Going Green? Economic Valuation of a MultipurposeWater Infrastructure in Northern Italy∗

By Arnaud Reynaud†, Denis Lanzanova,Bruna Grizzetti and Camino Liquete ‡

A contingent valuation approach is used to estimate how house-holds value different multipurpose infrastructures (grey or green)for managing flood risk and water pollution. As a case study weconsider the Gorla water park located north of Milan, in the Lom-bardy Region of Italy. The Gorla park is a neo-ecosystem recentlybuilt on the shore of the Olona River in an area previously usedfor poplar plantation. A novel aspect of our research is that it notonly considers the values people hold for different water ecosystemservices (pollution removal, recreative use, biodiversity, flood riskreduction), but also their preferences for how those outcomes areachieved (through grey or green infrastructures). The results indi-cate that the type of infrastructure delivering the ecosystem services(grey of green) does have an impact on individuals’ preferences forfreshwater ecosystem services. By considering the type of infras-tructures within the choice model, we gain a richer understandingof the relationship between social welfare and freshwater ecosystemservices. Keywords: ecosystem services, green infrastructure, val-uation, contingent valuation

I. Introduction

The term “Green infrastructure” refers to the living network of green spaces,water and other environmental features in both urban and rural areas. It is oftenused in an urban context to cover benefits provided by trees, parks, gardens,road verges, allotments, cemeteries, woodlands, rivers and wetlands. Indeed, justas built infrastructure like roads and utilities is necessary for modern societies,green infrastructure provides the ecosystem services (cleaning the air, filteringand cooling water, conserving and generating soils, pollinating crops and otherplants, regulating climate, sequestering carbon, protecting areas against storm,etc.) that are equally necessary for people’s well-being. Despite the abundantliterature in urban planning, economic analysis of green infrastructures remains

∗ This work is a part of the FP7 project OPENness (Operationalisation of natural capital and ecosys-tem services) founded by the European Union’s Seventh Programme for research, technological develop-ment and demonstration under grant agreement No 308428.† Corresponding author. European Commission DG Joint Research Centre, Institute for Environment

and Sustainability (IES), Unit H01 - Water Resources, arnaud.reynaud@jrc.ec.europa.eu‡ European Commission DG Joint Research Centre, Institute for Environment and Sustainability

(IES), Unit H01 - Water Resources

1

2

still quite limited. Our paper aims at filling this gap with a specific focus on greeninfrastructures dedicated to flood risk management and water pollution removal.

In this paper, we use a contingent valuation approach to estimate how house-holds value different multipurpose infrastructures (grey or green) for managingflood risk and water pollution. As a case study we consider the Gorla water parklocated north of Milan, in the Lombardy Region of Italy. The Gorla park is a neo-ecosystem recently built on the shore of the Olona River in an area previously usedfor poplar plantation. A novel aspect of our research is that it not only considersthe values people hold for different water ecosystem services (pollution removal,recreative use, biodiversity, flood risk reduction, ), but also their preferences forhow those outcomes are achieved (through grey or green infrastructures). Theresults indicate that the type of infrastructure delivering the ecosystem services(grey of green) does have an impact on individuals’ preferences for freshwaterecosystem services. By considering the type of infrastructures within the choicemodel, we gain a richer understanding of the relationship between social welfareand freshwater ecosystem services.

The remaining of the paper is organized as follows. Section II describes therelevant literature on valuation of green infrastructures. Section III describesour case study in Italy, the design of the contingent valuation survey and itsadministration. The results of the econometric model are reported in Section IV,and Section V concludes the paper.

II. Relevant literature

A. Ecosystem services and benefits provided by green infrastructures

Tzoulas, Korpela, Venn, Yli-Pelkonen, Kazmierczak, Niemela, and James (2007)review the literature on green infrastructure in relationship with ecosystem health,human health and human well-being.

Wang, Bakker, de Groot, and Wortche (2014) review the literature from differ-ent disciplines to synthesize the current knowledge on the effects of green infras-tructures on the indoor environment and human comfort in urban areas.

B. Valuation of services provided by green infrastructures

Jim and Chen (2006) have used a the contingent valuation method to evaluatethe recreational amenities of urban green spaces in Guangzhou, China.

Lopez-Mosquera and Sanchez (2011) examine whether green space users reflecttheir own personal values through the benefits and attributes they perceive inthis type of good.

Mell, Henneberry, Hehl-Lange, and Keskin (2013) use a contingent valuationapproach to value the development of green infrastructure investments (trees) inthe urban core of Manchester, Benefits and costs of street trees have been also

3

assessed in Lisbon, Portugal (Soares, Rego, McPherson, Simpson, Peper, andXiao 2011) and in Portland, US (Donovan and Butry 2010).

Wilker and Rusche (2014) have used a contingent valuation approach to valuedifferent types of green infrastructures in Esslingen, Germany. They analyze howthe elicited willingness to pay can be integrated in regional planning policies.Use of economic valuation to create public support for green infrastructure isalso discuss in Vandermeulen, Verspecht, Vermeire, Huylenbroeck, and Gellynck(2011).

The perspective of Baptiste, Foley, and Smardon (2015) is a little bit differentsince the authors focus on the factors that influence the public’s willingness toimplement green infrastructure on private properties.

III. The contingent valuation survey

In this section, we describe the Gorla Maggiore constructed wetland (North-ern Italy), together with the design of the contingent valuation survey and itsadministration.

A. The Gorla Maggiore constructed wetland

The Gorla water park is a constructed wetland located in Gorla Maggiore, asmall municipality north of Milan, in the Lombardy Region. The Gorla water parkis a neo-ecosystem built on the shore of the Olona River in an area previously usedfor poplar plantation, see Figure 1. The Gorla water park has been developed,under the sponsorship or the Lombardia Regional Authority and the co-fundingby Fondazione Cariplo, through a participatory process. The whole area surfaceis about 3 ha. It includes a flood prevention area (1 ha), a pollutant removal area(0.4 ha of phragmites reed bed and 0.3 ha of natural-like multispecies wetland)and a leisure and recreational area (1.3 ha of park).

The municipality of Gorla Maggiore operates a combined sewer system de-signed to collect rainwater runoff, domestic sewage, and industrial wastewater inthe same pipe network. Most of the time, the combined sewer system transportsall of wastewater to a sewage treatment plant, where it is treated and then dis-charged into water bodies including the river Olona. During periods of heavyrainfall, however, the wastewater volume in the combined sewer system can ex-ceed the capacity of the sewer system or treatment plant. In that case, excesswastewater is discharged directly to the river Olona. These overflows contain notonly stormwater but also untreated human and industrial waste, toxic materials,and debris. To address this issue, an innovative green infrastructure has beendeveloped (the first one of this type in Italy). It consists of four sand filters ver-tical beds aimed to treat the first flush coming from the combined sewer overflowdevice and from an extended retention basin for the accumulation and the slowrelease in the river of the second flush, with the main aim of flood protection but

4

Figure 1. The Gorla Maggiore water park

also with some benefits in terms of pollution control. The green infrastructureaddresses then the issue of pollution and flood control.

The Gorla water park is a multi-purpose infrastructure since it also includes aleisure and recreational area which dedicated to wide range of activities includingeducational activities, biking, running, picnicking (appropriate infrastructures areinstalled), animal watching. A wide range of educational services related to thepresence of fauna are available on the site (water birds and small amphibians) andadvertised by informational panels. Flora is highlighted, especially for the plantsinvolved in the water purification processes. The accessibility is excellent (600meters from the town through a foot path). The park seems has been particularlywell designed for educative activities with a special pond in which frogs can be veryeasily observed, and with many informational panels. In addition, a small flux ispermanently diverted from the Olona river for feeding a small pool located on thebottom of the extended retention basin, so to permit the development of a highlydiversified ecosystem while offering multiple pathways for the removal/control ofseveral pollutants.

To summarize, the Gorla water park has been designed to provide four differenttypes of water ecosystem services:

• pollution control (reduction of pollution load discharged to Olona river bya combined sewer overflow),

• flood prevention (storage of rainwater and regulation of flow discharge tothe river),

5

• recreational use (use of the park by the local population),

• biodiversity enhancement (bird species richness).

B. Contingent valuation survey

A wide range of economic valuation techniques can be used to value ecosystemservices provided by the Gorla water park. We use here a contingent valuationapproach.

Development of the survey. — The survey has been developed by an inter-disciplinary team including ecologists, biologists, hydrologists and environmentaleconomists following the guidelines suggested by the NOAA panel (Arrow, Solow,et al. 1993). The starting point for developing the survey has been a field triporganized in July 2014 in the Gorla water park. We conducted there different sci-entific activities including sampling in the pond for macroinvertebrates, samplingin the river for macroinvertebrates, evaluating plant biodiversity in the artificialwetland and identifying the eco–recreational potential of the area. A first versionof the survey was then designed following this field trip. Following discussionswith the engineering company who had designed the park, a second version ofthe survey was developed and submitted for comments and discussions to somerepresentatives of the municipality of Gorla Maggiore. Some parts of the surveywere then amended (see the following discussion of vehicle payment) and the finalsurvey was then tested internally at the Joint Research Center of the EuropeanCommission and externally. The survey was finally translated in Italian by anItalian native speaker.

Structure of the questionnaire. — The questionnaire consists of three parts.In the first part, we measure how often individuals have visited the Gorla parkin the last 12 months. We also have collect information regarding the type recre-ational activities done by individuals when visiting the park. The second part isthe main contingent valuation section. In the third part, we collect some basicsocioeconomic information on respondents and identify protest answers.

Reference scenario. — We first describe a reference situation in which thewhole area is covered by a private poplar plantation. This situation which usedto prevail before the construction of the Gorla Maggiore park is defined as beingthe reference scenario and is described in the questionnaire as:

“There is a plot of private land where poplars are grown for the pro-duction of timber. The ecosystem produces wood, but does not offer alot of ecological services.”

6

Since a crucial issue is the good understanding by respondents of the characteris-tics of the reference scenario, we describe explicitly in the questionnaire the levelof ecosystem services provided (in terms of pollution reduction, recreational ac-tivities, biodiversity and flood protection) associated to this scenario. Both thephrasing and quantification ecosystem service associated to the reference scenario(and also to the four the alternative scenarios) have been discussed and validatedby biologists and by IRIDRA, the engineering private firm which was responsiblefor the design and the construction of the Gorla park. The script we have usedfor the reference scenario is the following:

- “Low pollution reduction: during rainfall events some of the wastewatercoming from the municipality of Gorla Maggiore is not treated and a largepart of pollutants are transported to the river Olona.

- Few recreational activities for the community: people can not access to thearea because it is private.

- Low biodiversity: only one type of plant is present and often treated withchemical products.

- Low protection against floods: lacks an retention area in the case of highflow rates.”

The verbal description of quantification ecosystem service associated to the refer-ence scenario was accompanied by visual aids for facilitating a full understandingof the valuation scenarios. As Mitchell (2002) points out, visual aids play a vi-tal role both in illustrating the verbal information and in holding respondents’attention during the presentation of scenarios. We have used two types of visualaids. First, each ecosystem service (pollution reduction, recreational activities,biodiversity and flood protection) has be identified by a specific pictogram. Sec-ond, the level of service provision by associated to a specific color (green for goodlevel, yellow for medium level and red for bad level).

Contingent valuation scenarios. — We have then proposed sequentially fourdifferent contingent valuation scenarios. Theses alternative scenarios have beendiscussed and validated by IRIDRA and by the representatives of the Gorla mu-nicipality. Respondents have been asked to evaluate these scenarios in comparisonto the reference scenario (private poplar plantation). We have used the followingscript:

“Suppose that the Park Gorla has not been built and instead there isa private poplar plantation, as it was in the past. Compared to thisreference scenario, we ask you to choose the project that you prefer toprevent the waste water of the municipality of Gorla Maggiore to pol-lute the river Olona. We ask you to evaluate four possible alternativesin comparison to the poplar plantation.”

7

Each scenario has been obtained by combining a type of infrastructure dedicatedto treat wastewater of the municipality of Gorla Maggiore (either a green or aconventional infrastructure) with the possibility to have or not a recreationalpark around this infrastructure (either a recreational park or a private poplarplantation). The script we have used for describing the green and the conventionalwastewater infrastructures, and the surrounding area is the following:

- “Green infrastucture: This infrastructure consists of a set of constructedwetlands with a wet retention pond.

- Conventional infrastructure: The infrastructure consists of a flush tank(buried and covered by grass) and a dry retention pond.”

- Park: Park with trees designed for recreational activities.

- Poplar plantation: Private plantation of poplar (non accessible for recre-ational activities).”

By combining the type of infrastructure and the type of area surrounding we getthe four contingent valuation scenarios:

- P1: green infrastructure & park;

- P2: green infrastructure & poplar;

- P3: conventional infrastructure & park;

- P4: conventional infrastructure & poplar.

To make people more clearly understand the meaning of these four scenarios,each of them was described by two pictures (one for the infrastructure and thisother one for the surrounding area). The pictures which have been shown to therespondents for each scenario are presented in Figure 2.

8

Figure 2. Visual description of each contingent valuation scenario

P1: green infrastructure & park P2: green infrastructure & poplar

P3: conventional infrastructure & park P4: conventional infrastructure & poplar

9

Table

1—

Desc

rip

tio

nof

the

four

contin

gent

valuatio

nsc

enario

s

P1:

Gre

enin

frast

ruct

ure

&p

ark

P2:

Gre

enin

frast

ruct

ure

&p

op

lar

P3:

Conven

tion

al

infr

ast

ruct

ure

&p

ark

P4:

Conven

tion

al

infr

ast

ruct

ure

&p

opla

r

Poll

uti

on

Hig

hp

oll

uti

on

contr

ol:

macr

op

hyte

s(w

etla

nd

pla

nts

)

neu

tralize

pollu

tants

Hig

hp

oll

uti

on

contr

ol:

macr

op

hyte

s(w

etla

nd

pla

nts

)

neu

tralize

pollu

tants

Hig

hp

oll

uti

on

contr

ol:

the

flu

sh

tan

kst

ore

sth

ew

ast

ew

ate

rfr

om

Gorl

a

du

rin

gst

orm

s.A

fter

afe

wd

ays

this

wate

ris

pu

mp

edb

ack

toth

ew

ast

ew-

ate

rtr

eatm

ent

pla

nt

tob

ecl

ean

edb

e-

fore

its

fin

al

dis

charg

ein

the

river

Hig

hp

oll

uti

on

contr

ol:

the

flu

sh

tan

kst

ore

sth

ew

ast

ew

ate

rfr

om

Gorl

a

du

rin

gst

orm

s.A

fter

afe

wd

ays

this

wate

ris

pu

mp

edb

ack

toth

ew

ast

ew-

ate

rtr

eatm

ent

pla

nt

tob

ecl

eaned

be-

fore

its

fin

al

dis

charg

ein

the

river

Rec

reati

on

al

Hig

hrecreati

on

level:

the

pon

dp

rovid

esn

ice

aes

thet

icvie

ws

an

dfi

shin

gop

port

un

itie

s

an

dth

eare

ah

as

bee

nre

store

d

wit

hri

pari

an

tree

san

dis

suit

-ab

lefo

rou

tdoor

act

ivit

ies

Med

ium

recreati

on

al

level:

the

pon

dp

rovid

esn

ice

aes

-th

etic

vie

ws

an

dfi

shin

gop

por-

tun

itie

sb

ut

the

pop

lar

are

ais

not

acc

essi

ble

Med

ium

recreati

on

al

level:

the

surr

ou

nd

ing

are

ah

as

bee

nre

store

dw

ith

rip

ari

an

tree

san

dis

suit

ab

lefo

r

ou

tdoor

act

ivit

ies

Low

recreati

onal

level:

the

infr

as-

tru

ctu

reco

ver

edw

ith

gra

ssca

nb

em

ud

dy

aft

erfl

ood

ing

an

dth

ep

op

lar

are

ais

not

acc

essi

ble

Bio

div

ersi

tyH

igh

bio

div

ersi

ty:

the

are

a

incr

ease

sh

ab

itat

availab

ilit

y

for

bir

ds,

dra

gon

flie

s,am

ph

ib-

ian

sin

clu

din

gse

ver

al

end

an

-

ger

edsp

ecie

s

Hig

hb

iod

iversi

ty:

the

are

a

incr

ease

sh

ab

itat

availab

ilit

y

for

bir

ds,

dra

gon

flie

s,am

ph

ib-

ian

sin

clu

din

gse

ver

al

end

an

-

ger

edsp

ecie

s

Low

bio

div

ersi

ty:

the

vari

ety

of

the

livin

gorg

an

ism

sis

rela

tivel

ylim

ited

Low

bio

div

ersi

ty:

the

vari

ety

of

the

livin

gorg

an

ism

sis

rela

tivel

ylim

ited

Flo

od

contr

ol

Hig

hfl

ood

contr

ol:

the

are

a

can

store

wate

ran

dco

ntr

ibu

te

tod

ecre

ase

the

flood

imp

act

dow

nst

ream

Hig

hfl

ood

contr

ol:

the

are

a

can

store

wate

ran

dco

ntr

ibu

te

tod

ecre

ase

the

flood

imp

act

dow

nst

ream

Hig

hfl

ood

contr

ol:

the

pon

dca

n

store

wate

ran

dco

ntr

ibu

teto

dec

rease

the

flood

imp

act

dow

nst

ream

Hig

hfl

ood

contr

ol:

the

pon

dca

n

store

wate

ran

dco

ntr

ibu

teto

dec

rease

the

flood

imp

act

dow

nst

ream

.

10

The level of ecosystem services provided (in terms of pollution reduction, recre-ational activities, biodiversity and flood protection) associated to each scenariowas also verbally and graphically presented. For the graphical representation weuse again some pictograms with a color representing the level of service provi-sion (green for good level, yellow for medium level and red for bad level). Forthe verbal description, we have used the script presented in Table 1. It shouldbe mentioned that the four scenarios allow to achieve a high level of pollutioncontrol (the primary objective of the Gorla Maggiore municipality). However,the technical way to achieve pollution control significantly differ depending onthe green or the grey infrastructure. The provision of recreational services variesacross scenarios from low in the scenario P4 (conventional & poplar) to high inthe scenario P1 (green & park). The two other scenarios provide an intermediatelevel of recreational services. The level of biodiversity is assumed to be high in thescenario P1 and P2 (green & park and green & poplar ) and low in the scenarioP3 and P4 (conventional & park and conventional & poplar). Lastly, the fourscenarios result in high flood control.

Format of the contingent valuation questions. — We have chosen to usean open–ended payment card (PC) approach, one of the most popular methodfor eliciting willingness to pay in environmental valuation where respondents arepresented with a series of ordered payment amounts, or bids, and typically areasked to circle the maximum of the series they would pay for the good undervaluation.1 In our case, the bid structure was constructed based on experts’suggestions and based on the actual construction and maintenance costs of theGorla park. It covers a ranges going from zero euro per household and per yearto more than 75 euros per household and per year.

The choice for the payment vehicle is a crucial element in applications of the con-tingent valuation method because it provides the context for payment (Morrison,Blamey, and Bennett 2000). Our pre-tests and the discussions we have had withthe representatives of Gorla Maggiore municipality leaded to the conclusion thatusing a tax increase for funding the infrastructure could not be considered in thecurrent economic context in Italy. Hence, due to the economic crisis, a lot of peo-ple may be per principle opposed to any taxation increase. We have then decidedto use the municipality budget as a payment vehicle making explicit that anymoney dedicated to fund the proposed infrastructure would then not be availablefor funding the provision of other municipal public goods. The script used forexplicating the payment vehicle is presented in Figure 3. This figure also givesopen–ended PC question used for the first contingent valuation scenario (green& park).

1The PC method was first developed by Mitchell and Carson (1981). The main advantages anddisadvantages of the PC format as opposed to other methods are fully discussed in (Mitchell and Carson2013) and some specific examples of empirical comparisons between WTP through PC and through otherformats include (Blaine, Lichtkoppler, Jones, and Zondag 2005)

11

Figure 3. Contingent valuation question for P1 (green & park)

12

C. Implementation and first results

Survey administration. — Mitchell and Carson (2013) have argued that the pre-ferred mode of administration for CV surveys is in–person interviews conductedin the respondent’s home.2 The main rational is the need to explain complex sce-narios using visual aids with control over pace and sequence.Mitchell and Carson(2013) have however acknowledged that mail survey may be suitable for surveyingrespondents who have familiarity with the good (e.g. recreational users). This istypically the case here. As a result, the survey has been distributed by mail to allhouseholds living in the municipality of Gorla Maggiore beginning of 2015. Thequestionnaire has been included to the newsletter which is regularly sent by themunicipality to all households, and it has been directly advertised on the web siteof the Gorla Maggiore municipality. Then households were given the choice eitherto directly fill in the questionnaire and to put it back into a dedicated urn at thetownhall of the municipality, or to fill the questionnaire online on a dedicated website (EU-survey).

Sampling issues. — 1,600 questionnaires have been distributed to householdsliving in Gorla Maggiore. We have received 71 filled questionnaires (25 from EU-survey, the dedicated web site and the remaining from the dedicated urn at thetownhall of the municipality). This translate to a low response rate (4.4%) butwhich is not surprising for this kind of mail survey. This raises however someissues related to the representativeness of our sample we discuss now.

Table 2—Socio-economic characteristics of our sample

Variable Italy Lombardy Gorla Maggiore Our sample

Household size (in 2014) 2.34 2.26 2.45 2.86

Female (in 2014) 51.5% 51.2% 50.2% 38.0%Average age population above 18 (in 2014) 51.1 51.3 51.5 54.8

Household annual income (in 2012) 29,436 34,097 29,120* 30,794

Population economically active (in 2011) 50.8% 54.8% 53.8% 53.5

*: for municipalities in Lombardy with less than 2,000 inhabitants

Socio-economic data for Italy, Lombardy and Gorla Maggiore come from ISTAT.

In Table 2 we compare some selected socioeconomic characteristics of our re-spondent sample with the characteristics of population in Gorla Maggiore, inLombardy and in Italy. On average the household size in our sample is higher

2The mode of administration for CV surveys has been highly debated in environmental economics.See Lindhjem and Navrud (2011) for a recent survey.

13

than what is reported by the Italian National Institute for Statistics (Istat) for themunicipality of Gorla Maggiore in 2014 (2.86 verus 2.45 persons per household).With 38.0% only, females are under-represented in our sample. On average, ourrespondents are slightly older than inhabitants in Gorla Maggiore. The averageannual household income in our sample is 30,794 euros. This amount is in betweenwhat is reported for Italy (29,436 euros) and for Lombardy (34,097 euros). Lastlythe percentage of respondents considered as economically active (i.e currently em-ployed and unemployed) in our sample matches very well the data reported byISTAT for Gorla Maggiore in 2011. Although we don’t claim that our sample isrepresentative of the population living in the municipality of Gorla Maggiore, theprevious analysis suggests no indication of strong differences with ISTAT datafor the municipality of Gorla Maggiore based on the observable characteristicspresented in Table 2 (at the exception of the share of females).

Use and perception of the Gorla Maggiore park. — The first part of thequestionnaire has been dedicated to collect data related to the way the GorlaMaggiore park in used and perceived by the respondents. On average, each re-spondent has visited the park a little bit more than 25 times over the last 12months (SD is 31.89). In our sample, the annual number of visits varies from 0(for 5 respondents) to more than twice a week (for 7 respondents). The averagetypically size of the group when the respondent goes to the park is 2.43 (SD is1.27), varying from 1 (for 14 respondents) to more than 5 (for 4 respondents).Respondents leave at proximity of the park. The average distance to the park is1.38 km (SD is 0.74). For 27 respondents the distance to the park is less than 1km.

Table 3—Frequency of recreational activities in the Gorla Maggiore Park

Activity Never Sometimes Often Sometimes or Often

Walking or dog walking 5 16 36 52

Running or biking 10 19 17 36Educating children to nature 18 11 8 19

Playing with kids 19 12 5 17

Picnicking 30 4 0 4Watching wildlife (birds/frogs) 8 18 18 36

Sightseeing / enjoying nature 1 22 32 54Sunbathing 27 9 1 10

Next, each respondent has been proposed a list of 8 possible recreational activi-ties, see Table 3. Each respondent has then been asked how often he has practicedeach activity in the last 12 months. Sightseing and walking / dog walking are by

14

far the two types of recreational activity which are the most often undertaken bypark visitors. 36 respondents have declared that they go to the park at least timeto times for running or biking, or for watching wildlife. Educational or leisureactivities with kids are also mentioned by some respondents. The main insight weget from Table 3 is that the Gorla Maggiore is used for wide range of recreationalactivities.

Preliminary analysis of answers to contingent valuation scenarios . —

Now we move to the answers given by the respondents to the four contingentvaluation scenarios P1, P2, P3 and P4 described above.

Table 4—Statistic descriptive on WTP per contigent valuation scenario

Mean Std. Dev Min Max Observations

Full sample

P1: green infrastructure & park 26.20 20.45 0 75 71

P2: green infrastructure & poplar 9.28 12.13 0 45 58

P3: conventional infrastructure & park 5.39 12.46 0 75 61P4: conventional infrastructure & poplar 3.20 10.28 0 75 61

Sample without false zeros

P1: green infrastructure & park 28.19 19.83 0 75 66

P2: green infrastructure & poplar 10.15 12.34 0 45 53

P3: conventional infrastructure & park 5.88 12.90 0 75 56P4: conventional infrastructure & poplar 3.48 10.69 0 75 56

Willingness to pay in euro per year and per household.

Table 4 gives some statistics on the maximum amount of money each respondentis ready to allocate to each contingent valuation scenario (in euro per year andper household for the following twenty years). We interpret this amount of moneyas a willingness to pay (WTP) for the corresponding scenario.

In a contingent valuation analysis, it is important to make the distinction be-tween the “true zero bids” corresponding to respondents having indicated thatthey are not willing to pay anything because they are truly averse or indifferentto the good for which a WTP is solicited from “false zero bids” which correspondto respondents having reported a zero WTP even though her true value for thegood in question is positive, (Hanley, Wright, and Alvarez-Farizo 2006). Falsezero bids may be categorized into three types. The first are “protest bids”, wherethe respondent reports a zero bid for reasons other than the respondent placinga zero value on the good in question. The second are “don’t know” responses,

15

where the respondent is simply uncertain about the amount they are willing-to-pay, noting that this amount could of course be zero. Third, some respondentmay have stated a zero bid because the task of selecting options is too complex(i.e., they have difficulties understanding or answering the choice questions).

To identify protest answers, respondents having reported zero WTP for the fourproposed scenarios have been asked if they agree or disagree with the six followingreasons: “1- I am not confident that the money will be used efficiently by themunicipality”, “2- I am against any tax expenses”, “3- I prefer the money to bespent on more important things”, “4- I cannot afford to pay any tax”, “5- I believethat the park should not be paid by me but directly by a central administration”and “6- I will never go to the park”. All respondents have also been asked tostate if the survey was clear, which is the case for 95.6% of respondents. Amongthe 6 respondents having reported zero WTP for the four proposed scenarios,5 who have selected at least one of the reason 1-, 2- or 5- can be classified as“false zeros”. In Table 4 we then report some statistics on willingness to pay perscenario first based on the full sample and second on a subsample excluding “falsezeros’.

Table 4 calls for a few comments. First, whatever the sample considered thereare significant differences across the WTP per scenario which varies from around3 euros per household and per year for scenario P4 corresponding to the conven-tional infrastructure with poplar plantation to 26 to 28 euros for P1 correspondingto the green infrastructure with park. Second, for a given surrounding area re-spondents have a higher WTP if the infrastructure is green compared to theconventional one. Considering the sample without “false zeros”, passing from aconventional to a green infrastructure increases the WTP by 6.67 euros per re-spondent and per year for a surrounding made of poplar and by 22.31 euros eurosper respondent and per year for a surrounding made of a park. Third, comparedto the three other scenarios, we find a much higher WTP for P1 which corre-sponds to the green infrastructure with park (the one which has been built in theMunicipality of Gorla Magiore). This may be viewed has an evidence of a strong“endowment effect”. The “endowment effect” refers to the theory that explainsobserved gaps between WTP and willingness to accept (WTA) by some featureof human preferences that leads owners to resist selling goods because (a) sellingis perceived as “losing” the endowed good, and (b) individuals are generally lossaverse (Plott and Zeiler 2005). The “endowment effect” has been highly docu-mented in contingent valuation studies, see Tuncel and Hammitt (2014). Oneshould lastly point out that there may be some other explanations for the higherWTP attributed by respondents to P1. These possible explanations include thepresence of an income effect, of transaction costs, the absence of credible substi-tutes to the existing park and the limited incentives to learn about preferencesfor a hypothetical transaction.

To gain some insight on how WTP differ across individuals, we provide inTable 5 the WTP for scenario P1 (green infrastructure & park) for several sub-

16

samples.

Table 5—WTP for scenario P1 (green infrastructure & park) by subsample

Mean Std. Dev Min Max Observations

Number of visits per year

– None 21.40 19.93 0 45 5

– [1,20] 24.54 15.12 0 62.5 38– >20 35.70 24.75 0 75 23

Distance to the park (in km)

– ≤ 1 27.14 19.30 0 75 25

– ]1,2] 28.46 20.20 0 75 32– >2 30.11 22.11 0 75 9

Level of appreciation of the park

– Low 24.80 31.15 2 75 5– Medium 27.01 15.89 0 62.5 30

– High 30.40 21.59 0 75 29

Age of respondent (in years)

– ≤ 40 35.28 23.34 2 75 16– ]40,50] 34.03 17.02 10 75 15

– >50 22.44 17.88 0 62.5 35

Household income (in euros per year)

– ≤ 15,000 16.40 18.38 0 45 15

– ]15,000 to 30,000] 31.14 19.95 2 75 29

– > 30,000 32.34 18.22 0 75 22

Sex of respondent

– Female 28.56 21.93 0 75 24

– Male 27.98 18.80 0 75 42

Willingness to pay in euro per year and per household, false zeros excluded.

As expected, the WTP increases with number of visits to the park during thelast 12 months, from 21.40 for respondents reporting no visits to 35.70 euros forthose having visited the park more than 20 times. The WTP for respondentslocated less than 1km from the park and for those located more than 2km fromthe park are not statistically different at 5%. The WTP does not appear to varywith the distance to the park. Respondents who have a low appreciation of theoverall quality of the Gorla Maggiore Park report a low WTP (24.80 euros) but

17

they represent only a small part of the sample (5 respondents).

Concerning the socioeconomic characteristics, we find a significantly lower WTPfor the oldest respondents. The average WTP for respondents over 50 yearsis only 22.44 euros per year. One should however be careful with interpretingthis result as a pure age effect since oldest respondents may have some specificscharacteristics affecting their WTP (i.e. low income or low frequency of parkvisit). We find a significant income effect, especially for the poorest respondents.The average WTP for households reporting an annual income lower than 15,000euros is only 16.40 euros per household. It is approximatively equal to half of ofthe WTP reported by wealthier households. Lastly, our result don’t reveal anysignificant difference between female and male WTP. This result is importantsince, as discussed previously, females are under-represented in our sample. Sincesex doesn’t matter we don’t anticipate that the under-representation of femaleswill affect our final estimates of the WTP.

IV. Econometric analysis of WTP

A. Estimation procedure

When analysing data obtained from a PC contingent valuation survey, it isunclear what assumptions should be made regarding respondent’s true willingnessto pay. A standard approach is to assume that the WTP follows a lognormaldistribution, thus avoiding negative values (Mahieu, Riera, and Giergiczny 2012).The valuation function can then be represented by:

(1) logWTP ∗i = X′iβ + εi

where WTP ∗i denotes the true WTP for respondent i, Xi a vector of explanatoryvariables and εi a random component following a normal distribution with meanzero and standard deviation σ.

A standard procedure to estimate Equation (1) is to assume that the true will-ingness to pay is the midpoint between the highest amount to which the respon-dent said “yes” and the lowest amount to which she said “no” (Cameron 1987).This approach allows direct estimation of willingness to pay, thus no assumptionsare made regarding the functional form of respondents’ utility or the error struc-ture of the data. A straightforward analysis consists then in simply regressingthe stated willingness to pay on various explanatory factors but (Cameron 1987,Cameron and Huppert 1989) have showed however that this type of analysis isgenerally not efficient.

An alternative is to explicitly consider the structure of data obtained froma PC contingent valuation survey. Since respondents are asked to select themaximum amount of money they would pay for the good under valuation, itmeans that the individual’s WTP is bounded by the largest amount they agreed

18

to pay and the smallest amount they refused (interval censoring).3 The usualparametric approach to estimate the valuation function with censored data in thedependent variable is the “interval data model” (Cameron and Huppert 1989).When considering the interval data model, the contribution of each response tothe likelihood function is given by the probability that the latent WTP value fallswithin the chosen interval. This probability is then found by taking the integral ofthe conditional probability density function over the range of WTP indicated bythe interval response, but the specific form for the probability depends upon thetype of censoring in the interval data model (interval censoring, right–censoringor left–censoring). Interval censoring corresponds to the case where WTP∗ lies inthe bracket bounded by the payment chosen and the next amount in proposed listdenoted tli and tui. In the right–censoring case, WTP∗ is greater than tli whereasthe left–censoring case correspond to a WTP∗ lower than tui. The conditionalprobability of observing each case for respondent i writes:

(2) P (WTP ∗i |Xi) =

Φ(tui−X

′iβ

σ

)− Φ

(tli−X

′iβ

σ

)if interval-censoring

1− Φ(tli−X

′iβ

σ

)if right-censoring

Φ(tui−X

′iβ

σ

)if left-censoring

where Φ is the cimulative standard normal density function. The correspond-ing log–likelihood function is made of three parts, which correspond to interval–censoring, left–censoring and right–censoring observations.

Since each respondent is asking to pass several CV questions, our approachrequires further generalization of the model presented above. Multiple responsesper individual are likely to induce some degree of correlation within responses(Moore, Holmes, and Bell 2011). To control for potential intra-individual correla-tion, we used a random effects panel model, which assumes that intra–individualcorrelation is randomly distributed over the sampled population. A random ef-fects model with normally distributed errors and latency in the dependent variableyields

(3) logWTP ∗ij = X′ijβ + ui + εij

with ui follows a normal distribution with mean zero and standard deviation σuand εij follows a normal distribution with mean zero and standard deviation σε.In Equation (3), WTP ∗ij is the latent value known to individual i in response tothe jth question but unobserved by the researcher, Xij is a vector of the data forthat response, and β is a vector of coefficients. In the random effects model the

3If the highest payment is chosen, the WTP is assumed to be located somewhere above this payment(right–censoring). If the lowest payment is chosen, the WTP is supposed to be below this payment(left–censoring).

19

error is decomposed into two components. The term ui is a random error thatvaries across individuals but is constant within an individual’s set of responses.The term εij is a random error that can vary across individuals and responses.The two error components, ui and εij , are assumed to be IID and independentof each other. The conditional probability of observing a sequence of choice forindividual i for all CV questions is obtained from Equation (2) by multiplyingthe corresponding probabilities. The model was then estimated using the randomeffects interval data model (xtintreg) in the Stata statistical package.

B. Estimation results

Holistic approach. — In the holistic approach, it is assumed that respondentsvalue the environmental good proposed as a whole (in our case, a green/greyinfrastructure with/without a recreational park) without considering the levelsof the attribute. The holistic approach relies on the fact that it has been shownthat respondents often use very simple heuristics for processing complex attribute-based choice tasks. We present in Table 6 some random–effects regression models.Model 1 only include the type of infrastructure valued. Model 2 include in ad-dition some socioeconomic characteristics of respondents. The two first columnscorrespond to the full sample whereas in columns 3 and 4 the false zeros havebeen excluded.

Three dummy variables have been introduced for describing the scenario un-der consideration. Green infrastructure is a dummy variable equal to 1 if theinfrastructure considered is green (the reference category is a conventional infras-tructure). Park is a dummy variable equal to 1 if surrounding area is a recreationalpark (the reference category is a private poplar plantation). Since the previousanalysis has suggested that there might be a premium for the scenario combiningthe green infrastructure and a recreational park, third dummy variable has beenadded to account for this situation. As explanatory variables, we have introduceda dummy variable for respondent indicating that there is at least one child below18 year in their hoiusehold and another dummy variable equal to 1 if the respon-dent is over 50 years old. Household income is introduced in logarithm and wealso control for the frequency of visits to the park.

Table 6 call for some remarks. First both the sign and the order of magni-tude of the estimated coefficients are very similar across models. Concerning thecharacteristics of the scenarios under study, we find a positive and significant will-ingness to pay for the green infrastructure (compared to the conventional one).Depending upon the model considered the WTP for a green infrastructure variesfrom 6.3 to 7.1 euros per household and per year (for a twenty year time horizon).We also find a positive (but not significant) WTP for a park varying from 2.2to 2.4 euros per household and per year. The most interesting finding is givenby the positive and highly significant coefficient for the interaction between thegreen infrastructure and the park. There is a specific premium for a project com-

20

Table 6—Random–effects regression model

M1Full M2Full M1NoFalse M2NoFalse

b/se b/se b/se b/se

Green infrastructure (0/1) 6.295*** 6.585*** 6.875*** 7.113***(1.82) (1.87) (1.89) (1.92)

Park (0/1) 2.165 2.275 2.359 2.444

(1.79) (1.84) (1.85) (1.89)Green infrastructure & Park (0/1) 14.716*** 15.523*** 15.905*** 16.468***

(2.53) (2.60) (2.62) (2.67)

Dummy if children below 18 (0/1) -0.408 -0.732(2.74) (2.73)

Dummy respondent age over 50 (0/1) -2.938 -2.104

(2.56) (2.59)Dummy for annual number visit > 20 (0/1) 6.734*** 7.841***

(2.43) (2.50)ln household annual income (euros) 6.833*** 6.228***

(2.33) (2.39)

Constant 2.578 -68.200*** 2.637 -62.751**(1.68) (24.16) (1.73) (24.74)

sigma u

Constant 8.822*** 7.627*** 8.759*** 7.588***(1.10) (1.04) (1.14) (1.06)

sigma e

Constant 9.768*** 9.789*** 9.677*** 9.670***(0.57) (0.45) (0.56) (0.48)

R-squared

N. of obs. 249.000 237.000 229.000 221.000

Source: auto.dta

bining a green infrastructure together with a recreational park. This premium isquite significant in terms of amount of money since it varies from 14.7 to 16.5euros per household and per year, depending upon the model considered. Ourresults suggest that people in Gorla Maggiore don’t put any specific value on apark if it associated with a conventional infrastructure. On contrary the parkwill be highly valued if is associated with the green infrastructure. One possibleinterpretation of this result is that the park and the green infrastructure may beperceived as two highly complementary goods. Another possible explanation isthe “endowment effect” we have discussed previously.

Concerning the characteristics of the respondents, the WTP appear to be sig-nificantly impacted by respondent’s income and respondent’s frequency of visitsto the Gorla Maggiore park. The higher is the household income, the higher willbe the WTP. In addition respondents who report a that that went to the parkat least 20 times during the last 12 months have a additional WTP which variesbetween 6.7 and 7.8 euros per household and per year.

Since there is a mapping between the four contingent valuation scenarios and theattributes describing the level of ecosystem services provided (pollution reduction,recreational activities, biodiversity and flood protection see Table 1), our estimatemy directly be interpreted in terms WTP per attribute. More specifically, the

21

WTP for high and a medium levels of recreational activities is estimated to be19.04 and 2.16 euros per household and per year (reference category is low levelof recreational activities). The WTP for a high level of biodiversity is estimatedto be 4.13 euros per household and per year (reference category is low level ofbiodiversity). Finally the joint WTP for a high level of pollution control and ahigh level of flood control is estimated to be 2.57 euros per household and peryear.

C. Cost benefits analysis

We perform in this section some back-of-the-envelope calculation to provideestimates of net benefits from implementing the four different contingent valuationscenarios.

Table 7—Cost benefit analysis of contingent valuation scenarios

Construction cost Maintenance cost Discounted net benefitsScenario (a) (b) (a) (b) WTP 2% 3% 4%

P1 : green & park 820,000 80,000 2,617 1,000 28.2 5,121 -68,101 -132,415

P2 : green & poplar 820,000 0 2,617 0 10.15 -514,840 -539,413 -560,997P3 : conventional & park 794,749 50,000 11,797 3,600 5.88 -885,598 -881,030 -877,017

P4 : conventional & poplar 794,749 0 11,797 0 3.48 -861,014 -854,674 -849,104

(a): infrastructure(b): landscaping

Both the construction and the maintenance costs presented in Table 7 have beenprovided by IRIDRA, the engineering private firm which was responsible for thedesign and the construction of the Gorla Maggiore park. Both the green and theconventional infrastructure have a life expectancy equal to 20 years. In Table 7 wereport the discounted net benefits for each scenario for 3 different interest rates(2%, 3% and 4%). One should stress at this point that net total benefits havebeen computed by multiplying the individual WTP by the number of householdsin the municipality of Gorla Maggiore (2,045 households in 2014). Since thepark is opened to anybody, our measure of benefits should be view as a lowerbound of the true value. Whatever the interest rate considered, the scenario P1(green infrastructure & park) provides the highest discounted net benefits. Whenconsidering a discount rate equal to 2%, we get a positive discounted net benefitequal to 5,121 euros.

22

V. Conclusion

A contingent valuation approach has been used to estimate how householdsvalue different multipurpose infrastructures (grey or green) for managing floodrisk and water pollution. As a case study we have considered the Gorla water parklocated north of Milan, in the Lombardy Region of Italy. The Gorla park is a neo-ecosystem recently built on the shore of the Olona River in an area previously usedfor poplar plantation. A novel aspect of our research is that it not only considersthe values people hold for different water ecosystem services (pollution removal,recreative use, biodiversity, flood risk reduction), but also their preferences forhow those outcomes are achieved (through grey or green infrastructures). Theresults indicate that the type of infrastructure delivering the ecosystem services(grey of green) does have an impact on individuals’ preferences for freshwaterecosystem services. By considering the type of infrastructures within the choicemodel, we gain a richer understanding of the relationship between social welfareand freshwater ecosystem services.

We argue that willingness to pay surveys may be useful for regional plan-ning (Vandermeulen, Verspecht, Vermeire, Huylenbroeck, and Gellynck 2011).As demonstrated in our paper, the elicited willingness-to-pay may help decision-makers to prioritise their long term investment decisions. In addition, the surveycan be an important instrument of stakeholder participation in reonal spatialplanning (Wilker and Rusche 2014). In our case, both the representatives of theGorla Maggiore municipality and the Lombardy region have been directly in-volved into the design of the survey and the analysis of the results. Moreover, thecontingent questionnaire has served as a basis for discussions at school in GorlaMaggiore. We believe that both a good understanding of the benefits local popu-lations get for green infrastructures and involvement of local stakeholders in thedecision-process are two important components any welfare-enhancing regionalspatial planning.

REFERENCES

Arrow, K., R. Solow, et al. (1993): “Report of the NOAA panel on contin-gent valuation,” .

Baptiste, A. K., C. Foley, and R. Smardon (2015): “Understanding urbanneighborhood differences in willingness to implement green infrastructure mea-sures: a case study of Syracuse, NY,” Landscape and Urban Planning, 136(0), 1– 12.

Blaine, T. W., F. R. Lichtkoppler, K. R. Jones, and R. H. Zondag (2005):“An assessment of household willingness to pay for curbside recycling: A com-parison of payment card and referendum approaches,” Journal of EnvironmentalManagement, 76(1), 15 – 22.

23

Cameron, T. A. (1987): “The impact of grouping coarseness in alternativegrouped-data regression models,” Journal of Econometrics, 35(1), 37 – 57.

Cameron, T. A., and D. D. Huppert (1989): “OLS versus ML estimationof non-market resource values with payment card interval data,” Journal ofEnvironmental Economics and Management, 17(3), 230 – 246.

Donovan, G. H., and D. T. Butry (2010): “Trees in the city: Valuing streettrees in Portland, Oregon,” Landscape and Urban Planning, 94(2), 77 – 83.

Hanley, N., R. E. Wright, and B. Alvarez-Farizo (2006): “Estimatingthe economic value of improvements in river ecology using choice experiments:an application to the water framework directive,” Journal of EnvironmentalManagement, 78(2), 183 – 193.

Jim, C., and W. Y. Chen (2006): “Recreationamenity use and contingent valu-ation of urban greenspaces in Guangzhou, China,” Landscape and Urban Plan-ning, 75(12), 81 – 96.

Lindhjem, H., and S. Navrud (2011): “Using internet in stated preferencesurveys: a review and comparison of survey modes,” International Review ofEnvironmental and Resource Economics, 5, 309–351.

Lopez-Mosquera, N., and M. Sanchez (2011): “The influence of personalvalues in the economic-use valuation of peri-urban green spaces: An applicationof the means-end chain theory,” Tourism Management, 32(4), 875 – 889.

Mahieu, P.-A., P. Riera, and M. Giergiczny (2012): “Determinants ofwillingness-to-pay for water pollution abatement: A point and interval datapayment card application,” Journal of Environmental Management, 108(0), 49– 53.

Mell, I. C., J. Henneberry, S. Hehl-Lange, and B. Keskin (2013): “Pro-moting urban greening: Valuing the development of green infrastructure in-vestments in the urban core of Manchester, {UK},” Urban Forestry & UrbanGreening, 12(3), 296 – 306.

Mitchell, R., and R. T. Carson (1981): “An experiment in determining will-ingness to pay for national water quality improvements,” .

Mitchell, R. C. (2002): “On Designing Constructed Markets in ValuationSurveys,” Environmental and Resource Economics, 22(1-2), 297–321.

Mitchell, R. C., and R. T. Carson (2013): Using surveys to value publicgoods: the contingent valuation method. Routledge.

Moore, C. C., T. P. Holmes, and K. P. Bell (2011): “An attribute–based ap-proach to contingent valuation of forest protection programs,” Journal of ForestEconomics, 17(1), 35 – 52.

Morrison, M., R. Blamey, and J. Bennett (2000): “Minimising PaymentVehicle Bias in Contingent Valuation Studies,” Environmental and ResourceEconomics, 16(4), 407–422.

Plott, C. R., and K. Zeiler (2005): “The willingness to pay-willingness to

24

accept gap, the “endowment effect”, subject misconceptions, and experimen-tal procedures for eliciting valuations,” The American Economic Review, 95(3),530–545.

Soares, A., F. Rego, E. McPherson, J. Simpson, P. Peper, and Q. Xiao(2011): “Benefits and costs of street trees in Lisbon, Portugal,” Urban Forestry& Urban Greening, 10(2), 69 – 78.

Tuncel, T., and J. K. Hammitt (2014): “A new meta-analysis on theWTP/WTA disparity,” Journal of Environmental Economics and Management,68(1), 175 – 187.

Tzoulas, K., K. Korpela, S. Venn, V. Yli-Pelkonen, A. Kazmierczak,J. Niemela, and P. James (2007): “Promoting ecosystem and human healthin urban areas using Green Infrastructure: A literature review,” Landscape andUrban Planning, 81(3), 167 – 178.

Vandermeulen, V., A. Verspecht, B. Vermeire, G. V. Huylenbroeck,and X. Gellynck (2011): “The use of economic valuation to create publicsupport for green infrastructure investments in urban areas,” Landscape andUrban Planning, 103(2), 198 – 206.

Wang, Y., F. Bakker, R. de Groot, and H. Wortche (2014): “Effect ofecosystem services provided by urban green infrastructure on indoor environ-ment: A literature review,” Building and Environment, 77(0), 88 – 100.

Wilker, J., and K. Rusche (2014): “Economic valuation as a tool to supportdecision-making in strategic green infrastructure planning,” Local Environment,19(6), 702–713.