. 2
( 11)


in hypothetical markets. Such ˜strategic overbidding™ may occur where respondents
feel that the amount they will actually have to pay will be related to some sample
measure, such as mean WTP, rather than their own statements. In such a case, if
formulated WTP exceeds expected mean WTP, the respondent may in¬‚ate stated
WTP (up to the expected mean) in an effort to improve the probability of provision.
Carson et al. (1999) extend this theoretical analysis in the context of OE for-
mat responses noting that in a case where there is uncertainty over the provision
of a good, individuals have a strategic incentive to overstate their willingness to
contribute to subsequent costs of provision, as the very nature of an OE response
tells respondents that these amounts are unlikely to bind them subsequently. Em-
pirical support for such a model is provided by Foster et al. (1997) who compare
CV responses to actual donations for public goods, in this case the preservation of
various UK bird habitats. This study found that while OE CV bids were on average
not signi¬cantly different from those which could be expected in the real payment
context, individuals presented with a hypothetical market were signi¬cantly less
likely to opt out of making a bid than those faced with making real donations.
Contrasting these ¬ndings with the evidence for understatement in OE WTP
responses (reviewed above) suggests that, in practice, some people respond to
the OE elicitation method by free-riding, while others strategically overstate.17

17 Further evidence for such a view can be gleaned from the relatively low degree of ¬t attained by statistical
models of OE CV responses. If individuals are responding in diametrically opposed ways to these questions,
then models are inevitably going to struggle to explain such data.
Recreation: valuation methods 25

However, meta-analyses of CV studies commonly report that OE-format analy-
ses record signi¬cantly lower WTP amounts than do those using other elicitation
methods (see, for example, Brouwer et al., 1999, and the study presented sub-
sequently in this volume). Therefore, on balance, OE formats appear to result in
under- rather than overstatement and may, in the absence of superior measures, be
justi¬ed as providing conservative estimates of underlying values.
(ii) ˜Good respondents™. Orne (1962) points out that the relationship between
analyst and respondent is an interactive process with the interviewee seeking clues
as to the purpose of the experiment. If this purpose is inadequately conveyed then the
respondent may react in one of two ways: either she will not give the questions due
consideration or she will attempt to guess the ˜correct™ answers, i.e. she will try to
be a ˜good respondent™ and give the answers which she feels that the analyst wants.
The problem of limited involvement may be assessed by recording and analysing
both the numbers of respondents who refuse to take part in the survey and the
length of interview. The good respondent problem may be exacerbated where the
interviewer is held in high esteem by the respondent (Harris et al., 1989), resulting
in responses which differ from true WTP. Desvousges et al. (1983) found little
evidence of such a bias but it should be noted that this study employed professional
interviewers, a potential solution to such problems. Tunstall et al. (1988) further
recommend that interviewers follow the wording of the questionnaire exactly and
that respondents be presented with a choice of prepared responses so as to minimise
over- or understatement of true evaluations.
In our own empirical work considerable emphasis has been placed upon minimis-
ing such sources of bias at the design stage. Experienced practitioners (including
several of those cited above) were consulted regarding the construction of ques-
tionnaires and execution of surveys.
(iii) Upward rounding. Bateman et al. (1993b) argue that respondents in a DC
format survey may have an incentive to accept bids which are in excess of true WTP
if the difference between the two amounts is relatively small. The deviation caused
by such an effect will only operate in an upward manner, i.e. the respondent will
not refuse to pay a bid level which is just below their true WTP. However, provided
that the respondent believes in the payment obligation (i.e. she does not engage in
strategic overbidding) this should be a relatively minor effect.
(iv) Anchoring. Kahneman et al. (1982), among others, have argued that respon-
dents faced with an unfamiliar situation (particularly where the good is also poorly
described) will interpret the DC bid level to be indicative of the true value of the
good in question (Kahneman and Tversky, 1982; Roberts et al., 1985; Kahneman,
1986; Harris et al., 1989; Green et al., 1998). Here the introduction of a speci¬c
bid level raises the probability of the respondent accepting that bid. Proponents of
this idea argue that this ˜anchoring™ effect may occur where a respondent has not
26 Applied Environmental Economics

previously considered her WTP for a resource (which is likely with regard to public
or quasi-public goods) and/or is unclear in her own mind about the true valuation.
In such cases the proposed bid level may provide the most readily available point
of reference onto which the respondent latches. There is no a priori presumption
about the direction of such an anchoring effect.18
(v) Starting point effects. Several studies have suggested that the use of an initial
starting point in iterative bidding (IB) games may signi¬cantly in¬‚uence the ¬nal
bid; for example, the choice of a low (high) starting point leads to a low (high) mean
WTP (see Desvousges et al., 1983; Roberts et al., 1985; Boyle et al., 1985; Navrud,
1989a; Green et al., 1990; Green and Tunstall, 1991). While the use of starting points
may reduce non-response and variance, commentators argue that such an approach
may lead respondents to take cognitive short-cuts to arrive at a decision rather
than thinking seriously about their true WTP (Cummings et al., 1986; Mitchell and
Carson, 1989; Loomis, 1990). It has also been noted that informing respondents
as to the construction costs associated with a proposed environmental change may
affect resultant bids (Cronin and Herzeg, 1982). One approach to this problem is
to allow the respondent to choose a bid from a range shown on a payment card
(Rowe et al., 1996). However, in some instances the choice of payment range on a
card may affect reported WTP bids (for example, if respondents assume that such
a range implies information about the ˜correct™ valuation response; see discussions
in Kahneman and Tversky, 1982; Roberts and Thompson, 1983; Kahneman, 1986;
Harris et al., 1989; Dubourg et al., 1997).
In summary, we can see that different elicitation formats may in theory result
in either understatement or overstatement of values. We now consider empirical
evidence concerning elicitation effects.

Elicitation effects: empirical evidence
Our own studies of elicitation effects have been conducted for both woodland and
other resources. Three elicitation methods (OE, DC and IB) were assessed19 in a
CV study examining users™ WTP for environmental preservation in the Norfolk
Broads, a unique wetland area located in East Anglia, UK (Bateman et al., 1995a,
1999b). Following a pilot survey (discussed subsequently in relation to payment
vehicle effects), a main survey sample of about 3,000 visitors was obtained through
face-to-face on-site interviews. This sample was divided up to permit suf¬cient

18 A related problem in DC (and potentially other) formats is the phenomenon of ˜yea-saying™ or ˜nay-saying™,
whereby the respondent decides ex ante to answer positively or negatively irrespective of the actual bid presented
(see Kanninen, 1995; Alberini and Carson, 2001).
19 An iterated exercise, in which an initial DC question was supplemented with two follow-up amounts, was also
conducted; see Langford et al. (1996) and Bateman et al. (2001b). This latter analysis supports the existence
of anchoring effects within iterated DC designs.
Recreation: valuation methods 27

Table 2.2. WTP for preservation of the Norfolk Broads
using various elicitation methods

Mean WTP 95% con¬dence interval
Elicitation method (£ per annum) (£ per annum)
DC 144 75“261
IB 75 70“81
OE 68 60“75

responses to test the various elicitation formats under investigation. WTP results
from these formats are summarised in Table 2.2.
Inspection of Table 2.2 shows that the results conform to our prior expectations
concerning potential elicitation effects. The overstatement and understatement in-
centives inherent in the DC and OE formats, respectively, seem to be re¬‚ected in the
ordering of derived valuation measures with mean DC WTP being more than twice
the OE estimate. Of course, it could be that while OE responses were downwardly
biased by free-riding (of which some, although not pervasive, evidence was found),
DC responses were unbiased. However, statistical analysis of the determinants of
WTP amounts provided evidence of a strong link between the starting point in the
IB bidding game and the ¬nal valuation amount stated. This in turn suggests that
DC bid levels might well be interpreted in a similar fashion, i.e. as anchoring points
which respondents used as heuristic indicators of the ˜correct™ valuation of the good
under investigation.
This analysis shows that we cannot reject the hypothesis that all elicitation for-
mats are, in one way or another, biased instruments for obtaining WTP values.
Which, if any, should be used in our subsequent research on woodland values? The
answer to such a question is still unclear and the subject of considerable ongoing
research within the CV community world-wide. For the purposes of the research
described here we have adopted a simple rule that, wherever possible, we should
employ lower-bound assumptions and conservative techniques, thus enhancing the
robust nature of derived results.20 As a consequence, in most of the woodland
evaluation research presented in Chapter 3, we adopt an OE elicitation format
for our CV studies (with a comparison against a payment card approach in one
study) on the grounds that such a choice is likely to produce conservative estimates
of WTP.

20 Such an objective accords well with the practice of H.M. Treasury with regard to its evaluations of non-market
woodland recreation values. The guidelines for best practice in CV studies given by the US NOAA Blue Ribbon
Panel (Arrow et al., 1993) also emphasise conservation design although notably they recommend the use of
DC-style referendum elicitation formats because of their desirable theoretical incentive properties.
28 Applied Environmental Economics

Table 2.3. Payment vehicle analysis results

Payment Sample Zero WTP Mean WTP S.E. Median WTP Coeff. of
vehicle size (%) (£) mean (£) variation (%)
DONATE 157 46.5 25.60 3.18 10.00 156
FUND 65 23.1 47.60 17.40 10.00 296
TAX 211 11.8 89.22 9.98 40.00 162

Payment vehicle effects
The idea that the way in which a payment is made is liable to affect an individual™s
willingness to make that payment (and implicitly the size of payment) is self-
evident from the expansion of credit and payment options schemes within modern
Western society. That such payment vehicle effects should arise in the purchase
of public goods is therefore not surprising and indeed can be seen as evidence
that CV respondents act as if hypothetical markets are binding. Payment vehicles
can usually be described in terms of two characteristics: collection mechanism and
temporal extent.
Considering ¬rst the commonly adopted approach of asking survey respondents
annual payment questions a number of tax-based and donation-based collection
mechanisms appear in the CV literature. The impact of varying these was studied
through an earlier survey of visitors to the Norfolk Broads (Bateman et al., 1993b).
Here a sample of over 400 respondents were presented with one of three payment
vehicles: (i) an unspeci¬ed charitable donation (the DONATE vehicle); (ii) a pay-
ment to a hypothetical charitable fund speci¬cally set up to facilitate ¬‚ood defence
work in the Norfolk Broads (FUND); and (iii) payments made via direct taxation
All payment vehicles were applied using an OE WTP elicitation method. Re-
sults, which are detailed in Table 2.3, show that both the DONATE and FUND
vehicles elicited large numbers of zero WTP bids (46.5 per cent and 23.1 per cent
respectively), which contradicts prior expectations regarding a sample of visitors
to the Broads who are expected to derive considerable value from the area. In con-
trast the TAX vehicle produced by far the lowest zero-bid rate (11.8 per cent) and
also performed better in terms of bid variability than the FUND vehicle, and about
as well as the DONATE vehicle. As no vehicle produced excessive evidence of
strategic bidding (large numbers of unreasonably high bids) this was not deemed a
All respondents in the Broads study were asked why they had responded in the
way they had. Many of those presented with the FUND and (especially) DONATE
vehicles commented that they were unhappy that such a vehicle would not be
binding upon all and that they were not con¬dent that payments via such vehicles
Recreation: valuation methods 29

would be fully channelled towards preservation work (trust funds were not to be
trusted!). Conversely, many of those responding to the TAX vehicle commented
that, while they disliked paying extra taxes, they had con¬dence that such money
would be spent ef¬ciently upon any ¬‚ood defence scheme.
In both the Norfolk Broads and woodlands studies a tax-based vehicle also has
the advantage of being the most likely method by which changes in the provision of
these quasi-public goods would be funded. This and the advantages outlined above
made such a payment vehicle the preferred approach for our subsequent woodland
studies. However, a remaining issue concerned the advantages of local taxes relative
to national ones, a topic which is discussed in Chapter 3.
Turning to consider variation in the temporal extent of payment vehicles, a num-
ber of researchers have experimented with per visit measures for which entrance fees
had been used. Several studies have noted differences in implicit values when both
per annum taxation and per visit entrance fee measures were obtained for the same
good (see, for example, Rowe et al., 1980, on landscape values; also Desvousges
et al., 1983; Brookshire and Coursey, 1987; Navrud, 1989b). Given this result we
felt it would be interesting to examine whether such a result was obtained when
applying the same design to UK woodlands, and if so why. Consequently, such a
comparison was made a further objective of our woodland valuation research.

Questionnaire design impacts: budget constraint and ordering effects
An area of particular interest was the impact which changes in the questionnaire
might have upon stated values when the valuation question itself was kept constant.
This was assessed through a joint consideration of two design issues: (i) the inclu-
sion or exclusion of a question (prior to the valuation question) asking respondents
to calculate their relevant annual recreational budget; (ii) the impact of changing
the order in which per annum and per visit valuation questions were presented to
The relevant economic theory concerning the budget constraint issue is pre-
sented in the mental accounting literature (Deaton and Muellbauer, 1980; Tversky
and Kahneman, 1981; Kahneman and Tversky, 1984) where total income is initially
allocated to various broad categories of expenditure (e.g. housing, food, recreation,
etc.), and then, in a second stage, subdivided among the speci¬c items which con-
stitute each category (e.g. the recreation category budget is allocated among forest
recreation, water recreation, etc.). Because of the hypothetical nature of the CV
market a potential problem may arise if respondents fail to consider all relevant
material such as the relevant category budget. Evidence on the impact of explicitly
asking respondents to consider income constraints prior to stating WTP sums is
mixed (Burness et al., 1983; Schulze et al., 1983; Willis and Garrod, 1993; Loomis
30 Applied Environmental Economics

et al., 1994) and consequently it was decided to make this an objective of subsequent
empirical investigation.
Question-ordering effects are considered by Brookshire et al. (1981), Tolley and
Randall (1983) and Hoevenagel (1990) who note that a good will elicit a higher WTP
response if placed at the top of a list of goods to be evaluated than if it is positioned
lower in the order. Similar evidence is presented by Kahneman and Knetsch (1992)
as part of a series of tests examining the extent to which CV responses are the
product of moral satisfaction (i.e. the ˜warm glow™ of contributing to a good cause)
rather than being linked to the characteristics of the good under evaluation. This
paper has triggered a wide empirical debate and stimulated theoretical research
arguing that, as the consumption of a given good in isolation is not identical to
the consumption of the same good as part of a larger set (because the other goods
in the set may be substitutes for, or complements to, the good under question),
then this phenomenon need not violate economic theory (see Carson et al., 1992,
1998; Randall and Hoehn, 1992, 1996; Carson and Mitchell, 1995; Rollins and
Lyke, 1998). While not attempting to establish whether variation is due to moral
satisfaction, warm glow or theoretically expected effects,21 we do use ordering
effect tests and simpler sensitivity analyses to establish the extent of such variation
in values. Furthermore, by combining the budget constraint and question-ordering
investigations within a split-sample design, a further analysis of the interactions
of these effects could be undertaken to see whether design effects might multiply
through a study.

Summary of woodland CV research objectives
CV is a widely applicable and widely applied monetary evaluation method with
a consistent basis in economic theory. Given the breadth of the current research
debate, we have deliberately focused our empirical investigations on a subset of
related issues which together examine the impact of differing designs upon elicited
values. The issues addressed in our subsequent woodland CV studies are:
(i) variations in WTP values between users and non-users of woodland recreation
(ii) the WTA compensation levels demanded by farmers for providing woodland recreation
opportunities on their land
(iii) elicitation effects (speci¬cally, a comparison between OE and payment card ap-
(iv) the choice of payment vehicle (in terms of both local versus national tax payment
collection mechanisms and annual versus per visit temporal extent of payments)
(v) budget constraint impacts
(vi) question-ordering effects.
21 Our recent research on the relationship between study design and such effects is reported in Bateman et al.
Recreation: valuation methods 31

Taken together it was intended that these studies would provide some insight into
the variability of valuation responses with changes in CV study design. These values
could then be compared to those derived from the travel cost method, to which we
now turn.

The travel cost method
Like the CV method, the travel cost (TC) approach relies upon a survey to gather
data. However, whereas a CV survey can, in principle, be applied in almost any
situation, a TC survey must involve at least a high proportion of users of the recre-
ational asset in question. Most typically this involves on-site surveys in which a
questionnaire is used to collect data on users™ place of residence; necessary demo-
graphic and attitudinal information; frequency of visit to this and other sites; trip
information such as purpose, length, associated costs, etc. From these data, visit
costs can be calculated and related, with other relevant factors, to the frequency of
visits in a ˜trip generation function™ (TGF) from which a demand relationship may
be established (for details see Freeman, 1993; Champ et al., forthcoming).22
As discussed by Bockstael et al. (1991), the literature can be divided into random
utility models (RUM), which examine the probability of a visit to a site given
information on all possible visit sites, and more basic TC models which predict
visits to a given site by utilising data collected from a survey of visitors at that site.
While theoretically more elegant, RUMs require more data than were available to
this research and so the more basic approach is employed here.
Two variants of this style of TC model can be identi¬ed depending on the de¬ni-
tion of the site visits variable (Bateman, 1993). The ˜individual travel cost™ (ITC)
method focuses on the number of site visits made by each visitor over a speci¬c
period, say one year. The ˜zonal travel cost™ (ZTC) method, on the other hand,
partitions the entire area from which visitors originate into a set of visitor zones
and then de¬nes the dependent variable as the visitor rate (i.e. the number of visits
made from a particular zone in a period divided by the population of that zone). In
both cases an uncompensated demand curve can be derived and consumer surplus
estimates of recreational value obtained.
The UK woodland recreation literature includes examples of both ITC and ZTC
applications which we review in Chapter 3. Results obtained from applying the two
22 Note that Randall (1994) provides a fundamental caution to those who assume that the revealed preference
nature of the TC approach gives it automatic ascendancy over expressed preference methods such as CV. He
notes that the TC method relies upon researcher-assigned visitation cost estimates rather than observable visit
prices and argues that these are inherently subjective, such that the method yields only ordinally measurable
welfare estimates. In essence, while the CV method at least presents respondents with hypothetical costs,
visitors never see the implicit travel costs used to calculate consumer surplus estimates in TC studies. An
empirical assessment of ˜Randall™s Dif¬culty™ is given by Common et al. (1999).
32 Applied Environmental Economics

Table 2.4. ZTC/ITC consumer surplus estimates for six UK forests

Travel cost CS/visitor Travel cost CS/visitor CS ratio:
Forest coef¬cient (£) coef¬cient (£) ZTC/ITC
’0.384 ’0.358
Brecon 2.60 1.40 1.86
’0.444 ’0.996
Buchan 2.26 0.50 4.52
’0.525 ’1.259
Cheshire 1.91 0.40 4.78
’0.694 ’0.327
Lorne 1.44 1.53 0.94
’0.702 ’0.215
New Forest 1.43 2.32 0.62
’0.396 ’0.386
Ruthin 2.52 1.29 1.95

Notes: All coef¬cients produced via OLS techniques and signi¬cant at 5%
level; travel cost de¬ned as full running costs; consumer surplus estimates at
1988 prices; n = 21 for all forests.
Sources: Garrod and Willis, 1991; Willis and Garrod, 1991a.

variants to the same data have been shown to be substantially different. Table 2.4
illustrates this point with regard to a joint ZTC/ITC study of six UK forest sites.
Using the same estimation procedure and cost assumptions throughout,23 estimates
of consumer surplus produced by ZTC ranged from almost 40 per cent smaller to
almost ¬ve times larger than those produced by ITC. As all the cost coef¬cients
produced by both methods are statistically signi¬cant this indicates some serious
problems for one or both of these approaches.
One limitation of the ZTC approach is the dif¬culty associated with the use of
an average value as a dependent variable. Employing a zonal visitor rate means
that it is impossible to use individual-speci¬c explanatory variables. For example,
membership of an environmental or outdoor pursuits association may well be a
highly signi¬cant predictor of recreational visits. However, information on such
individual characteristics cannot be used in the ZTC approach and a constructed
zonal average for such variables is likely to be highly inef¬cient (Brown and Nawas,
1973). Similarly, intrazonal variation is to a considerable degree lost in the ZTC
approach, as interzonal average effects dominate in curve-¬tting. An extreme case of
this occurs where concentric, circular travel time zones are used with no distinction
being made within the resultant circles for other variables such as socio-economic
or substitute availability measures.24
While concentric zones are common in earlier ZTC applications,25 other ap-
proaches to zonal de¬nition are perfectly feasible. The de¬nition of the width

23 See table notes for details. Cost de¬nition and estimation issues are discussed brie¬‚y later.
24 For an illustration, see Bateman (1993).
25 Furthermore, zones may be cut off at some ¬nite distance although the outer band may be in¬nite. Englin and
Mendelsohn (1991), in their study of rainforest tourism, analyse visits from all countries.
Recreation: valuation methods 33

and number of zones is typically either arbitrary or in¬‚uenced by the availability
of demographic data; for example, B¨ jo (1985) uses county boundaries to de¬ne
zones. In effect, each possible de¬nition of zones implies a different aggregation
of population and, in practice, almost certainly a different visitor rate. This, in turn,
will imply changes in the estimated demand curve and thereby different consumer
surplus estimates. Therefore, in reality, it is almost certain that an analyst could
respecify zones so as to either in¬‚ate or reduce valuation estimates as required.
This is an example of the more general phenomenon known as the modi¬able areal
unit problem (MAUP) (Openshaw, 1984). The extent to which valuation results
may alter is uncertain26 and there is active research into statistical aspects of the
MAUP issue (e.g. Batty and Longley, 1996).
A further problem for ZTC, which again does not af¬‚ict the ITC method, is that
R2 statistics will always be upwardly biased. This arises as a natural consequence
of aggregating individual responses across zones and so reducing the number of
curve-¬tting points to the number of zones. Consequently the very high R2 values
recorded in many ZTC studies should be treated with extreme caution. Their only
real validity is as indicators of which model has relatively higher explanatory power
within any particular functional form; their absolute value should be disregarded
(and even not reported as it may well be misleading). This criticism does not apply
to the ITC for which goodness-of-¬t statistics are, in this respect, unbiased.
Given these problems, Brown and Nawas (1973) argue that the ZTC method
is inef¬cient and therefore prefer the use of ITC, a sentiment echoed in early
applications by Gum and Martin (1975) and Bowes and Loomis (1980). Indeed the
US literature over the past two decades has slowly moved from the use of ZTC to
employing ITC. However, the ITC approach is not without problems.
Dobbs (1991) points out that most ITC studies to date have incorrectly estimated
consumer surplus, in that they have ignored the inherently discrete nature of the
dependent variable. In such cases the integration of a smooth demand function may
lead to signi¬cant bias in consumer surplus estimates. However, Dobbs develops a
programmable approach to the computation of discrete dependent variable bene¬ts
which overcomes this problem.
A more fundamental problem with ITC occurs where a high proportion of visitors
make only one visit per annum or are ¬rst-time visitors (Freeman, 1979; Bowes
and Loomis, 1980). In such cases, statistical techniques commonly used in ITC
analyses may not have a suf¬cient spread of observations to make the approach
operational. In recent work we have addressed this problem through the application
of Poisson distribution models,27 details of which we present in Chapter 4.

26 See also Christensen (1985) and Price et al. (1986).
27 However, a Poisson regression will have problems of underdispersion with large numbers of low counts.
34 Applied Environmental Economics

In conclusion, the decision to use either zonal or individual TC approaches may
have a substantial impact upon the results obtained. While there are a number of
methodological problems associated with the application of both, these seem more
tractable in the case of the ITC approach, which also has theoretical advantages
over ZTC (Bockstael et al., 1991). Consequently we adopt the ITC method for use
in the valuation studies presented in Chapter 3. However, the recently developed
literature on bene¬ts transfer has used the ZTC approach as a readily tractable
technique for estimating the numbers of visitors arriving at a given site (Loomis
et al., 1995). The area visit rates used in the ZTC provide as much information
regarding which areas do not yield visitors (e.g. those which are distant from the
site, are socio-economically disadvantaged, etc.) as those which do. Therefore the
technique yields demand functions which can readily be applied across a study area
of which the site distance, socio-economic and other characteristics are known to
yield defensible estimates of the number of arrivals expected at a site.
As a consequence, while we use the ITC approach to estimate the value of a
recreational trip to a woodland, our model of the number of trips made to woods
and the latent demand for trips to potential new woodlands (presented in Chapter 4)
owes more to the zonal-based approach of the ZTC method.

Focal methodological issues
As before, our methodological review focuses exclusively upon those issues which
are addressed in our subsequent empirical work, with the interested reader being
referred to the previously cited literature for wider discussions.

Calculating travel costs
Travel costs are composed of two principal elements: direct travel expenditure
(e.g. petrol costs) and the opportunity cost of time.

Travel expenditure
Two issues are pertinent here: measurement and valuation. Accurate measurement
is a vital ingredient of valid welfare estimation. However, we have shown elsewhere
that a number of questionable simpli¬cations are commonly adopted in the distance
calculations underpinning expenditure estimates (Bateman et al., 1996a, 1999a;
Brainard et al., 1999). For example, rather than using the actual point from which a
visitor starts their journey, many ITC studies use centre points or ˜centroids™ of cities
(Rosenthal et al., 1986) or counties (Mendelsohn et al., 1992) as outset origins.28
This may cause a systematic error given that the very basis of the TC method

28 Note that we are referring here to a problem with ITC studies, although all ZTC studies, by their very nature,
also use zonal outset areas.
Recreation: valuation methods 35

is that individuals are mindful of costs in determining their choice of recreation,
i.e. we would expect, ceteris paribus, that within any area there would be more
visitors from outset locations nearer to the study site than from further away. The
use of a centroid will partly mask that variation as all visitors within the boundary
of the outset area will be assumed to travel from the common central point. This
should, on average, lead to an overestimation of the travel costs faced by visitors
from a given area as, within that area, most visitors come from locations which
are closer to the study site than is the centroid point. The larger the outset area
used, the greater we would expect any resultant error to be. A further measurement
issue concerns assumptions regarding routing. The use of constant road speeds or
straight-line distances ignores the extent and quality of the road network which
underpins true travel distances and times (Rosenthal et al., 1986). We address all
of these aspects of the measurement issue directly in our empirical studies through
the application of GIS techniques.
Turning to the valuation issue, a variety of alternative approaches can be identi-
¬ed; for example Boj¨ (1985) simply refers to the economy class rail fare. However,
such a simple approach is less applicable to car travel, where three cost calculation
options exist:

(i) petrol costs only (marginal costs)
(ii) full car costs: petrol, insurance, maintenance costs, etc.
(iii) perceived costs as estimated by respondents.

Clearly, using option (ii) will raise visit costs above that of option (i) and ul-
timately increase consumer surplus estimates, a result con¬rmed in comparisons
of these approaches undertaken by both Hanley and Common (1987) and Willis
and Benson (1988). Price (1983) and Christensen (1985) argue that the correct cost
measure is that which visitors perceive as relevant to the visit. It may well be that
visitors are poor at perceiving daily insurance and maintenance cost equivalents or
that they see these as sunk costs which do not enter the TGF, i.e. they only consider
the marginal cost of a visit, equating this with marginal utility. As a result of this
apparent con¬‚ict we adopt a sensitivity analysis approach in our empirical work,
testing all three of the above cost de¬nitions.

Time costs
Time enters the visit cost function through the travel time and on-site time variables.
However, theoretical analysis (McConnell, 1975, 1999; Freeman, 1979; Wilman,
1980; Johannson, 1987; Shaw and Feather, 1999a, 1999b; Berman and Kim, 1999)
shows that the relevant opportunity costs per hour need not be the same for these two
items. Furthermore, determination of these opportunity costs raises considerable
36 Applied Environmental Economics

Travel time values are particularly dif¬cult to analyse in that, as noted previously,
we have no de¬nite a priori notion of whether travel time utility is positive or
negative. If travel time has positive utility, then using some general travel time cost
as a price will overestimate the consumer surplus of a visit. This will be the case
for ˜meanderers™ who gain utility primarily from the journey itself (Cheshire and
Stabler, 1976). Boj¨ (1985) does not include a travel time cost (i.e. implicitly he
gives such time an opportunity cost of zero) on the grounds that 80 per cent of survey
respondents expressed a positive utility for travel time to the site under analysis.
This approach assumes that ignoring residual travel time costs only leads to a minor
underestimate of the true consumer surplus.29 However, this approach is far from
standard. Indeed static optimisation of any conventional utility function (subject to
income and time constraints) would indicate that the marginal rate of substitution
between labour and leisure (i.e. the value of recreational travel time) is equal to the
wage rate. However, when individuals are not able completely to vary the number
of hours worked the substitution of time for money becomes constrained and the
direct relation between the value of time and the wage rate breaks down (Johnson,
1966; McConnell, 1975).
Early applied investigations of the relationship of wages to travel time were
undertaken by Cesario (1976) and Cesario and Knetsch (1970, 1976). These papers
examined commuters™ choice of transport to and from work (and relevant costs) to
estimate an implicit value of travel time. Cesario (1976) concludes that, ˜on the basis
of evidence collected to date, the value of time with respect to nonwork travel is
between one quarter and one half of the (individual™s) wage rate™ and subsequently
uses a value of one-third the wage rate to price travel time.30 However, this analysis
only considers commuting time and there is no necessary reason why the marginal
utility obtained should be applicable to recreation travel time.
Common (1973) and McConnell and Strand (1981) use an iterative process
whereby successive time values are substituted into the TGF, the ¬nal choice being
determined where the explanatory power (R2 ) of the model was maximised.
Desvousges et al. (1983) apply the value of time results of Cesario (1976),
McConnell and Strand (1981) and a full wage rate assumption to an ITC model of
individual visitation patterns at twenty-three water recreation sites in the USA.
Testing at the 10 per cent con¬dence level, Desvousges et al. (1983) reject the
McConnell and Strand (1981) approach, while the Cesario (1976) and full wage
assumptions perform equally well, both being rejected in approximately seven of
the twenty-three cases. On the basis of these results Smith and Desvousges (1986)
29 Johansson (1987) points out that if time costs are ignored, then ˜the estimated curve will be located inside and
be less steep than the “true” one, except possibly for those living very close to the recreation site, since the
underestimation of costs increases in relation to distance from the visitor™s zone of origin™.
30 An alternative approach is that of Nelson (1977) who calculates a marginal implicit price of proximity to the
central business district with housing data for Washington, D.C., from which he derives a value of time which,
when related to wage rates, falls within the Cesario range.
Recreation: valuation methods 37

conclude that ˜for practical purposes, there is no clear-cut alternative to . . . using the
full wage rate as a measure of the opportunity cost. Even though it may overstate
the opportunity costs . . . none of the simple adaptations are superior.™
Similar results are obtained in a completely different cultural setting by
Whittington et al. (1990) in a study of the value of time spent collecting water
in Kenya. Here two separate approaches are employed, both of which indicate
a value of time approximately equivalent to the wage rate for unskilled labour.
However, activities such as collecting water are qualitatively different from those
associated with recreation. In their TC study of UK forest recreation, Benson and
Willis (1992) employ three wage-rate-based value of time assumptions:

(i) 0, which assumes that visitors would not bene¬t from some alternative recreation
(ii) 25 per cent, the UK Department of Transport™s value of non-working time used in CBA
assessments of road proposals up to 1987
(iii) 43 per cent, the value of time used by the UK Department of Transport following its
review of non-work time in 1987 (Department of Transport, 1987).

While the Cesario approach is, on the surface, theoretically and practically ap-
pealing, a deeper analysis of the complexities of the work/leisure relationship high-
lights some important problems. In a thorough analysis, Bockstael et al. (1987)
note two major issues: (i) wage rates may vary with work hours; for example, a
second job may pay a lower rate than the primary one; (ii) individuals face uneven
time constraints, i.e. they may be restricted to working speci¬c hours in particular
jobs. As a result the wage rate may be an appropriate measure of time costs for
those (at interior solutions) who can fully vary their work hours, but it will be inap-
propriate for those who cannot (at corner solutions). While Bockstael et al. provide
a theoretically plausible approach to the valuation problem by incorporating time
and income constraints into a utility function, the empirical application of such
a technique is problematic. In particular the data requirements of such a model,
including information regarding each individual™s time constraints, are exacting.
For these reasons such complex approaches have not been widely adopted and no
published UK study has attempted such an analysis.
Shaw (1992) provides a number of suggestions as to how the value of time
problem might be addressed in a practical study. One suggestion is to use CV-type
questions to elicit WTP for recreation time,31 while another is to accept that there is
likely to be some rather unclear link with wage rate and therefore to use a sensitivity
analysis approach with a wide range of wage fractions.

31 We have employed a similar approach in a TC study of the Norfolk Broads (unpublished). Here respondents were
asked WTP to reduce travel time. However, many gave a zero response indicating that the journey contributed
positively to trip utility. Further direct questions con¬rmed this ¬nding.
38 Applied Environmental Economics

As far as the unit value of on-site time is concerned, if the length of time spent
on site were a constant for all visits to a particular site, then such costs could
effectively be ignored as they would imply only an increase in absolute visit costs
but not a change in marginal relationships. Furthermore, in an empirical analysis,
Boj¨ (1985) ¬nds no evidence to refute an assumption of constant on-site time
costs, while Bockstael et al. (1987) omit on-site time from their empirical analysis
because of its potentially ambiguous effect upon demand arising from its inclusion
within both the utility function and the constraints.

Summary: treatment of travel costs
The treatment of travel and time costs within the TGF is one of the most crucial
issues in operationalising the TC method. The approach we have adopted in this
study is as follows.
Measurement. One fundamental issue concerns the measurement of linear and
temporal distance. We believe that the use of GIS to analyse digital road net-
works (incorporating road length, quality and average travel time by individual
road segment) in certain of our TC studies considerably enhances the accuracy of
measurement compared to that in most other published research.
Travel expenditure. Following the above review we adopt three de¬nitions of
monetary travel costs: petrol only; petrol plus standing charges (insurance, depre-
ciation, etc.); and respondents™ perceived travel cost.
Time costs. We adopt the suggestion of Shaw (1992) and perform a wage rate sen-
sitivity analysis upon travel time. Four wage rate values are employed: 0 (following
the argument of Benson and Willis, 1992); 43 per cent (the UK Department of
Transport™s value of time); 100 per cent (following the empirical ¬ndings of Smith
and Desvousges, 1986); and the variable wage rate percentage which provides the
best ¬t to the data (our preferred option). We recognise the limitations of such an ap-
proach and that the labour supply method of Bockstael et al. (1987) is theoretically
superior. However, such an analysis is both complex and demanding in terms of
data requirements. Given limited resources our approach should provide a reason-
able approximation, while yielding an analysis which is more rigorous than other
contemporary UK studies. In line with such research, we have omitted on-site time
from the cost function (although such data were collected and analysed), following
the argument that this may not signi¬cantly affect consumer surplus.32
Total travel costs. Given that travel and time costs are both functions of distance,
their inclusion together within the TGF is likely to create signi¬cant problems of
multicollinearity. Accordingly (and for additional reasons reviewed subsequently)
we use the common approach seen in studies from Cesario and Knetsch (1970)
32 Following the analysis of McConnell (1992a), who shows how on-site time may, in certain circumstances, be
a signi¬cant factor (and proposes a solution to its treatment), we intend to incorporate this into future studies.
Recreation: valuation methods 39

to the present day by adding together travel and time costs to produce total visit
Where pertinent we then multiply the total visit cost by the respondent™s stated
proportion of the total day™s enjoyment attributable to the site in question, thereby
allowing for that share of the day™s utility derived from other sites and the journey
itself. This adjusted visit cost is then entered as an explanatory variable within the

Other explanatory variables
Demand for site visits is likely to be a response to the quality and attributes of
a site, yet multicollinearity problems may make the incorporation of numerous
such attributes within a single function dif¬cult. Early TC studies tackled this issue
through the construction of single-variable quality indices (Ravenscraft and Dwyer,
1978; Talheim, 1978). However, such an index cannot be adequately de¬ned without
full knowledge of the functional relationship between demand and site attributes.
As this relationship is dictated by individual preferences for different attributes, the
creation of a truly representative index is impractical.
Subsequent research has attempted to tackle the problem via multisite studies
(Vaughan and Russell, 1982) or by adopting two-stage estimation procedures in
which collinear quality attributes are omitted from the ¬rst stage (an otherwise
conventional TGF) but then used as explanatory variables in a series of second-
stage models which predict each of the independent variables used in the initial
analysis (Smith and Desvousges, 1986). Such a two-stage procedure is modi¬ed
for use in our models of agricultural value presented in Chapter 8.
In our own analyses we initially omit consideration of site quality impacts, con-
centrating instead on the development of improved measures of the principal ex-
planatory variable, travel cost, through use of GIS techniques. However, in our
discussion of ongoing work presented in Chapter 4, we detail recently developed
models which use these same GIS techniques to incorporate detailed site quality
variables into our TC models. A similar approach is taken to the issue of substitute
sites, consideration of which is omitted from the models presented in Chapter 3
but included in Chapter 4, where we show how GIS techniques can produce highly
detailed variables quantifying accessibility to alternative sites, measures which are
readily incorporated into TC models.

Functional form
Analysts are faced with a variety of functional forms under which the TGF can be
speci¬ed (typically linear, quadratic, semi-log, double-log and Box“Cox). None of
these has strong theoretical ascendancy over the others. However, speci¬cation of
a linear form produces a ¬rst derivative which will be a constant and is therefore
40 Applied Environmental Economics

theoretically problematic, implying as it does non-diminishing marginal utility for
additional trips to a site and thus that the individual cannot decide how many
trips to make in total. Log forms may be useful for elasticity estimates and have
the advantage of avoiding negative values for the dependent variable.33 However,
the double log form may also be criticised on theoretical grounds as its asymptotic
properties imply in¬nite visits at zero cost, an attribute which is particularly unlikely
for demand curves for on-site experience (see Everett, 1979).
An altered functional form (even if it has similar explanatory power) can have
a highly signi¬cant impact upon the demand curve and resultant consumer surplus
estimates. In an early TC study of recreational ¬shing in Grafham Reservoir (UK),
Smith and Kavanagh (1969) found that both semi-log (dependent variable) and
double-log functions ¬tted the data very well (R2 = 0.91 and 0.97 respectively).34
However, when the resultant demand curves were examined it was found, at a zero
admission price, that the semi-log form predicted 54,000 annual visits while the
double-log form estimated over 1,052,000, with obvious consequences for con-
sumer surplus estimates. Subsequent re-estimation made little difference to this
From a statistical viewpoint the most appropriate functional form may be eval-
uated by examining relative degrees of explanation. However, R2 tests are strictly
non-comparable where the dependent variable changes. A more valid test is to
compare visitor rates predicted by the model with observed visitor rates using
either a large sample, Wilcoxon signed rank test35 or a Mann“Whitney U test36 as
Because of its large potential for disturbing consumer surplus estimates, we see
the functional form issue as one of the most serious problems affecting the TC
approach (as pointed out, it may potentially have far more impact than substitute
site or congestion effects). Consequently we have made this a priority issue in our
applied research. We investigate a variety of functional forms38 and estimation pro-
cedures (see below) with regard to both the valuation models detailed in Chapter 3
and the prediction of arrival numbers discussed in Chapter 4.

Estimation procedure
Pearce and Markandya (1989) point out that a truncation bias may be introduced
where ordinary least squares (OLS) estimation techniques are employed with ITC
33 See, for example, Ziemer et al. (1980); Vaughan et al. (1982); Desvousges et al. (1983); Smith and Desvousges
(1986); Hanley (1989); Benson and Willis (1990).
34 This was a ZTC study, for which R2 ¬gures are, as previously discussed, upwardly biased.
35 Wilcoxon (1945); see Mendenhall et al. (1986: p. 806).
36 Mann and Whitney (1947); see Kazmier and Pohl (1987: p. 496).
37 Box“Cox approaches to ¬tting functional forms are arguably superior to standard form approaches.
38 The use of various functional forms such as log models also partially addresses the issue of heteroscedasticity
(Maddala, 1988).
Recreation: valuation methods 41

data. The normal error distribution inherent in this technique does not allow for
the fact that in such studies the dependent variable can only take positive values.
This problem has been tackled through the use of procedures such as maximum
likelihood (ML) estimation, where the function can be speci¬ed so as to explicitly
allow for this truncation.
Empirical studies come to differing conclusions regarding the extent of variance
between OLS (truncated) and ML (non-truncated) estimates of consumer surplus.
While some ¬nd relatively small differences (Balkan and Kahn, 1988), others ¬nd
that bene¬t estimates differ substantially (Smith and Desvousges, 1986; Garrod and
Willis, 1991). Given this debate we have employed both OLS and ML estimation
techniques in our ITC studies although valuation estimates from ML models are
preferred and only these are used in the CBA presented at the end of this volume.
Although the theoretical case against OLS methods still applies for ZTC models,
in practice such an approach should produce accurate results where the de¬nition of
zones is such that all have a substantial positive visitor rate (e.g. when relatively few,
often large, zones are used). However, if this is not the case then truncation effects
will again make OLS techniques inappropriate (e.g. where many, often small, zones
are used, some having zero visit rates). While we do not include a ZTC model in
our valuation studies, such an approach is applied to our models of visitor arrivals
at unsurveyed sites (Chapter 4), with OLS techniques being adopted in a study with
relatively few large zones and a Poisson regression model (allowing for truncation)
being implemented in a study with many small zones.

Summary of woodland TC research objectives
The TC method is a potentially useful valuation tool producing uncompensated
consumer surplus estimates of use value. While the zonal (ZTC) approach is seen
as providing a useful basis for prediction of the number of individuals expected
to visit an existing or proposed woodland site (an approach which is developed
in the potential demand models presented in Chapter 4), a review of the literature
indicates an increasing preference for the use of the ITC variant for valuation
purposes. Consequently the TC-based valuation studies presented in Chapter 3 use
the ITC method. The above review has identi¬ed a number of research objectives
for these studies which we summarise as follows:

(i) to investigate the impact of different strategies for measuring travel time and travel
distance upon resultant consumer surplus estimates; in particular, utilising the analyt-
ical capabilities of a GIS, we examine the impact of improving the resolution of the
de¬ned journey outset location and the effect of moving from simple to sophisticated
approaches for modelling journey routing
42 Applied Environmental Economics

(ii) to conduct a sensitivity analysis across a variety of de¬nitions of travel expenditure
and time cost
(iii) to examine the impact of various estimation procedures and functional forms upon
resulting consumer surplus estimates.

The research objectives outlined above are in harmony with those de¬ned previously
for our CV applications in that all of these analyses essentially examine the impact
of varying study design and execution upon derived values. Convergent validity
testing via comparison of CV and TC results provides a further research objective
for our valuation studies.
Recreation: predicting values

While typically unpriced, recreational time is often the most valuable part of any
day (Broadhurst, 2001). This chapter discusses applications of the CV and TC
methods to the valuation of unpriced, open-access recreation in UK woodlands. The
following section presents a review of the existing literature, after which we describe
analyses undertaken as part of this research. We conducted three separate woodland
recreation valuation studies, all in the UK: two in Thetford Forest, East Anglia,
and one in and around Wantage, Oxfordshire. These are subsequently referred to
as the Thetford 1, Thetford 2 and Wantage studies. The design of these studies
re¬‚ected both the previous ¬ndings and research objectives set out in Chapter 2
(i.e. to investigate the validity and sensitivity of measures) and the desire to obtain
values which were of use within our wider CBA. In Chapter 4 we consider the
transferability of these ¬ndings to our wider study area of Wales.

Review of the literature
In the UK there have been more applications of the CV and TC methods to the
evaluation of woodland recreation than of any other open-access recreational good.1
A review of the literature identi¬ed over forty relevant papers containing over
a hundred monetary evaluation estimates (see details in Bateman, 1996). These
included studies calculating national-level values, estimates based on household
once-and-for-all payments and various other measures which were of little use in
our wider study. However, a smaller number of studies provided per person per
visit values which can be readily utilised in valuing the woodland visit numbers
1 We have excluded non-UK studies as we believe that the uncertainties surrounding relevant cultural and socio-
economic differences between countries such as the USA (where the majority of evaluation work has been
conducted) and the UK make such extrapolations of highly dubious value. Loomis (1996) provides a review of
non-UK evaluations of forestry preservation bene¬ts conducted using the CV method.

44 Applied Environmental Economics

estimated in Chapter 4. Our review of previous studies is categorised according to
the various valuation methods employed (ITC, ZTC and CV).

ITC studies
Prior to the present research, the work of Willis and Garrod (1991b) was the only ITC
study giving per person per visit estimates of UK woodland recreation bene¬ts. This
study provided estimates for six sites across the UK. However, problems concerning
sample size and functional form (detailed in Bateman, 1996) mean that we have
reservations about the transferability of these particular results to a wider context
and prefer our own ITC measures discussed later in this chapter.

ZTC studies
Table 3.1 presents results from three separate ZTC studies2 but is dominated by the
multisite analysis of Benson and Willis (1992). The ¬gures reported for this partic-
ular study are from their ˜Standard Model™ where travel expenditure is calculated
using full costs of 33p per mile and travel time is valued at 43% of wage rate (see
the discussion of travel cost de¬nitions in Chapter 2). Consumer surplus values are
given for both the study year and as a 1990 equivalent, the latter being the base year
for our wider CBA study.
The utility of these ¬ndings for estimating recreation bene¬ts at other sites is
discussed later in this chapter.

CV studies
The majority of potentially useful UK woodland recreation studies have been con-
ducted using the CV method. All of the results summarised in Table 3.2 were
derived from WTP questions concerning per person per visit recreation values.
These studies all employed an entrance fee payment vehicle, although a variety of
elicitation methods were used as were both direct ˜use™ and ˜use + option™ value
formats (see Chapter 1), as indicated.

Bene¬ts transfer
To what extent can the results summarised above (and indeed those from our own
studies) be applied to other woodland areas? This issue of transferring bene¬t

2 The study by Christensen (1985) is reviewed in Bateman (1996) but is not included here because of problems,
highlighted by Christensen, regarding the quality of data employed.
Recreation: predicting values 45

Table 3.1. Forest users™ per person per visit recreation values from ZTC studies

Forest Study-year value (£) Study year 1990 value (£)

Benson and Willis (1992)
New Forest 1.43 1988 1.69
Cheshire 1.91 1988 2.26
Loch Awe 3.31 1988 3.91
Brecon 2.60 1988 3.07
Buchan 2.26 1988 2.67
Durham 1.64 1988 1.94
North Yorkshire Moors 1.93 1988 2.28
Aberfoyle 2.72 1988 3.21
South Lakes 1.34 1988 1.58
Newton Stewart 1.61 1988 1.90
Lorne 1.44 1988 1.70
Castle Douglas 2.41 1988 2.85
Ruthin 2.52 1988 2.98
Forest of Dean 2.34 1988 2.76
Thetford 2.66 1988 3.14
mean (all forests) 1.98 1988 2.34
Hanley (1989)
Aberfoyle 1.70 1987 2.14
Everett (1979)
Dalby 0.41 1976 1.30

estimates has in recent years developed into a major area of research.3 The advan-
tages of a rigorous approach to bene¬ts transfer are clear. The costs, both ¬nancial
and temporal, of conducting individual valuation exercises at each site involved in a
policy decision would be prohibitive. Consequently the US Environmental Protec-
tion Agency and, more recently, several UK government organisations, including
the Department of the Environment, Food and Regional Affairs and the Environ-
ment Agency, have shown considerable interest in this avenue of research. However,
as several eminent researchers acknowledge, the problems involved in formulat-
ing and conducting a successful bene¬ts transfer are numerous and formidable
(Desvousges et al., 1992, 1998; Atkinson et al., 1992; McConnell, 1992b; Smith,
1992; Downing and Ozuna, 1996; Kirchhoff et al., 1997; van den Bergh et al.,
1997; Brouwer and Spaninks, 1999).
We can identify two basic approaches to bene¬ts transfer: unit value transfer and
function transfer (discussed subsequently). At the extreme, unit value transfer may
simply involve assuming that, say, a per visit value estimated at one ˜source™ site

3 Loomis (1992) actually traces research into bene¬ts transfer back to 1962. However, it was only in the late 1980s
that this became a major focus of research. See the review by Bateman et al. (2001d).
46 Applied Environmental Economics

Table 3.2. Forest users™ per person per visit recreation values from CV studies

Elicit. Study-year 1990 value
Forest Value type value (£) Study year (£)

Whiteman and Sinclair (1994)
Mercia use OE 1.00 1992 0.93
Thames Chase use OE 0.71 1992 0.66
Great Northern use OE 0.81 1992 0.75
Hanley and Ruffell (1992)
various use OE 0.93 1991 0.88
Hanley and Ruffell (1991)
Aberfoyle use OE 0.90 1991 0.85
Aberfoyle use IB 1.21 1991 1.14
Aberfoyle use PC 1.39 1991 1.31
Aberfoyle use DC 1.49 1991 1.41
Bishop (1992)
Derwent Walk use OE 0.42 1989 0.46
Derwent Walk use+option OE 0.97 1989 1.06
Whippendell use OE 0.54 1989 0.59
Whippendell use+option OE 1.34 1989 1.46
Willis and Benson (1989)
New Forest use OE 0.43 1988 0.47
Cheshire use OE 0.47 1988 0.51
Loch Awe use OE 0.50 1988 0.55
Brecon use OE 0.46 1988 0.50
Buchan use OE 0.57 1988 0.62
Newton Stewart use OE 0.73 1988 0.80
Lorne use OE 0.72 1988 0.79
Ruthin use OE 0.44 1988 0.48
mean use OE 0.53 1988 0.58
New Forest use+option OE 0.88 1988 0.96
Cheshire use+option OE 0.72 1988 0.79
Loch Awe use+option OE 0.76 1988 0.83
Brecon use+option OE 0.66 1988 0.72
Buchan use+option OE 0.79 1988 0.86
Newton Stewart use+option OE 1.18 1988 1.29
Lorne use+option OE 1.02 1988 1.12
Ruthin use+option OE 0.63 1988 0.69
mean use+option OE 0.82 1988 0.90

Hanley (1989)
Aberfoyle use OE 1.24 1987 1.53
Aberfoyle use PC 1.25 1987 1.55

Willis et al. (1988)
Castle Douglas use OE 0.37 1987 0.46
Recreation: predicting values 47

Table 3.2. (cont.)

Elicit. Study-year 1990 value
Forest Value type value (£) Study year (£)
South Lakes use OE 0.39 1987 0.48
North Yorkshire Moors use OE 0.53 1987 0.66
Durham use OE 0.31 1987 0.38
Thetford use OE 0.23 1987 0.28
Dean use OE 0.28 1987 0.35
Castle Douglas use+option OE 0.80 1987 0.99
South Lakes use+option OE 0.86 1987 1.06
North Yorkshire Moors use+option OE 1.03 1987 1.27
Durham use+option OE 0.56 1987 0.69
Thetford use+option OE 0.41 1987 0.51
Dean use+option OE 0.63 1987 0.78

Notes: 1 Valuation categories investigated are as follows: use = use value; option =
option value (the extra WTP to ensure conservation of the site for future use).
Elicitation methods are: OE = open-ended; IB = iterative bidding; PC = payment card;

DC = dichotomous choice.

can be applied to the ˜target™ or ˜policy™ site for which values are required. This
is clearly very crude and so a considerable literature has developed applying the
principles of ˜meta-analysis™ to bene¬t estimates.4 Here researchers have related
measures such as the mean bene¬t value reported in each of a set of source site
studies to a series of simple (usually binary) explanatory variables detailing, for
example, the evaluation method employed, the type of resource studied, the mea-
surement unit and the elicitation method used (see, for example, Smith and Kaoru,
1990; Walsh et al., 1992; Rosenberger and Loomis, 2000). Our bene¬t transfer
study of reviewed articles derives directly from such a meta-analysis approach.
Given that we are only considering woodland recreation studies, we do not need to
de¬ne variables detailing the type of good evaluated,5 and other explanatory factors
are incorporated by de¬ning relevant binary variables as in the studies cited above.
Before considering results from our meta-analysis we need to consider the alter-
native bene¬t function transfer approach, which in many ways is more theoretically
appealing. Here, as before, a set of source site studies are gathered together, but

4 For an introduction to the principles of meta-analysis, see Glass et al. (1981) and Wolf (1986). Note that these
sources show that the form of analysis found in the bene¬t valuation literature and in this volume is, strictly
speaking, only a partial meta-analysis dictated by the constraint of studies which were not designed with such
cross-study analyses in mind (e.g. de¬nitions of variables typically vary between bene¬t studies). Guidelines
for a common standard of design and reporting for future studies to facilitate such meta-analyses are set out in
Bateman et al. (2002).
5 In a separate study we present a simple analysis of valuations across differing recreational experiences, noting
that the results were logically related to both the substitutability of the environmental resource concerned and
the magnitude of the change in provision considered (Bateman et al., 1994).
48 Applied Environmental Economics

rather than using summary results, such as mean values, the raw data are used to
estimate a general bene¬t value function. This is then used to estimate values for
the target site by holding the estimated coef¬cients of the function constant and
changing the explanatory variable values in line with the characteristics of the target
site. So, for example, if the bene¬t transfer function estimated from source sites
included a coef¬cient linking recreational values to the size of a site, then one of the
elements in predicting values for the target site would be to multiply its size by the
estimated coef¬cient in the transfer function. Undertaking this operation for all the
explanatory variables in the transfer function provides the overall estimate of values
for the target site.
This approach need not be con¬ned solely to the estimation of values, and in
Chapter 4 we apply it to the estimation of visitor numbers, showing that the method
works quite acceptably in such an application. However, in empirical trials, the
function transfer approach does not fare so well in the estimation of values for target
sites (Loomis, 1992; Bergland et al., 1995; Downing and Ozuna, 1996; Brouwer
and Spaninks, 1999). In a study combining data from a single survey questionnaire
applied at source sites in ¬ve countries, Brouwer and Bateman (2000) ¬nd that the
function transfer approach yields higher bene¬t value estimation errors for target
sites than does a simpler, meta-analysis style, unit value transfer. One possible cause
of such ¬ndings is that bene¬t value functions differ more substantially between
sites than do functions predicting arrival numbers (which the results presented in
Chapter 4, as well as ongoing research, suggest are comparatively simple).6 Value
functions may differ in terms of which explanatory variables are pertinent and/or
in coef¬cient estimates for those variables (i.e. what in¬‚uences bene¬t values,
and how, varies across sites). While these effects may not be that profound when
viewed as a whole (making simpler unit bene¬t value transfers reasonably valid),
the function transfer approach may give undue weight to these differences, leading
to unreliable value estimates.
Given the above, we adopt a function transfer approach for estimating the num-
ber of arrivals to target sites (see Chapter 4), but a simpler meta-analysis transfer
approach to the estimation of values. Consideration of the ZTC studies reviewed
above (ITC studies being discarded for the reasons given) indicated that these re-
sults were not suitable for entry in such a meta-analysis both because of a lack
of observations and because our own TC work (see discussion of the Thetford 2
study later in this chapter) suggested that the travel expenditure and travel time cost
assumptions used in the Benson and Willis (1992) ˜Standard Model™ were liable
to produce overestimates of bene¬t values. Given our desire to emphasise defensi-
ble lower-bound values, the estimates given in Table 3.1 were not used for further
6 This work has been carried out at a variety of woodland and non-woodland sites (e.g. waterways, beaches, built
attractions, etc.) and is funded by the Forestry Commission, British Waterways and others.
Recreation: predicting values 49

analysis.7 Thus we argue that only the CV studies detailed in Table 3.2 provide a
suitable concentration of observations for further cross-study analysis.

A meta-analysis of previous CV studies
Our meta-analysis of previous UK CV studies yielding per person per visit values
for woodland recreation follows the approach of Smith and Kaoru (1990), Walsh
et al. (1992) and Rosenberger and Loomis (2000). Table 3.2 lists seven studies
yielding forty-four estimates. To this we have added one compatible value from
the Thetford 2 study discussed later in this chapter.8 While this list represents the
largest set of estimates for any UK natural resource, it is still considerably smaller
than those used by Smith and Kaoru (77 studies of which 35 were used to yield
some 400 estimates) and Walsh et al. (120 studies yielding 287 estimates of which
129 were obtained using the CV method) in their meta-analyses of US resources.
This underlines the difference in available, comparable studies in the US and UK
and reinforces our opinion that the major barrier to successful bene¬t transfer in
the UK is the lack of suf¬cient, high-quality valuation studies. The analysis we
conducted here was therefore intended to be illustrative rather than de¬nitive.
Our database of valuation estimates yielded the following simple explanatory
WTP = study mean willingness to pay (£/person/visit)
OPTION = 1 if the study asked WTP for use plus option value; 0 if the study
asked WTP for use value alone
ELICITAT = elicitation method (categorical variable): 1 = open-ended; 2 =
iterative bidding; 3 = payment card; 4 = dichotomous choice
OE = 1 if open-ended elicitation method; 0 if other elicitation method
AUTHOR = authorship (categorical variable)
Following Glass et al. (1981) an early concern was to ensure the comparability
of studies. A number of reviewed studies were excluded from Table 3.2 due to
design, implementation or gross reporting problems (see Bateman, 1996). To some
extent, further design effects are incorporated within analysis of the AUTHOR
variable, which identi¬ed individual study designs. Although a generalised linear
model9 (Aitken et al., 1989) analysis did reveal some differences, these were highly
correlated with the OPTION and OE variables and the AUTHOR variable had to

7 Such analysis is given in Bateman (1996) which concludes that these results are upper-bound values for woodland
8 This value is obtained from the sample in the Thetford 2 study who faced an entrance fee question not preceded
by budget or tax questions. This sample is comparable with the other studies examined in Table 3.2.
9 The estimated model was speci¬ed so as to explicitly permit the use of categorical variables such as AUTHOR
within linear regression models with each level of the variable being treated in a manner analogous to the use
of individual dummy variables.
50 Applied Environmental Economics

be omitted from further analysis. Analysis of unusual design effects was therefore
conducted by identifying statistical outliers (as discussed below).
Clearly the variables ELICITAT and OE cannot be included within the same
model, as one is derived from the other. Analyses of variance showed that the
numbers in categories 2, 3 and 4 of the ELICITAT variable were too small to allow
for meaningful individual treatment. However, when these categories were amalga-
mated to form the OE variable, a signi¬cant difference (at the 5 per cent signi¬cance
level) between results from these and the open-ended studies was observed. Follow-
ing these preliminary analyses we concluded that the most conservative approach
was to investigate a simple model of WTP, relating it to just the OPTION and OE
Estimation of this model identi¬ed two statistical outliers, which may indicate
the presence of unusual design effects.10 These observations were excluded and the
¬nal model was:
WTP = 1.3525 ’ 0.7571 OE + 0.3120 OPTION (3.1)
(14.04) (’7.28) (5.02)
R 2 = 61.1% R 2 (adj.) = 59.2% n = 43
Figures in brackets are t-statistics
A number of interesting observations arise from Equation (3.1). The overall ¬t
of the model is good (given that we are dealing with socio-economic data) with
about 60 per cent of total variation explained. However, the strongest explana-
tory variable is the constant, a ¬nding which may re¬‚ect a common perception
among respondents regarding an appropriate response to a per visit WTP question.
Responses may be re¬‚ecting a mixture of respondents™ notions of a socially fair
level of WTP and prior experience of payments for comparable goods (entrance
fees, car parking fees, etc.). Such motivations move bids away from the underlying
value they are intended to measure. In effect, such measures may be more akin to
prices than values.
The sign and signi¬cance of both of the explanatory variables is as anticipated.
Relative to other approaches the use of an OE elicitation technique results in lower-
bound WTP sums, while asking respondents to assess both their use and option value
produces higher bids than when use values alone are considered. By combining
these two factors we can use the coef¬cients of Equation (3.1) to predict cross-
study estimates for the four types of per person per visit values shown in Table 3.3.
Furthermore, by referring to information regarding the number of persons in an
average visitor party we can infer the various per party per visit values also shown
The Bishop (1992) OE use + option value for Whippendell Wood and the Hanley (1989) OE use value for
Aberfoyle. For further details, see Bateman (1996).
Recreation: predicting values 51

Table 3.3. Woodland recreation values from a cross-study analysis of CV estimates

Per person per visit Per party per visit
values1 (£, 1990)
values (£, 1990)

OE elicitation Other elicitation OE elicitation Other elicitation
Value type method method method method
Use value 0.60 1.35 1.82 4.12
Use + option value 0.91 1.66 2.78 5.06

Note: 1 Assuming a mean party size (from Thetford 2 study) of 3.05 persons per party.

in the table (sensitivity analysis on these estimates is given in the summary at the
end of this chapter).

Our review of UK monetary evaluations of woodland recreation suggests that,
while the literature is developing fast, the body of consistent, high-quality papers
necessary for advanced bene¬t transfer does not exist to date (although it is ar-
guable whether this is even true of the more advanced US literature). Consequently
we have conducted a fairly simple cross-study meta-analysis concentrating on re-
sults from just one valuation method, the CV approach. While this is suf¬cient
to demonstrate our wider methodology, it does mean that the results should be
treated with some caution. We attempt to remedy this in the following sections,
which examine a number of methodological and theoretical issues across both
chosen valuation methods, as well as providing further bene¬t estimates for the
wider study.

The ¬rst Thetford CV/TC study
Our initial woodland recreation study was conducted in Thetford Forest, East Anglia
(providing the user sample) and the city of Norwich (about twenty-¬ve miles distant;
providing the non-user sample) in the summer of 1990 (hereafter referred to as the
Thetford 1 study). The research consisted of both CV and TC analyses. The CV
study involved a split-sample design examining payment vehicle and elicitation
effects across both users and non-users, while the TC study (which used the ITC
variant) concentrated on visit cost assumptions and the impacts of varying functional
form. On account of space constraints, only principal results are presented here,
with full details being given in Bateman (1996).
52 Applied Environmental Economics

The Thetford 1 CV study: elicitation, payment vehicle and user
versus non-user effects
The CV study asked respondents for their WTP for the recreational services and
facilities available at Lynford Stag, a major woodland recreation site within Thetford
Forest. In total seven subsamples were gathered. These can be divided into two
(i) whether respondents were users or non-users
(ii) whether an annual tax or per visit payment vehicle was used.

In all the annual tax payment (but not entrance fee) treatments it was decided
to inform respondents, prior to any WTP question, of the current average level of
annual per household payments to support the Forestry Commission, which was
estimated at approximately £2.60 per annum.11 This approach followed contempo-
rary practice in UK CV studies, particularly as pioneered in the work of Turner and
Brooke (1988), a study which had recently been approved (as part of a wider CBA)
by H.M. Treasury. However, subsequent studies, such as that reported by Baron
and Maxwell (1996), indicate that cost information provided to CV respondents
may be construed as indicating the value of the good in question (see subsequent
results regarding payment card effects and the discussion of starting point bias in
Chapter 2). This suggests that in the Thetford 1 study, cost information may have
anchored WTP responses towards this sum. Consequently we must treat the abso-
lute level of WTP results from this experiment with some caution although relative
differences remain of interest (the subsequent study in Wantage abandoned this
approach and so provides some evidence of the magnitude of the anchoring effect).
Table 3.4 details WTP results across the three annual tax format samples.
Per annum WTP responses were elicited using an OE question while per visit
responses were obtained using a payment card. While this precludes strict compa-
rability across samples (study resource constraints meant that further subsamples
could not be gathered at that time), such an approach was chosen to facilitate further
testing of design effects as follows:
(i) For the tax format, while both users and non-users were presented with a general tax
payment vehicle, a further subsample of non-users was presented with a community
charge (poll tax) vehicle. At the time of the study the imposition of a poll tax was the
major political issue of the day and this vehicle was deliberately chosen to examine
the potential magnitude of payment vehicle effects. Non-users were identi¬ed as the
group who might have the most ill-de¬ned preferences and so provide the most extreme
responses to such effects.

11 Based upon Forestry Commission (1985a).
Recreation: predicting values 53

Table 3.4. Summary WTP responses for the Thetford 1 CV study

Payment Payment Sample Elicitation Mean WTP 95% C.I. Median
period vehicle (£) (£) (£)
Per annum General tax Users OE 5.14 1.48“8.81 2.00
Per annum General tax Non-users OE 3.51 1.13“5.88 0.70
Per annum Poll tax Non-users OE 7.09 2.68“11.50 0.00
Per visit Entrance fee Users PCL 1.21 0.99“1.43 1.00
Per visit Entrance fee Users PCH 1.55 1.19“1.92 1.25
Per visit Entrance fee Non-users PCL 1.45 1.15“1.75 1.25
Per visit Entrance fee Non-users PCH 2.37 1.98“2.76 2.00

Note: 1 OE = open-ended; PCL = payment card (low range); PCH = payment card (high

(ii) For the per visit format two payment cards were used, the ¬rst showing a payment range
from £0 to £3 in increments of 50p and the second ranging from £2 to £5 using the
same increments. Both cards also explicitly stated that respondents were free to select
any other amount.

All samples were collected using face-to-face interviewing of randomly selected
respondents.12 Sample size was ¬fty for most subsamples falling to a minimum of
forty-six. While not large, the continuous nature of the valuation responses meant
that these samples were generally suf¬cient to perform rudimentary statistical and
validity analyses. Summary WTP statistics are reported in Table 3.4.
Because of the differences in elicitation method (and the use of existing payment
information in the per annum questions) we cannot meaningfully compare per
annum with per visit results and must con¬ne ourselves to comparisons within these
subgroups. Considering the per annum results we can see that, as expected, when all
other factors are held constant (i.e. when a general tax vehicle is used), both mean
and median WTP is higher for users than for non-users (although the high response
variability typical of OE studies combined with relatively small sample size means
that these differences are not statistically signi¬cant in this case). Analysis shows
that, although all non-user samples are socio-economically similar, the user group
enjoys signi¬cantly higher income levels, a ¬nding which somewhat complicates
the interpretation of this result. However, comparison of these ¬ndings with results
obtained using the poll tax vehicle shows that the latter has a clear and strong effect
on univariate WTP measures. The ¬rst point to note is that while refusal to pay rates
are similar across the two general tax subgroups (both about 15 per cent), just over
50 per cent of those faced with a poll tax vehicle refuse to pay. Just as interesting are

12 The authors wish to thank Joanne Wall (formerly of the University of East Anglia) for managing the survey.
54 Applied Environmental Economics

the ¬ndings that, despite this, the poll tax sample recorded the highest mean WTP
amount. In effect, while most respondents reject the use of poll tax as a suitable
vehicle for funding the public good under evaluation, a minority are strongly in
favour of such an approach and state comparatively large WTP sums, resulting in
a relatively high mean.
Consideration of the per visit values detailed in Table 3.4 shows that for both
user and non-user samples the higher-range payment card results in higher mean
and median WTP sums (although these differences are not statistically signi¬cant
in the present samples). Both the non-user samples record higher WTP sums than
their user group counterparts. One plausible explanation of this ¬nding is that
non-users see the use of entrance fees as a method of moving funding costs away
from themselves and onto users; we therefore have to discount the validity of such
responses as indicators of underlying values.
A number of socio-economic variables were collected in all surveys so as to
facilitate regression analysis of underlying bid functions (full results are reported
in Bateman, 1996).13 These functions14 suggested that a consistent set of factors
underpinned valuation responses across formats, with higher WTP values being as-
sociated with higher incomes,15 clear knowledge of, or living near, the area under
evaluation. For those facing per annum questions, WTP was positively associated
with the number of visits made to Thetford annually, while for those facing per visit
questions, regular visitors stated relatively lower amounts. However, for this latter
group, when the number of annual visits is considered, this equated to a higher
than average total WTP. These ¬ndings conform to prior expectations. However,
it was noticed that bid functions for all the per visit subsamples were dominated
by a highly signi¬cant intercept term, suggesting that responses were subject to
some prior notion of a ˜correct™ (or ˜social norm™) answer, most probably in¬‚u-
enced by experience of entrance fees at comparable attractions (for example, car
parking fees at National Trust sites). While this again conforms to prior expec-
tations, it undermines the validity of these particular answers as a source of valid
In conclusion, while the CV exercise carried out as part of the Thetford 1 study
produces a number of results which conform to prior expectations, its major ¬nd-
ings highlight the potential impact of design effects, so providing valuable pointers

13 In the case of the on-site interviews with forest users, variables collected included: home address; sex; age;
employment; whether the interviewee was a pensioner; income; precise interview location; preference for
natural or urban recreation; history and frequency of visits to the speci¬c site and forest entirety; time spent
on site; and use-value WTP. Similar variables were elicited from the non-user samples with the addition of
questions regarding respondents™ knowledge of the forest and integral visitor sites.
14 In each case a log (dependent) functional form satis¬ed an n-scores normal distribution test. All functions ¬tted
the data to an acceptable degree, with R2 values ranging from 0.15 to 0.50.
15 This relationship was proxied in some cases by a negative association between reported WTP and the respondent
being a pensioner.
Recreation: predicting values 55

towards improved study design. In particular, the highly signi¬cant impact of
changing the payment vehicle indicates that considerable care is needed if future
studies are to elicit usable estimates of recreation value (rather than estimates of
how respondents perceived the payment vehicles themselves). Furthermore, results
from the entrance fee experiment suggest that payment cards have the potential
to impact upon stated values. The possibility of entrance fees themselves causing
respondents to resort to simple heuristics rather than to preferences in determining
values also arose but could not be adequately assessed and so was made an objective
of subsequent work.
We now turn to consider the ITC analysis carried out as part of the Thetford 1

The Thetford 1 TC study: functional form effects
Responses from the 129 parties of visitors (comprising almost 400 individuals)
interviewed at Thetford Forest were used to undertake an ITC study of recreational
values. In addition to the variables discussed previously, data regarding the distance,
cost and duration of visits, substitutes and further socio-economic variables which
might explain visits were collected. OLS estimation techniques were employed
(a comparison with maximum likelihood techniques was conducted as part of the
Thetford 2 study described subsequently) and initial analysis considered the correct
speci¬cation of the dependent variable for our trip generation function (TGF). A
series of correlation and regression tests con¬rmed that a log dependent variable was
clearly superior. This decision was not so clear-cut when speci¬cation of the cost
variable was considered. Following the discussion in Chapter 2, three de¬nitions of
travel expenditure cost (marginal (petrol only); petrol plus insurance; full running
costs) and three de¬nitions of travel time cost (zero (respondents enjoy travelling);
the Department of Transport (DoT) wage rate; full wage rate) were investigated.
All linear and logarithmic permutations of these costs were considered in de¬ning
total travel costs, and statistical tests indicated that a cost function using the full
running cost estimate of travel expenditures and a zero travel time cost assumption
provided the most signi¬cant travel cost variable. A considerable advantage of
using a cost function which is not (via time costs) linked to wage rates is that the
visitor™s income may be entered as a separate explanatory variable without inducing
collinearity problems.
Further explanatory variables were investigated through stepwise regression anal-
ysis of the full range of socio-economic variables collected in the survey. Of these,
only the respondents™ household income proved signi¬cant. This ¬nding again
echoes the results of earlier UK TC studies (Willis and Benson, 1988, 1989) which
report TGFs relating visits to cost and some indicator of socio-economic status.
56 Applied Environmental Economics

Table 3.5. Thetford 1 TC study: consumer surplus estimates for
three functional forms

CS per person per
visit (£, 1990)
Travel cost CS per
1 child = 1 child =
Functional R adj. coef¬cient. party per visit children
form (%) (t-ratio) (£, 1990) 1 adult 0.5 adult omitted
Double-log 44.2 3.37 1.07 1.19 1.34
Semi-log 39.9 7.40 2.40 2.67 3.00
(dep.) (’7.42)
Linear 21.0 27.42 8.88 9.87 11.10

The impact of changing the functional form was investigated16 and Table 3.5
reports summary ¬ndings and consumer surplus estimates per party visit and per
individual visit. The latter results are subdivided to consider different treatments of
child visitors.
Inspection of Table 3.5 shows the double-log functional form gives the best
¬t to the data17 and resultant valuation estimates accord well with prior ex-
pectations. Clearly misspeci¬cation of functional form leads to substantial error
in consumer surplus estimates (e.g. adopting a linear form very substantially
overestimates recreation values). The ¬nal four columns of Table 3.5 consider
the issue of whether to report per party or per person values. These are highly
responsive to the treatment of children within the sample. Our proposed solution,
which we adopt in subsequent work, is to concentrate upon the party as the basic
unit of valuation, thereby avoiding subjective decisions regarding individual level
In conclusion, this study was generally satisfactory and provided useful guide-
lines for our future TC studies. At ¬rst glance it also generated a defensible valuation
of woodland recreation bene¬ts. However, during the course of this analysis we be-
came increasingly conscious of the theoretical problems associated with applying
OLS estimation techniques to ITC data and therefore made an analysis of potential
estimation effects a feature of our subsequent TC work, reported in the Thetford 2

16 These are all parametric functional forms and so impose corresponding assumptions upon our analysis. Cooper
(2000) considers non-parametric and semi-parametric approaches to TC analysis.
17 This function yields higher explanatory power than both those reported in the Willis and Garrod (1991b) ITC
studies of UK woodland recreation and higher than all but two of the twenty-two comparable OLS estimated
functions reported by Smith and Desvousges (1986) in their ITC studies of water-based recreation in the United


. 2
( 11)