lable at ScienceDirect

Journal of Environmental Management 91 (2010) 1963e1971

Contents lists avai

Journal of Environmental Management

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

Valuing algal bloom in the Black Sea Coast of Bulgaria: A choiceexperiments approach

Tim Taylor a,*, Alberto Longo b

aDepartment of Economics, University of Bath, Claverton Down, Bath, UKbGibson Institute for Land, Food and the Environment, School of Biological Sciences, Queen’s University Belfast, Belfast, BT9 7BL, UK

a r t i c l e i n f o

Article history:Received 17 July 2009Received in revised form26 March 2010Accepted 20 April 2010Available online 2 June 2010

Keywords:Algal bloomValuationChoice experimentsBulgaria

* Corresponding author.E-mail addresses: ecstjt@bath.ac.uk (T. Taylor), a.lo

1 Previous studies (e.g. ECOHARM) have examinedtourists from the presence of algal blooms (see Stolteasked how much tourists were willing to pay to impsite under hypothetical scenarios, in the event thatback to the same site in the future. We deem such aweak, due to the difficulties in designing incentive-costated preferences questionnaires targeting tourists.

0301-4797/$ e see front matter � 2010 Elsevier Ltd.doi:10.1016/j.jenvman.2010.04.007

a b s t r a c t

Increased interest in water quality in coastal and marine areas stemming from the Water FrameworkDirective and the Marine Strategy Framework Directive has led to important questions in relation topolicies that address nutrient loadings. This paper presents the results from a choice experiment study toassess the recreational damage associated with algal blooms caused by nutrients flows into Varna Bay,Bulgaria. Varna Bay is an important beach destination on the Black Sea coast of Bulgaria. Algal bloomevents have been experienced frequently in the area. A choice experiment questionnaire was developedand applied in the Varna Bay area to assess the extent to which the quantity of algal blooms and theduration of the bloom affect recreational activities. The amount of bloom was found to be important, asrespondents were on average willing to pay a one off tax of 18.97 Leva (V9.73) for a program thatprovides beaches free from algal blooms.

� 2010 Elsevier Ltd. All rights reserved.

1. Introduction

Nutrient loadings in marine ecosystems leading to algal bloomsmay cause detrimental impacts on fisheries, biodiversity and waterclarity. Algal blooms, caused by nutrients loading fromwastewatertreatment, agriculture and industry, have been reported along theBlack Sea coast at Varna, one of the major coastal resorts ofBulgaria, since the 1980s (Moncheva, 2006). Though the prevalentspecies of algae are non-toxic, the reduced visibility in the coastalwaters does affect the area by spoiling the beauty of the landscapeand impacting on the enjoyment of recreational activities.

In this paper we assess the monetary values that local residentsattach to changes in the visibility in coastal waters affected by algalblooms in Varna Bay.1 We only focus on the effects that algalblooms have on recreational values. Our study does not capture allthe impacts related to eutrophication. Other forms of use value e

e.g. commercial fishing, tourism activities e and non-use value arenot included.

ngo@qub.ac.uk (A. Longo).the welfare loss suffered byl et al., 2003). These studiesrove the water quality at thethe respondent would comen approach to be particularlympatible payment vehicles in

All rights reserved.

The paper is structured as follows. First, we present an overviewof Varna Bay, along with a description of the major issues caused byalgal blooms in the area. We then look at previous studies that havevalued the external costs resulting from algal blooms. The choiceexperiments methodology used to value the external costs of algalblooms in Varna Bay, and the associated questionnaire and itsimplementation, are then presented. The paper concludes witha discussion of the results and their implications for the imple-mentation of the Water Framework Directive.

2. Varna Bay e overview

Varna Bay is an important coastal resort on the Black Sea coast ofBulgaria. It is located south of the Danube delta. The largest touristresort on the Bulgarian coast, it faces significant environmentalpressures (Ketsetzopoulou and Moncheva, undated). The Varna Bayarea has been the subject of a number of algal bloom events in recenthistory. In the period 1983e2002 there were 117 incidents of bloom,with 7 incidents of bloom in 2001. A number of species have bloomedin recent history, including Skeletonema costatum, Cerataulina pelag-ica, Prorocentrum minimum and Gymnodinium sp (Moncheva, 2006).Thewatershed, including a river flowing into the bay and a small lakehosting a wastewater treatment plant, is shown in Fig. 1.

The area of Varna Bay is not much affected by nutrient inputsfrom the Danube delta, due to prevailing currents. The majorsources of nutrient inputs are from the watershed around Varna

Fig. 1. Varna Bay and Catchment. Source: Moncheva (2005).

T. Taylor, A. Longo / Journal of Environmental Management 91 (2010) 1963e19711964

itself. Fig. 2 gives an indication of the frequency of algal bloomevents in Varna. The frequency of blooms depends on the quantityof nutrient loads from farming, industrial activities, tourism activ-ities and waste water, and are more frequent in hot and drysummers. Algal blooms affect a number of activities, includingrecreational activity e which is the focus of this paper.

The link between nutrients and algal bloom density, and conse-quently sight depth in water, is complex. Petrova and Velikov (2005)examined the impact of the biochemical status of the water andEmiliania huxleyi blooms in the case of the Bulgarian coast of the BlackSea, finding that depleted biogenic elements make favourable condi-tions for the development of E. huxleyi. Some studies have showna relationship between nutrient concentrations and algae (forexample, James and Havens,1996; Krause-Jensen et al., 2007; Pertolaet al., 2004). Others have examined the influence of algal density orchlorophyll-a concentrationsonsightdepth (McQuatters-Gollopet al.,2007; Megard and Berman, 1989). The difficulties in transferringresults of algal bloom influences on secchi depth, or nutrients onsecchi depth directly, from the existing literature to the Varna case aremanifold. Differences in the coastal morphology, light conditions,climatic conditions, coastal water currents and the like will all affectthe levels of algal bloom or visibility attributable to an increase innutrients. In addition, the species mix of algae may change withdifferent levels of nutrients orwith the Redfield ratio. Despite the lackof scientific models that accurately predict algal blooms, stated pref-erences methods, like the one employed in this paper, are useful toassess how much society values the negative effects on recreationalactivities caused by algal blooms. Such estimates can be used bygovernments together with models that take into account the uncer-tainties in the relationships between nutrients and algal blooms.2

3. Previous valuation studies on eutrophication and algalbloom in marine environments

A number of studies have used the contingent valuation method(Mitchell and Carson, 1989) to value the environmental damage

2 Estimates from stated preferences methods are based upon hypotheticalscenarios that respondents deem credible. If policy makers want to compare thecosts and benefits of policies aimed at reducing the amount of nutrients flowing incoastal waters, monetary estimates from stated preference methods should then beweighted by the probability that nutrients caused by anthropogenic activities leadto algal blooms events. We did not want to bring uncertainty into the choiceexperiments exercises, as it would have made the exercises more complex for ourrespondents and increased the variance of the estimates.

caused by eutrophication. A major study on the Baltic Sea usedsimilar contingent valuation questionnaires in Sweden (Soderqvist,1995) and Poland (see Markowska and Zylicz, 1999). In both cases,the questionnaire included: questions on how people use the BalticSea, information on eutrophication and its effects, questions onpeoples’ knowledge of these effects, a description of the valuationscenario and Willingness To Pay (WTP) questions. A typicalcontingent valuation dichotomous choice format was employed, inwhich respondents were confronted with the following question:“If there were a referendum in Sweden (Poland) about whether tolaunch the action plan or not, would you vote for or against theaction plan, if the resulting environmental tax you pay amounted toSEK [X] per year for a period of 20 years?” Seven different amountsfor X were randomly used in the question. The surveys measuredchanges in Total Economic Value e i.e. direct use plus indirect useplus non-use values (Soderqvist, 1995). Both surveys were designedas mail questionnaires. The Swedish survey was sent to 600randomly selected adults, of which about 60 percent responded.Likewise, the Polish survey was sent to 600 randomly selectedadults yielding a 50 percent response rate. Themean annualWTP ofSwedish respondents was about V726.90 per adult (or V406.57 ifnon-respondents to the survey are assumed to have a zero WTP).Gren et al. (1995) suggest rounding the figures down to V677.61and V369.61 per adult to reflect the fact that an open-ended WTPsurvey, undertaken as part of the same overall study, resulted inmuch lower estimates of mean WTP.3 The mean annual WTP perPolish adult was V102.92 (or V52.69 if non-respondents to thesurvey are assumed to have a zero WTP).

Le Goffe (1995) investigated the costs of eutrophication, usinga contingent valuation study to measure the benefits of reducingeutrophication in coastal waters near Brest, France. The survey,which was administered to over 600 Brest residents, elicitedinformation for two goods: local pollution problems frommicrobes(including bathing and shellfish impacts) and local pollution fromhigh concentrations of nutrients in the harbour and its conse-quences for the marine ecosystem. The respondents were asked fortheir maximumWTP for “salubrity”, defined as being able, withoutrisk, to bathe and to consumewild shellfish in the harbour bay area.Respondents were also asked for their maximum annual WTP for“ecosystem”, defined as preventing the asphyxiation of the harbourwaters fromhigh concentrations of nutrients. The averageWTPwas

3 Unconverted figures for the suggested rounding down are from SEK 5900 toSEK 5500 and from SEK 3300 to SEK 3000 respectively.

Fig. 2. Bloom events in Varna (1982e2002). Source: Based on Moncheva (2006).

5 Choice experiments and other stated preference (SP) techniques have recentlyemerged as a complement to revealed-preference (RP) techniques. While RPevaluate economic agents’ behaviours in real markets, SP involve choice responsesevoked in hypothetical markets. The interest in hypothetical behaviour ineconomics arises for different reasons, such as the necessity to investigateeconomic agents’ preferences for new policies that might be implemented, for thedevelopment of a new product or good, or for evaluating goods that are not tradedin real economic markets. All these examples make it clear that it is not possible toestimate agents’ preferences using RP.Choice experiments analysis allows a greatdeal of flexibility because researchers can explore how a change in the hypotheticalscenario influences people’s responses, and compare the current scenario withmany hypothetical alternatives. This is particular helpful for informing policydecisions before the policy itself has been decided upon.Usually RP data fromregular marketplaces (such as the labour and the housing market) contain infor-mation about actual market equilibria for the behaviour of interest, and can be used

T. Taylor, A. Longo / Journal of Environmental Management 91 (2010) 1963e1971 1965

V38.90 and V28.95 per household per year, respectively, for thegoods “salubrity” and “ecosystem”. The WTP figure for “salubrity”represents 10 percent of the annual water bill of residents (whichwas the payment vehicle used in the survey). No information wasgiven in the study on the physical benefit (e.g. improvement inwater clarity) which was valued. Again, the welfare gains cannot berelated to a specific change in nutrient loading.

Stoltel et al. (2003) investigated how Harmful Algal Blooms(HABs) impact on tourism in Europe. The authors conducteda contingent valuation exercise in 4 countries: Riccione (Italy),Galway (Ireland), Hanko (Finland), and Hyéres, Les Pradet andCarquieranne (France). They sampled 780 individuals over all thelocations during the months of June, July and August. In the fourcountries, they investigated tourist experience with problemscaused by algal blooms and their willingness to pay to mitigatethese problems. Using a one and half-bounded dichotomous choiceformat,4 they asked respondents if they would be willing to paya tax to finance investments that would prevent algal blooms fromforming. When analysing each location separately, the authorsfound the HAB impact on tourism to be between 0.92 and 4.9million Euro per year in Riccione (Italy), between 9.0 and 16.4million Euro per year in Galway (Ireland), between 85 and 539thousand Euro per year in Hanko (Finland), and between 4 thou-sand and 442 thousand Euro per year in Le Pradet, Hyéres andCarquieranne (France). Surprisingly, the authors found that WTPwas not affected by respondents’ previous experience and contactwith algal blooms. Such relationship was only found when the datafrom the 4 case studies were pulled together in a single model. Thisis perhaps compounded by the low sample size relative to thenumber of locations studied.

Sandstrom (1996) used a multi-site travel cost model andapplied a random utility model (RUM) of Swedish seaside recrea-tion to estimate the benefits (damages avoided) from reducedeutrophication of seas around Sweden. Sight depth data fromaround the Swedish coast was used as an indicator of water qualityrelated to eutrophication, as the author suggested that sight depthmay be directly related to the recreationist’s perception of waterquality, in addition to being highly correlatedwith nutrient load; anincrease in the content of nutrients reduces the transparency ofwater. To link the sight depth variable with nutrient loading,a relationship was estimated between sight depth and

4 Respondents are first asked if they would pay X. If they answered ‘yes’ (‘no’)they are then asked if they would pay 50% more (less), in a second contingentvaluation question.

concentrations of total phosphorus and total nitrogen. Sandstromfound that a 1 percent reduction in nitrogen content improvedsight depth by 0.63 percent; a 1 percent reduction in phosphoruscontent increased sight depth by 0.18 percent. It is then possible tomeasure the benefits of policies to reduce total nitrogen and totalphosphorus, in terms of increases in consumer surplus. The bene-fits of a 50 percent reduction in nutrient loading along the entireSwedish coastline was estimated to range between V29.6 millionand V66.5 million, depending on the model specification used(between V17.25 and V38.81 per trip).

To our knowledge, choice experiments have not been previouslyapplied to the case of algal blooms affecting recreational activities incoastal waters. They have been applied in riverine systems (e.g.Hanley et al., 2006), in coastal areas for tourist preferences fordifferent beaches (Brau and Cao, 2006), and in identifying prefer-ences for improved coastal water quality (Eggert and Olsson, 2003).In the next sectionwepresent an overviewof the choice experimentmethodology, before applying it to the algal bloom case in Bulgaria.

4. Application of conjoint choice experiments

A useful tool to assess the monetary valuation of environmentalgood is provided by the conjoint choice experiments technique.5

Such analyses are also known in the literature as conjoint choiceanalysis or choice experiments. In a typical conjoint choice survey,respondents are shown various alternative representations ofa good, which are described by a set of attributes, and are asked tochoose the most preferred (Hanley et al., 2001). The alternativesdiffer from one another in the levels taken by two or more of theattributes.6 This approach has the advantage of simulating realmarket situations, where consumers face two or more goodscharacterized by similar attributes, but different levels of theseattributes, and are asked to choose whether to buy one of the goodsor none of them. In a single choice experiment exercise researcherslearn only which alternative is the most preferred, but the result ofthe exercise does not tell us anything about the preferences for theoptions that have not been chosen. A single choice experimentexercise does not offer a complete preference ordering. Therefore, ifresearchers want to know a complete ordering of preferences it isnecessary either to ask a respondent to domany choice exercises, orto survey more respondents varying the levels of the attributes.

To motivate the statistical analysis of the responses to conjointchoice experiment questions, it is assumed that the choice betweenthe alternatives is driven by the respondent’s underlying utility. Therespondent’s indirect utility is broken down into two components.

to infer short-term departures from the current equilibria. In contrast, SP data fromresponses to choice experiment questions are especially rich in attribute trade-offinformation. Therefore, SP data are useful in estimating future changes in agents’behaviour (Louviere et al., 2000).

6 For example, a common attribute of a policy is the price, and this may takedifferent levels such as V10, V20 and so forth.

7 An example of a situation where the IIA would not be plausible is the blue bus/red bus example given by McFadden (1974). Consider commuters initially choosingbetween two modes of transportation, car and red bus. Suppose that a consumerchooses between the car and bus with equal probability, 0.5. Now suppose a thirdmode, blue bus, is added. Assuming bus commuters do not care about the colour ofthe bus, consumers will choose between these with equal probability. But then IIAimplies that the probability of each mode is 1/3; therefore, the fraction ofcommuters taking a car would fall from 1/2 to 1/3, a result that is not very realistic.While this example is admittedly extreme (in practice, one would group the bluebus and red bus into the same category), it indicates that the IIA property canimpose unwanted restrictions in the conditional logit model (Wooldridge, 2002).

8 The multinomial probit model requires burdensome computation compared tothe family of multinomial choice models derived from the Gumbel distributedutility function, since it involves multi-dimensional integration in the estimationprocess. Despite more powerful computers becoming available, applications of themultinomial probit model are still scarce, as convergence to satisfactory results inapplications with more than three choices appears to require several restrictions onthe standard deviations and correlations (see Greene, 2008).

9 A copy of the questionnaire is available from the authors on request.

T. Taylor, A. Longo / Journal of Environmental Management 91 (2010) 1963e19711966

The first component is deterministic, and is a function of theattributes of alternatives, characteristics of the individuals, anda set of unknown parameters, while the second component is anerror term. Formally (see Alberini et al., 2007),

Vij ¼ V�xij;b

�þ eij (1)

where the subscript i denotes the respondent, the subscript jdenotes the alternative, x is the vector of attributes that vary acrossalternatives (or across alternatives and individuals), and e is an errorterm that captures individual- and alternative-specific factors thatinfluence utility, but are not observable to the researcher. Equation(1) describes the random utility model (RUM).

Respondents are shown two or more representations of thegood. The alternative they select is the one that gives them thehighest utility. Because the observed outcome of each choice task isthe selection of one out of K alternatives, the appropriate econo-metric model is a discrete choice model expressing the probabilitythat alternative k is chosen. Formally, the probability that respon-dent i chooses alternative k is given by:

Prðk is chosenÞ ¼ PrðVk > V1;Vk > V2;.;Vk > VKÞ¼ Pr

�Vk > Vj

�cjsk; (2)

If the error terms e in Equation (1) are independent and identicallydistributed and follow a standard type I extreme value distribution,it can be shown (Train, 2003) that the probability that respondent ipicks alternative k out of K alternatives is:

PrðkÞ ¼ expðwikbÞPKj¼1 exp

�wijb

� (3)

Where wij is the vector of all attributes of alternative j. Equation (3)is the contribution to the likelihood in a conditional logit model.The full log likelihood function of the conditional logit model is:

log L ¼Xni¼1

XKk¼1

yik$log Prði chooses kÞ; (4)

where yik is a binary indicator that takes on a value of 1 if therespondent selects alternative k, and 0 otherwise, and Pr(i choosesk) is equal to Pr(k) in Equation (3).

For large samples and assuming that the model is correctlyspecified, the maximum likelihood estimates bb are normallydistributed around the true vector of parameters b, and theasymptotic variance-covariance matrix, U, is the inverse of theFisher information matrix.

Once model Equation (4) is estimated, the rate of trade-offbetween any two attributes is the ratio of their respective b coef-ficients. The marginal value of attribute l is computed as thenegative of the coefficient on that attribute, divided by the coeffi-cient on the price or cost variable:

MPl ¼ �bblbb2

: (5)

The willingness to pay for a commodity is computed as:

WTPi ¼ �xibbbb2

; (6)

where x is the vector of attributes describing the commodityassigned to individual i. It should be kept in mind that a properWTP can only be computed if the choice sets faced by the indi-viduals contain the “status quo” e the current situation, in whichno commodity is acquired, and the cost is zero.

The conditional logit model described by Equations (1)e(6) iseasily amended to allow for heterogeneity among the respondents.

Specifically, one can form interaction terms between individualcharacteristics, such as age, gender, education, etc., and all or some ofthe attributes or of the constant terms (the alternative-specificconstants), andenter these interactions in the indirectutility function.

Whether or not interaction terms are included, implicit in theconditional logit model is the assumption of Independence ofIrrelevant Alternatives (IIA), which states that the ratio of the oddsof choosing any two alternatives depends only on the attributes ofthe alternatives being compared, and is not affected by the attri-butes of other alternatives. Formally,

PrðkÞPrðhÞ ¼ expðwikbÞ=

Pj exp

�wijb

�expðwihbÞ=

Pj exp

�wijb

� ¼ expðwikbÞexpðwihbÞ

: (7)

An implication of the IIA is that, as shown in Equation (7), addinganother alternative, or changing the characteristics of a third alter-native, does not affect the relative odds between alternative k and h.IIA generally imposes restrictive substitution patterns among thealternatives. A change in the attributes of one alternative, therefore,changes theprobabilities of theother alternativesproportionately tosatisfy the conditional logit’s requirement that the ratio of theseprobabilities remains the same (Train, 2003). This implies that theconditional logit is not well suited for alternatives that individualsperceive as close substitutes of one another.7 Researchers are thusadvised to test for violations of this assumption using the appro-priateHausman test, and to considermodels that relax it, such as thenested logit model, the multinomial probit model and randomparameters logit models, where the coefficients are assumed to becomposed by two parts, a fixed component common to all respon-dents and a respondent specific component. The latter is generallyassumed to be distributed following the normal, lognormal, ortriangular distribution (see Train, 2003).8

5. Design of the questionnaire

A questionnaire was designed using the choice experimentsapproach for in-person interviews to elicit the preferences ofindividuals for a reduction in algal bloom in Varna Bay. Focusgroups and pretesting were used to test the questionnaire, and toidentify appropriate payment vehicles and levels for the tax.

The structure of the questionnaire was as follows.9 First, therespondents were asked about their use of beaches, includinggeneral length of stay at the beach, what they did at the beach, whotraveled with them and the importance they placed on variousattributes in choosing which of the beach resorts to visit (includingenvironmental attributes). The respondents were then informedabout the algal bloom issue in theVarna Bayarea and askedwhether

Fig. 3. Example Choice Set for Algal Bloom Prevention in Varna.

10 This representation was the result of lengthy discussions among the studyteam, and focus groups. The colour of the bloom used in the pictures was decidedupon in consultation with scientists in Varna.

T. Taylor, A. Longo / Journal of Environmental Management 91 (2010) 1963e1971 1967

theyhadexperiencedabloomepisode andwhether thebloomcouldaffect their activities at the beach. The questionnaire then turned toissues relating to policy options to reduce algal bloom. A descriptionof the impacts of algal blooms was provided ewith clarity of waterbeing the major impact in Varna. The fact that algal bloom in theVarna case is not harmful to health was explicitly mentioned toreduce bias due to perception of a possible health risk. A policyscenario was presented, with the payment vehicle being a one offtax. The wording of the payment vehicle was as follows:

“The government is currently assessing ways of reducing thealgal blooms in Varna Bay. Possible options include, for example,projects for the improvement of wastewater treatment in theregion of Varna.All the projects under consideration aim to reduce the intensityand the duration of the algal blooms. They have proven to beeffective at other beaches, and are currently under considerationin Spain and Belgium as well.However, these projects cost money. The current budget of thegovernment could cover part of the costs, such as the mainte-nance of the cleanup measures. However, given the limitedbudget, the remaining part of the costs to allow the program tostart will be covered with a one-time tax that each household inVarna will have to pay next year (and next year only).The amount of the one time tax varies from one project to thenext. The projects’ efficiency in reducing the quantity and theduration of the algal blooms also differ. And we have reasons tosuspect that some of the projects may influence the number ofpeople who come to the beach.So in the next few questions, we will describe the expectedimpacts of some of the projects based on these characteristics.And we will ask you to tell us which one you would prefer,between two potential projects and the current situation.”

A series of four choice sets was then presented to eachrespondent. In total we used 28 choice sets, with differentrespondents being allocated to one of seven blocks of four choicequestions. An example is given below in Fig. 3. The attributes usedin the choice experiments were:

(i) visibility, expressed as the depth in metres to which it wouldbe possible to see into the water during a bloom event, itslevels being: 0.5 m low visibility, 1 mmedium visibility (statusquo), 1.5 m high visibility;

(ii) the duration of the bloom in weeks, its levels being 1 (statusquo), 4, 6 weeks;

(iii) congestion, described as distance from the closest person atthe beach, its levels being within 3 m e high congestion,between 3 and 30 m e medium congestion (status quo), andbeyond 30 m e low congestion;

(iv) the tax amount, its levels being 0 for the “status quo”, and 12,25, 35, 50, 75, and 100 Leva for the other alternatives.

In each choice set, respondents could choose between twohypothetical projects and a “noproject” scenario, where no change tothecurrent situationwasenvisagedand therewasnoadditional tax topay. In addition toverbal descriptions, for visibilityand congestionweused visual representations. For visibility, a pictorial representationwas used with different water-colour shadings to show the depth ofthe visibility in the water.10 For congestion, a photo of a beach wasmanipulated to show a given density of people at a beach.

After the choiceexperimentsquestionsweaskeda setofdebriefingquestions to assess whether respondents believed the hypotheticalscenarios and whether they had considered all the attributesdescribing the choice sets. The questionnaire then concluded by col-lecting the usual socio-economic characteristics of the respondents.

6. Results

6.1. Descriptive statistics

Wefirst explore the characteristics of the sample of the Bulgarianresidents interviewed. Tables 1 and2 report thedescriptive statisticsof our sample. It was quite balanced in terms of gender, welleducated and had an average netmonthly household income of 605Leva. This income was above the national average annual grossincome for Bulgaria, which was 5863 Leva per household in 2006(National Statistical Institute, 2010a). Based on the 71 percent ofpeoplewho knew the amount of their water bill, the average annualbill was 128 Leva. About 81 percent of the respondents had a job, 8percentwere retired, about 9 percentwere students, 2 percentwerehomemakers, and only about 3 percent were unemployed. In 2006,

Table 1Descriptive statistics: dummy variables.

Variables N Mean Std Dev

Activities at the beachGoes swimming 850 0.77 0.42Any type of walking 850 0.36 0.48Sunbathing 850 0.85 0.36Does Thalasso therapy 850 0.23 0.42

People at the beach withGoes with family 850 0.49 0.50Goes with friends 850 0.66 0.47Goes with partner 850 0.16 0.36Goes on his/her own 850 0.17 0.37

Ranking of attributesCrowd is very important 826 0.41 0.49Access is very important 836 0.41 0.49Distance to the beach is not important 830 0.62 0.49Water quality is very important 835 0.68 0.47Cleanliness at the beach is very important 836 0.70 0.46Facilities at the beach are very important 830 0.40 0.49Family tradition is not important

when deciding a beach822 0.65 0.48

Experience and perception of algal bloom eventsHas been at the beach with algal blooms 850 0.87 0.34Algae affect recreation experience 849 0.40 0.49

Scenario understandingRepeated scenario description 850 0.06 0.24Knows the household’s water bill 850 0.71 0.45Considered all attributes

in the choice experiment850 0.67 0.47

More importance to visibilityin the choice experiment

850 0.04 0.20

More importance to durationin the choice experiment

850 0.03 0.18

More importance to congestionin the choice experiment

850 0.04 0.18

More importance to costin the choice experiment

850 0.22 0.42

Socio-economic variablesMale 850 0.48 0.50Full time worker 850 0.63 0.48Employed part time 850 0.09 0.29Self-employed 850 0.09 0.28Unemployed 850 0.03 0.17Retired 850 0.08 0.27Student 850 0.09 0.28Looking after the home full time 850 0.02 0.15University or PhD qualified 850 0.40 0.49

Table 2Descriptive statistics.

Variable N Mean Std Dev Min Max

Days at the beach in past 12 months 815 17.47 14.02 0 210Time at the beach (minutes) 819 147.18 120.80 30 420Travel time 834 24.13 12.19 0 150Water bill value (per year) 850 128.03 94.87 0 400Perceived difficulty of choice exercise

(1 ¼ very easy; 5 ¼ very difficult)842 1.92 0.90 1 5

Visibility rank 825 2.18 1.11 1 4Duration of algal bloom rank 826 2.54 0.99 1 4Number of people at the beach rank 836 2.76 1.04 1 4Amount of the tax rank 831 2.39 1.26 1 4Age 825 41.00 12.54 16 87Number of children 850 0.88 0.86 0 4Children living at home 850 0.58 0.81 0 4Net monthly household income

(upper income was set at 1000)839 605.15 252.60 100 1000

Annoyance level 850 1.85 0.89 0 5

T. Taylor, A. Longo / Journal of Environmental Management 91 (2010) 1963e19711968

the unemployment rate in urban areas was 7.5 percent (NationalStatistical Institute, 2010b), suggesting that our survey had a largerportion of economically active people than the national average. Theaveragenumber of childrenwithin a household in our survey is 0.58,just above the national average (OECD, 2008).

Turning to the attitudes and behaviour of our respondents at thebeach, Tables 1 and 2 show that people spent an average of 2.5weeks at the beach every year and on the average trip they stayedabout 2.5 h. Respondents average travel time to the beach was quiteshort, only about 24 min, as almost all respondents (98%) lived inVarna. Most respondents traveled by car (68%), but also bus andwalking were used by about one third of respondents respec-tively.11 Among the favourite activities at the beach, our respon-dents liked sunbathing (85%), swimming (77%), walking along thebeach (36%), and doing thalasso therapy (23%). Most peoplewent tothe beach with friends (66%) and family (49%), and only about17percent and 16 percent respectively went on their own or witha partner. Among the factors affecting the decision to go to thebeach, the water quality and the cleanliness of the beach wereimportant elements for about 70 percent of the respondents, whilecrowding, quality of the services, access and family tradition playedminor roles in respondents’ choices. It is remarkable that almost 90percent of the respondents had seen algal blooms at the beach, buttheir presence had affected the recreational experience of only 40percent of the respondents.

6.2. Choice experiments questions

Prior to analysing our choice experiments questions we checkedthe quality of the answers. Before asking the choice questions, onlyabout 6 percent of the respondents asked the enumerator to repeatthe description of the program, suggesting that our questionnairewas quite clear. In addition, after the choice questions, the ques-tionnaire presented a series of debriefing questions to identifyrespondents who failed to understand the choice exercise,answered strategically, or found the questions too difficult.

Most respondents found the choice exercise quite easy: on a 1e5Likert scale question about the difficulty of the choice exercise,where 1 means very easy and 5 very difficult, the average answerwas 1.92. About 70 percent of the respondents gave equal weight toeach attribute, while about 20 percent gavemore importance to thecost attribute and few respondents gave more importance to theother attributes.12 We further asked respondents to rank theimportance of the attributes in the choice questions, and the resultsshow that they gave a similar importance to the presence and theduration of the algal bloom, followed by the cost, whilst thecrowding at the beach was the least important attribute. About 23percent of the respondents always chose the status quo, and,among these, the reasons to vote against the proposed policieswere stated as the already high expenses the household faced for47 percent of them, the already high taxes for 29 percent of them,doubt in the usefulness of the program for 28 percent of them, thenegligible nuisance caused by the algae for 27 percent of them, andother reasons for 6 percent of them.13 Finally, at the end of thequestionnaire, the enumerators reported on the quality of theinterview. According to the interviewers, about 97 percent of therespondents understood the choice exercise, only about 5 percent

11 The total does not sum to 100 because some respondents used more than onetravel mode. Only 5 respondents used a moped, and 24 a bicycle.12 We deem this latter result as encouraging, as many observers have suggestedthat in choice experiments respondents overlook the price attribute. Indeed ourresults do show that price is one of the most important attributes.13 The total does not sum to 100 because respondents could volunteer more thanone answer.

Table 3Choice experiments models results: conditional logit models (CLM) and random parameters logit model (RPLM).

Variable Model 1 (CLM) Model 2 (CLM) Model 3 (CLM) Model 4 (RPLM)a

Coeff. t-ratio Coeff. t-ratio Coeff. t-ratio Coeff. t-ratio

Cost �0.0192 �16.78 �0.0187 �16.12 �0.0201 �16.28 �0.0188 �15.92Duration �0.0758 �5.26 �0.0774 �5.08 �0.0869 �5.42 �0.0773 �5.13Low congestion 0.0870 2.00 0.0699 1.58 0.0922 1.98 0.7564 1.69Medium congestion 0.2025 5.23 0.1978 5.09 0.2122 5.21 0.1971 4.90High visibility 0.5042 12.75 0.3464 3.98 0.3278 3.60 0.3582 4.04Medium visibility 0.0282 0.68 0.1360 1.56 0.1421 1.54 0.1232 1.47(High visibility)*(duration) 0.0417 1.97 0.0466 2.10 0.0392 1.80(Medium visibility)*(duration) �0.0279 �1.25 �0.0256 �1.08 �0.251 �1.31ASC1 �0.4136 �4.58 �0.4647 �3.70 �1.2031 �5.57 �0.4632 �3.69ASC2 �0.3310 �3.51 �0.4149 �2.89 �1.3895 �6.06 �0.4129 �2.86(Income)*(ASC1) 0.0015 7.66(Full time worker)*(ASC1) �0.2140 �2.19(Uni or phd)*(ASC1) 0.3598 3.59(Travel time)*(ASC1) �0.0117 �2.98(Dummy for swimming)*(ASC1) 0.2474 2.08(Days at the beach)*(ASC1) �0.0019 �0.51(Income)*(ASC2) 0.0018 8.98(Full time worker)*(ASC2) �0.1557 �1.60(Uni or phd)*(ASC2) 0.4786 4.83(Travel time)*(ASC2) �0.0179 �4.40(Dummy for swimming)*(ASC2) 0.2871 2.42(Days at the beach)*(ASC2) 0.0024 0.77Observations 3396 3396 3172Log-likelihood function �3283.51 �3281.27 �2948.55 �3281.24

a The random parameter logit model was run with random parameters for all variables, except for the ASCs and cost, using 500 Halton draws. The derived standarddeviations for the random parameters were assumed to follow a triangular distribution with spread constrained to be equal to the mean of the coefficient. We thereforeomitted from the table the estimates of the standard deviations as these are equal to the absolute value of the coefficients estimates.

T. Taylor, A. Longo / Journal of Environmental Management 91 (2010) 1963e1971 1969

were annoyed or very annoyed during the interview, and generallythe interviews had no interruptions caused by lack of under-standing, annoyance, or other reasons. This first analysis of the dataallows us to conclude that the questionnaire was well accepted byrespondents, they engaged in the choice questions, and did notshow significant protest behaviours. We therefore ran our econo-metric models with the full sample of observations.

In running the econometric models, we use effect code dummyvariables (see Louviere et al., 2000) for the levels of the attributes ofcongestion and visibility, and continuous variables for tax andduration. The models were estimated both in NLOGIT 4.0 and inBiogeme (Bierlaire, 2003). Table 3 reports the results of threeconditional logit model specifications. Model 1 is the basic modelwith the four attributes and the Alternative-Specific Constants(ASCs) for the two hypothetical scenarios, to explain the variationin utility not captured by the attributes. As we expected thatrespondents might have considered the combined effect of dura-tion of the bloom and visibility, in Model 2 we add interactionterms between the two visibility dummies and the duration of thealgal bloom. The third model further adds interaction terms withthe ASCs and some socio-economic characteristics of the respon-dents. Likelihood ratio tests show that Model 3 outperforms bothModels 1 and 2.14 The coefficients of the interaction terms suggestthat our model is internally consistent: people with higher income,that have a high education degree, and that enjoy swimming at thebeach are more likely to choose the hypothetical programs.Respondents that are employed full time and that live further away

14 When computing the LR tests to compare models 1e3, we needed to re-runModels 1 and 2 with the same number of observations as in Model 3. This was doneafter deleting the missing observations for the socio-economic variables used inModel 3. The LR tests show that Model 3 outperforms Model 1 (LR ¼ �2*(�3061.49�(�2948.55)) ¼ 225.888) and Model 2 (LR ¼ �2*(�3059.72�(�2948.55)) ¼ 222.33). When we ran our models after cleaning the dataset from abnormal observa-tions, such as ‘protest’ answers, and respondents who found the questionnairedifficult or appeared annoyed during the interview, the estimated coefficients donot change more than 10% compared to the full sample.

from the beach, and therefore have less opportunities to visit thebeach, are less likely to choose the hypothetical programs. Quitesurprisingly, the number of days spent at the beach does not seemto have a significant effect on the probability of choosing thehypothetical programs over the status quo.

Before proceeding with the interpretation of the output fromModels 1e3, we tested whether the estimates from the conditionallogit models of Table 3 were independent from irrelevant alterna-tives. We followed the test developed by Hausman and McFadden(1984) that suggested that if a subset of the choice set truly isirrelevant, omitting it from the model will not change parameterestimates systematically. The test followed a chi-squared distribu-tion with degrees of freedom equal to the number of coefficients tobe estimated in the restricted model:

c2 ¼�bbs � bbf

�0hbV s � bV f

i�1�bbs � bbf

�(8)

where s indicates the estimators based on the restricted subset (afterdeleting one ormore alternatives), f is the estimator based on the fullset of choices, and bV s and bV f are the respective estimates of theasymptotic covariance matrices. We carried out the test by firstdropping the “status quo” alternative from the model. As the “statusquo” was chosen 1668 times by our respondents, we needed to runthemodel without those observations. In addition, we needed to runthe model without the variables that do not change across alterna-tives. The result of the test with two coefficients in the model (dura-tion, visibility) reported a value of 369.06. We therefore rejected thehypothesis that the odds ratios for the other two alternatives areindependent from the current situation based on these results, as thestatistic exceeds the critical value of a chi-squared distribution withtwo degrees of freedom (Critical chi-squared[2] ¼ 5.99.15

15 We reached the same conclusions when we dropped the first alternative, A, orthe second alternative, B. The chi-squared statistic after dropping the first alter-native is 196.47, and after dropping the second alternative is 203.35.

Fig. 4. WTP for different levels of visibility (model 4).

17 It is important to note that a negative willingness to pay is not equivalent toa willingness to accept compensation estimate.18 At the time of the survey, 1 Euro was exchanged for 1.95 Leva.

T. Taylor, A. Longo / Journal of Environmental Management 91 (2010) 1963e19711970

Our analysis, therefore, proceeded with Model 4, a randomparameters logit model, that relaxed the IIA property. In a randomparameters logit model (see Revelt and Train,1996), the parametersare not assumed to be fixed across individuals:

bik ¼ bk þ uik (9)

In equation (9), the parameter b of attribute k for respondent i iscomposed by a fixed part, bk common to all individuals, and bya part that varies across individuals, uik, where u is a randomcomponent. In Model 4 we allowed all parameters, except those forcost and the two ASCs, to be random. Following Hensher et al.(2005), we specified triangular distributions for the randomparameters, with the spread restricted to be equal to the mean toguarantee the same sign for the parameter estimates across theentire distribution.

The sign of the estimates of the attributes coefficients suggeststhat people preferred a low level of congestion, compared toa medium, and a high level of congestion (the latter coefficient iscalculated as the negative of the sum of the coefficients for low andmedium congestion). In terms of algal blooms, the negative andstrongly significant sign of the coefficient of duration shows thatrespondents did not like a program that entails a long period ofalgal bloom. In terms of quantity of algae, respondents preferredprograms that offer high visibility levels, followed by medium andthen low visibility levels. However, to have a clear understanding ofthe effects of the attributes on the choices, it is necessary to look atthe interaction terms between the quantity of algae and the dura-tion of the bloom. Fig. 4, where we plot WTP for different levels ofvisibility and duration of bloom, helps us to analyse this point. Thegraph limits its investigation to 6 weeks, as WTP predictionsoutside of the range of the attributes levels used in the choiceexperiments questions would produce highly uncertain results(Hensher et al., 2005). Respondents were on average willing to pay18.97 Leva (s.e. 4.58)16 for not having algal blooms at all, a scenariowhere visibility is high and duration of the bloom is set to zero. Thefigure also shows that our respondents werewilling to paymore forprograms that entail a high visibility compared to programsoffering low or medium visibility levels, independent of thenumber of weeks of algal bloom. When we look at the number ofweeks, we can see that respondents’ WTP increased for programswith high visibility, while it decreased and became negative formedium and low levels of visibility. Respondents showed a WTPclose to zero for a scenario of medium foam lasting one week,which is consistent with the current situation. For low visibility

16 Standard errors are calculated with the Delta method.

levels, respondents always had a negative WTP,17 which is under-standable as such a scenario would be worse than the currentsituation. The graph also shows that when comparing algal bloomslasting for 6 weeks, the longest period of bloom considered in ouranalysis, our respondents still had a positive WTP of 6.86 Leva (s.e.5.40) for a program that offered high visibility levels, and hada negativeWTP for medium and low levels of algal bloom. This is aninteresting result, that shows that what really concerned residentsof Varna Bay is first of all the visibility of thewater, and secondly thelength of the bloom.

7. Conclusions

Eutrophication is an important issue in the area around VarnaBay, Bulgaria, with a number of occurrences of algal blooms inrecent times. This paper uses the choice experiment approach toelicit the value placed on algal blooms by residents in the Varnaarea. The advantage of the use of choice experiments in this case invaluing different attributes of the bloom (both visibility impair-ment and duration) is that it should facilitate better understandingof the way changes in bloom characteristics affect the values placedon them. This paper shows that there are significant welfare lossesfrom algal blooms. The key findings are:

i) The amount of bloom is important: respondents are willing topay 18.97 Leva (V9.73) for a program that entails no algalbloom, whereas, for the case of a medium level of bloom forone week, WTP falls to zero18;

ii) Duration is important: respondents are willing to pay morefor programs that offer shorter duration of algal bloom. Forhigh visibility and 6 weeks of duration, the willingness to payis 6.86 Leva (V3.52); and

iii) Congestion of the beach may be significant: people prefera low level of congestion, compared to a medium, and a highlevel of congestion.

These results are lower than those of the previous studiesreported above, but it must be borne in mind that this study did notconsider toxic algal blooms and was based in Bulgaria, whereaverage household income is about V3670.19 The attitudinalaspects of the survey show that respondents were more willing toswim in discoloured water than was the case in a similar study inMallorca (Torres et al., 2007). Aggregating the above results acrossthe population of Varna suggests that residents would be willing topay 2.5 million Leva (V1.3 million)20 for algal blooms to be elimi-nated This is not an insignificant sum, particularly bearing in mindwe do not include the additional welfare costs of tourists and alsoother economic effects of the blooms, and is a value that Bulgarianpolicy makers should consider when looking at the costs forimplementing the Water Framework Directive.

Further work is needed to link the values to changes in nutrientlevels, which is a difficult task given the complex nature of ecosys-tems. The uncertainty in the relationships between nutrient loadsand algal blooms make cost benefit studies difficult. How can wereconcile our valuation results, based upon clearly defined algalbloomsscenarios,with theuncertainties in thenutrient loadsealgalblooms relationships? Our study has estimated the benefits of not

19 National Statistical Institute of Bulgaria e 2006 data on income from all sources.20 The population of Varna is 357,270 inhabitants and the average household sizein Bulgaria as 2.7 persons.

T. Taylor, A. Longo / Journal of Environmental Management 91 (2010) 1963e1971 1971

having algal blooms. These benefits should be compared with thecosts associated with reducing nutrients flows, considering theprobabilities of algal bloom events given different levels of nutrients.

It is important to be able to value algal blooms and other effectsof nutrient inputs on riverine and coastal systems as policies areincreasingly being put in place to reduce loadings (e.g. through theWater Framework Directive). Without knowing the values placedon blooms it may be difficult to evaluate the extent to whichinvestments should be targeted to reducing bloom events,compared to other issues in the coastal zone e for example beachcleaning or measures to improve accessibility to the coast. Thisstudy presents a first estimate for the value placed on reducing theimpacts of algal blooms in the Bulgarian case.

Acknowledgements

The authors would like to thank Snejana Moncheva and IvelinaMoncheva of IO-BAS for their involvement in this study. The effortsand co-operation of the team at IO-BAS were essential for thesuccessful completion of this task.

The authors would also like to acknowledge the contributions ofMarta Petrucci, David Barton, Veronique Choquette, Walter Hecq,Antoni Riera, Dolores Garcia and Cati Torres in the formulation ofthe questionnaire. Additional research assistance fromEmmaNorthand Steve Arnold is acknowledged. Comments from Wan JungChou, Nick Dale and anonymous referees have improved the work.

All errors are, of course, our responsibility.The European Union (Contract No. 003933) is gratefully

acknowledged for its financial support in this study, which formspart of the Thresholds Integrated Project.

References

Alberini, A., Longo, A., Veronesi, M., 2007. Basic statistical models for conjoint choiceexperiments. In: Kanninen, B. (Ed.), Valuing Environmental Amenities usingChoice Experiments: A Common Sense Guide to Theory and Practice. Springer.

Bierlaire, M., 2003. BIOGEME: a free package for the estimation of discrete choicemodels. In: Proceedings of the 3rd Swiss Transportation Research Conference,Ascona, Switzerland. Available at http://transp-or.epfl.ch/page63023.html.

Brau, R., Cao, D., 2006. Uncovering the Macrostructure of Tourist’s Preferences: AChoice Experiment Analysis of Tourism Demand to Sardinia. In: FEEM WorkingPaper 33.2006. www.feem.it Available online from.

Eggert, H., Olsson, B., 2003. Heterogeneous Preferences for Marine Amenities: AChoice Experiment Applied to Water Quality. Working Paper. Department ofEconomics, Göteborg University, Sweden. http://swopec.hhs.se/gunwpe/abs/gunwpe0126.htm.

Greene, W.H., 2008. Econometric Analysis, sixth ed. Prentice Hall, Boston, MA, USA,p. 1216.

Gren, I.-M., Elofsson, K., Jannke, P., 1995. Costs of nutrient reductions to the BalticSea. Beijer International Institute of Ecological Economics. Royal SwedishAcademy of Sciences, Stockholm.

Hanley, N., Mourato, S., Wright, R.E., 2001. Choice modelling approaches: a superioralternative for environmental valuation? Journal of Economic Surveys 15,435e462.

Hanley, N., Wright, R.E., Alvarez-Farizo, B., 2006. Estimating the economic value ofimprovements in river ecology using choice experiments: an application to the

Water Framework Directive. Journal of Environmental Management 78 (2),183e193.

Hausman, J., McFadden, D., 1984. Specification tests for the multinomial logit model.Econometrica 52 (No. 5), 1219e1240.

Hensher, D.A., Rose, J.M., Greene, W.H., 2005. Applied Choice Analysis: A Primer.Cambridge University Press, New York.

James, R.T., Havens, K.E., 1996. Algal bloom probability in a large subtropical lake.Water Resources Bulletin 32 (5), 995e1006.

Ketsetzopoulou, M., Moncheva, I., undated. Measuring Sustainable TourismCapacity and Development of the Coastal Zone of Black Sea Region in Bulgaria.

Krause-Jensen, D., Carstensen, J., Dahl, K., 2007. Total and opportunistic algal coverin relation to environmental variables. Marine Pollution Bulletin 55, 114e125.

Le Goffe, P., 1995. The benefits of improvements in coastal water quality: a contin-gent approach. Journal of Environmental Management 45 (no. 4), 305e317.

Louviere, J.J., Hensher, D.A., Swait, J.D., 2000. Stated Choice Methods: Analysis andApplications. Cambridge University Press, Cambridge.

Markowska, A., Zylicz, T., 1999. Costing an international public good: the case of theBaltic Sea. Ecological Economics 30, 301e316.

McFadden, D.L., 1974. The measurement of urban travel demand. Journal of PublicEconomics 3, 303e328.

McQuatters-Gollop, A., Raitsos, D.E., Edwards, M., Pradhan, Y., Mee, L.D., Lavender, S.J., Attrill, M.J., 2007. A long-term chlorophyll data set reveals regime shift inNorth Sea phytoplankton biomass unconnected to nutrient trends. Limnologyand Oceanography 52 (2), 635e648.

Megard, R., Berman, T., 1989. Effects of algae on the Secchi transparency of thesoutheastern Mediterranean Sea. Limnology and Oceanography 34 (8),1640e1655.

Mitchell, R.C., Carson, R.T., 1989. Using Surveys to Value Public Goods: TheContingent Valuation Method. Resources for the Future, Washington D.C.

Moncheva, S., 2005. Black Sea Case Study. Presentation to Thresholds Meeting,Kalmar.

Moncheva, S., 2006. Database on algal bloom events in Black Sea case study.Contribution to D3.1.1 from Thresholds Project.

National Statistical Institute, 2010a. Household budgets for the 1998e2008 period.http://www.nsi.bg/BudgetHome_e/BudgetHome_e.htm (accessed online21.01.10.) at.

National Statistical Institute, 2010b. Labour force, employed and unemployed byplace of residence and sex in 2006. http://www.nsi.bg/Labour_e/Labour_e.htm(accessed online 21.01.10.) at.

OECD, 2008. Family database: SF1: family size and composition. Available online at.http://www.oecd.org/dataoecd/62/22/41919509.pdf.

Pertola, S., Kuosa, H., Olsonen, R., 2004. Is the invasion of Prorocentrum minimum(Dinophyceae) related to the nitrogen enrichment of the Baltic Sea? HarmfulAlgae 4 (3), 418e492.

Petrova, D., Velikov, V., 2005. The Emiliania Huxley along the Bulgarian Black SeaCoast. In: Ozhan, E. (Ed.), Proceedings of the Seventh International Conferenceon the Mediterranean Coastal Environment, MEDCOAST 05, 25-29 October2005, Kusadasi, Turkey. MEDCOAST Secretariat, Ankara, Turkey, pp. 591e600.

Revelt, D., Train, K., 1996. Incentives for Appliance Efficiency: Random ParametersLogit Models of Households’ Choices. Manuscript. Department of Economics,University of California, Berkeley.

Sandstrom, M., 1996. Recreational Benefits from Improved Water Quality: ARandom Utility Model of Swedish Seaside Recreation. Stockholm School ofEconomics, Economics Research Institute. Working Paper No. 121.

Soderqvist, T., 1995. The Benefits of Reduced Eutrophication of the Baltic Sea:a Contingent Valuation Study. Stockholm School of Economics and BeijerInternational Institute of Ecological Economics, Stockholm.

Stoltel, W., Scatasta, S., Granéli, E.1, Weikard, H.P., van Ierland, E., 2003. ECOHARM:The Socio-economic Impact of Harmful Algal Blooms in European MarineWaters. European Commission, Brussels (EVK3-CT-2001e80003).

Torres, C., Garcia, D., Riera, A., 2007. Valuing Environmental Costs of Algal Blooms:An Application to Santa Ponça Bay, Mallorca. Deliverable Report 1.4, FP6Thresholds Project. Prepared for the European Commission.

Train, K.E., 2003. Discrete Choice Methods with Simulation. Cambridge UniversityPress, Cambridge.

Wooldridge, J.M., 2002. Econometric Analysis of Cross Section and Panel Data. JohnWiley and Sons, Chichester.