Assessment of regeneration projects in urban areas of environmental
interest: a stated choice approach to estimate use and quasi-option values
Elisabetta Strazzera∗
DRES and CIREM, University of Cagliari, Italy
Elisabetta Cherchi,
DIT and CIREM, University of Cagliari, Italy
Silvia Ferrini,
DEPFID, University of Siena, Italy, and CSERGE, University of East Anglia, UK
Abstract
This study adopts an attribute-based stated choice approach to evaluate public preferences
over planning alternatives for an urban site of environmental interest. Since such projects
involve some uncertainty and irreversibility, a special attention is devoted to the estimation of
the quasi-option values which are associated to project development. Two distinct measures
for the quasi-option value are estimated, and both coefficients indicate that the public places a
significant value on reduction of the possibility of adverse irreversible effects: a more prudent
development strategy is valued about four times more than a procedure that provides a lesser
hedge against undesired outcomes. Furthermore, the study involved elicitation of
intertemporal preferences over projects with different time spans, and estimation of the
implicit discount rates: the values obtained seem high if compared with standard discount
rates applied to public projects, but not far from interest rates on consumption found in the
market.
∗
Corresponding author: [email protected]. The present work is part of the project “The Economic Valuation in
Urban and Environmental Regeneration Projects” financed by the Italian Ministry of University and Research,
PRIN 2005.
1
1. Introduction
The procedures for land use planning in European Union countries have been
substantially modified after signature of the Århus Convention on June 1998, and the
introduction of the criterion of Strategic Environmental Assessment (SEA, directive
2001/42/EC) as a tool to evaluate land planning actions and enhance the participation level of
local communities to the decision process. Rigorous adoption of the required measures at the
national and local levels would ensure democratic control over decision making, with
beneficial effects also in terms of efficacy of the planning action, since, as discussed by
Beierle and Cayford (2002), empirical evidence shows that active community participation in
the planning process increases its effectiveness.
Implementation of public participation in urban planning is not an easy task, though. Innes
and Booher (2004) evaluate different methods used in recent planning processes, and find that
a convenient procedure is to operate with small task groups, where stakeholders interact with
facilitators and possibly with urban planners to propose and discuss issues and requirements
for the prospective change. The problem is that small task groups may not be sufficiently
representative of the relevant population, especially if the latter is large and heterogeneous.
As discussed by Willis (2006), quantitative surveys of the population at large could be used as
a complementary instrument to assess alternative planning options expressed by task groups.
In this paper we outline a scheme for public participation processes, designed to assess public
preferences over different urban planning options in an area of environmental interest. The
scheme involved two stages: in the first, selected task groups discussed the relevant issues for
the plan, and proposed specific interventions. The outcome from this stage, combined with
2
other technical information, served as a basis to design a few planning options, which were
presented for assessment in the second stage. This stage involved a quantitative survey,
designed to identify the preferred project options in a representative sample of the population.
In particular, we adopted the stated Choice Experiments approach, which is a Stated Choice
method which allows evaluation of specific attributes of the planning options. This technique
has been intensively used in marketing and transport analysis to assess trade-offs between
different attributes of a good or service: the interested reader is referred to Louviere et al.
(2000), Ortúzar and Willumsen (2001) and Hensher et al. (2005) for comprehensive
overviews. Applications to land and urban planning are relatively more recent, but fast
growing. Oppewal et al. (1997) analyse planning choices relative to size and characteristics of
shopping malls; Earnhart (2001) assesses the value of environmental facilities at residential
locations; Scarpa et al. (2001) and Garrod et al. (2002) assess the impact of a traffic calming
program in rural towns in England; Alvarez-Farizo and Hanley (2002) and Bergmann et al.
(2006) evaluate the impact of renewable energy investments on the quality of landscape;
Alberini et al. (2005) analyse different options for remediation and development of
brownfield sites; Willis (2006) estimates the acceptability of different land planning options
subsequent to development of mining activities, Rambonilaza and Dachary-Bernard (2007)
analyse preferences expressed by tourists and residents over different planning actions on a
rural landscape.
The stated Choice Experiments method allows to attach a monetary value to single
components of a project, and to measure marginal rates of substitution (trade-offs) between
different elements of the planning options. In the present work, two different stated Choice
Experiments settings have been designed. The first is intended to eliciting public preferences
for specific project scenarios in an urban area of environmental interest. In addition, we wish
3
to explore if there is an interest for individuals in participating in further steps of the planning
activity: not just having a say on the planning options, but also some control on how the
project is implemented. The second Choice Experiment, more innovative with respect to
previous literature, is designed to assess the citizens’ preferences regarding the procedures
selected for implementation of the project. Since implementation of projects in fragile areas
inevitably involve some uncertainty and irreversibility, a special attention is devoted to the
estimation of the value that the public assigns to the adoption of more precautionary methods
in the implementation of a specified urban planning project. We will adopt two different
characterisations for the value of information (quasi-option value) useful to reduce
uncertainty when undertaking an irreversible action. Moreover, we will attempt to assess
intertemporal preferences over projects with different time spans, in order to provide an
estimate of the implicit discount rates that the citizens attach to a specified public project.
The paper is structured as follows: the next section discusses the concept of quasi-option
value and the issues related to social discounting of public projects; section 3 presents the
case study and a description of the survey; section 4 exposes the econometric methods
employed in this work; sections 5 and 6, respectively, contain results from the econometric
estimates obtained from the two Choice Experiments of our study; finally, section 7 concludes
the paper.
2. Quasi-option values and social discounting for public projects
Once a planning programme is approved, possibly after lengthy negotiations with the
stakeholders, the planning process still has a long way to go: difficulties may occur when
4
ideas are turned into action, and it is possible that eventually the implementation is not
consistent with the plan. A specific problem inherent in urban planning in areas of
environmental interest is that if the development of the project produces undesired effects,
there might be an irreversible loss of environmental values.
In situations where high value sites are at stake, it could be worth to adopt a precautionary
strategy at the development stage. For example, one possible strategy could be to defer
development until all relevant information is gathered in order to reduce uncertainty about the
result. The value of potential future learning on the effects of the development of a project can
be interpreted as a quasi-option value, along the lines of Arrow and Fisher (1974), Henry
(1974a, 1974b), and Hanemann (1989). Zhao and Kling (2004) further build on the notion of
option value and value of information, suggesting the notion of commitment cost, which is the
value of potential future learning on the effects of the development of a project. If the subject
expects that she can learn about the value, then she may choose to wait for more information
before making a decision, and will be willing to pay something less to have the project
developed today rather than next year. Corrigan et al. (2008) empirically test this hypothesis
in a Stated Preference (Contingent Valuation) setting, finding that commitment costs
effectively influence WTP for a program of water quality improvement1.
Of course ex ante information will never be complete, and only after implementation of the
project the effects will actually be revealed. So, as put forth by Fisher and Hanemann (1987),
“if the information about the consequences of an irreversible development action can be
obtained only by undertaking development, this strengthens the case for some development”.
In many situations it is possible to adopt a precautionary strategy by choosing gradual
development, rather than developing all at once. In a different context (climate policy
1
An alternative approach would be to assess willingness to pay to obtain useful information while delaying
development of a project. This is the approach that will be taken in the present work.
5
choices) Ha-Duong (1998) defines a strategy which entails some act to learn as “sequential
decision framework”, as opposed to “one-shot decision framework”. A sequential method of
execution allows learning: in our context, a sequential strategy may involve development of
small parcels of the area. After completion of a section, planners and stakeholders can see if
the results are satisfactory or not, and, if required, revise their procedures for the subsequent
parcel. Also in this case it is possible to think of a value of information, or quasi-option value,
which is associated with the choice of a procedure that allows to obtain useful information
before full development is completed.
Both types of quasi-option values mentioned above will in general entail some additional
costs in the project development procedure: the cost of delay and collection of new
information in one case; the cost of producing at a smaller scale, facing demanding quality
constraints, in the second case. It may be argued that public preferences could guide the
decision makers’ choice when facing a trade-off between cost and security in development, as
well as they should guide decision making in choosing between different planning options. If
the public expresses a concern that precautionary measures should be considered in order to
avoid irreversible undesired effects in the development of the project, it seems that such
preferences should be taken into account. In this circumstance, it would be useful to assess
the value that the public attach to adoption of such precautionary measures, in order to
quantify the additional cost that the community is willing to sustain. The present work
explores this issue, i.e. it aims at eliciting the public willingness to pay for information that
can be obtained through a precautionary development procedure, in terms of the two types of
quasi-option values discussed above.
6
Finally, we observe that land planning projects generate cost and benefits flows in the long
run, and their assessment involves a decision on which procedure should be used to translate
future values into present values (EPA, 2000). After a thorough analysis of the issue, Lind
(1990) concludes that different discount rates should be applied, depending on the specific
program: in some cases the government’s borrowing rate should be applied, in others the
market borrowing rate would be more suitable. Arrow et al. (1996) suggest that the rates
applied in the discount procedure should be dependent on the temporal horizon of the project,
and be based on the rate at which individuals are willing to trade off present for future
consumption, which, in general, will not equal the rate of return in private investments.
However, as reported by Frederick et al. (2002, p.379), a number of empirical and
experimental studies show that the range of rates at which individuals are willing to trade off
present for future consumption can be extremely wide. Moreover, a large number of studies
suggest that short run discount rates are much higher than those employed in the long run (see
Frederick et al., 2002, for a thorough discussion). These findings cast doubts on the validity of
the standard exponential discounting method, and different models have been proposed (e.g.
the hyperbolic model, first axiomatised by Harvey, 1986; the gamma discounting model,
proposed by Weitzman, 2001). An interesting variety of positions regarding the appropriate
model for the social rate of discount to be applied to public programs can be found in Portney
and Weyant (1999), Henderson and Bateman (1995), and Groom et al. (2005). As commented
by Freeman (2003), it seems clear that economists have not yet reached consensus on the
issue.
Many empirical and experimental studies dealing with estimation of discount rates have
focused on intertemporal preferences regarding private goods, either material (consumption
goods) or immaterial (health). Fewer studies have focused on estimation of discount rates
7
related to public projects: a recent application can be found in Viscusi et al. (2008), which
uses Stated Choice experiments to elicit discount values for a project aimed at improving the
quality of water bodies, with effects starting immediately, or after some delay (maximum 6
years). A further aim of the present study is to estimate the implicit discount rate that the
public assigns to investments characterized by time spans usually associated to public projects
(up to 20 years). Our results will hopefully provide some new material for the debate around
the appropriate social rate of discount to be applied to public projects.
3. The case study and the survey
Our application deals with a stretch of seafront (“Poetto Beach”), bordering an area occupied
by saltpans and a lagoon (“Molentargius Marsh”): a major nature conservation area, situated
within the urban structure of the metropolitan area of Cagliari, the capital town of Sardinia,
where pink flamingos nest along with a variety of other bird species.
It is the seafront that largely bears the environmental problems. During the summer season, in
particular, tourists and especially locals flock in droves to the beaches, making an estimated
100,000 trips/day, 60% of which are concentrated in the morning period and only in particular
spots along the seafront. The most critical spot along the entire seafront is the section nearest
the urban centre, which, besides residential housing, also accommodates several commercial
and recreational activities. This intensive anthropogenic pressure is one of the factors
underlying the environmental degradation that afflicts this stretch of coast: erosion of the
sandy shore that in few decades has changed the “face” of the seafront. A restoration project
was carried out in 2002, which included beach sand replenishment, the construction of a new
8
main road and an increase in the number of parking places, but it had disastrous
environmental effects on the beach: the replenishment had altered the quality of the sandy
shore, the white and extremely fine sand being submerged by dark sand with very different
grain size characteristics.
A technical committee appointed by the Regional Government of Sardinia to analyze the
present conditions and evaluate the opportunity of further action, outlined three alternative
planning options for the seafront: 1) action aimed at increasing use values (tourist and
entertainment facilities), 2) action aimed at keeping a balance of use and non use values, 3)
action aimed at increasing non use values (environmental quality). Which choice is best suited
to the case at hand is a matter of public preferences over these alternative options: elicitation
of such preferences is one aim of the present study.
The survey was preceded by an intense preparatory study. The first step involved a qualitative
analysis (one Metaplan and two Focus Groups), which had the purpose to stimulate a debate
around the planning options sketched by the Technical Committee, and match them up with
proposals and issues raised by lay persons and their attitudes toward environmental problems.
The qualitative phase has been fundamental for the design of scenarios and attributes to be
presented in the stated Choice Experiment (CE).
The quantitative survey was composed by two parts. The first section was dedicated to collect
information on socio-economic characteristics and individual habits, as well as to filter the
sample (the stated CE exercise was submitted only to car drivers). The questionnaire collected
information about travel length, time spent to park the car, characteristics of the visit,
frequency of visits in the summer months, other use values attached to the seafront area. The
second part comprised two stated CEs to be submitted to each respondent.
9
First Choice Experiment: CE-Project
The first stated Choice Experiment, named CE-project, was built to elicit preferences over
three planning options characterized by an environmental regeneration of the corridor
alongside the beach, but also a restriction in terms of accessibility by car. The characteristics
of the planning options were summarized in three attributes: environmental quality, extra
parking time and cost to access (Tab.1). Each attribute attains three levels defined as
differences with respect to the actual situation (status quo). The full combination of attributes
and levels described eighty-one different planning options and from those we considered only
the nine main effects choice tasks.2 The final design was randomly divided in three blocks of
three choice tasks each. The number of choice tasks was kept low because a few pilot studies
made it clear that interviews too long would induce weariness, as individuals were asked to
work through two different CE sets, and a lengthy questionnaire regarding personal
characteristics and activities performed. Table 1 illustrates the attributes used and their levels.
***INSERT TABLE 1. ABOUT HERE
In particular, the environmental quality attributes levels were defined as follows:
•
The Urban Damaged scenario corresponds to the Status Quo option, which did not
include any improvement with respect to the current situation.
2
The effect of single factors on the responses is called “main effects” and the key assumpion is that only the
attributes and not their interaction are relevant for the model. We can demonstrate that this design is optimal for
linear models and further discussion regarding non-linear models can be found in Ferrini and Scarpa (2007).
10
•
At the Urban Regenerated level the road will be asphalted anew, a tiled sidewalk will
be constructed; both private and public transport will be allowed but car parking will
be restricted to only one side of the road.
•
At the Intermediate level the road and the pedestrian sidewalk will be paved with
ecological (non petroleum-based) materials; private motor vehicles would not be
allowed, while public transport will still be available along the promenade; and
bicycles will have a reserved track.
•
The Natural scenario entails removal of both private and public traffic, which are
displaced to a parallel avenue; a wooden walkway; a dirt road for bikers; more
vegetation along the promenade.
All regeneration scenarios, alternative to the Status Quo option, include a regular maintenance
and security service.
To measure the value of car access in the area, two types of cost were used: the first is
measured in time, i.e. the additional time required for parking the car (additional with respect
to the time currently experienced by each individual); the second is measured in money unit
(Euro), and it is the road price that car drivers are required to pay to enter the marine area
(which comprises not only the seafront, but also the avenue parallel to the seafront, and the
connecting roads where access and parking would still be available under all planning
options). The road pricing was selected as our economic instrument instead of other options
(such as a local tax or a park fee) for several reasons: in particular, it can be easily associated
to the damage produced by the cars on the environment of the entire area; it is an out-ofpocket cost, hence directly associated to each specific trip; it is independent of the duration of
the stay and of the specific location of the parking area; and finally, the road pricing is a
11
relatively novel form of payment which has been successful to alleviate congestion problems
in some city centers (London is a well known example, recently Milan has followed suit).3
Since all examined projects implied some reduction of parking slots along the beach
promenade, it was important to take into account that this would have increased the time to
find a parking space. Usually drivers start looking for a parking space as close as possible to
the beach, and only if they fail they move away from the promenade. This motivated the
inclusion of the attribute “time to park” as an additional cost attached to the realization of a
project.
The CE-project exercise was presented as a binary choice between the status quo and one
project alternative. In particular, the scenario variable was defined throughout images (one for
each level) created with a rendering technique. This has been found to improve significantly
the comprehension of the urban/environmental scenario tested. More details on the images
used in the CE can be found in Cherchi and Strazzera (2008).
Second Choice Experiment: CE-Implement
The second stated Choice Experiment, named CE-implement, deals with different
implementation modes for a given project. This second exercise was reserved to respondents
who chose an intervention project in at least one of the first stated CE task. One scenario,
among those chosen in the first exercise, was selected, and the individual was asked to bear it
in mind as a reference in answering to the following questions about the implementation
procedures4. This scenario represented the Base procedure.
3
Testing the acceptability of the road pricing instrument in the context of our application was another aim of the
study.
4
The enumerators were instructed to select the environmental quality attribute (Regenerated Urban,
Intermediate, or Natural) which in the first set of exercises was associated to the highest monetary cost attribute,
12
Respondents could choose between the Base procedure or the Improved option, characterized
by the following attributes:
•
Control: a one-shot procedure that did not allow any control over the quality of the
work –and its correspondence with the approved project, versus a sequential mode that
allows such control and possibility to correct an unsatisfactory development
procedure;
•
Wait information: a procedure that involves immediate execution of the selected
project, versus one that delays operation by one year in order to allow gathering of
further information (e.g. technical investigations and public hearings) potentially
useful to improve the implementation of the project;
•
Duration: a procedure guaranteeing 10 years of maintenance, rather than 15, or 20
years of maintenance.
Finally, the cost (road price) associated to each implementation mode varied between €0.50
(associated only to the Base procedure) and €1.00, €1.50 or €2.00 associated to improved
implementation modes.
The Base procedure is characterized by no Control, no delay to gain information, 10 years of
maintenance and €0.5 of access cost. Combining the attributes and levels indicated in Table 2
we defined the Improved procedure option used in the second valuation task.
****INSERT TABLE 2 ABOUT HERE
if the project option was chosen; and ask the respondent to consider it as the project to be implemented. In the
Base procedure of the second set of exercises, the cost attribute is always a rebate of the amount that was
accepted to be paid in the first set.
13
Again, from the full combination of attributes and levels only the main effects were
considered selecting a total of 9 choices tasks randomly divided in three groups. Each
respondent received three choice sets; in each choice situation the respondent had to compare
the base procedure with an alternative option.
Both CE tasks were tested in several pilot studies. Each pilot study used a sample of 20-25
visitors in order to verify that the CE questions were understandable and estimable, and to test
the description of the attributes and the levels employed. The valuing exercises were further
controlled in four pre-test surveys, on samples of about 50 individuals each.
The main survey was administered through in-person interviews at destination (i.e. to people
in the beaches) in August 2006. The sample, randomly chosen among people who drove a car
to reach the beach area, consisted of 500 respondents who completed the socio-economic
section of the questionnaire and participated in the first CE-project exercise. As one fifth of
the respondents either chose the status quo in all the choice tasks, or never chose any option,
400 individuals participated in the second CE task, named CE-implement. .
In the following we report some descriptive statistics of the sample characteristics, while the
results of the CE exercises, as well as the results of the models estimated, will be discussed in
section 3.
The sample (500 individuals) is mainly composed by males (63.5%), heads of the family
(67.7%), active (66.6%), mainly as employees (81.7% of the active people). The age of the
sample is distributed between 19 and 84 years: 11.9% is younger than 30, 35.8% is between
31 and 45 years old, 40.8% between 46 and 65, and 11.5% older than 65. The variable
education is distributed as follows: 28% of the sample has primary education, 52% secondary
14
education and just 20% a higher education level. The sample is mainly composed by people
of average income. Excluding the interviewees (37%) who do not provide any answer, in the
residual sample 18.8% say they earn less than €1000/month, 70.5% between €1000 and
€2000, and 10.7% more than €2000. Analogous results were obtained at family level. The
percentage of people who do not provide an answer on the family income is about 21%.
Among those who give a response, less than 10% declare a family income higher than €3,500
(note that on average there are approximately 3 members per household).
As expected, since our sample is composed only by car drivers, 96.2% of the interviewees
own a car, which in the 40% of the cases is the only car available in the family, while 50.4%
of the families own 2 cars, and just the 9.6% own 3 or more cars.
The majority of individuals live in the metropolitan area (80% of the interviewees travelled by
car less than 20 minutes, 15% between 20 and 40 minutes and 5% more than 40 minutes),
hence the travelling cost by car for the specific trip to the beach is generally low: 64.8% of the
respondents pay less than €1, 21.8% between €1 and €2, and 13.4% more than €2.
Finally, and maybe more interestingly for the present work, we note that 81.6% of the
respondents said that they found a parking space very close to the beach so that they had to
walk from the parking space to the final destination a “perceived” time of four minutes or
less; another 16.6% indicated a walk time of exactly five minutes; the remaining 1.8% said
they walked 6 minutes or more. Moreover, for the large majority of respondents (81%) the
“perceived” time to find a parking space for their car is approximately zero: it seems that,
when the survey was administered, the offer of parking space was more than sufficient, even
in a situation in which substitution between transport modes was almost absent. It can be
noticed that 94.8% of the car users declared that they use always their car to go to the beach
area. Among the 5% of the sample who sometimes use some other transport modes, a 50%
15
uses a motorbike, a 30% uses the bus and only 15% a bicycle. In line with these results, 92%
of the interviewees has never used the bus in the current season to go to the beach, and more
than 85% has never considered the possibility of going to the beach by bus.
4. The econometric models
Following the classical formulation of discrete choice models (Domencich and McFadden,
1975), individuals are assumed to choose among several available options on the basis of an
index of preference (called utility) that depends on the vector of specific characteristic of
alternative j and individual q:
'
U qj = U ( X qj )
(1)
The formation of individual preferences typically rely on compensatory rules5, so that there is
'
a trade-off among the different characteristics ( X qj ) depending on their relative importance
(θ qj ) . As modellers are able to observe only a subset ( X qj ⊂ X qj' ) of the vector of real
'
attributes (Manski, 1977; McFadden, 1981), the random utility can be rewritten as:
U qi = U qi ( X qi ,θ qi ,ε qi )
(2)
where εqj is the error term, representing heterogeneity sources which are not explicitly
included in the utility function.
A widely accepted hypothesis to operationalize equation (2) is that random utility can be
treated as the sum of the systematic, representative or observable part (Vqj), which is a
function of the attributes Xqj, and the random component, such that utility is given by:
U qj = Vqj ( X qi ,θ qi ) + ε qj
5
(3)
A renewed interest for non-compensatory models have emerged in the recent literature. See for example
Cantillo and Ortúzar (2005)
16
To derive a specific discrete choice model, instead, assumptions on the error terms
distribution are necessary because in general we can write:
(
p qj = prob (U qj ≥ U qi ) = prob Vqj − Vqi ≥ ε qi − ε qj
)
∀ i Ai ∈ A(q), i ≠ j
(4)
this can be summarised into:
pqj =
∫
f (ε q )d ε q
RN
ε qi ≤ ε qj + Vqj − Vqi ∀i Ai ∈ A(q ), i ≠ j
RN = 
Vqj + ε qj ≥ 0
(5)
and different discrete choice models will be obtained, depending on the distribution of ε:
multinomial logit, MNL (Domencich and McFadden, 1975); nested logit, NL (Williams,
1977; Daly and Zachary, 1979); multinomial probit, MNP (Daganzo, 1979); mixed
multinomial logit, MMNL (Ben-Akiva and Bolduc, 1996; McFadden and Train, 2000; Train,
2003), GEV model framework (e.g., Bierlaire, 2001; Daly, 2001; Wen and Koppelman, 2001,
Daly and Bierlaire, 2006; Bierlaire et al., 2008).
As well known, the underlying hypothesis of the MNL is that the error components are
independent and identically distributed Extreme Value type 1 (EV1) with zero mean and
variance (σ 2 = π 2 6λ 2 ) depending only on λ, the scale parameter of the distribution, constant
over individuals and alternatives. The MNL is the simplest discrete choice model available
but the assumptions of Independence of Irrelevant Alternatives, homoskedasticity,
heterogeneity in preferences and correlation over choices and time, may prove unsatisfactorily
in real contexts. Nevertheless it is crucial as a basis for comparison. More general, but also
complex, models that allow to account for the above effects are the mixed multinomial logit
and the probit multinomial model.
17
The probit model is based on the assumption of Normal distributed errors. The complexity of
this formulation resides in the fact that the integral distribution of the difference of two
normal variables does not have a closed form, so the Probit probability is expressed as an
integral and required simulation methods to be solved. However, as the Normal distribution is
closed to addition and subtraction, in the case of binary choices (as in our case), the Probit
model becomes easily manageable because it reduces to the following closed form:
pq1 = Φ [ (V1 − V2 ) / σ ε ]
(6)
Where σ ε2 = σ 12 + σ 22 − 2 ρσ 1σ 2 is the univariate distribution (N(0, σε)) of the difference
between two Normal univariate variables N(0, σ1) and N(0, σ2), and Φ(⋅) is the cumulative
standard Normal distribution which has tabulated values. Under the further assumption of iid
Normal distribution (i.e. σ1 = σ2, and ρ = 0) the iid Probit model is obtained where σ ε2 = 2σ 2 .
It is important to remark that, while the iid probit model is homoskedastic, the general version
of the binary probit, although simple, allows to handle any temporal correlation pattern and
unobserved factors that are correlated over time or choices, as required when multiple
observations are available per individual. In particular, in the panel effect probit model the
error component (ηqj) is specified as the sum of an effect (εqj) independent over individuals
and alternatives and an effect specific for each alternative υj, that induces correlation across
observations of the same individual:
µqj = υ j + ε qj
where εqj
(7)
and υj are independent normal random variables, with the proportional (ρ)
contribution of the panel data component υj to the total variance equal to:
σ 2j
ρ= 2
.
σ j + σ qj2
(8)
18
where σ 2j is the variance of the individual effect υj and σ qj2 is the variance of the independent
effect εqj. It is important to note that if σ 2j is small compared to σ qj2 , ρ goes to zero, indicating
that the variance associated with the individual effect is relatively unimportant and therefore
there is no significant difference between the random effect and the standard probit model.
A more general specification for panel data can be obtained with the mixed multinomial logit
specification: the error term εqj is assumed iid EV1 distributed, while the second error term
(υj) can be chosen by the modeller among a range of distributions (Normal, Lognormal,
Normal Censored or Truncated, SB and so on), depending on the phenomenon s/he needs to
reproduce (Train and Sonnier, 2005). When the error term εqj is assumed to be distributed as a
Normal, the mixed logit specification is equivalent to the random effect probit model, with the
coefficients scaled by a constant.
The Mixed Multinomial Logit in its most general form is represented as follows:
∫
Pqi = Lqi ( δ ) f ( δ ;θ )dδ
(9)
where Lqj (δ ) is the kernel logit of individual q choosing alternative j, evaluated at parameters
δ, and f ( ⋅;θ ) is a density function, with parameters θ over the population, chosen by the
modeller. There are two alternative interpretations, random parameters and error components.
In the error components specification, the individual preference parameters are fixed, and the
error term has a mixed structure. In the random parameters model, one or more individual
preference parameters can be modelled as random variates, with possibly different
distribution functions, in order to account for preference heterogeneity across individuals6. As
demonstrated by McFadden and Train (2000), the Mixed Logit model can approximate any
6
Empirical identification problems may arise if there is not enough variability of the data among alternatives,
see Cherchi and Ortùzar (2008).
19
discrete choice models at any desired level of accuracy. All the econometric models described
above are usually estimated through maximum likelihood (ML) or maximum simulated
likelihood (MSL).
As well known, estimated coefficients can be used to derive the marginal rates of substitution
between attributes (or willingness to pay, WTP), which is given by the ratio between the
marginal utility of the attribute s and the marginal utility of the cost:
WTP ( s ) =
∂Vqj / ∂sqj
∂Vqj / ∂cqj
(10)
It is important to note that when the utility function is linear in income, the marginal utility of
the cost is equal to minus the marginal utility of income; hence, a change of one attribute from
one level to another can be valued in terms of Hicksian income variations, like compensating
or equivalent variation. Moreover, in the linear-in-the-parameter and in-the-attribute utility
function, a consistent estimator of individual willingness to pay for the attribute is simply
obtained replacing the unknown coefficients with corresponding estimators.
When a random parameters model is applied, the WTP is obtained as a ratio of random
variables. A simple case is obtained when the cost attribute is held fixed: in this case, the
resulting WTP for an attribute with random coefficient follows the same distribution of the
random coefficient (Revelt and Train, 1998). If also the price coefficient is a random variable,
the derivation of the WTP distribution is more complex, being the ratio of two random
variates, and care must be taken in the interpretation of the results (see Meijer and
Rouwendal, 2006; Hensher and Greene, 2003; Cherchi and Polak, 2005; Sillano and Ortúzar,
2005).
20
5. Estimation results from CE-project
This section reports the results of the model estimation using the first Choice Experiment data
set that refers to the choice of an urban project. Tables 3.a, 3.b, and 3.c report the frequency of
choices of scenarios, by level of monetary cost associated to the option.
****INSERT TABLES 3A, 3B, 3C ABOUT HERE
It seems quite clear that the Urban scenario is the least preferred, with more people choosing
the Status Quo option with respect to the other scenarios, and especially many more
respondents equally (un)satisfied by either option. Frequencies of choices over the other two
scenarios are fairly equally distributed, with a slight preference for the Intermediate scenario.
Obviously, the response when choosing between the Status Quo and the Project option was
also influenced by the other two attributes of the choice, i.e. the monetary (Road Price) and
time (additional time spent to search a parking spot) costs. It can be observed that for most
scenarios the acceptance of the Project option decreases as the cost increases, and that the
“Neither” option is selected more often when intervention is associated to higher levels of
cost.
Next, we show the estimates of five alternative models for the choice between keeping the
Status Quo situation or developing a Project. The cases where the individual did not choose
any option were treated as “protest” responses and removed from the sample7. All models are
estimated using only the choice attributes, i.e. the type of scenario, the additional time spent
7
Removal of protest responses may give rise to sample selection problems, and in this case a sample selection
model should be used to correct for selectivity bias (see Strazzera et al., 2003a and 2003b). We did not find any
significant selectivity effect in this data.
21
to park, and the road price8. The first model is a “pooled” Probit model, i.e. it is assumed that
all observations are independent. However, this assumption may be unwarranted, since one
individual can generate up to three observations, and unobserved individual effects may
occur. Actually, if we compare the specifications with and without panel effect, the former are
clearly superior: a Likelihood Ratio test applied to compare the Panel Probit with the Pooled
Probit, and the Mixed Logit with the Pooled Logit reject the restricted models. In particular,
column 4 reports the estimates obtained from a Panel Random Effect Probit, where the
contribution of the panel data component to the total variance is given by the correlation
coefficient; while column 5 reports the results from the Panel Mixed Logit estimation, where
the standard deviation of the alternative specific error component is estimated to explicitly
account for the correlation among the observations of the same individual. It is interesting to
note that although the Panel Probit and the Mixed Logit provide different coefficient
estimates (because of the different scale implicit in the two model structures, see Ortúzar and
Willumsen, 2001), they give the same log-likelihood value and WTP estimates (see Table 5),
as the WTP is scale free.
****INSERT TABLE 4 ABOUT HERE
The WTP estimates obtained from the two pooled models are identical, and a bit lower than
those obtained from the panel specifications, especially the WTP for the Regenerated Urban
scenario attribute. All models indicate that the Intermediate scenario is the one valued most,
8
Other models with covariates inserted as interaction terms with alternative specific constants and attributes
were estimated for both experiment sets. These models do not improve the estimates of the attribute coefficients,
and for brevity are not reported here. Some results can be found in Strazzera et al. (2008) and Cherchi and
Strazzera (2008).
22
followed close by the Natural scenario, while the value attached to the Urban scenario is
significantly lower.
****INSERT TABLE 5 ABOUT HERE
6. Estimation results from CE-implement
If the interviewee in the first exercise selected at least one intervention project, a second
exercise was proposed to elicit preferences over different ways to implement the project, as
was described in Table 2. The individuals’ sample size at this stage is 400, since exactly 100
respondents never chose any intervention option in the first exercise. Also in this exercise in
some cases individuals did not select any of the two alternatives proposed, being equally
unsatisfied by either option: hence, the number of valid observations is 1084. The following
tables show the frequencies of responses obtained across choices and attributes.
****INSERT TABLES 6A, 6B. AND 6C. ABOUT HERE
Just as with the CE-Project data, we first estimate two simple “pooled” Logit and Probit
models, which are reported in columns two and three of Table 7. The estimates from these
two models are equivalent, producing the same WTP estimates, as reported in Table 8. A
Panel Probit with random effects is then estimated to account for correlation among responses
obtained from the same individual, and the hypothesis of correlation is supported by a
Likelihood Ratio test. In column five we report the Mixed Logit specification analogous to
the Panel Probit; again, the Likelihood Ratio test rejects the restricted Logit model.
23
Unfortunately, also the Panel specifications do not provide a precise estimate for the Duration
attribute coefficient, which is not significant, so we adopt a different specification of the
Mixed Logit model, fitting a Random Parameter Logit where all parameters, but the Cost
parameter, are modelled as random Normal variables. The last column in Table 7 reports the
estimates of this model, where the mean coefficient is significant at 10% (P-value: 0.078); the
other coefficients are all statistically significant at 1%. The Likelihood Ratio test can be used
to select between the Random Parameter model and the Logit model (the restricted
specification is rejected), while the two Mixed Logit models are not nested and their
likelihoods are not directly comparable. In the Random Parameter specification the coefficient
of the Cost attribute is held fixed to allow simple calculation of the WTP: when the price
coefficient is fixed, and the attribute coefficient is distributed as a Normal, the resulting WTP
estimate for a random parameter is a random variable, which follows a Normal distribution.
Its mean is given by the ratio of the mean estimated coefficient and the price coefficient; and
the standard deviation is the ratio of the standard deviation of the estimated coefficient, and
the price coefficient (Revelt and Train, 1998). Therefore, the WTP for the Control attribute is
distributed as a Normal (µ: 1.38; σ2: 4.64); the WTP for the Wait Information attribute is a
Normal (µ: 0.73; σ2: 2.34); and the WTP for the Duration attribute is a Normal (µ: 0.04; σ2:
0.05).
****INSERT TABLE 7 ABOUT HERE
Table 8 reports the WTP estimates for the three attributes of the CM-implement exercise. The
estimates differ slightly across models, but it is quite clear that our respondents place a
significant value on the two quasi-option variables –especially the Control attribute.
24
****INSERT TABLE 8 ABOUT HERE
The value attached to the attribute Duration is estimated as €0.04 by the Random Parameter
Logit model. We choose this value for calculation of the implicit discount rate, since it was
derived from a relatively more significant estimate of the attribute coefficient. The result is
interpreted as follows: for each additional year of maintenance, in the range from 10 to 20
years, an individual is just willing to pay 4 cents more on top of the 50 cents required to
access the area by car.
To calculate the implicit discount rate we follow a procedure adopted in Keller and Strazzera
(2002). We should find the r that solves:
T
T
PBt
PI t
=
∑
∑
t
t
t =1 (1 + r )
t =1 (1 + r )
(16)
where PB is the flow of net benefits –benefits minus costs– under the Base implementation
mode, PI is the corresponding flow under the Improved mode, and T is the time span of
payments. The consumer’s benefits of the maintenance services are measured in terms of
WTP related to each implementation mode, and they last 10, 15, or 20 years; while the costs
are measured as the price due for the number of years in which the individual is expected to
keep visiting the area, since the introduction of the access charge. For example, for T=30,
solving this polynomial equation gives us a discount rate r=0.23 for a period of 15 years of
maintenance, and r=0.27 for 20 years. While these discount rates are certainly higher than the
current official interest rate, they seem compatible with discount rates accepted by individuals
in ordinary consumption choices: the Bank of Italy Report indicated for the year 2005, when
25
the survey was administered, an average interest rate of 16% for credit cards, and an average
20% interest rate for a popular financial service aimed at employees.
7. Conclusions
The aim of the present research was twofold. First, we were interested in analyzing public
preferences over alternative planning choices, characterized by different use values and
environmental quality. A stated Choice Experiment approach was taken in order to evaluate
the rates of substitution across different attributes (project scenarios, monetary and time
costs). The second objective was to analyse the attitude of the public to participate not only in
the evaluation of alternative project options, but also in the assessment of the development
procedure to be chosen for the selected project. This is an aspect that is often overlooked in
discussions on democratic participation in planning decisions, but in our opinion it can be an
important tool to help control on the way the planning decisions are realized in practice.
The estimation results have shown that the citizens have a clear preference ranking of the
alternative planning options proposed: the preferred scenario, i.e. the Intermediate scenario,
which involves some improvement in environmental quality (no private traffic on the
promenade) while enhancing some use values (public transport, bike track) is valued about
50% more than the least preferred scenario. Moreover, the results of our survey show that
there is indeed a strong interest from the public in ensuring that more careful and conservative
methods should be selected, even though they are more expensive: the stakeholders would, on
average, be willing to pay as much as four times (about €2, summing up the two quasi-option
values), of the amount that would be paid for a development procedure that provides a lesser
hedge against irreversible effects in a site of high environmental value.
26
Finally, the implicit social discount rate was estimated, at values relatively high if compared
to standard discount rates applied to public projects (7%-13%, depending on specific cases,
and if in developed or developing countries), but in line with interest rates on consumption
found in the market in the period immediately preceding the survey. This result is helpful in
signalling that the preferences elicited through our stated Choice Experiment exercises are
consistent with actual market behaviour. It remains to be seen if these consumption interest
rates are indeed good candidates for use in cost-benefit analysis of public projects.
References
Alberini, A., Longo, A., Tonin, S., Trombetta F., Turvani M., 2005. The role of liability,
regulation and economic incentives in brownfield remediation and redevelopment:
evidence from surveys of developers. Regional Science and Urban Economics 35(4),
327-351.
Alvarez-Farizo, B. and Hanley, N., 2002. Using conjoint analysis to quantify public
preferences over the environmental impacts of wind farms, Energy Policy 30(2), 107–
116.
Arrow, K.J., Cropper, M.L., Eads, G.C., Hahn, R.W., Lave, L.B., Noll, R.G., Portney, P.R.,
Russell, M., Schmalensee, R., Smith, V.K., Stavins, R.N., 1996. Is There a Role for
Benefit-Cost Analysis in Environmental, Health, and Safety Regulation? Science 272,
221-222.
Arrow, K.J., Fisher, A.C., 1974. Environmental preservation, uncertainty, and irreversibility.
Quarterly Journal of Economics 88, 312-319.
27
Beierle, T., Cayford, J., 2002. Democracy in Practice. Public Participation in Environmental
Decisions. Washington, DC: Resources for the Future Press.
Ben-Akiva, M.E., Bolduc, D., 1996. Multinomial probit with a logit kernel and a general
parametric specification of the covariance structure. Working Paper, Dept. of Civil
Engineering, MIT.
Bergmann, A., Hanley, N. , Wright, R., 2006. Valuing the attributes of renewable energy
investments. Energy Policy, 34(9), 1004-1014.
Bierlaire, M., 2001. A general formulation of the cross-nested logit model, Proceedings of the
1st Swiss Transport Research Conference, Monte Verita, Ascona, Switzerland.
Bierlaire, M., Bolduc, D., McFadden, D., 2008. The estimation of generalized extreme value
models from choice-based samples. Transportation Research B 42(4) , 381–394.
Cantillo, V., Ortúzar, J. de D., 2005. A semi-compensatory discrete choice model with
explicit attribute thresholds of perception. Transportation Research 39B, 641-657.
Cherchi, E., Ortúzar J. de D., 2008. Empirical identification in the mixed logit model:
analysing the effect of data richness. Network and Spatial Economics 8, 109-124.
Cherchi, E., Polak, J. W., 2005. The assessment of user benefits using discrete choice models:
implications of specification errors under random taste heterogeneity. Transportation
Research Record 1926, 61-69.
Cherchi, E., Strazzera, E., 2008. La valutazione economica di interventi di restrizione all’uso
dell’auto mirati alla riqualificazione urbana. In: Astarita, V., d’Elia, S. Festa, C., (Eds.),
Interventi e metodologie di progetto per una mobilità sostenibile. Franco Angeli (in
press).
28
Corrigan, J.R., Kling, C.L. Zhao J., 2008. Willingness to Pay and the Cost of Commitment:
An Empirical Specification and Test. Environmental and Resource Economics 40(2),
285-298.
Daganzo, C.F., 1979. Multinomial Probit: The Theory and Its Applications to Demand
Forecasting. Academic Press, New York.
Daly, A., 2001. Recursive nested extreme value model. Working Paper, Institute for Transport
Studies, University of Leeds.
Daly, A., Bierlaire, M., 2006. A general and operational representation of generalised extreme
value models. Transportation Research B 40(4), 285–305.
Daly, A., Zachary, S., 1979. Improved multiple choice models. In: Hensher D., Dalvi, Q.,
(Eds.), Identifying and measuring the determinants of mode choice. Teakfield, London.
Dixit, A.K., Pindyck, R.S., 1994. Investment Under Uncertainty. Princeton, NJ: Princeton
University Press.
Domencich, T., McFadden, D., 1975. Urban Travel Demand - A Behavioural Analysis. North
Holland, Amsterdam.
Earnhart D., 2001. Combining Revealed and Stated Preference Methods to Value
Environmental Amenities at Residential Locations. Land Economics 77(1), 12-29.
EPA, 2000. Guidelines for Preparing Economic Analyses, U.S. Environmental Protection
Agency.
Ferrini, S., Scarpa R., 2007. Designs with a priori information for nonmarket valuation with
choice experiments: A Monte Carlo study. Journal of Environmental Economics and
Management 53(3), 342-363.
Fisher, A.C., Hanemann, W., M., 1987. Quasi-option value: Some misconceptions dispelled.
Journal of Environmental Economics and Management, Elsevier 14(2), 183-190.
29
Frederick, S., Loewenstein, G., Donoghue, T.O., 2002. Time Discounting and Time
Preference: A Critical Review. Journal of Economic Literature XL, 351-401.
Freeman, A.M., 2003. The measurement of environmental and resource values: theory and
methods. Resources for the Future Press, Washington, DC.
Garrod, G., Scarpa, R., Willis, K., 2002. Estimating the benefits of traffic calming on through
routes: A choice experiment approach. Journal of Transport Economics and Policy 36 (2),
211-231.
Groom, B., Hepburn, C., Koundouri, P., Pearce, D., 2005. Discounting the future: the long
and the short of it. Environmental and Resource Economics, 32, 445-493
Ha-Duong, M, 1998. Quasi-option value and climate policy choices. Energy Economics, 20,
599-620.
Hanemann, W.M., 1989. Information and the concept of option value. Journal of
Environmental Economics and Management 16, 23-37.
Harvey, C.M., 1986. Value Functions for Infinite-Period Planning. Management Science
32(9), 1123–1139.
Henry, C., 1974a. Investment Decisions under Uncertainty: The ‘Irreversibility Effect.
American Economic Review 64, 1006–1012.
Henry, C., 1974b. Option Values in the Economics of Irreplaceable Assets. Review of
Economic Studies Symposium 196, 89–104.
Hensher, D., Rose, J., Greene, W., 2005. Applied choice analysis: A primer. Cambridge
University Press, Cambridge, UK.
Hensher, D.A., Greene, W.H., 2003. The Mixed Logit Model: The State of Practice.
Transportation 30, 133-176.
30
Krinsky, I., Robb, A.L., 1986. On Approximating the Statistical Properties of Elasticities.
Review of Economics and Statistics 68, 715–19.
Innes, J., Booher, D., 2004. Reframing public participation: strategies for the 21st century.
Planning Theory & Practice 5(4), 419 – 436.
Keller, R.L., Strazzera, E., 2002. Examining predictive accuracy among discounting models.
Journal of Risk and Uncertainty 24 (2), 143–160.
Lind, R.C., 1990. Reassessing the Government's Discount Rate Policy in Light of New
Theory and Data in a World Economy with a High Degree of Capital Mobility. Journal of
Environmental Economics and Management 18, S8-S28.
Louviere, J.J., Hensher, D.A., Swait, J.D., 2000. Stated Choice Methods: Analysis and
Application. Cambridge University Press, Cambridge.
Mansky, C., 1977. The structure of random utility models. Theory and Decisions 8, 229-254.
McFadden, D., 1981. Econometric Models of Probabilistic Choice. In: Manski C., McFadden
D., (Eds.), Structural Analysis of Discrete Data with Econometric Applications.
Cambridge, MA: MIT Press.
McFadden, D., Train, K., 2000. Mixed MNL Models for Discrete Response. Journal of
Applied Econometrics 15(5), 447-470.
Meijer, E., Rouwendal, J., 2006. Measuring welfare effects in models with random
coefficients. Journal of Applied Econometrics 21(2), 227-244.
Mensink, P., Requate, T., 2005. The Dixit-Pindyck and the Arrow-Fisher-Hanemann-Henry
Option Values are not Equivalent. Resource and Energy Economics 27(1), 83-88.
Oppewal, H., Timmermans H., Luoviere I.J., 1997. Modelling the effects of shopping centre
size and store variety on consumer choice behaviour. Environment and Planning A 29(6),
1073-1090.
31
Ortúzar, J. de D., Willumsen, L.G., 2001. Modelling Transport. John Wiley & Sons,
Chichester.
Portney, P.R., Weyant, J.P. (Eds.), 1999. Discounting and Intergenerational Equity. Resources
for the Future, Washington, DC.
Rambonilaza, M., Dachary-Bernard J., 2007. Land-use planning and public preferences: What
can we learn from choice experiments method?, Landscape and Urban Planning, 83,
318-326.
Revelt, D., and K. Train, 1998. Mixed Logit with Repeated Choices: Households' Choices of
Appliance Efficiency Level. Review of Economics and Statistics 80 (4), 647-657.
Scarpa, R., Willis, K., Garrod, G., 2001. Estimating benefits for effective enforcement of
speed reduction from dichotomous-choice CV: the case of rural trunk routes.
Environmental and Resource Economics 20 (4), 281-304.
Sillano, M., Ortúzar, J. de D., 2005. Willingness-to-pay estimation with mixed logit models:
some new evidence. Environment and Planning 37A, 525-550.
Strazzera E., Cherchi, E., Ferrini, S., 2008. A Choice Modelling Approach for Assessment of
Use and Quasi-Option Values in Urban Planning for Areas of Environmental Interest,
Fondazione Eni Enrico Mattei Note di lavoro, 63/2008.
Strazzera, E., Genius, M., Scarpa, R., Hutchinson G. 2003a. The effect of protest votes on the
estimates of willingness to pay for use values of recreational sites. Environmental and
Resource Economics 25(4), 461-476.
Strazzera, E., Scarpa, R., Calia, P., Garrod, G., Willis, K. 2003b. Modelling zero values and
protest responses in contingent valuation surveys, Applied Economics, 35(2), 133-138.
Train, K., 2003. Discrete Choice Methods with Simulation. Cambridge University Press.
32
Train, K., Sonnier, G., 2005. Mixed logit with bounded distributions of correlated partworths.
In: Scarpa, R., Alberini, A. (Eds.), Applications of Simulations Methods in
Environmental Resource Economics. Kluwer Academics Publisher, Dordrecht, The
Netherlands.
Viscusi, W.K., Huber J., Bell J., 2008. Estimating discount rates for environmental quality
from utility-based choice experiments. Journal of Risk and Uncertainty 37, 199–220.
Weitzman, M., 2001. Gamma Discounting. The American Economic Review 91, 260-271.
Wen, C., Koppelman, F., 2001. The Generalized Nested Logit Model”. Transportation
Research B 35(7), 627-641.
Willis, K.G., 2006. Assessing Public preferences. The use of stated preference methods to
assess the impact of varying planning conditions. Town Planning Review 77(4), 485-505.
Zhao, J., Kling, C.L., 2004. Willingness to Pay, Compensating Variation, and the Cost of
Commitment. Economic Inquiry 42, 503-517.
33
Table 1. Attributes levels for CE-project
Status quo
Environmental quality
Scenario
Extra parking time respect to the
experienced in the current trip
Out-of-pocket Cost to access the
environmental area
Urban
Damaged
Project
Urban Regenerated
Intermediate
Natural
0 min
0 min
2 min
5 min
0.50 €
0€
1.00 €
2.00 €
Table 2. Attributes levels for CE-implement
Base
Control over the implementation
Wait information
Duration of the project
implementation
Out-of-pocket Cost to access the
environmental area
No
0 yrs
Improved
No
Yes
0 yrs
1 yrs
10 yrs
10 yrs
15 yrs
20yrs
1.00€
0.50 €
1.50€
2.00€
34
Table 3.a Scenario Urban Regenerated
choice
Cost
Total
S-Quo
Project
Neither
0.50€
52
84
31
167
1.00€
56
66
45
167
2.00€
59
62
45
166
167
212
121
500
Total
Table 3.b Scenario Intermediate
choice
Cost
Total
S-Quo
Project
Neither
0.50€
26
127
13
166
1.00€
40
102
26
168
2.00€
56
79
31
166
122
308
70
500
Total
Table 3.c Scenario Natural
choice
Cost
Total
Total
S-Quo
Project
Neither
0.50€
34
115
18
167
1.00€
39
113
15
167
2.00€
52
71
43
166
125
299
76
500
35
Table 4. Results of Models estimated for the CE-project.
Attribute
Urban
Intermediate
Natural
Cost
Park time
Pooled
Panel
Logit
Probit
Probit
1.016***
0.620***
1.339***
2.334***
(0.173)
(0.105)
(0.207)
(0.340)
1.723***
1.054***
2.063***
3.610***
(0.182)
(0.108)
(0.234)
(0.393)
1.651***
1.007***
1.963***
3.422***
(0.179)
(0.105)
(0.223)
(0.368)
-0.558***
-0.340***
-0.641***
-1.125***
(0.099)
(0.060)
(0.100)
(0.176)
-.0569*
-0.034*
-0.055**
-0.093**
(0.030)
(0.018)
(0.027)
(0.046)
Logit
(0.036)
Dev.
Par.
2.885***
Distr. (Const.)
Observations
(0.280)
1233
1233
Individuals
LogL
Mixed
0.731***
Rho
Std.
Panel
Pooled
-756.18
-756.07
1233
1233
475
475
-642.25
-642.25
Standard errors in parentheses
*** 1% significance; ** 5% significance; * 10% significance
36
Table 5. Willingness to pay for CE-project Attributes (€)
Attribute
Regenerated
Pooled
Pooled
Panel
Panel
Logit
Probit
Probit
Mixed Logit
1.82
1.82
2.09
2.07
(1.11;2.52)
(1.16;2.51)
(1.22;3.00)
(1.22;2.98)
3.09
3.09
3.22
3.21
(2.58;3.69)
(2.59;3.68)
(2.73;3.85)
(2.62;3.99)
2.96
2.96
3.06
3.04
(2.49;3.50)
(2.50;3.52)
(2.57;3.85)
(2.52;3.75)
-0.10
-0.10
-0.09
-0.08
(-0.26; 0.04)
(-0.27;0.04)
(-0.17;-0.01)
(-0.20;0.02)
Urban
Intermediate
Natural
Park time
Krinsky-Robb (1986) confidence interval (5%;95%) based on 10.000 simulations
37
Table 6.a. Frequencies of choices per attribute: Control
Choice
Control
Total
Base
Improved
Neither
No
190
146
64
400
Yes
286
462
52
800
476
608
116
1200
Total
Table 6.b. Frequencies of choices per attribute: Wait information
Choice
Wait
information
Total
Base
Improved
Neither
0
204
270
51
525
1 year
272
338
65
675
476
608
116
1200
Total
Table 6.c. Frequencies of choices per attribute: Duration
Choice
Duration
Total
Total
Base
Improved
Neither
10
171
198
38
407
15
155
219
37
411
20
150
191
41
382
476
608
116
1200
38
Table 7. CE-Models Results of Models estimated for the CE-implement
Attribute
Control
Wait information
Duration
Cost
Pooled
Pooled
Panel
Panel Mixed
Panel Logit
Logit
Probit
Probit
Logit
RP
0.870***
0.541***
0.996***
1.176***
2.594***
(0.121)
(0.075)
(0.122)
(0.221)
(0.499)
0.273**
0.168**
0.356***
0.623***
1.362***
(0.133)
(0.082)
(0.144)
(0.225)
(0.424)
0.019
0.012
0.021
0.037
0.076*
(0.015)
(0.009)
(0.017)
(0.029)
(0.043)
-0.619***
-0.383***
-0.726***
-1.273***
-1.874***
(0.127)
(0.078)
(0.138)
(0.237)
(0.405)
0.775***
Rho
St.
(0.033)
Dev.
Par.
3.286***
Distr. (Const.)
St.
Dev.
(0.331)
Par.
4.040***
Distr. (Control)
(0.754)
St.
2.868***
Dev.
Par.
Distr. (Wait)
St.
Dev.
(0.699)
Par.
0.403***
Distr. (Duration)
Observations
(0.096)
1084
1084
Individuals
LogL
-717.45
-717.44
1084
1084
1084
393
393
393
-589.14
-588.72
-665.14
Standard errors in parentheses*** 1% significance; ** 5% significance; * 10% significance
39
Table 8. Willingness to pay for Implementation Attributes (€)
Attribute
Control
Wait
Pooled
Pooled
Panel
Panel Mixed
Panel Logit
Logit
Probit
Probit
Logit
RP
1.41
1.41
1.37
1.38
1.38
(1.08;1.74)
(1.08;1.74)
(1.08;1.72)
(1.08;1.73)
(0.63;2.13)
0.44
0.44
0.49
0.49
0.73
(0.05;0.80)
(0.07;0.80)
(0.16;0.81)
(0.20;0.78)
(0.37;1.08)
0.03
0.03
0.03
0.03
0.04
(-0.15;0.17)
(-0.16;0.17)
(-0.13;0.13)
(-0.14;0.14)
(-0.19;0.21)
information
Duration
Krinsky-Robb (1986) confidence interval (5%;95%) based on 10.000 simulations
40