Valuing enhancements to forest recreation using choice experiment and contingent behaviour methods

ARTICLE IN PRESS

Journal of Forest Economics 13 (2007) 75–102 www.elsevier.de/jfe

Valuing enhancements to forest recreation using choice experiment and contingent behaviour methods Michael Christiea,, Nick Hanleyb, Stephen Hynesc a

Institute of Rural Sciences, University of Wales Aberystwyth, Aberystwyth, Ceredigion, Wales, SY23 3AL, UK b Department of Economics, University of Stirling, Stirling, Scotland, FK9 4LA c Teagasc, Athenry, Ireland Received 3 October 2006; accepted 15 February 2007

Abstract This research utilises two valuation techniques (a frequency-based choice experiment model and a contingent behaviour model) to value a range of improvements to recreational facilities in forest and woodlands in Great Britain. We provide the first comparison in the literature of welfare results from these two approaches. Four groups of forest users are targeted in this research: cyclists, horse riders, nature watchers and general forest visitors, and look also at ‘‘sub-groupings’’ within these classes of forest user. We found that heterogeneity of preferences exists within each of these groups. In particular, more specialist forest user groups attain generally higher values for improvements than general users. For example, downhill mountain bikers were willing to pay more for the provision of dedicated downhill courses than family cyclists for easy cycle trails. It is also argued that the use of a frequencybased choice task in the choice experiment has advantages over the more traditional choice tasks for applications such as forest recreation since a frequency-based task better reflects

Corresponding author.

E-mail address: [email protected] (M. Christie). 1104-6899/$ - see front matter r 2007 Elsevier GmbH. All rights reserved. doi:10.1016/j.jfe.2007.02.005

ARTICLE IN PRESS 76

M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

actual behaviour and encourages respondents to pay closer attention to the ‘‘distance travelled’’ attribute. r 2007 Elsevier GmbH. All rights reserved. JEL classifications: Q51; Q26; Q23 Keywords: Choice modelling; Contingent behaviour; Forest recreation

Introduction Forests and woodlands are an important and significant destination for outdoor recreation trips. For example, in Great Britain, there are an estimated 252 million leisure days trips made to forest and woodlands each year: this accounts for 20% of all leisure day trips in the countryside (TNS Travel and Tourism, 2004). People have a wide range of motivations for visiting forests. Some will be attracted by the forest itself and the wildlife it provides a home to, while more local users may simply find it a convenient place to walk their dog. The more adventurous visitor may be attracted by specialist recreation facilities provided within the forest, which might include, for example, dedicated walking or mountain bike trails, or wildlife hides. In Great Britain, numerous visitor surveys have been undertaken (predominantly by the Forestry Commission) to collect information on the levels of participation in forest recreation and the motivations of forest visitors (Forestry Commission, 2004a, 2005a). The management of forests for recreation and in particular the provision of facilities is costly. In 2003–04, the Forestry Commission spent £38 million on ‘recreation, conservation and heritage’1 within its forests (Forestry Commission, 2005b). Clearly such levels of spend need to be justified in terms of ensuring that investments are best targeted to attain the greatest marginal gains. In terms of cost-benefit analysis, the Forestry Commission’s income from ‘recreation, conservation and heritage’ in 2003–04 was £13 million (Forestry Commission, 2005b). Thus, in accountancy terms, the Forestry Commission made a deficit of around £25 million on the provision of recreation opportunities. However, market prices generally do not exist for many aspects of forest recreation due to the legal and physical impracticality of excluding users. Economic evaluation techniques have therefore been developed to measure the value (consumers’ surplus) derived from recreational use of environmental resources such as forests. These techniques include stated preference methods, revealed preferences methods, and combined methods. To date, over 30 studies have been undertaken to measure the value of recreation in UK woodland, yielding over 100 separate benefit estimates (see Jones et al., 2003 for a review and meta-analysis of these studies). In these studies, it has been estimated that the annual national aggregate consumers’ surplus associated with recreation in UK woodlands range between £40 million (Bateman, 1996) to over 1

Note that information specifically on the level of expenditure on recreation alone is currently not available from the Forestry Commission.

ARTICLE IN PRESS M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

77

£50 million (Benson and Willis, 1992) in 2003 prices. Taking account of these nonmarket values suggests that investment in forest recreation generates a net annual benefit to society of between £15 million and £25 million, although this ignores any opportunity costs of devoting forests to recreational use rather than felling. The above argument has been used over the past few decades to justify continued investment in forest recreation. However, such argument only partially addresses issues relating to the future investment in forest recreation. A more pertinent question, which has yet to be rigorously addressed, is what types of investment in recreation facilities generate the greatest welfare gains. The majority of current research has either simply valued forest recreation in a generic sense (e.g. Bishop, 1992) or forest recreation as a single attribute of wider forest values (Christie, 1999; Hanley, 1989; Hanley and Ruffell, 1993; Willis and Benson, 1989; Willis et al., 1988). Moreover, none of the existing research explores the heterogeneity of consumers’ surplus values for forest recreation between alternative uses and user. This research aims to fill this knowledge gap. Existing studies that have valued forest recreation have used both revealed preference (RP) methods (e.g. the travel cost method) and stated preference (SP) methods (e.g. contingent valuation) (Jones et al., 2003). The two approaches have very different merits. For example, RP approaches have the advantage that values are grounded on actual behaviour (Herriges and Kling, 1999), but the approach is unable to value resource provision beyond current levels. SP approaches overcome these limitations, enabling valuations to go beyond existing levels of provision; however, they may be subject to a wide range of potential biases (Mitchell and Carson, 1989). A relatively new addition to the suite of SP approaches is the choice experiments (CE) method, which we adopt in this research. A key attraction of CE is that it is capable of valuing the component attributes of forest recreation, and the method has not been used to date for forest recreation in a UK (although it has been used to look at the value of forest landscapes: Hanley et al., 1998). The CE approach relates here to routine behaviour (site choice as a function of site attributes), and makes use of a realistic and familiar payment vehicle, namely distance travelled to a forest (which is subsequently converted to a cost of travel coefficient). Furthermore, in this research we utilise a frequency-based choice task in the choice experiment. We argue that a frequency-based choice task is better suited to the valuation of environmental goods with repeated uses than the traditional choice-based choice task. To our knowledge, this is the first application of a frequency-based choice task in an environmental valuation literature. Recent developments in valuation research has also focused on combining RP and SP data sources and therefore developing methodologies that potentially draw on the best elements of both approaches (Adamowicz et al., 1994, 1997; Cameron, 1992). An example of a combined RP-SP method is the contingent behaviour model (Englin and Cameron, 1996; Hanley et al., 2003b), which is also adopted in this research. Contingent behaviour models are capable of assessing (i) the extent to which the number of trips made to a forest changes when forest facilities change and (ii) the consumers’ surplus value associated with that change.

ARTICLE IN PRESS 78

M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

In this research we utilise both choice experiments and contingent behaviour models to value improvements to the forest recreation resource. The use of the two methodologies allows direct comparison of the benefit estimates, as well as demonstrating the relative merits of the two methods.

Improving facilities for forest recreation In this research, we are interested in identifying which types of improvements to the forest recreation resource may result in the greatest gains in welfare, and how these welfare gains are distributed across different uses and user groups. Existing research highlights that there are a wide range of recreation activities currently taking place within forests (Forestry Commission, 2004a). Following a series of focus group meetings with forest visitors and forest managers, four recreation activities were identified as being most suitable for study: cycling, horse riding, nature watching and general forest visitors. The selection criteria used to identify these activities included: those activities that attract a significant number of users; those activities where the number of users is expanding most rapidly; those activities that have specific facility/infrastructure/management needs; and those activities that are important to future forest policy.





Cycling Cycling represents an interesting and unique case study for this research. Over the past 10 years or so, there have been a number of significant investments in ‘mountain bike centres’, which provide a range of specifically built technical single track trails for mountain biking. More recently, this provision has extended to provide more ‘hard core’ facilities such as downhill, 4 cross and dirt jumps. These centres attract a high number of users; many of whom are willing to travel long distances (Forestry Commission, 2002). This fact suggests that mountain bikers attain high levels of utility from these centres. Restricting the examination of cycling to mountain biking alone, however, would be unwise since forests also attract large numbers of casual and family cyclists. These cyclists are likely to have different needs to mountain bikers and are therefore also examined. Horse riding Horse riding presents a second interesting case study for this research. Horse riding currently takes place in 22% of the GB’s main forest sites (Forestry Commission, 2004a), and accounts for around 2% of forest users (TNS Travel and Tourism, 2004). Currently, horse riding takes place on existing forest trails, but further horse specific facilities are generally not provided. One prominent exception to this is at the forests of Dyfnant and Lake Vyrnwy. Here, the local riding group, in partnership with the Forestry Commission, have developed a suite of facilities aimed to meet the specific needs of horse riders. Facilities provided include corrals, horse box friendly parking, and horse friendly trail surfaces. The facilities are now in much demand and serve to demonstrate the potential of providing horse specific facilities. It was suggested during developmental

ARTICLE IN PRESS M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102





79

focus groups with forest managers that efforts to promote horse riding in forests could potentially follow the success of the mountain biking centres and therefore there was interest to examine whether this is the case or not. Nature watching Nature watching was also identified as an activity of interest for this research. Nature watching currently takes place in 15% of forests (Forestry Commission, 2004b); however, there are also opportunities within most forests to view nature. An interesting finding from forest visitor studies is that whilst only a small proportion of visitors visit forests specifically to watch nature, many visitors report that seeing nature enhanced their visit. Thus, the third case study examined public values for enhancing nature and nature watching facilities in forests, which might include formal facilities such as interpretation centres, wildlife hides and nature walks, as well as less formal provisions such as forest management that increases opportunities for wildlife and for viewing wildlife. General forest visitors Finally, it was recognised that limiting the study only to cover the three activities outlined above might be restrictive in that the views and values of other types of forest users would not be captured. It was therefore considered that it would be appropriate to group all other non-specialist forest users into a single ‘general forest visitor’ category, which would also cover forest visitors undertaking alternative or multiple activities during the trip to the forest.

Methods As outlined above, two methodologies (the frequency-based choice experiment and the contingent behaviour models) were utilised in this research. We now provide an overview of these two methods, and then outline details of the survey instrument used in the research. The frequency-based choice experiments model CE are a stated preference valuation technique which is capable of measuring the economic benefits associated with the component attributes of an environmental policy (Hensher et al., 2005; Louviere et al., 2000). CE has been used to value a wide range of environmental goods including recreation (Hanley et al., 2002), biodiversity (Christie et al., 2006) and landscapes (Bullock et al., 1998). The CE method relies on surveys to gather data. Within the survey, respondents are presented with a series of choice tasks in which they are asked to choose their preferred policy option from a list of (usually) three options; one of which normally includes the status quo or ‘‘do nothing’’ option. An alternative approach is to use frequency of use data in the choice task instead of choice data. In the frequency-based application, respondents are asked to allocate their next five recreation trips to forests (to be undertaken within a year) between two hypothetical forests and a stay at home option – see Fig. 1 for an example of a typical choice task used in this research. It is argued that the

ARTICLE IN PRESS 80

Type of trail

M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102 Forest A Multi-use trails + dedicated way marked, long distance (+ 20 miles) cross country bike trails.

Forest B Multi-use trails + dedicated technical single track mountain bike trails. +

+ Optional trail obstacles

A range of optional trail obstacles No optional trail obstacles. north shore.

Bike wash facilities

Bike washing facilities available.

No bike wash facilities.

Changing and shower facilities

Changing / shower facilities available.

No changing / shower facilities.

General facilities

Facilities included car parking, toilets, BBQ / picnic area, café and forest shop.

Facilities include car parking and toilets only.

Information Only basic information on the forest, trails, and wildlife provided.

Detailed and up-to-date information on the forest, trails, and wildlife provided at on website

Surrounding

Distance1

Forest not managed to increase opportunities to view wildlife, points of interest and view points.

Forest enhanced to increase opportunities to view wildlife, features of interest and view points.

Forest located 300 miles from your home.

Forest located 150 miles from your home.

I would allocate my next Forest A 5 trips (to be taken within [] the next year) to:

Forest B []

Stay at home []

Fig. 1. Example of a frequency-based CE model choice task. Notes (1). The ‘distance’ travelled was subsequently converted to a ‘travel cost’ parameter in the choice model.

frequency-based approach potentially has a number of advantages for this application over the more traditional choice-based question both in terms of realism and efficiency of data collection. In terms of realism, the frequency-based question appears to be better suited to capture the dynamic nature of people’s actual recreation behaviour, which often involves numerous trips to a number of different forests over a finite planning period. Also, it allows respondents to demonstrate a limited intended use of the forest by, for example, only allocating two trips to the hypothetical forests and allocating the remaining three trips to the stay at home option. Thus, the frequency-based choice task has the capacity to capture a much richer depth of information on forest visitor’s intended behaviour. The analysis of a frequency-based CE model is based on Hanemann’s (1994) random utility maximisation (RUM) theory. Accordingly, it is assumed that the forest visitor’s utility function can be broken down into two parts, one deterministic

ARTICLE IN PRESS M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

81

(V) and one which is a random error term (e). It is also assumed that the forest visitor chooses how often to visit each site from the set of all possible sites (C) according to the relative utility they obtain from each site, which in turn depends on the characteristics of that site. The deterministic part of utility is assumed to be a linear function of site characteristics: V ij ¼ b1 þ b2 X ij2 þ b3 X ij3 þ    þ bn X ijn þ mðY i  TC ij Þ;

(1)

where Xij represent site attributes, Y is income and TCij is travel costs of visiting site j for individual i (estimated here by multiplying the return distance travelled to the forest by £0.15 per mile cost of travel). The probability that site j will be selected over other sites in C depends, for individual i, on: pi ðjÞ ¼ Prob½V ij þ ij XV ik þ ik ;

8k 2 C:

(2)

If we assume that the error term is independent and identically distributed (IID) with an Extreme Value Gumbel distribution, then this gives us the familiar conditional logit model. Assuming that the Independence of Irrelevant Alternatives (IIA) hypothesis holds, then: pi ðjÞ ¼ P

expðV ij Þ . k2C expðV ik Þ

(3)

Compensating variation welfare measures can now be obtained, using the standard Hanemann (1994) formula, where V0 is (deterministic) utility in the initial situation, and V1 is utility in some different situation: (e.g. when forest characteristics have been improved at one site within the choice set): " # X X 1 lnð expðV j0 ÞÞ  lnð expðV j1 ÞÞ ; (4) CV ¼  bTC j2C j2C Utility changes are converted into money-metric using the inverse of the marginal utility of income, which is here the parameter on the travel cost variable (bTC). The forest recreation attributes and levels examined in the CE study were developed from information gathered during a review of relevant literature and from forest user and forest manager focus groups. For each of the four recreation activities examined in this research, seven attributes were specified (see Table 1 for a summary of attributes and attribute levels). There are a number of important issues to highlight regarding the design of attributes and levels. First, the number of attributes and the number of levels of each attribute was consistent across all four recreation activities. This allowed the same experimental design to be used across the four user groups. Furthermore, four of the attributes (‘general facilities’, ‘information’, ‘surroundings’, and ‘distance’2 attributes) and their respective levels were identical over all four user groups. This was done to allow comparisons of values for these attributes and levels across all survey respondents. The four activityspecific attributes (i.e. the first four attributes in Table 1), however, were designed to be specific to the respective recreation activities. So for example, for cycling, the 2

Which, in the analysis, is converted to a ‘travel cost’ parameter.

ARTICLE IN PRESS 82

M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

Table 1. Summary of attributes and levels used in the CE model Cycling

Horse riding

Nature watching

General forest users

Type of trails Multi-purpose trails Cross country trails Technical single track Downhill trails

Type of trails Multi-purpose trails Horse friendly trail Carriage driving trails Long distance trails

Trails/routes Multi-purpose trails All access nature trail Interpretative trails

Walking trails Multi-purpose trails All access trails Sculpture trails

‘Off the beaten track’

Long distance routes

Optional trail obstacles No obstacles Obstacles provided

Optional trail obstacles No obstacles Obstacles provided

Wildlife hides

Mountain bike trails

Not provided Provided

Not provided Provided

Bike wash facilities

Parking

Horse riding trails

No bike wash Bike wash facilities

Normal Horse-box friendly

Wildlife viewing centers Not provided Provided

Not provided Provided

Showers

Guided nature walk

Nature trails

Not provided Provided

Corrals and tie up point Not provided Provided

Not available Available

Not provided Provided

General facilities Car park+toilet only +picnic areas +cafe´ & shop +play areas

General facilities Car park+toilet only +picnic areas +cafe´ & shop +play areas

General facilities Car park+toilet only +picnic areas +cafe´ & shop +play areas

General facilities Car park+toilet only +picnic areas +cafe´ & shop +play areas

Information Basic only Detailed information

Information Basic only Detailed information

Information Basic only Detailed information

Information Basic only Detailed information

Surroundings No change Managed for wildlife

Surroundings No change Managed for wildlife

Surroundings No change Managed for wildlife

Surroundings No change Managed for wildlife

Distance (travel cost) 20 miles 75 miles 150 miles 300 miles

Distance (travel cost) 20 miles 75 miles 150 miles 300 miles

Distance (travel cost) 20 miles 75 miles 150 miles 300 miles

Distance (travel cost) 20 miles 75 miles 150 miles 300 miles

attributes were ‘type of trail’, ‘optional trail obstacles’, ‘bike wash facilities’ and ‘changing and shower facilities’. In the ‘general forest user’ study, these activity specific attributes related to improvements in the main recreation activities undertaken at forests, i.e. cycling, horse riding and nature watching. Finally, the cost of each trip alternative relative to staying at home was expressed as a one-way travel distance, giving a monetary ‘travel cost’ attribute with which welfare estimates

ARTICLE IN PRESS M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

83

could be obtained once this variable had been converted from miles into out-ofpocket travel costs (Hanley et al., 2002). In this study, there are five attributes with two levels of provision and three attributes with four levels of provision. A complete 25  43 factorial design would involve 2048 combinations. Following standard practice in choice modelling, the attributes and levels were allocated to choice tasks according to an orthogonal main effects fractional factorial design. Following Street et al. (2005) such a design produced 16 choice cards, which were subsequently split into four sub-samples each containing four choice tasks. A similar experimental design was applied to all four recreation activities investigated.

The contingent behaviour model Contingent behaviour models, as used in this project, combine elements of revealed preference and stated preference methods. This allows the researcher the advantage of grounding welfare estimates in actual behaviour, yet of also being able to look beyond the existing range of variability in environmental quality across sites, and/or beyond the existing structure of access prices for recreational resources. Each person i in each data set yields two responses. The first is the number of trips (Vij) she makes to a given forest j per year, as a function of travel costs to that forest (TCij), travel costs to other, substitute sites (TCsub ij), income (Yi), the gender, age and education level of the respondent (Si, Ai, Ei), and a vector of dummy variables representing unobserved quality differences for each site in the sample (D1yD6). The second observation is how many extra trips (if any) she says she would make if a specified improvement in recreational facilities at the site occurs. Such models have been used before to predict changes in both consumers’ surplus and the number of visits made due to improvements in environmental quality, such as reductions in coastal water pollution at bathing beaches (Hanley et al., 2003a). Four groups of forest users were again studied: cyclists, horse riders, nature watchers, and general forest visitors. In each group, people were asked how their behaviour, in terms of intended number of trips to the site where they were sampled, would change, if one of two improvements were undertaken. These improvements are described in Table 2. It is important to note that the improvements described in the contingent behaviour model relate directly to the improvements examined in the choice experiment. Several econometric considerations arise in this data. First, repeated observations across individuals imply a possibility of error correlation between responses. To take account of the panel nature of the data, we use a random effects specification (Hanley et al., 2003a), comparing this to a pooled model which ignores error correlation. Second, the integer nature of the visits variable implies that one should make use of a count data model (Hellerstein, 1991). Third, to allow for possible overdispersion of the data, we employ a negative binomial specification, which is a generalisation of the Poisson model (Haab and McConnell, 2002).

ARTICLE IN PRESS 84

M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

Table 2. Scenarios used in contingent behaviour exercise Cyclists A—‘‘suppose that next year a range of new optional trail obstacles were built along side the existing mountain bike trails at this forest. The types of challenges would include:  jumps,  drop-offs and,  sections of ‘northshore’ (raised wooden bike trails).’’ Cyclists B—‘‘suppose that new shower and changing facilities were built at this forest next year. These facilities would be free to use and would include:  showers,  changing room and,  secure lockers.’’ Horse riders E—‘‘suppose that next year a range of new optional trail obstacles were built along side the existing horse trails at this forest. The types of challenges would include jumps and ditches. The severity of these challenges would range from easy to difficult. all challenges would be situated on a short loop off the main horse trail and therefore would not directly affect the difficulty of existing trails.’’ Horse riders F—‘‘suppose that a new horse-friendly parking facility was built at this forest next year. This facilities would be free to use and would include:  Horse box friendly parking facilities that had plenty of room to park and manoeuvre large vehicles with horse boxes,  Safe horse corrals (pens) and tie up points.’’ Nature watchers I—‘‘suppose that next year several new wildlife viewing hides were built at various locations within this forest. The hides would be built throughout the forest in areas where various types of wildlife are known to congregate. All of the hides are likely to be located at least 1 mile from a car park and several will be built in quiet remote areas of the forest over 5 miles from a car park. Although all hides would be accessible by trails, these trails generally would not be suitable for pushchairs/wheel chairs.’’ Nature watchers J—‘‘suppose that next year a new wildlife viewing centre was built at a central location within this forest. It is expected that you would be able to see a variety of birds and some large mammals from the centre. Active wildlife management (including the use of feeding stations) would be used to attract the wildlife to the centre. The viewing centre would be built near a main car park in the forest. The viewing centre would be accessible using a short ‘all access’ path suitable for pushchairs and wheelchairs’’. Forest visitors M—‘‘suppose that next year a new art/sculpture trail was built within this forest. The Art/Sculpture trail would be approximately 1 to 2 miles long. The art/sculpture exhibits would depict images of the forest/countryside and be built with materials that blend in with the forest. The actual trail would be suitable for people of all abilities.’’ Forest Visitors N—‘‘suppose that next year a new family play facility was built at a central location within this forest. The play facilities would include play equipment for all ages including:  An enclosed safe play area for toddlers,  Traditional and ‘adventure’ play facilities for older children, and  High wire ‘Go Ape’ facilities for teenagers (and the odd adult!). All facilities would be built with material that blends in with the forest.’’

The survey instrument Data for this research was collected using on-site, in-person interviews between May and September 2005. Interviews were undertaken at seven forests

ARTICLE IN PRESS M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

85

throughout Great Britain: Glentress, Dyfnant, Cwm Carn, Thetford, New Forest, Rothiemurchus, and Whinlatter. These forests were selected to cover the range of recreational activities examined in the research. Interviews were conducted during both week days and weekend days, as well as during all daylight hours. The format of the survey questionnaire followed standard guidelines for the design of the valuation survey instruments (Arrow et al., 1993; Bateman et al., 2002; Hensher et al., 2005; Herriges and Kling, 1999). Survey respondents were therefore provided with some background information on the study and then asked to outline how they use forests for recreation. Next, they were presented with information on how the forest where they were sampled might be improved for recreation. Respondents were then presented with a single contingent behaviour scenario (out of the eight listed in Table 2) and asked to identify the extent to which their number of planned trips to the forest in the next 12 months would change if the stated improvements were made. Respondents were then presented with a series of four CE choice tasks. Before making their choices, respondent were presented with a series of statements that reminded them to answer truthfully and account for their personal budget constraints. A follow-up question was also asked to identify respondent’s motivations when answering the choice tasks. Finally, socio-economic, demographic and attitudinal data was collected from the respondents.

Results Descriptive statistics A total of 1568 on-site, personal interviews were undertaken during this research. Table 3 provides a breakdown of where the interviews were undertaken by forest and by recreation activity. Table 3 also provides a summary of the main recreation activity undertaken during trips to the respective forests. General forest users accounted for just under half (47.5%) of the total sample, while cyclists accounted for 37.3% of the sample. Horse riders and nature watchers were less well represented accounting for 7.1% and 8.1% respectively. These low numbers reflect the fact that there were often very few people in the forests undertaking these two activities. In Glentress and Cwm Carn, the majority of people interviewed were cyclists; reflecting the fact that both these forests were managed for mountain biking. There were also a significant number of cyclists interviewed at Thetford. The majority of horse riders were interviewed at Dyfnant and the New Forest. Nature watchers were found at all sites, but in low numbers. Finally, general forest users were found in large numbers at Thetford, the New Forest and Rothiemurchus. It should be noted that approximately three-quarters of the people approached by interviewers completed a questionnaire. Analysis of non-respondents indicated that their basic demographics were not significantly different from those respondents who completed the survey.

ARTICLE IN PRESS 86

M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

Table 3. Summary of the number of interviews undertaken by forest and main activity Forest

Glentress Dyfnant Cwm Carn Thetford New Forest Rothiemurchus Whinlatter Total

Main recreation activity undertaken

All respondents

Cycling

Horse riding

Nature watching

General forest recreation

221 5 260 70 17 8 4 585

6 37 0 2 60 0 6 111

27 29 5 24 5 30 7 127

47 61 1 205 161 267 3 745

301 132 266 301 243 305 20 1568

Choice experiments results A series of CE models were run for the forest recreation data. These models are reported in Tables 4–7 respectively for cyclists, horse riders, nature watchers and general forest users. The implicit prices reported in the results below relate to per person per trip values for visiting a forest that has the specified facilities provided relative to not visiting the forest (i.e. staying at home). Cyclists Table 4 reports a series of CE models for cyclists. Model 1 includes all respondents who indicated that cycling was the main activity undertaken during the trip to the forest, while Models 2, 3, and 4 disaggregate this sample into different sub-groups of cyclists. Model 2 is restricted to include ‘leisure cyclists’ only; i.e. those who indicated that they were not riding on cross country trails, technical single track, downhill or dirt jumping. Model 3 represents ‘mountain bikers’; defined as those cyclists who indicated that they were riding on either cross country trails or on technical single track. Finally, Model 4 includes ‘Downhill riders’; defined as those riders who indicated that they were downhill cycling or dirt jumping. For expedience, the Models reported in Table 4 only show data on the forest attributes and an Alternative Specific Constant (ASC) based on the ‘stay at home’ option; socioeconomic and attitudinal variables are not shown in these models. In Model 1 all of the cycle-specific attributes were significant and positive, suggesting that cyclists consider these attributes to be important and that they are more likely to visit a forest if these attributes are provided. The highest ‘per person per trip’ implicit prices (estimated by dividing the coefficient of the attribute with the negative of the coefficient on the travel cost attribute) were found for forests with ‘dedicated downhill trails’ (£9.74 per person per trip), ‘dedicated single track trails’ (£8.40), and ‘obstacles’ (£7.56). Other significant attributes included ‘dedicated cross country trails’ (£5.81), ‘bike wash facilities’ (£4.27) and ‘changing and shower

Table 4. Choice experiment models—cyclists Model 3 Mountainbikers

Model 4 Downhill riders

ASC Trails (dedicated cross country) Tails (dedicated single track) Trails (dedicated downhill) Obstacles (jumps and drop-offs) Bike wash facilities Changing and shower facilities Parking, toilets, picnic Parking, toilets, picnic, cafe´, shop Parking, toilets, picnic, cafe´, shop, play areas Detailed information Enhanced surroundings Travel cost

0.324*** (6.246) 0.106*** (4.550) 0.154*** (6.488) 0.178*** (7.334) 0.138*** (11.512) 0.078*** (6.386) 0.029** (2.451) 0.014 (0.497) 0.019 (0.669) 0.029 (0.991)

0.532*** (2.800) 0.033 (0.386) 0.083 (0.981) 0.053 (0.597) 0.030 (0.681) 0.092** (2.010) 0.001 (0.028) 0.031 (0.320) 0.010 (0.108) 0.103 (1.064)

0.369*** (6.544) 0.125*** (4.934) 0.182*** (7.122) 0.162*** (6.15) 0.134*** (10.346) 0.073*** (5.572) 0.031** (2.444) 0.023 (0.734) 0.006 (0.181) 0.033 (1.035)

0.427*** (4.639) 0.018 (0.456) 0.002 (0.060) 0.342*** (8.298) 0.192*** (9.365) 0.045** (2.160) 0.011 (0.544) 0.051 (1.004) 0.042 (0.842) 0.033 (0.633)

0.003 (0.269) 0.004 (0.335) 0.018*** (35.555)

0.004 (0.096) 0.039 (0.871) 0.028*** (13.962)

0.007 (0.556) 0.011 (0.842) 0.018*** (6.544)

0.050** (2.437) 0.069*** (3.358) 0.015*** (17.371)

LL model LL (Constants only) LL ratio test (w2) p-value Pseudo-r2 Correct predictions Number of respondents Hausman test for IIA w2 [13] Pr (C4c)

11181.10 12149.13 1936.00 0.000 0.08 0.429 566

1061.04 1202.45 282.81 0.000 0.12 0.418 54

9416.237 10244.06 1655.65 0.000 0.08 0.414 480

3607.905 3888.635 561.46 0.000 0.07 0.411 183

30.199 0.004

18.488 0.149

24.476 0.027

12.462 0.490

87

Notes: t-stats in parenthesis. * significance at p ¼ 0.1. ** significance at p ¼ 0.05. ***significance at p ¼ 0.01.

ARTICLE IN PRESS

Model 2 Leisure Cyclists

M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

Model 1 All cyclists

88

Table 5. Choice experiment models—horse riders Model 3 Endurance riders /carriage drivers

ASC Trails (dedicated horse riding) Trails (dedicated carriage driving) Trails (dedicated long distance trails) Obstacles (jumps and ditches) Horse friendly parking Horse corrals and tie-ups Parking, toilets, picnic Parking, toilets, picnic, cafe´, shop Parking, toilets, picnic, cafe´, shop, play areas Detailed information Enhanced surroundings Travel cost

0.067 (0.474) 0.062 (0.982) 0.235*** (3.894) 0.155** (2.360) 0.033 (1.055) 0.004 (0.138) 0.018 (0.557) 0.426*** (6.654) 0.219*** (3.103) 0.181*** (2.721) 0.028 (0.893) 0.063** (1.997) 0.027*** (18.718)

0.009 (0.061) 0.092 (1.390) 0.220*** (3.483) 0.165** (2.378) 0.027 (0.830) 0.009 (0.277) 0.031 (0.897) 0.506*** (7.556) 0.309*** (4.187) 0.236*** (3.360) 0.020 (0.609) 0.028 (0.837) 0.026*** (17.456)

1.425*** (2.957) 0.089 (0.410) 0.212 (0.893) 0.110 (0.576) 0.151 (1.404) 0.131 (1.166) 0.177* (1.708) 0.640*** (2.789) 0.701*** (2.944) 0.638*** (2.865) 0.466*** (4.056) 0.426*** (3.718) 0.040*** (7.433)

LL model LL (Constants only) LL ratio test (w2) p-value Pseudo-r2 Correct predictions Number of respondents Hausman test for IIA w2 [13] Pr (C4c)

2022.282 2296.042 547.52 0.000 0.12 0.422 105

1829.171 2075.314 492.29 0.000 0.12 0.421 95

217.618 284.432 133.63 0.000 0.23 0.500 10

18.052 0.156

21.146 0.070

— —

Notes: t-stats in parenthesis. *significance at p ¼ 0.1. **significance at p ¼ 0.05. ***significance at p ¼ 0.01.

ARTICLE IN PRESS

Model 2 Family/leisure riders

M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

Model 1 All horse riders

Table 6. Choice experiment models—nature watchers Model 3 Nature watchers—centres & nature trails

ASC Trails (dedicated easy access nature trails) Trails (dedicated nature trails with information) Trails (‘off the beaten track’ nature trails) Wildlife hides Wildlife viewing centres Guided nature walks Parking, toilets, picnic Parking, toilets, picnic, cafe´, shop Parking, toilets, picnic, cafe´, shop, play areas Detailed information Enhanced surroundings Travel cost

0.450*** (4.098) 0.008 (0.150) 0.075 (1.348) 0.115** (2.087) 0.121*** (4.180) 0.100*** (3.406) 0.010 (0.357) 0.084 (1.400) 0.009 (0.162) 0.033 (0.527) 0.029 (1.003) 0.064** (2.199) 0.018*** (15.163)

0.462*** (3.642) 0.005 (0.083) 0.046 (0.714) 0.117* (1.832) 0.152*** (4.504) 0.131*** (3.783) 0.019 (0.557) 0.106 (1.537) 0.053 (0.777) 0.012 (0.171) 0.002 (0.047) 0.071** (2.067) 0.020*** (14.222)

0.622*** (4.685) 0.010 (0.137) 0.058 (0.806) 0.114 (1.599) 0.102*** (2.695) 0.181 (4.723) 0.036 (0.987) 0.095 (1.195) 0.074 (1.006) 0.054 (0.671) 0.002 (0.064) 0.071* (1.851) 0.018*** (12.044)

LL model LL (Constants only) LL ratio test (w2) p-value Pseudo-r2 Correct predictions Number of respondents Hausman test for IIA w2 [13] Pr (C4c)

2496.812 2654.437 315.25 0.000 0.06 0.385 123

1895.11 2039.41 288.60 0.000 0.07 0.396 95.0

1578.538 1683.511 209.95 0.000 0.06 0.396 79.0

7.254 0.888

19.326 0.113

9.942 0.697

89

Notes: t-stats in parenthesis. *significance at p ¼ 0.1. **significance at p ¼ 0.05. ***significance at p ¼ 0.01.

ARTICLE IN PRESS

Model 2 Nature watchers— general

M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

Model 1 All nature watchers

90

Table 7. Choice experiment models—general forest users Model 3 Walkers

Model 4 Non-active general visitors

ASC Trails (Easy access) Trails (Art/sculpture walks) Trails (Long distance walking) Single track mountain bike trails Horse riding trails Nature trails/wildlife hides Parking, toilets, picnic Parking, toilets, picnic, cafe´, shop Parking, toilets, picnic, cafe´, shop, play areas Detailed information Enhanced surroundings Travel cost

0.153*** (3.368) 0.011 (0.455) 0.055** (2.384) 0.015 (0.620) 0.088*** (7.351) 0.004 (0.288) 0.030** (2.473) 0.133*** (5.410) 0.112*** (4.450) 0.084*** (3.367)

0.024 (0.395) 0.018 (0.562) 0.048 (1.501) 0.080** (2.420) 0.098*** (5.966) 0.017 (0.995) 0.015 (0.908) 0.157*** (4.617) 0.102*** (2.952) 0.043 (1.231)

0.068 (1.310) 0.003 (0.099) 0.040 (1.510) 0.022 (0.816) 0.071*** (5.211) 0.028** (2.022) 0.025* (1.848) 0.187*** (6.737) 0.160*** (5.530) 0.087*** (3.059)

0.164*** (2.801) 0.039 (1.346) 0.088*** (3.031) 0.042 (1.396) 0.088*** (5.894) 0.030* (1.897) 0.039** (2.571) 0.061** (1.963) 0.051 (1.620) 0.107*** (3.496)

0.002 (0.181) 0.052*** (4.263) 0.019*** (38.601)

0.029* (1.730) 0.063*** (3.776) 0.019*** (27.402)

0.018 (1.347) 0.038*** (2.757) 0.019*** (33.191)

0.007 (0.489) 0.050*** (3.263) 0.020*** (31.822)

LL model LL (Constants only) LL ratio test (w2) p-value Pseudo-r2 Correct predictions Number of respondents Hausman test for IIA w2 [13] Pr (C4c)

14281.98 15235.14 1906.32 0.000 0.06 0.387 706

7428.155 7916.58 976.85 0.000 0.06 0.385 365.5

11038.29 11741.4 1406.22 0.000 0.06 0.386 544

9114.99 9770.897 1311.81 0.000 0.07 0.390 452.3

16.238 0.236

14.524 0.338

17.157 0.192

8.826 0.786

Notes: t-stats in parenthesis. *significance at p ¼ 0.1. **significance at p ¼ 0.05. ***significance at p ¼ 0.01.

ARTICLE IN PRESS

Model 2 Active general Visitors

M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

Model 1 All general Visitors

ARTICLE IN PRESS M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

91

facilities’ (£1.58). Interestingly, the ‘multipurpose trails’ attribute3 was significant but negative (implicit price ¼ £23.95). This suggests that cyclists did not want to share trails with other users. None of the other non-cycle-specific attributes (i.e. ‘general facilities’ such as picnic sites, cafe´s, play areas, ‘information’ and ‘enhance surroundings’) were significant in Model 1, suggesting that cyclists did not consider these attributes to be important to their choice of forest. Finally, the travel cost (distance) attribute was (as expected) significant and negative; i.e. cyclists would be less likely to go to a forest if the distance required to travel was greater. The goodness-of-fit of each model was estimated using the pseudo-r2 statistic, which is estimated using: psuedor2 ¼

LLEstimated model . LLBase model

The psuedo-r2 value for Model 1 is 0.08. This value is lower than the minimum recommended acceptable value of 0.1 (Louviere et al., 2000). A further approach to measuring how well the model is performing is to determine the proportion of choices correctly predicted. This is achieved using data from a contingency table of the predicted choice outcomes for the sample based upon the estimated model versus the actual choice outcomes as they exist within the data (Hensher et al., 2005). In Model 1, 42.9% of predictions are correct. Finally, we also report the Hausman test for the IIA assumption used in the conditional logit model. In Model 1, the p-value ¼ 0.04, indicating that there is a violation of IIA assumption. Although we recognise that this violation can be addressed using a less restrictive model specification such as a nested logit or a random parameters logit models (Hensher et al., 2005), we report conditional logit results for all groups of recreationalist in this paper, since IIA violations were generally not found in the other models examined in this research. Models 2, 3, and 4 relate to different sub-groups of cyclists. Model 2 is based on ‘leisure cyclists’. Within this Model, only three parameters appear to be significant: the ASC (stay at home), the ‘bike wash facilities’ and the ‘travel cost’ parameters. The parameter on the ‘bike wash facilities’ was positive (implicit price ¼ £3.32 per person per trip), indicating that leisure cyclists would prefer forests with bike wash facilities. The parameter on the ‘travel cost’ attribute is significant and negative indicating that respondents were less likely to choose an option if the distance required to travel to the forest was greater. None of the other forest attributes were significant in Model 2. Model 3 is based on mountain bikers. Here, all of the cycle-specific attributes are positive and significant. Positive implicit prices were therefore found for ‘dedicated single track trails’ (£10.07), ‘dedicated downhill trails’ (£8.93), ‘obstacles’ (£7.39), ‘bike wash facilities’ (£4.04) and ‘changing and shower facilities’ (£1.71). 3

Of the four levels of the ‘Type of trails’ attribute, three were included in the model as dummy variables using an effects coding matrix. The coefficient for the fourth attribute level (i.e. ‘multi-purpose trails’) was estimated as the negative sum of the other coefficients. A similar matrix was used for the ‘General facilities’ attribute.

ARTICLE IN PRESS 92

M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

‘Multipurpose trails’ was again significant but negative (implicit price ¼ £25.90), providing evidence that mountain bikers in particular did not want to share trails with other users. The ‘travel cost’ attribute was again significant and negative as expected. None of the general facilities attributes were significant in this model. Finally, in the ‘downhill riders’ model (Model 4), the ‘dedicated downhill trails’, ‘obstacles’ and ‘bike wash facilities’ attributes were all significant and positive. Of note is the fact that the ‘per person per trip’ implicit prices associated with these facilities (£23.34, £13.14, and £3.05 respectively) were the highest over all models and activities. The ‘detailed information’ attribute was also positive and significant (£3.41), while the ‘enhanced surroundings’ attribute was significant but negative (£4.71) suggesting that downhill riders did not want enhanced opportunities to view wildlife and points of interest (presumably to avoid distractions whilst travelling downhill very quickly!). The ‘travel cost’ attribute was again significant and negative. In terms of the significance of these models, LL ratio tests indicated that the inclusion of the forest attributes significantly improved the overall LL function of the three models. The pseudo-r2 values were 0.12, 0.08 and 0.07 respectively for Models 2, 3, and 4, and the number of correct predictions of all models was around 41%. Finally, the Hausman test for the IIA assumption was not rejected in Models 2 and 4 (p ¼ 0.140 and 0.490 respectively), but was rejected in Model 3 (p ¼ 0.027). Overall, these results suggest that there was overwhelming support for further investments to create and enhance mountain bike centres, and in particular to provide additional ‘hard core’ facilities such as downhill courses and optional obstacles such as jumps and drop-offs on existing trails. Furthermore, there was general support for the provision of bike wash facilities at forests where any form of cycling takes place. Horse riders Table 5 reports the CE models for horse riders. Model 1 reports the findings from all horse riders, while Models 2 and 3 respectively report models for ‘family/leisure riders’ and ‘endurance riders or carriage drivers’. In Model 1 (all horse riders) the ‘travel cost’ attribute is significant and negative as expected. In terms of the provision of horse specific facilities, the provision of ‘dedicated carriage driving’ facilities was positive and significant, increasing utility by £8.88 per person per trip, while the provision of ‘dedicated long distance routes’ was found to reduce utility (£5.85). The other horse riding specific attributes (‘obstacles’, ‘horse friendly parking’ and ‘horse corrals and tie-up’) were insignificant in the model and therefore appear not to influence the decision-making of horse riders. The ‘general facilities’ attribute was significant; however, there appears to be some inconsistencies within this attribute. The provision of ‘parking, toilets and picnic areas’ increased utility (£16.06), while the added provision of a cafe´/shop reduced utility (£8.25). A further increase in general facilities to include play areas then increased utility again (£6.83). Although it is unclear why these inconsistencies exist, it is interesting to note that the signs associated with the alternative levels of provision of general facilities are consistent with that found for the general forest

ARTICLE IN PRESS M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

93

visitors—see below. A possible explanation is that visitors prefer forests that have either minimal facilities, or ones that have a wide range of facilities. The enhancement of the forest surroundings for viewing features of interest also increased utility (£2.39). The LL ratio test statistic was significant at p ¼ 0.05, indicating that the inclusion of the forest attributes significantly improves the overall LL function of the model. The pseudo-r2 measure of goodness-of-fit was 0.12, and approximately 42% of choice outcomes were correctly predicted. The Hausman IIA test indicated that the IIA assumption was not violated in this model. Model 2 (family and leisure horse riders) was very similar to Model 1 both in terms of attribute coefficients, implicit prices and goodness-of-fit. This, however, was not surprising since the majority of horse riders (95 out of the 105 in the full sample) were included in this sub-group. Perhaps the main difference in this model was that the ‘enhanced surroundings’ was not significant in Model 2. Model 3 included only those riders who were participating in endurance rides or carriage driving. Only 10 respondents were included in this sub-sample and therefore it is unlikely that there is sufficient data to allow meaningful analysis of this data. Overall, the findings from the horse riders sample indicate that there was little evidence in support of the provision of horse specific facilities within forests. Information gathered in debriefing interviews indicates that the main reason for this lack of demand stems from the relative difficulties associated with transporting horses to and from forests. Nature watchers Table 6 reports the CE models for nature watchers. In Model 1 (all nature watchers), the ‘wildlife hides’ (£6.83 per person per trip), ‘wildlife viewing centres’ (£5.65) ‘enhanced surroundings’ (£3.62) and ‘off the beaten track nature trails’ (£6.48) were positive and significant. As expected, the ‘travel cost’ attribute was negative and significant in the model. Model 2 was restricted to only include casual nature watchers, i.e. those who did not use specific nature watching facilities (such as hides) during their visit to the forest. The significant attributes within this model were the same as in Model 1. Finally, Model 3 was restricted to include those nature watchers who stated that they were using a viewing centre, nature trail or guided walk. Within this Model only the ‘Wildlife hides’ and ‘Enhanced surroundings’ attributes were significant and positive. Implicit prices for these attributes were similar to those found in Models 1 and 2. In terms of the overall performance of the models, all of the LL ratio tests were significant; indicating that the inclusion of the attributes increased the performance of the models compared to the constants only model. The pseudo-r2 values for the models were around 0.06, which is rather low. All of the Hausman IIA tests indicated that the IIA assumption was not violated in any of the nature watcher models. General forest users The final sub-group of forest users were the general forest visitors. Four choice models were generated for this group (Table 7). In Model 1 (all general

ARTICLE IN PRESS 94

M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

forest visitors), attributes that were significant and positive included ‘single track mountain bike trails’ (£4.59 per person per trip), ‘nature trails/wildlife hides’ (£1.56), ‘Art/sculpture trails’ (£2.90), and ‘enhance surroundings (£2.70), while the ‘travel cost’ attribute was significant and negative. Similar to the horse riders, we find inconsistent results in terms of the general facilities attribute: ‘parking and toilets’ (£5.45), ‘parking, toilets and picnic’ (£6.95), ‘parking, toilets, picnic, cafe´, and shop’ (£5.87), ‘parking, toilets, picnic, cafe´, shop and play areas’ (£4.37). Again, the underlying reasons for these values are unclear. Insignificant attributes included ‘easy access trails’, ‘long distance walking trails’, ‘horse riding trails’ and ‘detailed information’. Models 2, 3 and 4 were based on sub-groups of the general forest users. Model 2 (active general visitors) included those general visitors who indicated that they partook in either a cycling, horse riding or nature watching activity whilst in the forest (note that these activities were not considered to be the single main activity undertaken during the trip to the forest). Within this group, ‘single track mountain bike trails’ (£5.26), ‘parking, toilets and picnic’ (£8.43) and ‘enhanced surroundings’ (£3.37) were found to be significant and positive. ‘Long distance walking trails’, ‘parking, toilets, picnic, cafe´, shop and play areas’, ‘detailed information’, and ‘travel cost’ were all significant and negative. Model 3 was based on general forest users who indicated that the main purpose of their visit was to walk in the forest. Significant and positive attributes included ‘single track mountain bike trails’ (£3.79), ‘nature trails/wildlife hides’ (£1.36) , ‘parking, toilets and picnic’ (£10.03) , ‘parking, toilets, picnic, cafe´, shop and play areas’ (£4.65) and ‘enhanced surroundings’ (£2.03). Significant and negative attributes included ‘horse riding trails’, ‘parking, toilets, picnic, cafe´, and shop’ and ‘travel cost’. Finally, Model 4 was based on those general forest visitors who did not cycle, horse ride, nature watch or walk during their trip to the forest. Significant and positive attributes included ‘art/sculpture walks’ (£4.38), ‘single track mountain bike trails’ (£4.40), ‘nature trails/wildlife hides’ (£1.95), ‘parking, toilets and picnic’ (£3.04), ‘parking, toilets, picnic, cafe´, shop and play areas’ (£5.35) and ‘enhanced surroundings’ (£2.47). ‘Horse riding trails’, ‘parking, toilets, picnic, cafe´, shop’ and ‘travel cost’ were again significant but negatively valued. In terms of the overall performance of the general visitor models, all of the LL ratio tests were significant. The Psuedo-r2 values were around 0.06, and the numbers of correct predictions were around 38%. The Hausman test indicated that the IIA assumption was not violated. Contingent behaviour results Given the contingent behaviour scenarios described in Table 2, eight models were estimated; two for each group of recreationalist. In each case, we are interested in (i) whether the ‘travel cost’ parameter is significant (if not, then no welfare estimates can be made), and (ii) whether the dummy variable for the change in site quality (‘Contingent Behaviour’ dummy) is significant (if not, no prediction of the change in visitor numbers can be made). As noted above, the econometric approach taken is to use a panel data estimator, rather than simply pooling the data and using OLS.

ARTICLE IN PRESS M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

95

A panel data estimator takes into account the correlation in the errors between each person’s two choices—actual and intended behaviour. We use a random effects rather than a fixed effects specification, since this fits the nature of our data better. Finally, since the dependent variable is an ‘‘count’’ integer, we must test whether a Poisson or Negative Binomial panel estimator is appropriate. To exclude overnight visitors, for whom the calculation of trip costs would be difficult, we restricted the sample to those visitors who had travelled no more than four hours in total to visit the forest where they were sampled. We initially ran both Poisson and Negative Binomial versions of each of the eight models. In all cases, tests on the over-dispersion parameter showed that the Negative Binomial was preferred over the Poisson. We also tested whether a panel specification was preferred to a pooled specification in each case, and the LR test statistics in all cases confirmed the need for a panel rather than pooled regression. Table 8 thus gives the results for all eight recreational cases, but just for the Negative Binomial random effects panel specification. Variables used were as listed in Table 8, namely travel costs, travel costs to the nearest substitute forest site, income, gender, age, education, site dummies for each of the forests sampled, minus one; and a ‘Contingent Behaviour’ variable, which is a dummy representing whether the visits we are explaining are actual, with current facilities, or hypothetical, with improved facilities. Looking at the number of observations in each group, we see that there are very few observations, relatively speaking, for both horse-riding groups. Furthermore, the ‘Contingent Behaviour’ variable is insignificant in both cases. We thus constrain our attention to the cyclist, nature watchers and general forest visitor groups, since in all six cases, the ‘Contingent Behaviour’ variable is significant at the 95% level or better. A point of interest is the strong gender effect in both cycling groups—men make many more trips, both actual and intended, than women—whilst income effects within the samples seem weak. In all six models, travel costs are significant and correctly signed. This is also true of the contingent behaviour (CB) dummy variable, which is significant and positive, indicating that hypothetical improvements in recreational facilities have the effect of increasing planned trips on average. The Wald statistic reported in Table 8 shows a high level of significance for each of the models. To estimate the recreation benefits from these improvements, two steps are needed. First, predict trips under current and hypothetical conditions, in order to calculate the change in predicted trips. Second, use the travel cost parameter estimate from the panel models to value this increase in trips in monetary terms. Table 9 summarizes these two stages, giving the change in predicted trips in column 2 and the change in consumers’ surplus per visitor per year in column 3. Confidence intervals for this welfare measure are also shown. As may be seen, the largest proportional changes in trips come from investing in new family play areas for general visitors (10.2% increase); and investing in new trail obstacles for cyclists (5.0% increase). The largest increase in consumers’ surplus per annum would accrue to general visitors for new family play areas (£8.75 per visitor per year), and for nature watchers if new wildlife hides are constructed in the forest (7.89 per visitor per year).

96

Travel cost to nearest sub. site Income Female D1 Dyfnant Forest D2 Cwm Carn forest D3 Thetford forest D4 New forest D5 Rothiemurchus Forest D6 Whinlatter Forest

Cycling B

Horse riders E

Horse riders F

Nature watchers I

Nature watchers J

Forest visitors M

Forest visitors N

0.052 (4.97)** 0 (0.57) 0.109 (0.93) 0.667 (2.71)** 0.788 (0.85) 0.153 (0.72) 0.614 (1.91) 0.777 (1.61) 1.785 (2.80)** 1.666 (1.68)

0.075 (7.10)** 0.038 (5.87)** 0.219 (1.67) 0.705 (2.90)** 0.969 (0.62) 0.136 (0.57) 0.008 (0.03) 1.537 (2.23)* 2.652 (2.47)* 3.136 (2.98)**

0.084 (2.58)** 0.126 (0.44) 0.227 (0.57) 0.624 (1.09) 1.425 (1.2)

0.092 (3.10)** 0.011 (0.13) 0.632 (2.43)* 0.74 (1.87) 1.805 (2.50)*

0.024 (2.32)* 0.009 (2.21)* 0.032 (0.32) 0.047 (0.37) 0.237 (0.69)

0.861 (0.64) 2.453 (3.35)**

0.997 (2.96)** 0.178 (0.5) 1.32 (3.93)**

0.339 (1.2) 0.144 (0.47) 0.618 (2.26)*

1.944 (1.09)

0.985 (0.79)

0.112 (2.51)* 0.023 (1.9) 0.314 (1.61) 0.106 (0.27) 1.833 (4.15)** 0.799 (0.97) 1.89 (3.48)** 0.847 (0.64) 2.898 (5.73)** 2.248 (3.76)**

0.043 (3.80)** 0.007 (2.11)* 0.111 (1.13) 0.048 (0.35) 0.533 (1.43)

5.008 (2.66)** 3.845 (3.07)**

0.095 (3.74)** 0 (0.03) 0.244 (0.85) 0.158 (0.74) 0.526 (1.5) 0.184 (0.35) 0.514 (1.77) 0.733 (0.7) 1.535 (3.51)** 0.553 (1.26)

ARTICLE IN PRESS

Travel cost

Cycling A

M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

Table 8. Negative Binomial random effects panel models for the Contingent Behaviour exercise

Age Income squared

Degree Holder

Observations No. of Respondents Wald w2(16) Log likelihood Likelihood ratio versus pooled Model

0.388 (1.77) 0.013 (0.42) 0.435 (0.61) 1.043 (1.12) 1.863 (2.54)* 0.12 (1.47) 19.804 (0.05) 58 29 64.3 132 61.93

*significant at 5% ** significant at 1%. Values in parentheses are t statistics.

0.292 (2.09)* 0.037 (2.04)* 2.078 (3.37)** 1.326 (2.16)* 0.534 (0.86) 0.121 (1.85) 14.53 (0.02) 94 47 84.22 222 111.23

0.072 (0.57) 0.015 (0.79) 0.062 (0.14) 0.69 (2.35)* 0.163 (0.47) 0.751 (5.39)** 17.854 (0.02) 90 45 87.36 155 10.69

0.003 (0.02) 0.022 (1.55) 0.314 (0.54) 0.182 (0.4) 0.028 (0.06) 0.363 (3.07)** 17.003 (0.03) 82 41 83.31 145 60.97

0.014 (0.25) 0.008 (1.21) 0.014 (0.07) 0.139 (0.66) 0.392 (2.03)* 0.167 (3.19)** 3.723 (6.04)** 558 281 73.38 1151 661.70

0.119 (2.33)* 0.003 (0.47) 0.285 (1.4) 0.421 (1.99)* 0.257 (1.39) 0.209 (3.47)** 2.354 (4.36)** 549 278 48.79 1095 557.58

ARTICLE IN PRESS

Contingent behaviour dummy Constant

0.192 (1.71) 0.018 (2.03)* 0.841 (2.13)* 0.085 (0.2) 0.391 (0.98) 0.052 (2.31)* 15.784 (0.05) 406 203 149.37 1175 645.58

M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

Qualification below A Levels A Levels or Equivalent

0.075 (0.74) 0.01 (1.17) 0.296 (0.79) 0.524 (1.4) 0.063 (0.19) 0.183 (5.16)** 3.984 (6.10)** 436 218 111.78 1244 593.11

97

ARTICLE IN PRESS 98

M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

Table 9. Results from the contingent behaviour models Improvement scenario

Cyclists: New optional trail obstacles built alongside existing bike trails. Cyclists: New shower and changing facilities provided at the forest. General visitors: New art/sculpture trails. General visitors: New family play areas provided at the forest. Nature watchers: several new hides built in forest Nature watchers – new wildlife centre built a

Predicted % change in trips over base

Increase in annual consumers’ surplus per visitor (£)a

+5.0

3.46 (2.46–5.62)

+0.3

0.66 (0.52–0.90)

+4.5 +10.2

2.79 (1.83–5.71) 8.75 (4.77–12.70)

+4.5

7.89 (5.35–18.73)

+2.0

3.30 (1.85–15.42)

95% confidence interval in parenthesis.

It is interesting to note that although the percentage increase in visits for this nature watcher group (4.5% increase) is smaller than for the cyclists new trail obstacles scenario (5.0% increase), the consumers’ surplus per visitor per year is higher in the nature watchers’ group due to a higher absolute increase in trips due to a higher initial predicted trip count. In other words, in the contingent behaviour models the consumers’ surplus per visitor per year estimates are influenced by both the consumers’ surplus values per visitor and the predicted changes in the number of visits. In contrast, the choice experiment models only take account of the consumers’ surplus values per visit.

Discussion This research adopted two valuation methods, the frequency-based CE model and the contingent behaviour model, to value a range of enhancements to the forest resource. This data was also disaggregated to explore any heterogeneity that may exist between uses and users. In this discussion, we first discuss the validity of the results, before discussing the policy implications of our findings. Assessment of the reliability and validity of findings The reliability of a valuation study is related to the extent to which the variation in values is due to random sources. An indication of the reliability may be measured in terms of the goodness of fit of a model. In the CE models, the pseudo-r2 values were generally below the recommended values of between 0.1 and 0.2 (Louviere et al., 2000). It was thought that, to some extent, the poor fit of the models may be related to the small sample sizes, particularly for disaggregated models. The contingent

ARTICLE IN PRESS M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

99

behaviour models, in contrast, fitted rather well, since all the Wald chi square statistics were highly significant. In six out of eight cases, the two focus variables (travel costs and contingent behaviour dummy) were found to be strongly significant and correctly signed. One aspect of the validity of valuation results is convergent validity: how closely estimates match to other sources of information on values. Since the choice experiment estimates of welfare change are per trip, and the contingent behaviour estimates are per year given the endogenous change in trip frequency, we cannot make an exact comparison across these two methods. Values might instead be compared to other studies, but there are no exact like-for–like comparisons in the literature. The data reported by Jones et al. (2003) includes 44 value estimates from contingent valuation studies (range ¼ £0.46–£1.55), 9 individual travel cost studies that use ordinary least squares estimators (range ¼ £0.47–£2.74), 7 individual Travel Cost studies using maximum likelihood estimators (range ¼ £0.07–£1.20), and 17 zonal TC estimates (range ¼ £1.58–£3.91). Overall, our values appear to be towards the upper bounds of these earlier estimates. However, none of these studies are directly comparable to our results and therefore no firm test of convergent validity can be made. Policy implications This research has provided much information on the value of improvements to the forest recreation resource. We now highlight some of the key policy implications stemming from our research. First, it is clear from the choice experiment data that the more specialist users attain greater consumers’ surplus per trip from the provision of activity specific facilities than non-specialist users. For example, mountain bikers had higher consumers’ surplus values per trip than general cyclists and indeed general forest visitors. Also, more specialist facilities were generally found to attract higher consumers’ surplus values. However, specialist activities often only attract relatively low number of users (relative to all forest recreationalists), and therefore forest managers need to be aware of the trade-offs between maximising benefits to individuals and maximising the overall benefits from a facility. Indeed, evidence from the contingent behaviour models demonstrated that the highest annual increase in consumers’ surplus per visitor came from the provision of family play areas. In this case, the high annual values were predominantly the result of relatively large increases in the predicted number of trips made to the forests with new play areas, as opposed to high per trip values. These results highlight a number of important issues that forest managers (and other end-users of valuation studies) need to consider with respect to their choice of valuation method adopted and the interpretation of the results from valuation studies. In particular, forest managers need to be clear, in their own minds, of what information they need to make informed resource allocation decisions. The evidence presented here demonstrates that the choice experiments method is better suited to providing detailed value estimates of the component attributes of a resource, and that this can be done in a relatively efficient way. However, choice experiments say

ARTICLE IN PRESS 100

M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

little about the likely change in the demand for a resource. The contingent behaviour method, on the other hand, has the advantage of being capable of predicting how a change in the quality of a resource affects user behaviour and also allows the value of that change to be measured. The contingent behaviour method is, though, restrictive in terms of the number of attribute that can be assessed in a single survey instrument. A challenge for environmental economists is therefore to explore new ways in which these methods can be refined and developed to combine the advantages of both techniques in a single survey protocol.

Conclusions This research aimed to provide an insight into the value that different groups of forest users have for a range of enhancements to the forest recreation resource. Novel aspects of this research include the fact that this is one of the first studies to value forest recreation utilising a combined RP-SP method: here we adopted the contingent behaviour model. Such methodologies are considered to be an improvement on either traditional revealed preference or stated preference methods since the combined approach draws on the relative merits of both approaches (Hanley et al., 2003a). Also, this study is also one of the first valuation studies to utilise an attribute-based valuation method to value the component attributes of forest recreation. Furthermore, in our analysis we analyse this data according to different groups of forest users, thus providing significant detail on the heterogeneity of values for enhancements to forest recreation. Another novel aspect to this research is that we utilised a frequency-based choice task in the CE model. Finally, this research has produced a wealth of information on the relative values of a range of improvements to the forest recreation resource by different uses and user groups. It is considered that this information will be valuable to the future management of forests in terms of enabling forest managers to best target resources to different forests and forest users. Acknowledgements We thank the Forestry Commission for funding the work on which this paper is based, and two referees for comments on an earlier version of the paper. References Adamowicz, W., Louviere, J., Williams, M., 1994. Combining revealed and stated preference methods for valuing environmental amenities. Journal of Environmental Economics and Management 26, 271–292. Adamowicz, W., Swait, J., Boxall, P., Louviere, J., Williams, M., 1997. Perceptions versus objective measures of environmental quality in combined revealed and stated preference models of environmental valuation. Journal of Environmental Economics and Management 32, 65–84. Arrow, K., Solow, R., Portney, P. R., Learner, E. E., Radner, R., Schuman, H., 1993. Report of the NOAA Panel on Contingent Valuations. Resources for the Future: Washington, D.C.

ARTICLE IN PRESS M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

101

Bateman, I.J., 1996. An economic comparison of forest recreation, timber and carbon fixing values with agriculture in Wales: a geographical information systems approach. Ph.D. thesis. Department of Economics. University of Nottingham, Nottingham. Bateman, I.J., Carson, R.T., Day, B., Hanemann, W.M., Hanley, N., Hett, T., Jones-Lee, M., Loomes, G., Mourato, S., O¨zdemiroglu, E., Pearce, D., Sugden, R., Swanson, J., 2002. Economic Valuation With Stated Preference Techniques. A Manual. Edward Elgar, Cheltenham. Benson, J.F., Willis, K.G., 1992. Valuing Informal Recreation on the Forestry Commission Estate. Forestry Commission, Wetherby. Bishop, K.D., 1992. Assessing the Benefits of Community Forests: An Evaluation of the Recreational User Benefits of Two Urban Fringe Woodlands. Journal of Environmental Planning and Management 35, 63–76. Bullock, C.H., Elston, D.A., Chalmers, N.A., 1998. An application of economic choice experiments to a traditional land use—deer hunting and landscape change in the scottish highlands [full text delivery]. Journal of Environmental Management 52, 335–351. Cameron, T.A., 1992. Combining contingent valuation and travel cost data for the valuation of nonmarket goods. Land Economics 68, 302–317. Christie, M., 1999. An assessment of the economic effectiveness of recreation policy using contingent valuation. Journal of Environmental Planning and Management 42, 547–564. Christie, M., Hanley, N., Warren, J., Murphy, K., Wright, R., Hyde, T., 2006. Valuing the diversity of biodiversity. Ecological Economics 58 (2), 304–317. Englin, J., Cameron, T.A., 1996. Augmenting travel cost models with contingent behavior data. Poisson regression analyses with individual panel data. Environmental and Resource Economics 7, 133–147. Forestry Commission, 2002. Wales Mountain Bike Survey 2002. Forestry Commission, Edinburgh. Forestry Commission, 2004a. Forest Visitor Surveys 2002. Forestry Commission, Edinburgh. Forestry Commission, 2004b. Forestry Statistics 2004. Forestry Commission, Edinburgh. Forestry Commission, 2005a. All Forests Visitor Monitoring. Forestry Commission Wales, Aberystwyth. Forestry Commission, 2005b. Forestry Commission GB and England Annual Report and Accounts 2003–04. Forestry Commission, Edinburgh. Haab, T.C., McConnell, K.E., 2002. Valuing Environmental and Natural Resources. The Econometrics of non-market Valuation. Edward Elgar, Cheltenham. Hanemann, W.M., 1994. Valuing the environment through contingent valuation. Journal of Economic Perspectives, 19–44. Hanley, N., 1989. Valuing Rural Recreation Benefits: An empirical comparison of two approaches. Journal of Agricultural Economics 40, 361–374. Hanley, N., Ruffell, R., 1993. The Valuation of forest characteristics. In: Adamowicz, W.L., White, W., Phillips, W.E. (Eds.), Forestry and the Environment—Economic Perspectives. CAB International, Wallingford. Hanley, N., Wright, R., Adamowicz, W., 1998. Using choice experiments to value the environment: design issues, current experience and future prospects. Environmental and Resource Economics 11 (3-4), 413–428. Hanley, N., Wright, R.E., Koop, G., 2002. Modelling recreation demand using choice experiments: climbing in Scotland. Environmental and Resource Economics 22, 449–466. Hanley, N., Bell, D., Alvarez-Farizo, B., 2003a. Valuing the benefits of coastal water quality improvements using contingent and real behaviour. Environmental and Resource Economics 24, 273–285. Hanley, N., Shaw, W.D., Wright, R.E., 2003b. The New Economics of Outdoor Recreation. Edward Elgar, Cheltenham. Hellerstein, D.M., 1991. Using count data models in travel cost analysis with aggregate data. American Journal of Agricultural Economics. Hensher, D.A., Rose, J.M., Greene, W.H., 2005. Applied Choice Analysis. A Primer. Cambridge University Press, Cambridge. Herriges, J.A., Kling, C.L., 1999. Valuing Recreation and the Environment: Revealed Preference Methods in Theory and Practice. Edward Elgar, Cheltenham. Jones, A., Bateman, I.J., Wright, J., 2003. Estimating Arrival Numbers and Values for Informal Recreational Use of British Woodlands. Forestry Commission, Edinburgh. Louviere, J.J., Hensher, D.A., Swait, J.D., 2000. Stated Choice Methods. Analysis and Application. Cambridge University Press, Cambridge. Mitchell, R.C., Carson, R.T., 1989. Using Surveys to Value Public Goods: The Contingent Valuation Method. Resources for the Future, Washington, DC.

ARTICLE IN PRESS 102

M. Christie et al. / Journal of Forest Economics 13 (2007) 75–102

Street, D.J., Burgess, L., Louviere, J., 2005. Quick and easy choice sets: constructing optimal and nearly optimal stated choice experiments. International Journal of Research in Marketing. 22, 459–470. TNS Travel and Tourism, 2004. Leisure Day Visits: Report of the 2002–3 GB Day Visits Survey. The Countryside Agency, Cheltenham. Willis, K.G., Benson, J.F., 1989. Values of User Benefits of Forest Recreation: Some Further Site Surveys. Forestry Commission, Edinburgh. Willis, K.G., Benson, J.F., Whitby, M.C., 1988. Values of user Benefits of Forest Recreation and Wildlife. Forestry Commission, Edinburgh.