States Dept. of

Agr., Economic Research Service.

Hanley, N., Spash, C., 1994. Cost-Benefit Analysis and the Envir-

onment. Edwarg Elgar.

Hanley, N., Wright, R., Adamowicz, W., 1998. Using choice experi-

ment to value the environment. Environ. Resour. Econ. 11 (3–4),

413–428.

Hansen, L.T., Breneman, V.E., Davison, C., Dicken, W., 2002. The

cost of soil erosion to downstream navigation. J. Soil Water

Conserv. 57 (4), 205–212.

Hausman, J.A., McFadden, D., 1984. Specification tests for the

multinomial logit model. Econometrica 52, 1219–1240.

ICONA, 1982. bPaisajes erosivos en el Sureste espanol: ensayo de

metodologıas para el estudio de su cualificacion y

cuantificacionQ. Ministerio de Agricultura Pesca y Alimentacion,

Monografıa, Madrid.

Lancaster, K., 1966. A new approach to consumer theory. J. Polit.

Econ. 74, 132–157.

Lehtonen, E., Kuuluvainen, J., Pouta, E., Rekola, M., Li Chuan-

Zhong, 2003. Non-market benefits of forest conservation in

southern Finland. Environ. Sci. Policy 6, 195–204.

Lopez, A., 1992. La comarcas olivareras andaluzas. Publicacion del

Instituto Nacional de Investigacion Agraria.

Louviere, J.J., 1988. Analysing Decision Making: Metric Conjoint

Analysis, No 67, Quantitative Applications in the Social Science

Series. Sage Publications, Newburg Park, CA.

Louviere, J.J., Hensher, D.A., Swait, J., 2000. Stated Choice Meth-

ods, Analysis and Application. Cambridge University Press.

MAPA, 2000. Hechos y cifras del sector alimentario espanol, 2000.

V Edicion revisada y actualizada, Secretaria Tecnica, available

at http://www.mapya.es.

McFadden, D., 1973. Conditional logit analysis of qualitative

choice behaviour. In: Zarembka, P. (Ed.), Frontiers in econo-

metrics. Academic Press, New York, pp. 105–142.

Mitchell, R.C., Carson, R.T., 1989. Using Survey to Value Public

Goods: The Contingent Valuation Method. Resources for The

Future, Washington, DC.

Pastor, M., Castro, J., Saavedra, M., Humanes, M., Pajaron,M.,

Civantos, M., Alvarado, M., Caballero J., 1999. Cultivo del

olivar en zonas de especial proteccion ambiental. Informacion

tecnica, 65/99, Consejerıa de Agricultura y Pesca, Junta de

Andalucıa.

Poe, G., Welsh, M., Champ, P., 1997. Measuring the difference in

mean willingness to pay when dichotomous choice contingent

valuation responses are not independent. J. Environ. Econ.

Manage. 73, 255–267.

Train, K., 1998. Recreation demand models with taste differences

over people. Land Econ. 74 (2), 230–240.