Using Labels to Investigate Scope Effects in Stated Preference Methods

Abstract

Integrated catchment policies are widely used to manage natural resources in Australian catchments. Integration of environmental processes with socio-economic systems is often difficult due to the limitations of decision support tools. To support assessments of the environmental and economic trade-offs of changes in catchment management, fully integrated models are needed. This research demonstrates a Bayesian Network (BN) approach to integrating environmental modelling with economic valuation. The model incorporates hydrological, ecological and economic models for the George catchment in Tasmania. Choice experiments were used to elicit information about the non-market costs and benefits of environmental changes. This allows the efficiency of alternative management scenarios to be assessed.

Full text

Page 1

Environ Resource Econ (2009) 44:521–535
DOI 10.1007/s10640-009-9299-z
Using Labels to Investigate Scope Effects in Stated
Preference Methods
Mikołaj Czajkowski · Nick Hanley
Accepted: 16 June 2009 / Published online: 27 June 2009
© Springer Science+Business Media B.V. 2009
Abstract
good on offer) remains a major criticism of stated preference methods, and many studies
fail a scope test of some sort. Across a range of existing explanations for insensitivity to
scope (commodity mis-specification, embedding, warm glows) there seems to exist no clear
conclusion on how to deal with the problem. This paper provides an alternative explanation
for insufficient sensitivity to scope, based on re-definition of the determinants of value for
environmental goods within an attributes-based choice model. In the proposed framework
respondents’ Willingness To Pay need depend not only on physical characteristics of a good,
but may also depend on the ‘label’ under which the environmental good is ‘sold’ in the
hypothetical market. To investigate this problem, a Choice Experiment study of biodiversity
was conducted. We find that controlling for the effects of a label—in this case, national park
designation—leads to significant increase in the scope sensitivity of welfare measures.
Insufficient sensitivity to scope (variations in the scale of the environmental
Keywords
Contingent valuation · Biodiversity valuation · National park
Scope test · Embedding · Warm glow · Label effect · Choice experiment ·
1 Introduction
Stated preference methods remain the only source of estimates for non-use values for
environmental goods, and hence can provide valuable inputs to cost-benefit analyses of
environmental change. Despite the initial distrust many researchers held for such methods,
they have eventually entered the mainstream of economic science (Carson and Hanemann
2005), and are now routinely used as part of the policy analysis process. However, certain
M. Czajkowski (B )
Warsaw Ecological Economics Center, University of Warsaw, Warsaw, Poland
e-mail: miq@wne.uw.edu.pl
N. Hanley
Economics Department, University of Stirling, Stirling, UK
e-mail: n.d.hanley@stir.ac.uk
123

Page 2

522 M. Czajkowski, N. Hanley
importantissuesremaintoberesolvedintermsofenhancingthevalidityofwelfareestimates
obtainedfromstatedpreferenceapproaches.Theseissueshaveariseninthecontextoftesting
whether the results of valuation methods are in line with the predictions of economic theory.
There are at least two such tests. Firstly, it is expected that as the price increases, the
demand for an environmental good should decrease. For instance, in dichotomous choice
CVM designs, this means that the proportion of respondents who answer ‘yes’ to a bid ques-
tionshoulddecreasewithincreasesinthebidamount.Asecondtestconsistsofexaminingthe
prediction that respondents should be willing to pay more as the amount or quality of envi-
ronmentalgoodtobeprovidedincreases.Failuretopassthisscopetesthasbeentraditionally
one of the pivots of criticism of stated preference methods. The purpose of this paper is to
investigate sources of failure to pass the scope test. An alternative explanation is proposed,
based on redefinition of the value drivers for environmental goods. Empirically, we show
that respondents’ Willingness To Pay might depend not only on physical characteristics of a
good being valued, but partly also on the ‘label’ under which the good is being ‘sold’. This
provides an alternative explanation for problems with the scope test, in that failing to control
for the effects of a label can lead to a mis-interpretation of scope sensitivity.
Inthenextsection,webrieflyreviewtheliteratureonscopetests,andthenofferanalterna-
tive explanation in terms of labels. The review concentrates on applications of the contingent
valuationmethod,sincethishasbeenthefocusofscopetestswithinstatedpreferencestudies.
Section 3 describes the design of an empirical study to investigate this idea using a different
stated preference technique, namely choice experiments, whilst Sect. 4 contains results. The
finalsectionconcludes.Whilstourempiricalapplicationusesthechoiceexperimentmethod,
we argue that labelling is a relevant concern in all stated preference methods, since it con-
stitutes a potentially important aspect of the framing of the good being valued. However,
implementing our suggestion for controlling for labelling effects on scope sensitivity may be
harder with contingent valuation than with choice experiment.
2 Explaining Scope Effects, or Their Absence
2.1 Existing Literature
As noted above, the scope test consists in testing whether respondents are willing to pay sta-
tistically more for a larger amount of an environmental good, in terms of increased quantity
or quality. Even though the origins of this test can be tracked earlier, most of the debate was
based on the critique put forward by Kahneman and Knetsch (1992). Within a choice experi-
ment,thescopeofanenvironmentalimprovementcanbedetailedalongmultipledimensions
using the attributes of a good—for example, what fraction of a river is upgraded, what size of
populationofaspeciesisprotected,orforhowlongawildlifepopulation’shabitatisprotected
from development. Testing for scopein a choice experiment can then involvetesting whether
the parameters on such attributes are significantly different from zero (see, for example,
Hanley et al. 2003). Indeed, early authors of environmental choice experiments suggested
that one advantage of choice modelling was that it allowed the researcher to directly measure
scope effects (or, to put it another way, to directly control for embedding effects)1in this
way, unlike contingent valuation (Hanley et al. 1998). More recently, Goldberg and Roosen
(2007) find that choice experiments can lead to value estimates which are more sensitive to
1As Goldberg and Roosen (2007) note, the terms “scope insensitivity” and “embedding” are often used
inter-changeably.
123

Page 3

Using Labels to Investigate Scope Effects in Stated Preference Methods523
scopethancontingentvaluation,basedonastudyofhealthrisks,bothintermsofthenumber
of illnesses and the level of risk which individuals could avoid.
However, a great deal of the debate in the environmental economics literature on scope
effects has been conducted in the context of contingent valuation. Within contingent valu-
ation, two versions of the scope test exist—internal and external. In an internal version the
same respondents are asked to state their WTP for different levels of environmental good
improvements.Inanexternalversion,twodifferentlevelsarevaluedbydifferentrespondents
using split samples. There seem to be fewer problems with internal scope tests (e.g., Smith
and Osborne 1996); some studies argue, however, that this might be due to the urge of the
respondents to maintain ‘internal integrity’ of their answers (Heberlein et al. 2005). On the
other hand, according to Adamowicz et al. (1999) an internal scope test allows for a com-
parison of pairwise WTP estimates for each respondent in the sample, and thus controls for
heterogeneity amongst respondents. There seems to be no clear consensus in the literature
whether insufficient sensitivity to scope is a regularly observed phenomenon, or happens
occasionally, or on how its causes might be resolved and eliminated. Some of the best known
examples of estimates failing the scope test are given by Kahneman and Knetsch (1992) and
Diamond and Hausman (1994). On the other hand there are meta-analyses indicating that the
scope test is usually passed successfully (e.g., Carson 1997; Brouwer et al. 1999).
Scope tests might be failed due to one of the following reasons: (1) insufficient power of
the test, taking into consideration the difference in the level of provision of a public good
(Arrow and Leamer 1997); (2) errors resulting from invalid construction of the hypothetical
market (unclearly defined goods or changes in the level of their provision) which masks the
true, underlying sensitivity to scope (Carson and Mitchell 1995); (3) embedding (Kahneman
and Knetsch 1992); and (4) the warm glow effect (Becker 1974).
The warm glow effect involves respondents stating their WTP for public or environmental
goods as a way of ‘purchasing moral satisfaction’. The idea, inspired by the Olson (1965)
conceptof‘impurealtruism’,waslaterrenamedbyBecker(1974)asawarmgloweffect.The
concept results from a notion that a respondent derives utility from the act of contributing to
anincreaseinthesupplyofapublicgood,duetosocialassent,prestigeormoralsatisfaction.2
TheconceptwaslaterusedbyAndreoni(1989)todemonstratetheoreticallywhyprogressive
taxes might increase social and charitable spending and why government spending on these
does not crowd out private contributions. Accepting warm glow as an explanation for insuf-
ficient sensitivity to scope would mean that a welfare change resulting from implementation
of a particular scenario would not be fully and directly associated with quantity or quality
changes to the good in question. This would imply a need to take into account motives for
wanting a public good to be supplied when estimating values (Johansson-Stenman 1998).
Even though warm glows or the purchasing of moral satisfaction might contribute to insensi-
tivitytoscope(Cooperetal.2004),thereisamajorproblemwithacceptingthisexplanations
forallobservedproblemswithscopetest.Despitethefactthatwarmgloweffectsarerelatively
well described in the literature devoted to charitable and social organizations, its extrapola-
tion to the valuation of public goods would necessary mean that consumers derive higher
utility with an increase in the taxes they pay or bids they accept (Chilton and Hutchinson
1999). This would mean that there should be a positive relationship between declared WTP
values and utility level derived solely from ‘moral satisfaction’ reasons. However, virtually
all studies (to our knowledge) implementing warm glow as a component of utility function
2Accordingtosomeauthors,incaseofpublicgoodstheexistenceofwarmgloweffectmeansthatrespondents
are in fact expressing their ethical, rather than economic preferences (Ritov and Kahneman 1997).
123

Page 4

524M. Czajkowski, N. Hanley
do this by introducing a non-zero constant representing warm glow in utility functions if a
non-zero WTP is declared. This constant is obviously independent of WTP or bid level.
A conclusion one can draw from this brief overview of the literature on scope effects is
that more research seems to be required in order to broaden understanding of the reasons for
scope insensitivity. The next section offers an alternative explanation which does not depend
on warm glow motivations.
2.2 A New Approach to Thinking About Scope Effects
We put forward a hypothesis that the elicited value of an environmental good depends not
only on the physical characteristics of the good in question, but also on the ‘label’ under
which it is ‘sold’. The label under which an environmental good is sold can be viewed as one
attribute of that good (Lancaster 1966). A label is, however, different from other attributes
because it is independent from all the physical (quantifiable) characteristics of the good,
and depends instead on the respondent’s perception of that good. This notion, which clearly
appliestoprivategoodsaswell,isinlinewiththenotionofframingdependencesuggestedby
Kahneman and Tversky (2000). Indeed, within contingent valuation, a label can be thought
ofasoneaspectoftheframingofagood—thedescriptionthatresearchersprovidetorespon-
dents about the characteristics of the good, and how it can be provided. Thinking about the
value of a label allows for an alternative explanation of scope test problems
There is a vast marketing literature devoted to the importance of labels and brands, and
how their associated images influence choices. Consumers are known to have preferences
overlabelsinadditiontothephysicalcharacteristicsofgoods.Theinfluenceoflabelsoncon-
sumers’ utility and choices is also supported by neurological studies (Roe and Haab 2007).
A stark example is provided by McClure et al. (2004), who replicated ‘The Pepsi Challenge’
using Functional Magnetic Resonance Imaging of subjects’ brains to investigate preferences
and information processing in choices between two well-known soft drinks presented with
and without brand names. The respondents given two samples of the same soft drink sup-
pliedwithoutbrandnames(labels)wereessentiallyindifferent.However,whenoneofthetwo
identical samples was labelled as Coke or Pepsi, subjects systematically preferred Coke to
the unlabelled alternative (which they were told could be Coke or Pepsi), despite both drinks
being chemically identical. These results were also reflected in different neural responses of
the subjects, who seemed to have neurologically processed labels (brands) differently than
physical product characteristics. The neural evaluative process, and subsequent choice of the
product, may thus be altered by the presence of a label. This provides a clear demonstration
of the effects of labels on preferences.
The importance of labels to choice have also been investigated within choice experi-
ments. Blamey et al. (2000) used a split-sample design to investigate the effect of labelling
choices for policy options for conserving remnant vegetation in the Desert Uplands region
of Queensland. They found that the inclusion of policy labels reduced the attention respon-
dents gave to the physical attributes of a good, and caused a re-allocation of utility away
from part-worths for these attributes and towards a value for the label itself, although WTP
estimates between the labelled and un-labelled treatments were not significantly different
from each other. Drawing on this finding, we argue that a value of an environmental good
elicited using stated preference methods depends partly on its physical characteristics and
partly on the label under which it is presented to the respondents. To formalize the approach
let indirect utility v be a function of a price vector P, income y, and a vector of attributes of
an environmental (public) good Q (for simplicity assume one public good).
123

Page 5

Using Labels to Investigate Scope Effects in Stated Preference Methods525
v (P, y, Q)
(1)
The attributes of the good consist of a label qL and a set of remaining physical (quanti-
fiable) attributes Q−1. Assume qL to be a binary variable representing inclusion of label
in the description of the good, and the label to be desirable
since labels are independent from physical characteristics of the good we assume qLto be
additively separable from Q−1.
v (P, y,qL, Q−1)
This formulation allows us to derive compensating and equivalent surplus measures, where
superscript 0 denotes the initial, and 1 the post-change level of the good:
v?P0,q0
?
thus
∂v
∂qL> 0
?
. In addition,
(2)
L, Q0
L, Q1
−1, y0?= v?P0,q1
L, Q1
L, Q0
−1, y0− CV?
(3)
v?P0,q1
−1, y0?= v?P0,q0
−1, y0+ EV?
In a standard random utility setting (McFadden 1974), assuming a vector of socio-demo-
graphic and choice specific explanatory variables Z, respondent’s j indirect utility becomes:
?Pj,qL,j, Q−1,j, yj
where εjis unobservable, individual specific error term. Note that the label component of
utility is also individual-specific, since individuals might differ in their perception of the
significance or meaning of the label used. Assuming in turn additivity of the deterministic
and stochastic parts of indirect utility function and an IID property of error terms, we may
derive respondent’s j willingness to pay3from:
?Zj,q1
In the case of a linear WTP function the result might be then simplified to show that esti-
mated mean WTP?WTP?would be a function of two additive components—the set physical
WTP(Q) = g (Q−1) + h (qL)
The above formulation of mean WTP has some important implications. When two different
levels of change in an environmental good are offered to respondents in a CVM exercise,
and they are described using the same label, it is obvious that the estimated values of WTP
would differ only with respect to its physical attributes Q−1(for example the percentage of
habitatsafeguardedfromdevelopment,ortheareaofwoodlandtobeplanted),whilethelabel
componentqLof the mean WTP would remain constant. As a result, if the share of a valueof
the label in total WTP?WTP?was “sufficiently high”, then the observed WTP estimates for
for the label effect on WTP would we expect the estimated WTPs in two, alternative change
scenarios to pass the scope test. This is an idea which is amenable to empirical examination.
Including a label component in welfare estimates may be done by introducing an alter-
native specific constant (ASC) in the utility function associated with the choice alternative
which is associated with the label. A wider debate in choice experiments has developed on
the practice of including or excluding such an ASC in experimental designs and thus in
welfare estimates. Decisions over whether to include an ASC seem arbitrary even though
in some cases this causes significant changes in welfare measures. To give a few examples,
vj
?= Vj
?Zj,qL,j, Q−1,j, yj
?+ εj
(4)
Vj
L, Q1
−1, yj− WTPj
?+ ε1
j= Vj
?Zj,q0
L, Q0
−1, yj
?+ ε0
j
(5)
characteristics of the good Q−1, and the label which was used for describing the good qL:
(6)
twolevelsofenvironmentalgoodmightnotbe“sufficientlydifferent”.Onlyaftercontrolling
3Assuming an improvement of an environmental good this will be their compensating surplus.
123

Page 6

526M. Czajkowski, N. Hanley
Horne et al. (2005) and Birol et al. (2006) include an ASC in welfare estimates, whilst
Lehtonen et al. (2003) and Nielsen et al. (2007) do not.
In this light, it is worth discussing if the value of a label should be included in welfare
estimates. The answer to this question seems to depend on the research goal. If the study is
aimind at estimating the change in consumers’ welfare due to a change in provision of an
environmental good, it is safe to concludethat the value of the label should be included in the
total economic value estimates of the scenario. If ‘purchasing’ a good with a particular label
generatesadditionalutilityitshouldbeaccountedforinwelfareestimates.Ontheotherhand,
if the aim of the study is to estimate values for changes in environmental qualities (marginal
or discrete), it should probably be excluded and only the value resulting from the changes of
physical attributes should be included. As our case study demonstrates, this is possible using
the CE method.
Our formulation of economic value which includes the value of a label is partly in line
with the works of philosophers (Gunn 1980; Rescher 1980; Rolston 1988) who propose the
concept of an ‘intrinsic’ value of a species that exists irrespective of its actual representatives
(real animals). This, according to such writers, explains the apparent paradox that an animal
of an endangered species would be more valuable than an animal of the same species when it
isnotendangered.ThisviewwascriticizedbyRussow(1981)whoclaimedthatthedefinition
of a species is somewhat flexible and so species per se cannot have values, because defining
more species would simply increase the total value. Assigning part of the value to labels
recognized by the respondents deals with these problems and explains apparent paradoxes:
here, that people value protecting something which has been labelled as endangered. This is
because there may be only as many ‘premiums’ for species as there are recognizable labels.
3 Study Design
We employed a labelled Choice Experiment (Blamey et al. 2000) to test if welfare measures
of implementing an environmental policy incorporates the value of a label which is not asso-
ciated with any physical characteristics of the good. We then demonstrate how the value of
such a label can contribute to an apparent insensitivity to scope when alternative scenarios
for increases in the good are valued.4
3.1 The Label
The context of the empirical application is the protection of forest biodiversity. The case
study site was the Białowie˙ za Forest in Poland, which is considered to be one of the most
important remaining temperate natural lowland forests in Europe. Initial steps in identifying
attributes and a label involved focus groups and verbal protocols, with subjects representing
both the general population of Poland and local communities neighbouring the Białowie˙ za
Forest.CurrentlyNationalParkdesignationappliesto16%oftheBiałowie˙ zaForestarea,and
there is an ongoing debate about whether Park designation should be extended to the whole
area of the forest. In focus groups, we found that respondents had preferences for extending
the national park irrespective of what the extension would mean in terms of on-the-ground
protection and management of biodiversity. Thus, national park designation was chosen as
the label for providing environmental change, since it was not associated with any character-
istics ofthe goodtobe valued(forest biodiversity), andyet appearedto bewidely recognized
4Note that we are not trying to test for the effects on parameter values in the utility function from whether a
label is included or excluded.
123

Page 7

Using Labels to Investigate Scope Effects in Stated Preference Methods527
and desired by respondents.5Focus group studies, verbal protocols and pre-testing clearly
showed that respondents did not think that extension of the national park itself would bring
aboutanyotherchangesthantheonesdescribedbytheattributesinthechoiceoptions;itwas
none-the-less strongly highlighted to respondents in the questionnaire that both alternatives
would essentially provide the same changes to the environment. It was thus safe to conclude
that emotional associations of the label were not tied to any physical attributes of the good
included in the experimental design.6
3.2 Physical Attributes
In order to describe the likely changes in biological diversity of the forest, the most impor-
tant elements of biodiversity were identified in cooperation with biologists and ecologists.
This list of candidate attributes was then pre-tested on the general public using focus groups,
resulting in the final selection of 4 attributes at 3-4 levels. The selected attributes repre-
sented potential changes in the biodiversity viewed as important by ecologists which were
comprehensible to respondents.
The first attribute—natural ecological processes—represented the natural dynamics of
the Białowie˙ za Forest. This illustrated natural changes of the forest’s flora—processes of
succession and regression, fluctuation, degeneration and regeneration, as well as seasonal
changes. According to our specialists, and as explained in the questionnaire, improvements
inthisattributecouldbeachievedbypassiveprotectionofagivenpercentageofthetotalarea
of the Białowie˙ za Forest. Three possible levels of this attribute were: status quo—16% of the
areaprotected,partialimprovement—30%protected,andsubstantialimprovement—60%of
the area to be passively protected.
Rare species of fauna and flora represented the second attribute. It was underlined in
the description that this attribute represents not just known, but also yet-unknown species.
Examplesofbothflagshipandlesser-knownspecieswereprovided,togetherwithinformation
concerning the likelihood of yet-unknown species occurring in the forest, and their depen-
dence on protection. A short general explanation of the importance of different species to
ecosystemfunctioningwasprovided.Thepossiblelevelsofthisattributewere:statusquo—a
decline threatening total extinction, partial improvement—maintaining current populations,
and substantial improvement—maintaining and expanding current populations.
Ecosystem components was the attribute characterizing the existence of biotopes and eco-
logical niches, such as dead wood, natural ponds, streams and clearings. It was explained in
the questionnaire that improvements in this attribute may be achieved by active protection
of these components. This attribute could be important for respondents both for the exis-
tence of the components alone, as well as being a proxy for improved well-being of species
inhabiting the forest. The possible levels of this attribute were: status quo—the lack of some
components and decrease in the quality of existing ones, minor improvement—regeneration
of deteriorating components across 10% of the forest area, partial improvement—regener-
ation and protection across 30% of the forest, and substantial improvement—regeneration
and protection across 60% of the forest area.
The last attribute was monetary, representing the cost of an additional compulsory tax to
be paid for the following 10years by households across Poland.
5It’s worth noting that national parks in Poland have differing management regimes and protection goals;
thus ‘extending the national park’ is not associated with any specific set of actions or characteristics.
6Such untied associations are referred to as ‘freestanding emotions’ (Rossiter and Percy 1997).
123

Page 8

528M. Czajkowski, N. Hanley
Fig. 1 Example of a choice card
Each respondent was presented with four choice sets, each consisting of three alterna-
tives. The first alternative always represented a status quo option for which there was no
variance in attribute levels (including cost). The other two choice alternatives were designed
using orthogonal fractional factorial design. The first of these choice alternatives was always
labelled ‘extension of the national park’ while the second was labelled ‘other forms of pro-
tection’. The design consisted of 32 choice sets blocked into 8 questionnaire versions. An
example of the choice card is given in Fig. 1. The design we used was a LMAfactorial design
(Louviere et al. 2006). The first (constant, status quo) alternative was added to every choice
set (there was no variation in the attribute levels for this alternative). The attribute levels for
theothertwoalternativesweregeneratedusingtheSPSSOrthoplan.Severalpossibledesigns
were compared manually to select the design with balanced alternatives and maximum vari-
ability of attribute levels in each choice set. The attributes were assumed to be generic across
alternatives(noalternative-specificparametersintheutilityfunction,exceptfortheconstant)
with dummy-coded attribute levels. Additionally, we used a blocking variable to limit the
number of choice-sets faced by each person. The complete design of the experiment is avail-
able from the authors on request. We calculated the D- and A-errors of our design following
the approach proposed by Scarpa and Rose (2008). Using the parameters of the MNL and
NL models as priors we were able to calculate these errors ex-post. The D-error for the MNL
parameters is 0.128, while for the NL model it is 0.126. We have also ex-post followed the
approach proposed by Street et al. (2005) to verify whether our design was efficient for the
MNL model. The result showed that our design was nearly 70% efficient.
4 Results
The face-to-face CE surveys were conducted in June 2007 on a nationwide representative
quota sample of adult Poles by a professional surveying company. There were a total of 400
surveys collected, resulting in 1600 choice observations. The statistical analysis was con-
ducted using NLOGIT 4.0. A number of different model specifications were tried, includ-
ing Multinomial Logit, Error Components, Nested Logit, Heteroscedastic Extreme Value,
123

Page 9

Using Labels to Investigate Scope Effects in Stated Preference Methods 529
RandomParametersLogitandMultinomialProbit—eachinmanypossiblefunctionalforms.
Invirtuallyallmodelswefoundstatisticalsignificanceofthe“PARK”variable,whichrepre-
sented choice of the labelled alternative—providing protection of biodiversity by extending
thenationalpark.Itmaybeinterpretedasapremiumconsumersgetwhentheprotectionplan
is implemented through the extension of a national park—thus providing an environmental
good with a desirable ‘label’.
We test our hypothesis using the results of the Nested Logit Model with Covariance Het-
erogeneity (see Bhat 1997). This model was chosen because it provided one of the best ratios
of correct predictions from amongst the alternatives listed in the previous paragraph, and
since it allows one to relax the assumption of constant error variances within and between
choosersorchoicesituations.Incaseofthetwo-levelnestingtree,theprobabilityofchoosing
a particular alternative j out of Jmalternatives in the branch (assume M branches) is:
Pr(choice = jm) =
exp(Vjm/αm)
?M
??Jm
i=1exp(Vik/αk)
i=1exp(Vim/αm)
??Jk
?(αm−1)
k=1
?αk
,
(7)
where Viis the utility function associated with alternative i, and αmis the inclusive value
parameter for the branch m. The Covariance Heterogeneity Nested Logit model allows the
inclusive value parameters to be functions of a set of individual-specific socio-demographic
characteristics v, i.e., α∗
error terms of the individual- and alternative-specific utility functions.
Detailed results of the final model are given in Table 1. The explanatory variables are
dummies representing different possible improvements in the levels of the attributes, thus
allowing for non-linear marginal utilities (the status quo being the reference level). Because
there was no statistical difference between partial and substantial improvement in the Rare
species of fauna and flora attribute in most of the models, the two levels are jointly repre-
sented as an improvement. PARK, as already stated, is a dummy representing the alternative
specific constant for the labelled alternative ‘extension of the national park’ and Cost is the
monetary variable measured in PLN2007. As noted above, the significance of the PARK var-
iable shows the presence of a clear labelling effect. Individual heterogeneity of covariance is
explained by the following variables: hinc—household income, previs—a variable set equal
to one if the respondent has visited the Bialowieza forest before, and intvis, a variable set
equal to one if the respondent intends to visit the forest in the future.
ImplementingtheapproachsuggestedinLouviereetal.(2006),WTPvaluesforeachlevel
of the attributes were calculated, with reference to the status quo level of each attribute. The
results, given in euro,7are summarized in Table 2. Standard errors were calculated using the
Delta method.
Inordertodemonstratetheinfluenceofthelabelonscopesensitivity twopolicyscenarios
were considered, both of which are currently being considered for actual implementation. In
the first scenario (LO) only minimal improvements to all the attributes were included, while
in the second scenario (HI) the attributes would be provided at the highest levels used in the
CE design. The components of the two policies are summarized in Table 3.
The resulting mean welfare estimates—WTPLOand WTPHI, respectively—were calcu-
lated using the approach provided by Hanemann (1982), extended to the case of nested logit
model with covariance heterogeneity. Letting V0
(for the alternative i) for the status-quo and the improved state of the public good (new level
of the attributes), respectively, and β1be the parameter of the monetary attribute in the utility
m= αmexp?δ?v?. This allows one to introduce heterogeneity of the
iand V1
ibe the value of the utility function
7The values in euro were calculated using the following exchange rate: 1 EUR ≈ 3.6PLN.
123

Page 10

530M. Czajkowski, N. Hanley
Table 1 Covariance heterogeneity nested Logit model estimates from the choice experiment
Variable CoefficientStandard error
p-value
Natural ecological processes (1-level improvement)
Natural ecological processes (2-level improvement)
Rare Species (improvement)
Ecosystem components (1-level improvement)
Ecosystem components (2-level improvement)
Ecosystem components (3-level improvement)
PARK (alternative specific constant)a
Cost
Inclusive value parameterb
Covariates in inclusive value parameter
Hinc
Previs
Intvis
0.290**
0.503***
0.313***
0.326**
0.387***
0.438***
0.937***
−0.033***
0.682***
0.115
0.147
0.110
0.132
0.141
0.149
0.151
0.004
0.012
<0.001
0.005
0.014
0.006
0.003
<0.001
<0.001
0.128<0.001
−1.263***
−1.954**
−1.171**
0.657
0.888
0.515
0.055
0.029
0.023
∗∗∗,∗∗,∗Significance at 1, 5, 10% level, respectively
Number of observations 1,069
Log likelihood function −1044.495
Chi squared 242.758
Degrees of freedom 12
Pr
χ2> critical value
= <0.001
aASC PARK is the label for alternatives which implement the change in the form of national park extension
bThe inclusive value for the non-restricted branch of the tree. A value between 0 and 1 is typical for the
common component of random terms (Hensher and Greene 2002)
?
?
Table 2 Implicit prices of the attribute levels (EURO)
Attribute Implicit priceStandard error
p-value
Natural ecological processes (1-level improvement)
Natural ecological processes (2-level improvement)
Rare species (improvement)
Ecosystem components (1-level improvement)
Ecosystem components (2-level improvement)
Ecosystem components (3-level improvement)
PARK (alternative specific constant)
2.47∗∗
4.28∗∗∗
2.66∗∗∗
2.78∗∗
3.30∗∗∗
3.73∗∗∗
7.97∗∗∗
0.983
1.192
0.960
1.131
1.161
1.210
1.242
0.012
<0.001
0.006
0.014
0.005
0.002
<0.001
∗∗∗,∗∗,∗Significance at 1, 5, 10% level, respectively
Table 3 Components of the policy scenarios
AttributesPolicy scenario ‘LO’Policy scenario ‘HI’
Natural ecological processes
Rare species
Ecosystem components
1-Level improvement
Improvement
1-Level improvement
2-Level improvement
Improvement
3-Level improvement
123

Page 11

Using Labels to Investigate Scope Effects in Stated Preference Methods531
Table 4 Welfare estimates for policy scenarios including/excluding the label (EURO)
Policy Welfare estimate90% C. I. Standard error
p-value
Welfare estimates including the labela
LO15.49
HI 18.25
Welfare estimates excluding the labelb
LO
HI10.28
11.28–22.21
13.86–25.03
1.537
1.635
< 0.001
< 0.001
7.525.58–10.74
7.97–13.80
1.490
1.703
< 0.001
< 0.001
aProviding environmental changes in the form of extending the national park
bProviding environmental changes in some other way
function, the compensating variation associated with the change becomes:
?
= −1
β1
k=1
Once again, in case of the covariance heterogeneity model one allows the inclusive value
parameters to be functions of a set of observed characteristics of the individual or the branch.
In order to compare the influence of the label on welfare estimates of the two policies in
termsofscopesensitivity,thewelfareestimatesofthetwopolicieswerecalculatedincluding
the label and excluding it (Table 4), through the device of either including or omitting the
parameter on PARK in the compensating surplus calculations. Finding a scope effect would
mean rejecting the null hypothesis that the mean WTP for the “low” scenario was equal to
the mean WTP for the “high” scenario, since the scenarios differ in terms of the quantity of
biodiversity conservation on offer. The means, standard errors and confidence intervals are
based on parametric bootstrapping (Krinsky and Robb 1986).
We first analyze the difference in means between the two policy scenarios differing in
attribute levels using Park et al. (1991) method of non-overlapping confidence intervals.
Based on the bootstrapped empirical distributions of mean welfare estimates of policies LO
and HI we found that the p-value that would allow one to conclude that the welfare estimates
are different (significance level assuring confidence intervals not to overlap) is 0.337 when
the label is included, and 0.199 when the label effect is excluded. According to Poe et al.
(2005) however, using confidence intervals for testing difference in means is inappropriate.
Instead they propose a convolutions method, which consists in calculating all the possible
differences between elements of a vector bootstrapped from the distribution of the higher
mean and a vector bootstrapped from the distribution of the lower mean. As a result, the ratio
of number of outcomes that are less than zero to the number of all the possible outcomes
gives an exact p-value of the hypothesis that the mean of vector 1 is higher than the mean of
vector 2. This is referred to as complete combinatorial convolutions method. Following this
approach we have estimated p-values for the hypotheses that WTPHIis higher than WTPLO
for the estimates including the label and excluding it.8The p-value for the null hypothesis
that the welfare estimate of the policy scenario ‘HI’ is equal to that for the policy scenario
CV = −1
β1
E
?
?
max
j
U1
?
− E
?
max
j
U0
⎤
??
αk
⎛
⎝ln
M
⎡
⎣
Jk
?
i=1
exp?V1
ik/αk
?
⎦
− ln
M
?
k=1
⎡
⎣
Jk
?
i=1
exp?V0
ik/αk
?
⎤
⎦
αk⎞
⎠
(8)
8Foreachpolicyscenariowehavebootstrapped100,000welfaremeasurestoapproximateempiricaldistribu-
tion of it. Then, the two welfare measures are compared by subtracting each possible estimate of policy option
123

Page 12

532M. Czajkowski, N. Hanley
‘LO’ was estimated to be 0.277 for the case of including the label, and 0.121 with the label
excluded.
This comparison demonstrates the potential influence which including a label in welfare
estimates may have on measured sensitivity to scope of environmental policies. If the value
of the label was included in welfare estimates, the two analyzed policies were not signifi-
cantly different (the p-value close to 0.28). However, excluding the value of the label (i.e.,
controlling for its presence) substantially increased the significance level of the difference
(to 0.12). In our case excluding the label allows the researcher to accept the null hypothesis
of difference in welfare estimates of the two policies at the 12% significance level. Since
this implies a very large change (of more than 50%) in the p-value, we can conclude that
controlling for the label effect gives an appreciably different conclusion as to sensitivity to
scope in this instance.
5 Discussion and Conclusions
Our empirical study demonstrated that WTP for increased protection of an environmental
resource may include an additional component resulting from presenting the scenario using
a label which is recognized by the respondents as desirable. This result is in accordance with
the findings of the split-sample test carried out by Blamey et al. (2000). Here, however, we
show that controlling for such an effect potentially allows one to draw different conclusions
over the presence of scope effects.
Our results show that controlling for the value of a label in welfare estimates can increase
the sensitivity of these estimates to changes in the physical characteristics of environmental
goods(scope).Thisconclusionwouldhavegeneralapplicabilitywhen:(1)alabelconstitutes
an important share of total economic value and (2) covariances between the parameter of a
labelintheindirectutilityfunctionwithparametersofotherattributesarenon-negative.If(2)
doesnotholdthenexcludingthelabelfromwelfareestimateswouldcauseboththemeanand
standarderrortodecrease,whichcouldresultinuncertainimpactsonthelevelofsignificance
of the difference between welfare estimates of the two policies. In our case the value of a
label constitutes an important share of total value (roughly 50%). Thus our empirical study
confirms that the value of a label may be a considerable component of the total WTP. This
validates the relationship made explicit in formula (6), that WTP might depend not only on
the physical attributes of the good, but may also consist of a constant component associated
with the value of a label.
The label used in our study represented providing additional biodiversity protection in
a particular way—extending the national park. The preferences of respondents for national
parks, even if no particular protection regime or attributes are associated with them, seem to
beconfirmedbytheresultsofBartczaketal.(2008)andJacobsenandThorsen(2008).These
authors also observe premiums for the existence of national parks per se, irrespective of the
actual protection policy implemented there or the actual delivery of improved biodiversity
protection. Indeed, inspecting the mean WTP values for either policy scenario (LO or HI in
Table 4) shows that controlling for the label effect associated with national parks as a protec-
tion strategy produceslarge changesin meanvalues:from 15.49euroto 7.52euro/household
for the LO option, and from 18.25 to 10.28 for the HI option.
Footnote 8 continued
1 from the policy option 2. The number of negative results divided by the total number of results (100,0002)
is equal to the p-value of the hypothesis, that welfare measure of the policy 1 is higher than welfare measure
of the policy 2.
123

Page 13

Using Labels to Investigate Scope Effects in Stated Preference Methods533
Another interesting example that can be interpreted as an importance of the label is pro-
vided by Jacobsen et al. (2008) who observe different welfare estimates of the same conser-
vation programmes when using quantitative descriptions of results on the number of species
protected than when the actual names of the species to be protected are used for description.
Even if the respondents were not familiar with the names of the species to be protected, they
seemed to process the information differently, and the utilisation of the species name—its
label—was observed to have altered their choices.
Our results are presented in the context of a choice experiment, so it is important to con-
sider what implication they have for contingent valuation (CV). Since all CV studies employ
a specific scenario for valuation—the framing of the good to be valued—one may expect
that in some instances a label may be an implicit but un-identified component of WTP. This
would cause welfare estimates to be less sensitive to changes in scope of the valued good
if the value of the label is relatively high. Testing for the presence of a labelling effect in
contingent valuation would involve using valuation scenarios where a label is first included
and then omitted: using different quantities of the good in each treatment would then reveal
whether labelling impacts on measured scope sensitivity.
Important practical questions that arise are what might constitute a label for an environ-
mental policy or good, and how to identify and devise an appropriate label for valuation
scenarios. We believe that there are many possible labels depending on the particular good
and on respondents’ preferences. Identifying appropriate labels requires qualitative analy-
sis—pretesting, focus groups or verbal protocols, in a manner no different from selecting all
the other attributes, or other aspects of the framing of the good.
In summary, this paper adds to the literature by suggesting a new way of thinking about
failures to pass scope tests. Willingness to pay is conceived as consisting of two sub-com-
ponents: one, a function of a physical attributes of the good; and the other, a value of the
label, which is used for presenting the good or valuation scenario to respondents. Our study
provides evidence, reinforcing that offered by Blamey et al. (2000), that labels can be a
substantial constituent of estimated value, and demonstrates that this may be a reason for an
apparent insufficient sensitivity to scope of welfare estimates. Future work could evaluate
the usefulness of labels as a way of explaining scope effects by investigating their effects
in a contingent valuation setting, but also by carrying out split-sample tests within choice
experiments (note that our study did not use split samples). De-briefing of respondents as to
why labels are important to them, and how labels impact on choices would also be useful, as
would testing of whether people really believe that labels are not related to the quantity or
quality of experimental design attributes in choice scenarios.
Acknowledgments
thisstudypossible.ThecontributionbyMałgorzataBuszko-Briggsisalsogreatlyacknowledged.Theresearch
was funded by Polish Ministry of Science and Higher Education and the Foundation for Polish Science. Three
referees and an Associate Editor provided insightful comments on earlier versions of the paper.
TheauthorswishtothankTomasz˙Zyliczwhoseinvaluablesupportandassistancemade
References
AdamowiczWL,BoxallPC,LouviereJJ,SwaitJD (1999) Statedpreferencemethodsforvaluingenvironmen-
tal amenities. In: Bateman IJ, Willis KG (eds) Valuing environmental preferences: theory and practice
of the contingent valuation method in the USA, EC, and developing countries. Oxford University Press,
New York
Andreoni J (1989) Giving with impure altruism: applications to charity and Ricardian equivalence. J Political
Econ 97(6):1447–1458
123

Page 14

534 M. Czajkowski, N. Hanley
ArrowK,LeamerEE(1997)Commentno.87,submittedtoNOAAinresponsetoadvancednoticeofproposed
rulemaking (59 FR 1062, January 7, 1994)
Bartczak A, Lindhjem H, Navrud S, Zandersen M,˙Zylicz T (2008) Valuing forest recreation on the national
level in a transition economy: the case of Poland. For Policy Econ 10(7–8):467–472
Becker G (1974) A theory of social interactions. J Political Econ 82:1095–1117
Bhat CR (1997) Covariance heterogeneity in nested logit models: econometric structure and application to
intercity travel. Transp Res Part B Methodol 31(1):11–21
Birol E, Karousakisb K, Koundouric P (2006) Using a choice experiment to account for preference heteroge-
neity in Wetland attributes: the case of cheimaditida Wetland in Greece. Ecol Econ 60:145–156
Blamey RK, Bennett JW, Louviere JJ, Morrison MD, Rolfe J (2000) A test of policy labels in environmental
choice modelling studies. Ecol Econ 32(2):269–286
Brouwer R, Langford IH, Bateman IJ, Turner RK (1999) A meta-analysis of Wetland contingent valuation
studies. Reg Environ Change 1:47–57
Carson RT (1997) Contingent valuation and tests of insensitivity to scope. In: Kopp R, Pommerhene W,
Schwartz N (eds) Determining the value of non-marketed goods: economic, psychological, and policy
relevant aspects of contingent valuation methods. Kluwer, Amsterdam
Carson RT, Mitchell RC (1995) Sequencing and nesting in contingent valuation surveys. J Environ Econ
Manag 28:155–173
Carson RT, Hanemann WM (2005) Contingent valuation. In: Mäler KG, Vincent JR (eds) Handbook of
environmental economics. Elsevier, Amsterdam
Chilton SM, Hutchinson WG (1999) Some further implications of incorporating the warm glow of giving into
welfaremeasures:acommentontheuseofdonationmechanismsbyChampetal.JEnvironEconManag
37(2):202–209
Cooper P, Poe GL, Bateman IJ (2004) The structure of motivation for contingent values: a case study of lake
water quality improvement. Ecol Econ 50(1–2):69–82
Diamond PA, Hausman JA (1994) Contingent valuation: is some number better than no number. J Econ
Perspect 8(4):45–64
Goldberg I, Roosen J (2007) Scope insensitivity in health risk reduction studies: A comparison of choice
experiments and the contingent valuation method for valuing safer food. J Risk Uncertain 34(2):
123–144
Gunn AS (1980) Why should we care about rare species. Environ Ethics 2:17–37
Hanemann MW (1982) Applied welfare analysis with qualitative response models. University of California,
Berkeley
Hanley N, Wright R, Adamowicz V (1998) Using choice experiments to value the environment. Environ
Resour Econ 11(3):413–428
Hanley N, MacMillan D, Patterson I, Wright RE (2003) Economics and the design of nature conservation
policy: a case study of wild goose conservation in Scotland using choice experiments. Anim Conserv
6(02):123–129
Heberlein TA, Wilson MA, Bishop RC, Schaeffer NC (2005) Rethinking the scope test as a criterion for
validity in contingent valuation. J Environ Econ Manag 50(1):1–22
Hensher DA, Greene WH (2002) Specification and estimation of the nested logit model: alternative normali-
sations. Transp Res Part B Methodol 36(1):1–17
Horne P, Boxall PC, Adamowicz WL (2005) Multiple-use management of forest recreation sites: a spatially
explicit choice experiment. For Ecol Manag 207(1–2):189–199
Jacobsen JB, Thorsen BJ (2008) Where to put a national park and what to put in it? An a priori study of the
willingness-to-pay for coming national parks. University of Copenhagen, Copenhagen
Jacobsen JB, Boiesen JH, Thorsen BJ, Strange N (2008) What’s in a name? The use of quantitative measures
versus ‘Iconised’ species when valuing biodiversity. Environ Resour Econ 39(3):247–263
Johansson-Stenman O (1998) The importance of ethics in environmental economics with a focus on existence
values. Environ Resour Econ 11(3–4):429–442
Kahneman D, Knetsch JL (1992) Valuing public goods: the purchase of moral satisfaction. J Environ Econ
Manag 22:57–70
Kahneman D, Tversky A (2000) Choices, values, and frames. Cambridge University Press, Cambridge, UK
Krinsky I, Robb AL (1986) On approximating the statistical properties of elasticities. Rev Econ Stat
68(4):715–719
Lancaster K (1966) A new approach to consumer theory. J Political Econ 84:132–157
Lehtonen E, Kuuluvainen J, Pouta E, Rekola M, Li C-Z (2003) Non-market benefits of forest conservation in
southern Finland. Environ Sci Policy 6(3):195–204
Louviere JJ, Hensher DA, Swait JD (2006) Stated choice methods: analysis and applications. Cambridge
University Press, Cambridge
123

Page 15

Using Labels to Investigate Scope Effects in Stated Preference Methods535
McClureSM,LiJ,TomlinD,CypertKS,MontagueLM,MontaguePR (2004) Neuralcorrelatesofbehavioral
preference for culturally familiar drinks. Neuron 44:379–387
McFaddenD (1974) Conditionallogitanalysisofqualititativechoicebehaviour.In: ZarembkaP(ed)Frontiers
in econometrics. Academic Press, New York
Nielsen AB, Olsen SB, Lundhede T (2007) An Economic Valuation of the Recreational Benefits Associated
with Nature-Based Forest Management Practices. Landsc Urban Plan 80:63–71
Olson M (1965) The logic of collective action. Harvard University Press, Harvard
Park T, Loomis JB, Creel M (1991) Confidence intervals for evaluating benefits estimates from dichotomous
choice contingent valuation studies. Land Econ 67(1):64
Poe GL, Giraud KL, Loomis JB (2005) Computational methods for measuring the difference of empirical
distributions. Am J Agric Econ 87(2):353–365
Rescher N (1980) Unpopular essays on technological progress. University of Pittsburgh Press, Pittsburgh
Ritov I, Kahneman D (1997) How people value the environment. In: Bazerman M, Messick D, Tenbrunsel
A, Wade-Benzoni K (eds) Environment, ethics, and behavior. New Lexington Press, San Franciso
Roe B, Haab T (2007) Using biomedical technologies to inform economic modeling: challenges and opportu-
nitiesforimprovinganalysisofenvironmentalpolicies.PaperpresentedattheFrontiersinEnvironmental
Economics Conference. Washington, DC
Rolston H (1988) Environmental ethics: duties to and values in the natural world. Temple University Press,
Philadelphia
Rossiter JR, Percy L (1997) Advertising communications and promotion management. McGraw Hill, Sydney
Russow LM (1981) Why do species matter? Environ Ethics 3:101–112
ScarpaR,RoseJM (2008) Designefficiencyfornon-marketvaluationwithchoicemodelling:howtomeasure
it, what to report and why. Aust J Agric Resour Econ 52(3):253–282
Smith VK, Osborne LL (1996) Do contingent valuation estimates pass a “Scope” test? A meta-analysis.
J Environ Econ Manag 31(3):287–301
Street DJ, Burgess L, Louviere JJ (2005) Quick and easy choice sets: constructing optimal and nearly optimal
stated choice experiments. Int J Res Mark 22(4):459–470
123