E CO L O G I CA L E CO N O MI CS 63 ( 20 0 7 ) 4 2–5 3
a v a i l a b l e a t w w w. s c i e n c e d i r e c t . c o m
w w w. e l s e v i e r. c o m / l o c a t e / e c o l e c o n
ANALYSIS
Quantifying public preferences for agri-environmental policy in Scotland: A comparison of methods Dominic Morana,⁎, Alistair McVittiea , David J. Allcroftb , David A. Elstonc a
Land Economy Research Group, Scottish Agricultural College, West Mains Road, Edinburgh, EH9 3JG, Scotland, UK Biomathematics and Statistics Scotland, James Clerk Maxwell Building, The King’s Buildings, Edinburgh EH9 3JZ, Scotland, UK c Biomathematics and Statistics Scotland, The Macaulay Institute, Craigiebuckler, Aberdeen, AB15 8QH, Scotland, UK b
AR TIC LE I N FO
ABS TR ACT
Article history:
This paper compares two methods for determining policy priorities for reform of Scottish
Received 18 August 2005
agricultural support. Multifunctional agriculture attempts to establish a new balance
Received in revised form
between traditional commodity support and payment for the production of non-market
21 September 2006
goods and services that are increasingly demanded by the public. Supplying non-market
Accepted 21 September 2006
goods presents particular problems for optimal policy design, not least the elicitation of
Available online 28 December 2006
consumer demand for those goods. From public focus groups, a range of attributes was derived as central to the Scottish public’s preferences for future agri-environmental reform.
Keywords:
This information was then combined in two separate survey methods using the Analytical
Multifunctional agriculture
Hierarchy Process (AHP) and choice experiments (CE). Both applications suggest that the
Choice experiments
public has defined preferences and a willingness to pay (using general income taxation) to
Analytical Hierarchy Process
affect changes beyond the status quo, and that policy payments should be targeted towards both environmental and social benefits. The divergent preference orderings derived from the alternative methods can be considered in the light of previous methodological debates on question framing, bounded rationality and respondent uncertainty. We speculate about the validity of alternative methodologies for informing particular policy questions. © 2006 Elsevier B.V. All rights reserved.
1.
Introduction
A distinguishing characteristic of Ecological Economics is that it has challenged the utility theoretic basis of methods advanced by the prevailing neoclassical school of environmental economics, whose tools have, by default, been the hegemonic policy decision framework in many OECD countries. There is a conviction that monetary valuation methods and cost–benefit analysis often limit the decision making process and that alternative deliberative or multicriteria methods can be more fruitful in terms of the information that can be derived for policy making (Munda, 1996; Jacobs, 1997; Clark et al., 2000).
These criticisms vary in the extent to which they challenge the underlying theoretical validity (Spash, 1998; Rosenberger et al., 2001), and the extent to which they advance plausible alternatives for evaluating trade-offs (Toman, 1998). Ultimately they boil down to the deliberative nature of respective methods, and the extent to which participation is mediated through any other medium apart from money. This tension has led to something of a methodological divergence with some researchers (e.g. MacMillan et al., 2002) augmenting the neoclassical approach by providing more time, information and the opportunity to deliberate as part of the stated preference exercise. In the emerging ecological economic tradition, Kenyon et al. (2001) and McDaniels et al. (2003) investigate the
⁎ Corresponding author. SAC, Kings Buildings, West Mains Rd, Edinburgh, EH9 6GU, UK. Tel.: +44 131 535 4128; fax: +44 131 667 2601. E-mail addresses:
[email protected] (D. Moran),
[email protected] (D.J. Allcroft),
[email protected] (D.A. Elston). 0921-8009/$ - see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.ecolecon.2006.09.018
EC O LO G I CA L E C O N O M I CS 6 3 ( 2 00 7 ) 4 2–5 3
role of Citizen Juries in environmental decision making. These experiments have yielded mixed results and have not truly addressed the methodological gap that still exists in the Ecological Economics toolbox. Despite the alleged differences and respective methodological failings, there appear to be few papers assessing whether different preference elicitation methods actually provide comparable policy recommendations. Duke et al. (2002) come closest to the current application in applying conjoint analysis and multicriteria analysis to farmland preservation in Delaware. Hanley et al. (1998) had previously taken a more neoclassical approach in comparing contingent valuation and choice experiments in Scotland. This paper compares two methods at the policy level, which is a scale above the environmental features that are normally the object of revealed and stated preference methods. Specifically we consider the nature of preferences for the multidimensional agri-environmental reform, which is a contentious process across the EU. Multifunctional agriculture attempts to establish a new balance between traditional commodity support and payment for the production of non-market goods and services that are increasingly demanded by the public. Supplying non-market goods presents particular problems for economically efficient policy design; not least the fact that the preferences for many public goods are imprecise and that demand assessment methods are highly dependent on individual knowledge and information provided as part of the demand elicitation exercise. Both information sets may diverge from true picture that may emerge as a result of a policy reform process. By extension, the use of stated preferences to guide policy design needs to be mindful of how preferences can change with information on the environmental, economic and social consequences of reform. But this does not obviate the need for decision making to be informed by current preferences. In a recent paper, Hall et al. (2004) reviewed published evidence on how agri-environmental reforms might be matched to measured public preferences. The basic policy challenge faced in the reform process is that multifunctional agriculture lays greater emphasis on the supply of non-market goods, but that market failure handicaps the design of efficient policy that matches supply and demand. While government policies attempt to approximate assumed public preferences, an increasing emphasis on evidence-based value for money in all spending decisions suggests that some attention should be paid to the explicit measurement of public demand and the use of demand information in the budgetary process (Brubaker, 2004). Overall Hall et al. concluded that public preferences had never been consistently canvassed as part of the agenda of agri-environmental reform, and that the totality of existing studies provides only a partial evidence base for informing the trade-offs that might be relevant in policy design. This conclusion can be qualified by the fact that the task of summarising and conveying the range of issues and conjectures about agricultural reform is in fact highly complex, and that a single survey method is unlikely to yield a complete view of preferences. The paper did suggest that certain methods (i.e. multicriteria analysis and choice experiments) in combination could be worthwhile exploring to derive a consistent preference ranking.
43
Accordingly the, the aim of this paper is to report on extensive survey work applying the identified methods to determine the preferences of the general public in relation to agri-environmental reform in Scotland. This study starts from the premise that in all likelihood the public’s preferences for the range of market and non-market outputs are not well formed. Few people routinely transact the range of public and private goods or have an idea of the relevant trade-offs. From this basis, this study employs a range of methods, first to identify the range of issues and preferences, and then to determine an empirical ordering of public preferences that might be used to validate policy choices. The surveys explored trade-offs between the economic and environmental outputs from agriculture, whether these preferences were consistent across distinct regions of Scotland, and whether monetary and non-monetary preference elicitation methods would generate equivalent preference weightings for the attributes considered. The paper is structured as follows. In the next section we provide background to the policy reform agenda and the methods that we used to elicit policy preferences. This is followed by details of An application in Scotland comprising the sequence of design and administration phases: focus groups, survey design implementation and results. The results are presented for two separate methods prior to a conclusion.
2. The demand for agricultural and countryside outputs In common with other EU member states, Scottish agriculture is in transition as the system of agricultural support is reappraised. The reduction in production-related support payments and a move towards stewardship schemes and farmspecific land management contracts has led to wider debate about the purpose of sector support and the role of public preferences in determining the forms of aid that are extended to farmers. This debate has dovetailed with other public concerns arising from a series of food and animal related health scares. Overall the public has been sensitised to the wider impacts of agriculture on the rural environment and the fact that there are some unavoidable trade-offs to be considered as part of the policy design process. The Scottish Executive’s “A Forward Strategy for Scottish Agriculture” (Scottish Executive, 2001) and CAP reforms have placed greater emphasis on both the provision of environmental goods and measures for rural development. Emphasis on non-market goods, both environmental and social, marks a change from traditional support for market production. There are many stakeholders in the outcome of this change and it is important to understand the views that the public might assign to policies designed to deliver combinations of outputs. As part of the evolution of agri-environmental policy, governments have attempted to demonstrate the benefits of reform using an array of methods to measure the value of public goods from agriculture. Some research has also been directed towards the characterisation of the variety of public goods and other benefits, such as rural employment, local foods and the economic and social vibrancy of rural communities. While environmental economic techniques have been used to reveal the values attached to specific public goods, few
44
E CO L O G I CA L E CO N O MI CS 63 ( 20 0 7 ) 4 2–5 3
studies have attempted to gain insights into the nature of trade-offs that are inherent in public preferences over the range of policy outcomes. For example, how much more is the public willing to pay for water quality relative to rural employment and/or the production of local food? Noting the importance of the question, Hall et al. (2004) suggested that the variety of approaches in the existing body of studies eliciting public preferences did not provide a clear picture of what the public wants from agriculture and the countryside. From a methodological viewpoint it is of some interest to see whether methods from competing perspectives actually converge.
3.
An application
In an attempt to clarify the picture this study applies choice experiments (CE: Louviere et al., 2000) to the question of public preferences for agri-environmental support. We compare the results with those derived from a second method that is less exacting in terms of its links to economic theory and statistical rigour, yet allows us to frame the same choices in a less constrained way. We use a variant of multicriteria analysis called the Analytical Hierarchy Process (AHP). AHP allows us to set up a range of preference choice sets without including a price attribute. Respondents make pairwise comparisons between policy attributes and levels. From these observed choices, preference weights or scores can be derived. Both AHP and CE require preliminary definition of attributes through a process of focus group discussions. Overall this study is similar in spirit to Duke et al. (2002) that applied AHP and conjoint analysis. That study did not directly compare the methods; the conjoint study considered quantitative attributes whilst the AHP was qualitative. The authors suggest that the results “reinforced” each other. Duke et al. instead used the different methodologies to mitigate the shortcomings of each. In an attempt to improve the performance of both CE and AHP, the study adopted a four-stage process to define the relevant policy attributes. The four stages comprised an initial literature review, focus groups across Scotland, an initial ranking survey to narrow down the focus group output, and a main survey phase. The sample frame for the study comprised the adult population of Scotland. The country was divided into the commonest geographical distinction of South, Central and North; a division that would also allow for the estimation of any regional differences in preference. The South broadly corresponded to Dumfries and Galloway and the Borders. Central broadly covered the central belt including Edinburgh, Glasgow and Stirling. North included the Highlands and Aberdeenshire. The Scottish Household Survey (2002) definitions of urban and rural were used to select sample points (towns) from each area to ensure that the sample broadly reflected the spread of population. Sampling quotas were based on sex, age and social class to ensure that the sample was demographically representative. The surveys were administered in respondents’ homes by trained interviewers employed by NFO-Europe. The research focussed on the Scottish public although the non-market nature of the attributes considered could be generalised to a wider UK context.
3.1.
Focus groups
Six focus groups of between seven and nine participants were held, spread over three locations as indicated in Table 1. Focus group discussions were moderated by a professional market research company and lasted for 1½ h. Discussions were based around a pre-prepared topic guide developed from a literature review (see Hall et al., 2004). The focus group discussions allowed an in-depth exploration of participant opinions, and provided a selection of topics to explore in population-based surveys; in particular the range of economic and environmental attributes that underlie public attitudes towards the countryside and related economic and environmental tradeoffs. Specific focus was on the role of farming in the countryside and whether respondents associated many rural public goods with the presence and practice of farming. If farmers were identified as the suppliers of goods, should they be compensated and on what basis? In terms of public awareness of the issues the findings were enlightening. In the first instance the link between farming and the countryside was not always spontaneously drawn. But when a link was drawn it was generally a positive association. Respondents considered that other agencies beyond agriculture (e.g. the Forestry Commission) were also regarded as having some responsibility for the countryside. Respondents also recognised that in the current economic climate farmers were burdened by extra responsibility. Participants felt that public subsidy was justified if farmers were trading off production and thus their own livelihoods for the supply of public goods. In order to finance the aid to farmers, any price increases on food were widely rejected in favour of taxation to try to prevent the less well off being adversely affected. In other words, the supply of public goods was viewed as a collective responsibility; a factor that needs to be reflected in the choice of any hypothetical payment vehicle. Opinion was divided on the basis for distributing public funding to farmers. A number of options for allocating funding were discussed. These included allocation according to the number of visitors to an area, allocation by area with the most potential for supplying environmental and social goods, or allocation by discounting areas where financial aid would have little perceived impact. Overall, the small number of focus group participants suggested some preferences for changing the status quo mix of outputs, and a willingness to pay for these changes through general taxation. The empirical questions that followed were, did the general public support these changes
Table 1 – Overview of the focus groups, including location and principal participant characteristics Group number 1 2 3 4 5 6
Gender
Age
Location
Date
Mixed Mixed Mixed Mixed Mixed Mixed
20–34 35–55 20–34 35–55 20–34 35–55
Inverness Inverness Edinburgh Edinburgh Jedburgh Jedburgh
11/12/02 11/12/02 12/12/02 12/12/02 14/01/03 14/01/03
EC O LO G I CA L E C O N O M I CS 6 3 ( 2 00 7 ) 4 2–5 3
Table 2 – Attributes and levels used in the choice experiment Attribute
Level 1
Level 2
Environment
Current Enhance practices wildlife habitats Landscape Current Enhance and access practices landscape appearance Rural Current Maintain development practices farming communities Targeting of Current Towards social payments practices benefits
than conventional surveys of public opinion. As in any choice experiment, the initial task was to:
Level 3 Enhance the quality of lochs, rivers and wetlands Enhance public access
Promote locally grown food Towards environmental and landscape benefits
Additional (pilot) annual taxes £5 £10 £20 £40 £70 £100 (six levels) (main study) £5, £10 £25 £50 £100 £200
and would they be prepared to pay for them? These questions were the basis of a wider quantitative survey of the general public. Because the range of issues covered in the focus group was still very wide the output of these was then drafted into a small scale preliminary rating survey that was sent (by mail) to a different group of 170 respondents who were representative of the Scottish public. This survey allowed us to determine a short list of the range of statements made in the groups. At this point of the process an open-ended CV question was also included to gain a feel for the range of payments that might bound the overall willingness to pay for favoured policy changes. This information would be necessary for the design of the more focussed choice experiment. Two wider public surveys were then undertaken. The first applied a choice experiment (CE) the second applying the AHP.
3.2.
45
Choice experiments (CE)
In the choice experiment framework individuals are typically presented with 4 to 8 choice sets representing hypothetical scenarios consisting of a number of policy attributes. Each of these attributes has a number varying levels, one of which typically represents the status quo, or current policy situation. Respondents are asked each time to indicate their preferred option in each set. If the comparison is done on a pairwise basis then respondents must indicate whether they prefer attribute A or B. For example, one policy attribute for agri-environmental conservation might be wild plant species with three levels (‘stay the same’, ‘increase by 10%’, ‘decrease by 10%’). The attribute levels in the choice sets are varied to allow the researcher to infer the attributes that significantly influence choice, the implied ranking of attributes, marginal WTP for changes in attribute level, and WTP for a program which changes more than one attribute simultaneously. When well-designed, CE provide a statistically efficient means of estimating WTP for marginal changes in a range of attributes that are of policy interest such as endangered status, location of reserves, and habitat management. The design of the survey and its administration are more complex
• select the attributes (characteristics) of the resource management problem; • select the levels which these attributes could take in the experimental design, and • select the levels and distribution of the “price tags” to be attached to the policy scenarios. The selection of attributes and their levels was influenced by the focus group discussions, the preliminary rating survey and in terms of their practical link to policy. The latter decision
Box 1 Descriptions of attribute levels presented to respondents. Improving wildlife habitats: Farmers would receive additional payments to improve both the quantity and quality of wildlife habitats on their land. For example, work might include the planting or restoration of features such as hedgerows or field margins that act both as habitats and as “corridors” between areas of uncultivated land. Improving the quality of lochs, rivers and wetlands: Improving public access to the countryside: Farmers would receive additional payments to improve public access, for example, through the maintenance of paths, stiles and the provision of signposts. Improving landscape appearance: Farmers would receive additional payments for undertaking work such as the restoration of features like dry stone walls or traditional farm buildings. They can also be paid for environmental features such as woodlands and hedgerows, which have landscape impacts too. Maintaining farming communities: Farm policy would have the aim of maintaining farming communities and supporting rural employment. This would involve encouraging young farmers to stay in the industry and ensuring the viability of traditional smaller farms, which might be done through setting up local co-operatives to allow farmers to share machinery and labour. Promoting locally grown food: Farm policy would support efforts by farmers to promote their produce in local markets and to develop schemes such as labelling to add value to their products in wider markets. As well as the above options, farm payments can be targeted in the following two ways: 1. Where social and economic benefits are greatest, for example in the number of jobs being created or protected. 2. Where environmental and landscape benefits are greatest, for example in areas where there is potential for a higher number of different animals.
46
E CO L O G I CA L E CO N O MI CS 63 ( 20 0 7 ) 4 2–5 3
criterion was judged by a substantial input from a Scottish Executive project steering group that provided observations on the link between attributes and their practical policy relevance. Statistical estimation issues also restrict the number of attributes and the numbers of levels. Finally, for a given number of respondents predetermined by budgetary limits, there is a limit on the number of choice sets that each respondent can realistically cope with as well as the complexity in terms of the number of attributes and their levels. The five attributes finally selected and their respective levels are summarised in Table 2. This mix of attributes was intended to capture the most relevant features of the public perception of the trade-offs between different public goods. Note that the inclusion of environmental, landscape and rural development attributes reflected the current emphasis on multifunctional agriculture within policy development. Box 1 presents the attribute levels as described to respondents. Whereas previous valuations of agri-environment policy have considered specific features and in some cases quantitative changes, this study has taken a broader, qualitative, view. This was primarily due to constraints placed on the size of design, and the burden we could realistically place on respondents. However, it offered flexibility in policy response within these attributes and provided trade-off information between broad strands of rural policy. These attributes and levels were then combined into a series of two-way choices (see Box 2). In each pair, the respondent was offered two alternative policy designs and asked which they preferred. If the respondent preferred neither of these options, they were then asked which was their least preferred option, thus implying a preference for the other alternative. This
Box 2 Example of choice experiment scenario.
essentially forced a choice. Whereas some studies specifically offer a “neither” option, i.e. a preference for the status quo, this approach allowed the greatest statistical efficiency given the restrictions faced regarding sample sizes. It should be noted that only 4% of responses (not respondents) were for the “neither” option, a further 2% were “don’t know”. Regarding option A and option B of the choice set as distinct, the three levels within each attribute give six pairs of different levels, one of which must be selected for the design to have full efficiency for the main effects. With four such attributes, this leads to 64 = 1296 combinations, and allowing each of the six price levels to be associated with each option gives a full choice set of 66 = 46,656 questions, choosing to give each respondent six questions would require 7776 respondents. In order to reduce this to a more realistic size, we sifted the set of questions by a factor of 36, resulting in 1296 questions, hence six questions for each of 216 respondents in each of the three sample regions. The method of reduction used ensured a design that is still balanced in the respect that each of the 64 combinations of levels of the policy attributes occurs exactly once, and each of the 62 combinations of price levels occurs 36 times. A combination of Latin squares was used to group the questions into sets of 6 for the respondents in as balanced a way as possible. Note that the 1296 questions were all distinct, a departure from the common practice of selecting a small number of questions that allow estimation of main effects under the assumption that interactions are absent and then gaining adequate sample size by replication of this same small number of questions. A small pilot study (106 respondents) was conducted to see how well the exercise performed. This exercise used price
47
EC O LO G I CA L E C O N O M I CS 6 3 ( 2 00 7 ) 4 2–5 3
Table 3 – Scoring system used to determine relative importance between AHP criteria Rating 1 2 3 4 5 6 7 8 9
Explanation of relative importance Two options are equally important Between 1 and 3 Chosen option is slightly more important Between 3 and 5 Chosen option is moderately more important Between 5 and 7 Chosen option is much more important Between 7 and 9 Highest possible degree of importance of chosen option over the other
levels of £5, 10, 20, 40, 70, and 100. Analysis of these data suggested that while balanced choices were being made, the options carrying the highest WTP prices were being chosen more frequently than expected. This high acceptance provided evidence that we could go to higher price levels and so for the full survey the price levels used were £5, 10, 25, 50, 100, and 200. In each case an approximate doubling of prices was maintained in going from one level to the next. The same design was repeated in each of the study areas: South, Central, and North. The South region consisted of the Borders and Dumfries and Galloway; the Central area consisted of the Central Belt; and the North region incorporated Aberdeenshire, Moray, Inverness, and Caithness. The surveys were administered by a market research company using face-to-face interviews during July and August 2003. Three samples were used to cover the South (225 respondents), Central Belt (224) and North (224) of Scotland. In total including the pilot survey, 673 responses were collected. Within each of the samples, a quota was used to ensure representativeness in terms of gender, age, social group and urban or rural residency.
3.3.
The Analytical Hierarchy Process (AHP)
AHP is a variant of a family of methods collectively termed multicriteria analysis. The method uses a number of pairwise comparisons between quantitative or qualitative criteria to assess the relative importance of each criterion. These can be arranged in a hierarchical manner known as a value tree to form sets of attributes and qualities (levels) within these attributes. The simplicity of the AHP approach is that unlike conjoint methods such as choice experiments, the qualities (or levels) of different attributes are not directly compared, thus removing the need for complex survey designs and associated impacts on sample size. Indeed, the AHP can be applied to single person expert samples (Duke and Aull-Hyde, 2002). The majority of the small number of existing applications of AHP to environmental and natural resource management issues have involved small samples of experts, resource managers and stakeholders (Duke and Aull-Hyde, 2002), the aim having been to reach consensus on management decisions and priorities in a manner similar to Delphi exercises, but in a way that also elicits the relative “utilities” of different management options. Respondents first make pairwise comparisons of the qualities within each attribute before comparing each of the
attributes. Cognitive burden may also be reduced as comparisons are between two qualities or attributes rather than a larger bundle of attributes and levels. As a consequence, respondents are less likely to adopt a simplistic choice heuristic such as concentrating disproportionately on one attribute as may be the case in CE (Swait and Adamowicz, 2001). Evidence of whether this is indeed the case is discussed below. The pairwise comparison is framed in the form of a question: how important is option A relative to option B? Where the options are individual attributes or levels, the responses to these questions are typically coded along a nine-point scale as set out in Table 3. If, for example, B is considered to be much more important than A, then the reciprocal of the relevant rating is assigned (i.e. 1/7 as opposed to 7 if A were strongly more important than B). As it is assumed that a respondent is consistent in judgements about any one pair of criteria, this use of the reciprocal requires only n( n − 1)/ 2 comparisons to be made whereas there are n criteria. The ratings, and their reciprocals, are then collected in a comparison matrix: 2
3 1 7 9 4 1=7 1 25 1=9 1=2 1 Weights are then estimated, which are consistent with the relativities between the attributes or qualities contained in the matrix. Although there is consistency in the judgements made between any pair of criteria, this is not guaranteed in judgements between pairs, so estimated weights must be derived that aim to provide the “best fit” for the observations (DTLR, 2001). This can be achieved by calculating the geometric mean of each row and normalising these by dividing by the sum of geometric means for each row. For the above matrix the weights would be:
Criterion 1 Criterion 2 Criterion 3 Sum = 5.0193
Geometric mean (1 × 7 × 9)1/3 = 3.9791 (1/7 × 1 × 2)1/3 = 0.6586 (1/9 × 1/2 × 1)1/3 = 0.3816
Weight 0.7926 0.1312 0.0760 1.000
In comparison with the rigourous CE design, the AHP format is less exacting. The attribute levels used were the same as for the CE, with an additional rural development level: “Preserve rural character”. This level represents a more general rural development aim not necessarily associated with agriculture. Constraints on the number of levels in the CE due to design and sample sizes do not apply with the AHP. Thus a wider range of attributes/levels can be considered. However, the length of task we can realistically expect respondents to engage in remains a constraint in common with the CE. The AHP questionnaire was administered to 169 respondents throughout Scotland using face-to-face interviews.
3.4.
Choice experiment results
The choice set data were analysed using a generalised linear model in GenStat (VSN International Ltd), which has some attractive features in terms of the treatment of covariates such as socio-economic or attitudinal variables. The response
48 Table 4 – Choice experiment results by region and for combined sample (standard errors in brackets, t-statistics in italics, estimates for comparisons of attribute levels which are significant at the 10% level are given in bold) South 2v1 a
Landscape/ access Rural development Targeting Common standard error Constant
Price
a b c d e
0.2606 6.15 0.169 d
d
3v2 c d
0.2554 6.02 0.2184 d
−0.0052 − 0.12 0.0494
2v1
3v1
North 3v2
2v1
3v1 d
Combined sample 3v2
d
0.2511 5.92 0.0717 e
d
d
0.3001 7.08 0.1059 d
0.049 1.16 0.0343
0.1912 4.54 0.1405 d
0.2081 4.94 0.0824 d
2v1
3v1
3v2
0.0169 0.4 − 0.058
d
d
0.2338 9.58 0.1262 d
0.2537 10.4 0.1351 d
0.0198 0.81 0.0088
3.99 0.2326 d
5.15 0.39 d
1.17 0.1573 d
1.69 0.1834 d
2.5 0.2624 d
0.81 0.079 e
3.34 0.276 d
1.96 0.3712 d
− 1.38 0.0952 d
5.17 0.2298 d
5.54 0.3397 d
0.36 0.11 d
5.49 0.0995 d 2.35
9.2 0.1505 d 3.55 (0.0424)
3.71 0.0509 1.2
4.33 0.1214 d 2.86
6.19 0.1359 d 3.21 (0.0424)
1.86 0.0144 0.34
6.56 0.097 d 2.3
8.82 0.1337 d 3.18 (0.0421)
2.26 0.0367 0.87
9.42 0.1062 d 4.35
13.92 0.1391 d 5.7 (0.0244)
4.51 0.0329 1.35
0.0551 (0.0597) 0.92 −0.0046 d (0.000687) − 6.68
2v1: coefficient for moving from status quo to second level. 3v1: coefficient for moving from status quo to third level. 3v2: coefficient for moving from second level to third level. Significant at the 5% level. Significant at the 10% level.
0.1566 d (0.0596) 2.63 −0.0045 d (0.000687) − 6.53
0.0364 (0.0592) 0.61 − 0.0048 d (0.000686) − 6.96
0.0821 d (0.0343) 2.39 −0.0046 d (0.000395) − 11.62
E CO L O G I CA L E CO N O MI CS 63 ( 20 0 7 ) 4 2–5 3
Environment
3v1 b
Central
49
EC O LO G I CA L E C O N O M I CS 6 3 ( 2 00 7 ) 4 2–5 3
variable is binary (A vs. B) and so a binomial error structure was used with logit link function. The terms were fitted in such a way that allowed the levels within each attribute to be compared with each other. Estimates are presented for the welfare effects of attribute level differences. 2 vs. 1 and 3 vs. 1 (1 being current practices) and also 3 vs. 2. The statistical significance of each effect was assessed using the corresponding t-statistic. By forming ratios of these estimates to the estimated price coefficient, we estimated the monetary value represented, i.e. how much, on average, a person was willing to pay for one option over another. Inclusion of “current practice” levels for each attribute allows prices to be interpreted as how much the public are willing to pay for an attribute level as compared with the current situation and also allows comparisons to be made of the size of effects between attributes as well as within. For example, the estimated coefficient, and hence implicit price, for 2 vs. 1 represents the extent to which respondents prefer a policy offering level 2 of the attribute over the current policy situation represented by level 1. As “current practices” were included in all policy attributes we have a baseline that allows us to compare directly preferences for different attributes. Table 4 shows results for the basic model for each of the study areas separately. As there were no significant differences between the three areas, we also present the results for the combined sample. Price can be included in the model as a factor with discrete levels, or as a continuous variable, using either a linear or a log scale. Investigations showed that a continuous linear scale was the most suitable, with the added advantage of ease of interpretation. In each case the difference between the “current practice” level and either level two or three is positive and significant at the 5% level. This
demonstrates that in each case the public preferred both of the new policy levels to the status quo. Therefore we can say that there were positive public preferences for new policies consisting of combinations of the attributes considered. The extent to which preferences were expressed between the policy levels is indicated in the coefficients for the 3 vs. 2 comparison, i.e. the relative preferences for two policy levels within each attribute: • enhanced water quality was preferred to enhanced wildlife habitats, except in the South, for example the coefficient for the South was − 0.0052 compared to 0.0198 for the combined sample; • enhanced public access was preferred to enhanced landscape appearance, except in the North, the coefficient for the North was −0.058 compared to 0.0088 in the combined case; • promotion of locally grown food was preferred to maintaining farming communities; and • targeting of environmental and landscape benefits was preferred to targeting of social benefits. However, the differences were only statistically significant for preferences between the promotion of locally grown food and maintaining farming communities. This was the case in each of the regions and in the combined sample. From the information given in Table 4 we can use the relative sizes of the estimated differences between attribute levels and the price coefficient to calculate how much, on average, a person was willing to pay for the difference between attribute levels. Table 5 shows these estimated values and, because price was included in the model as a linear term, these are directly in pounds. Confidence intervals for these estimated values were calculated by simulating from the distributions of the coefficients from the fitted model. For each
Table 5 – Implicit prices for different attribute levels (upper and lower 95% confidence levels in italics, estimates for comparisons of attribute levels which are significant at the 10% level are given in bold) South
Central
North
Combined sample
Estimated values (£) 2v1 Environment Lower 95% CI Upper 95% CI Landscape/Access Lower 95% CI Upper 95% CI Rural development Lower 95% CI Upper 95% CI Targeting Lower 95% CI Upper 95% CI Standard errors Environment Landscape/access Rural development Targeting a b c
a
£56.81 £36.30 £87.34 £36.84 £18.31 £61.39 £50.71 £30.95 £79.21 £21.69 £3.60 £43.71 2v1 13.03 10.95 12.30 10.16
3v1
b
£55.68 £35.19 £86.16 £47.61 £28.06 £75.50 £85.02 £60.70 £124.02 £32.81 £14.31 £57.41 3v1 13.02 12.08 16.23 10.94
3v2
c
−£1.13 − £19.96 £17.59 £10.77 − £7.30 £31.03 £34.29 £16.06 £58.55 £11.10 − £7.01 £31.36 3v2 9.49 9.70 10.81 9.71
2v1
3v1
3v2
2v1
3v1
3v2
2v1
3v1
3v2
£55.97 £35.28 £86.99 £15.98 − £2.63 £37.31 £40.88 £21.65 £67.44 £27.06 £8.44 £50.80 2v1 13.22 10.12 11.65 10.75
£66.90 £44.57 £101.61 £23.61 £5.19 £46.09 £58.49 £37.29 £90.58 £30.29 £11.58 £54.95 3v1 14.59 10.38 13.63 11.01
£10.92 − £7.42 £31.60 £7.65 − £11.06 £27.63 £17.61 − £0.62 £39.31 £3.21 − £15.60 £22.82 3v2 9.88 9.79 10.14 9.72
£40.03 £21.92 £64.60 £29.41 £11.98 £51.41 £57.78 £38.16 £86.70 £20.31 £3.05 £40.89 2v1 10.87 10.01 12.37 9.59
£43.56 £25.20 £68.84 £17.25 £0.02 £37.00 £77.71 £55.32 £112.50 £27.99 £10.59 £50.04 3v1 11.11 9.38 14.60 10.01
£3.54 − £14.07 £21.59 −£12.14 − £31.23 £5.04 £19.93 £2.93 £40.08 £7.68 − £9.70 £26.38 3v2 9.03 9.19 9.42 9.14
£50.94 £38.94 £65.70 £27.49 £16.84 £39.56 £50.07 £38.20 £64.63 £23.14 £12.59 £34.95 2v1 6.82 5.78 6.73 5.69
£55.27 £42.90 £70.70 £29.43 £18.76 £41.70 £74.01 £60.03 £92.04 £30.31 £19.50 £42.84 3v1 7.08 5.84 8.17 5.95
£4.31 − £6.03 £14.92 £1.92 − £8.45 £12.42 £23.97 £13.51 £35.76 £7.17 − £3.18 £17.95 3v2 5.32 5.31 5.66 5.38
2v1: price for moving from status quo to second level. 3v1: price for moving from status quo to third level. 3v2: price for moving from second level to third level.
50
E CO L O G I CA L E CO N O MI CS 63 ( 20 0 7 ) 4 2–5 3
Table 6 – Effects of socio-economic and other factors on the choice experiment results: mean deviances and p-values Degrees of freedom
Attribute main effects Price Socio-economic variables: Gender Age (7 groups) Class (4 groups) Children in household (6 groups) Working (11 groups) Education (5 groups) Income (linear) Rural/urban
South
Central
Combined sample — main effects
North
Mean deviance
pvalue
Mean deviance
pvalue
Mean deviance
pvalue
Mean deviance
pvalue
8 1
20.49 46.89
b 0.001 b 0.001
13.95 44.79
b 0.001 b 0.001
16.41 51.32
b 0.001 b 0.001
48.63 142.12
b0.001 b0.001
9 54 27 45
1.66 0.92 1.08 1.05
0.093 0.637 0.35 0.382
1.02 1.3 0.89 0.85
0.419 0.067 0.623 0.713
0.5 1.31 0.62 1.14
0.876 0.061 0.937 0.243
0.98 0.85 0.68 1.05
0.454 0.773 0.889 0.385
90 36 9 9
1.82 1.58 1.09 1.07
b 0.001 0.014 0.362 0.382
0.83 0.92 2.36 0.94
0.849 0.599 0.012 0.493
1.8 1.09 2.21 0.81
b 0.001 0.332 0.019 0.606
1.33 1.1 2.86 0.91
0.025 0.307 0.002 0.515
pair of levels within attributes, 999,999 such ratios of level effects to the negative of the corresponding price coefficient were simulated and the upper and lower confidence limits derived from the distributions of simulated values. The implicit prices allow us to compare the relative preference for each of the attribute levels both within and between attributes. There is also additivity amongst the implicit prices. For example the difference in prices between moving from “current practices” to level 2 and from “current practices” to level 3 is the same as the difference between levels 2 and 3. For example, in the South region for the environment attribute, 2v1 = £56.81, 3v1 = £55.68 and 3v2 = − £1.13, i.e. o55:68−o56:81 ¼ −o1:13: It should be noted that the calculation of the implicit prices by dividing the attribute level comparison coefficients by the
inverse of the price coefficient inflates the errors inherent in those estimated coefficients. Consequently, more precise comparisons of relative preferences should be obtained directly from the estimated coefficients in Table 4. Overall, WTP was highest in the South with a mean implicit price across all attributes of £48.40 compared to £39.90 in the Central region and £39.26 in the North. For the combined sample the mean was £42.58. Analysis of the effects of socio-economic variables on choice indicated that these effects were an order of magnitude smaller when compared to the effect on choice of the attributes themselves. The mean deviances and associated p-values of these variables are presented in Table 6. One result of note was that whether respondents lived in rural or urban areas had no significant influence on preferences. A number of attitudinal questions were asked prior to the CE exercise, primarily to encourage respondents to begin
Table 7 – Interaction effects of attitudinal factors on attribute coefficients (t-statistics in italic) Factor “Favours rural development policy” Enhance wildlife habitats Enhance water quality Enhance landscape appearance Enhance public access Maintain farming communities Promote locally grown food Target social benefits Target environmental benefits Price
a b
“Negative impact if farming ceased”
“Positive attitude to the environment, landscape and access”
“Negative attitude to subsidies”
0.0323 0.66 0.0416 0.86 − 0.0263 − 0.54
−0.0074 − 0.2 −0.0203 − 0.54 0.0236 0.63
0.1148 a 2.47 0.1011 a 2.19 0.1074 a 2.34
0.0044 0.14 0.0332 1.06 −0.033 − 1.06
− 0.1038 a − 2.12 0.1113 a 2.31 0.2296 a 4.64 − 0.0294 − 0.61 − 0.079 − 1.63
0.019 0.51 0.0699 b 1.87 0.1504 a 3.97 0.0027 0.07 −0.0442 − 1.18
0.1649 a 3.57 0.0413 0.9 0.0621 1.34 0.0168 0.37 0.1163 a 2.54
0.019 0.61 − 0.0722 a − 2.33 − 0.1104 a − 3.48 0.023 0.74 0.0534 b 1.71
− 4.4E− 05 − 0.06
− 0.00142 a − 2.84
0.000206 0.27
Significant at the 5% level. Significant at the 10% level.
0.000803 1.29
51
EC O LO G I CA L E C O N O M I CS 6 3 ( 2 00 7 ) 4 2–5 3
thinking about the trade-offs inherent in agri-environment and rural development policy. We used a principal components analysis to find linear combinations of the attitudinal questions to identify factors that best summarise the variation in the attitudinal responses. Eight components were extracted from the data as explaining variation greater than expected by chance. Together, these accounted for 59% of the total variance of the attitudinal responses. After varimax rotation with Kaiser normalisation to align the space spanned by these components as closely as possible to the individual attitudinal questions, we derived eight attitudinal factors that can be broadly described as follows. 1. 2. 3. 4. 5. 6. 7. 8.
Favours rural development policy. Negative impact if farming ceased. Policy should help all aspects of rural areas. Positive attitude to the environment, landscape and access. Negative attitude to subsidies. Positive environmental impacts if farming ceased. Other industries would replace farming. Farmers do not deserve subsidies.
Initial analysis of the effects of these factors on choices made revealed that only two factors had a significant effect. An alternative approach was then taken using the respondents’ average score for each of the questions given high weight within each of the above eight factors. This resulted in a conceptually similar but simpler analysis, in which the initial examination of the mean deviances of each of the eight factors revealed that four factors had no significant effect; these were excluded from further analysis, although as factor analysis does not classify respondents there is a degree of overlap across the factor scores for each respondent. Those factors that had a significant effect on choice (1, 2, 4 and 5 above) are included in the results presented in Table 7. • Respondents who scored highly on factor 1, associated with a positive attitude to rural development, tended to have the strongest preferences for rural development policies, but the most negative feelings for enhanced public access. • Respondents who felt that there would be negative impacts if farming were to cease, factor 2, tended to have the strongest preferences for rural development. • Respondents with positive attitudes to environment and landscape, factor 4, tended to have the strongest preferences for the relevant attributes and also for the targeting of payments towards these benefits. • Respondents who scored highly on factor 5, associated with feelings that farm payments are too high and have negative impacts, tended to be least favourable towards rural development payments and have stronger preferences for targeting environmental benefits. Of the four factors, it is only this one that has a significant relationship with pricing: respondents that scored highly on factor 5 tended to have the most negative attitude towards paying more for the policy packages.
3.5.
AHP results
Table 8 presents the results for the AHP. Environment (wildlife habitats and water quality) was the highest weighted attri-
bute, followed by rural development and then landscape appearance and access. Quality weights were calculated from the within attribute comparisons. These were then multiplied with the attribute weights to determine overall weightings for each quality. The targeting of payment attribute was assessed separately from the other attributes, the purpose being to determine preferences for different targeting options for the other attributes. The weights were calculated on an individual level and then averaged to give a single score across the sample. This approach allows the use of bootstrapping techniques to estimate confidence intervals for the AHP weights and hence determine whether differences in weights can be considered to be statistically significant. Bootstrapping involves re-sampling with replacement from the observed data in order to build up a whole population of sample means, giving us a distribution-free measure of how much uncertainty there is about the actual sample mean. In this case we based calculations on the 146 respondents who gave complete responses to all the pairwise comparisons. From these Table 8 – AHP results, within attribute and overall weights and implied ranking of attribute levels (95% confidence intervals for attribute and overall quality weights in brackets) Attribute
Quality
Environment
Attribute Quality Overall Rank 0.452 (0.410– 0.504)
Improve wildlife habitats Improve water quality Landscape and access
Rural Development
3
0.511 (0.442– 0.584) 0.489 (0.425– 0.556)
0.072 (0.059– 0.087) 0.096 (0.073– 0.121)
6
0.519 (0.464– 0.586) 0.481 (0.415– 0.534)
0.198 (0.156– 0.237) 0.181 (0.143– 0.220)
2
1
5
0.380 (0.308– 0.437) Maintain farming communities Promote locally grown food
Targeting towards environmental and landscape benefits Targeting towards social benefits
0.196 (0.155– 0.233) 0.257 (0.217– 0.293)
0.168 (0.136– 0.199) Improve landscape appearance Improve public access
No targeting of payments
0.417 (0.355– 0.480) 0.583 (0.533– 0.635)
4
0.154 (0.128– 0.182) 0.380 (0.340– 0.421)
3
0.466 (0.425– 0.508)
1
2
52
E CO L O G I CA L E CO N O MI CS 63 ( 20 0 7 ) 4 2–5 3
the sample average could be calculated to give the point estimate of each weight. We then sampled with replacement from these 146 individuals to create another set of 146 weights. We then calculated the mean of this new sample. We repeated this a large number of times, here 9999, in order to build up a population of sample means. We then ordered the weights, quoting the 250th and 9750th ones as 95% confidence interval for the mean. This was being done simultaneously for all attributes/qualities. The confidence intervals for the attributes indicate that significant differences between the Landscape and Access attribute and the Environment and Rural Development attributes. This pattern is reflected for the qualities, with the overall weights for qualities within the Environment and Rural Development attributes forming a group where there are no significant differences between qualities, whilst the landscape and access qualities form a second group with significantly different and lower weights. As with the choice experiment, this does not imply that the general public have no preference for landscape and access attributes; it is just that they are lower relative to other possible policy attributes. An interesting result arises with respect to the targeting of farm payments, in that the targeting of social benefits has a higher and significantly different weight to the targeting of environmental and landscape benefits. This is contrary to both the choice experiment results and what might be inferred from the previous AHP results.
4.
Discussion
The results of the CE using the combined sample (Table 4) show that the policy options fell into four groups: Group 1 (Most preferred) promotion of locally grown food; Group 2 (Upper group) enhance the quality of lochs, rivers and wetlands, enhance wildlife habitats, maintaining farming communities; Group 3 (Lower group) targeting environmental and landscape benefits, enhance public access, enhance landscape appearance, targeting towards social benefits; Group 4 (Least preferred) status quo. Differences within these groups were never statistically significant at the 5% level, whereas differences between groups were all statistically significant. The implicit prices for the policy changes in groups 1, 2 and 3 vary in approximately the ratio 3:2:1 (see Table 5). There are two noteworthy conclusions to be drawn here. From the WTP evidence the first unsurprising conclusion is that the public has reasonably assigned greatest weight to public goods that can be linked to them in a direct use sense (i.e. food and water quality), whereas lesser weight appears to have been more abstract non-use attributes. Second is that both environmental and rural development aims are supported by the public relative to the status quo. This provides evidence in support of the current reconfiguration of agricultural policy and support towards both of these aims. While we do not necessarily expect the methods to converge in terms of the preference weights of features, we would expect preference ordering to be stable. However, the preference orderings resulting from the AHP differ from those of the CE, with the environment attribute being more favoured than rural development. Alternative explanations for the divergence can be found in a range of literature covering framing effects (Tversky and Kahneman, 1986), bounded rationality (Kahneman, 2002) or uncertainty and unfamiliarity with the goods
being valued (Hanemann and Kriström, 1995). The trade-offs between attributes being conducted in this study were not easy and are perhaps not as mutually exclusive as the design suggests. Nevertheless, at the level of policy discourse, taxpayers are commonly asked to make such stark trade-offs. More immediately, we can speculate that by simultaneously presenting respondents with all the attributes including a price constraint, CE offers a more realistic reflection of the trade-offs to be made, whereas the universe of trade-offs is narrowed down to pairwise comparisons in the AHP. At the policy level then, some consideration needs to be given to the dimensions of the choices at hand when choosing preference elicitation methods. While this difference may be an artefact of this study, our view is that CE and AHP are simply asking different questions that reflect different aspects of the reality of real world trade-offs. The inclusion of a monetary attribute in the CE variant arguably constrains choice behaviour within a private budget constraint. Put another way, the CE versus AHP divergence is perhaps the difference between what respondents privately would like to spend money on and what they would like to see provided collectively in the absence of any constraint on public choice. In policy terms, both response sets are legitimate and can inform different policy questions. A further a priori assumption was that the AHP would present a lower cognitive burden as respondents are faced with pairwise comparisons as opposed to the CE’s simultaneous assessment of different levels across five attributes. Following both the AHP or CE exercises respondents were asked to rate the degree of difficulty they experienced on a scale ranging from very easy to very difficult. This debriefing question found that 12% and 38% of AHP respondents found the task either very or fairly difficult compared to 9% and 34% respectively for the CE. Conversely, 10% and 35% of CE respondents found the task either very easy or fairly easy compared to 6% and 30% of AHP respondents. Overall then, respondents were divided over the demands placed on them by CE and AHP. A possible reason for the divergence of task difficulty is that the AHP involved a longer exercise, with 18 pairwise comparisons and importance ratings required compared to 6 choices in the CE.1 Further study is therefore required as CE exercises often vary in the burden presented. For example, between 4 and up to 16 choices may be required, which may offer more than the two alternatives used in this study. It might be the case that the CE approach is more intuitive as respondents, when acting as consumers, are familiar with making simultaneous trade-offs between the attributes of marketed goods. This higher cognitive burden associated with the AHP may result in greater respondent fatigue and lead to the adoption of a heuristic that will enable a swifter end to the survey. In this study the survey was undertaken as a pen and paper exercise in respondents’ homes. Care should therefore be taken in designing AHP exercises to reduce undue burden on respondents, 1
The original AHP exercise also involved comparisons of the attribute qualities with status quo levels. For the environmental qualities these were defined as being “protect” rather than “improve” for political reasons to reflect obligation on farmers not to damage the environment. However, “protect” was rated to be more important than “improve”, as might be expected, due to loss aversion, despite not representing a policy change.
EC O LO G I CA L E C O N O M I CS 6 3 ( 2 00 7 ) 4 2–5 3
for example through the use of computer interfaces. It might also be more appropriate to undertake such in-depth preference elicitation exercises in group settings such as those described by MacMillan et al. (2002), where a trade-off can be made between sample size and response quality. The AHP due to its lower sample size requirements has some potential in such settings that should be explored further.
Acknowledgement This review derives from the project ‘Beauty Beast and Biodiversity: What does the Public Want from Agriculture?’ A more detailed report can be located at: http://www.scotland.gov.uk/ Publications/2004/09/19892/42598. The project is funded by the Scottish Executive Environment and Rural Affairs Department. We would like to thank the inputs of a project steering group, and in particular Robert Henderson and Andrew Moxey. We would also like to thank two anonymous referees for comments on an earlier draft of this paper.
REFERENCES Brubaker, E.R., 2004. Eliciting the Public’s Budgetary Preferences: Insights from Contingent Valuation. Public Budgeting and Finance, pp. 72–95 (Spring). Clark, J., Burgess, J., Harrison, C.M., 2000. ‘I struggled with this money business’: respondents perspectives on contingent valuation. Ecological Economics 33, 45–62. DTLR, 2001. Multi-criteria analysis manual. Office of the Deputy Prime Minister, London, UK. Duke, J., Aull-Hyde, R., 2002. Identifying public preferences for land preservation using the analytic hierarchy process. Ecological Economics 42, 131–145. Duke, J.M., Ilvento, T.W., Hyde, R.A., 2002. Public support for land preservation: measuring relative preferences in Delaware. FREC Research Reports, FREC RR02-01. Department of Food and Resource Economics, University of Delaware. http:// www.udel.edu/FREC/PUBS/RR02-01.pdf. Hall, C., McVittie, A., Moran, D., 2004. What does the public want from agriculture and the countryside? A review of evidence and methods. Journal of Rural Studies 20 (2), 211–225. Hanemann, W.M., Kriström, B., 1995. Preference Uncertainty, Optimal Designs and Spikes. In: Johansson, Per-Olov, Ktriström, Bengt, Mäler, Karl-Göran (Eds.), Advances in Environmental Economics. Manchester University Press, England. Hanley, N., Macmillan, D.C., Wright, R.E., Bullock, C., Simpson, I., Parsisson, D., Crabtree, J.R., 1998. Contingent valuation versus
53
choice experiments: estimating the benefits of environmentally sensitive areas in Scotland. Journal of Agricultural Economics 49 (1), 1–15. Jacobs, M., 1997. Environmental valuation, deliberative democracy and public decision making institutions. In: Foster, J. (Ed.), Valuing Nature? Economics Ethics and Environment. Routledge, London. Kahneman, D., 2002. Maps of bounded rationality: a perspective on intuitive judgement and choice. Nobel Prize Lecture. http:// nobelprize.org/economics/laureates/2002/kahnemann-lecture. pdf. Kenyon, W., Hanley, N., Nevin, C., 2001. Citizens’ juries: an aid toenvironmental valuation? Environmental Planning C. Government and Policy 19 (4), 557–566. Louviere, J., Hensher, D., Swait, J., 2000. Stated choice methods: Analysis and application. Cambridge University Press, Cambridge, UK. MacMillan, D., Philip, L., Hanley, N., Alvarez-Farizo, B., 2002. Valuing non-market benefits of wild goose conservation: a comparison of interview and group-based approaches. Ecological Economics 43, 49–59. McDaniels, T., Gregory, R., Arvai, J., Chuenpagdee, R., 2003. Decision structuring to alleviate embedding in environmental valuation. Ecological Economics 46, 33–46. Munda, G., 1996. Cost–benefit analysis in integrated environmental assessment: some methodological issues. Ecological Economics 19 (2), 157–168. Rosenberger, R., Peterson, G., Clarke, A., Brown, T., 2001. Dispositions for Lexicographic Preferences of Environmental Goods: Integrating Economics. Psychology, and Ethics RESEARCH PAPER, vol. 2001-12. http://www.rri.wvu.edu/pdffiles/ rosenberger2001-12.pdf. Scottish Executive, 2001. A forward strategy for Scottish agriculture. Scottish Executive Environment and Rural Affairs Department, Edinburgh, UK. Scottish Household Survey, 2002. http://www.scotland.gov.uk/ Topics/Statistics/16002/4049. Spash, C.L., 1998. Investigating individual motives for environmental action: Lexicographic preferences, beliefs and attitudes. In: Lemons, J., Westra, L., Goodland, R. (Eds.), Ecological Sustainability and Integrity: Concepts and Approaches. Kluwer Academic Publishers, Netherlands, pp. 46–62. Swait, J., Adamowicz, W., 2001. The influence of task complexity on consumer choice: a latent class model of decision strategy switching. Journal of Consumer Research 28 (1), 135–148. Toman, M., 1998. Sustainable Decision making: The State of the Art from an Economics Perspective. Discussion Paper, vol. 98-39. Resources for the Future, Washington D.C. http://www.rff.org/ Documents/RFF-DP-98-39.pdf. Tversky, A., Kahneman, D., 1986. Rational choice and the framing of decisions. Journal of Business 59 S251-0S278.
