Valuing public goods of the agricultural landscape: a choice experiment using reference points to capture observable heterogeneity
评估农业景观的公共产品：一个利用参考点捕捉可观察异质性的选择实验

2.1 A large-scale mail survey

This CE is designed to estimate the marginal values of 8 grassland types, 10 types of field elements and 9 agri-environmental quality attributes. In total, the MWTP for 28 attribute levels were estimated. The attributes were designed to be as policy relevant as possible.

The first phase of the study involved literature surveys, interviews and meetings to design and select the most policy relevant landscape attributes and their choice levels. Discussions with scientists of related disciplines and policy implementers at the responsible regional or national boards Footnote2 generated a list of candidate attributes. Then, by iterative ranking a final set of attributes was agreed upon by the research team in collaboration with the consulted experts.

An aim of the study was to develop a set of attributes covering the entire agricultural landscape in terms of object types and major quality aspects. Attributes were designed to cover all TEV-motive types Footnote3 to avoid the problems of partial analysis. To be as adaptable to varying site conditions and policy relevant as possible, the attributes were defined in terms of public goods, not in terms of any production technology.

Since the policy measures and the type of environmental services and qualities differ between permanent grasslands and field elements, the CE was divided into two separate surveys. The attributes and their levels are presented in Table 1 and described more in detail below.

The survey about permanent grassland was divided into two versions. Each version concerned five common attributes out of a total of seven. Five types of grasslands (open grazing land, shore meadow, oak-wooded pasture, birch-wooded pasture and dry meadows with junipers) were included. A binary experimental design was adopted for these objects as there was no prior consensus on which type to use as a baseline. In the permanent grassland surveys, levels of the environmental quality attributes were stated in absolute terms (e.g. given amount of brushwood, or given number of red-listed species).

The field element survey was also divided into two versions with, respectively, one unique attribute and three attributes that were common to both versions. Version 1 had type of linear field elements as a design specific attribute, while the version 2 had type of point elements as a unique attribute. The three common attributes (biodiversity, cultural heritage and visibility characters) were each stated at two relative levels of magnitude (i.e. more or less). The use of relative levels was viewed as more intuitive by our focus groups than the use of levels expressed by percentages or absolute numbers. Applying this attribute design approach, the biologists and policy implementers may then interpret what corresponds to ‘More biodiversity’ with respect to all variables of species composition, landscape context, region, etc.

Every respondent received an introductory letter, a fact folder and a questionnaire. In the introductory letter the respondents were briefly presented with the background and aims of the study from environmental, policy and fiscal perspectives, as well as matters of confidentiality. The fact folder presented first attributes and the respective choice levels by short texts and photos, with wording and motives as neutral and balanced as possible. The questionnaires had three sections. The first had questions about the respondent's relationship to the agricultural landscape, the second section had the CE, and the third had questions about the respondent's socio-economic status. Throughout the questionnaire, the respondents were reminded they had to consider their budget constraints as the choices involved increased income taxes. To reduce the probability of a hypothetical bias, we adopted the cheap-talk script from Carlsson et al. (Citation2005). The choice sets were created simultaneously in that the design was selected from the collective factorial (Louviere Citation1988). From the full collective factorial, we used the D-optimal linear design routine in SAS (Kuhfeld Citation2001) to select 36 choice sets. These were then randomly blocked into four versions, each containing nine choice sets. Footnote4 Each choice set had two alternatives. The CE did not include any opt-out alternative, that is, a possibility to choose ‘Neither alternative 1 nor 2’ or to choose a ‘Don't Know’ (DK) option, in the choice sets. However, since we were primarily interested in estimating the marginal WTP of the attributes, this would be an appropriate design. Footnote5

There are several potential disadvantages of including an opt-out or DK alternative. Even though it is argued that the use of opt-out or DK alternatives will recognise respondent behaviour or respondent uncertainty it can, on the other hand, be used as ‘an easy way out’ by respondents facing a complex and difficult choice task or a choice set that does not contain enough distinct options. A similar type of escaping by the opt-out alternative could possibly be expected in a CE where respondents exhibit lexicographic preferences. This is possible if, for example, respondents have preferences for levels of landscape attributes that are higher than those included in the choice set. A further problem is that respondents who would choose an opt-out alternative may be different from those making trade-offs between attributes, in the sense that they make different trade-offs on the margin. If this was the case, then an opt-out option would indeed affect the marginal willingness to pay as well.

The CE was carried out by mail surveys following the standard procedure with testing by focus groups and a pilot study of 500 randomly selected inhabitants in Sweden. All questionnaires were sent out in the spring 2008 to a sample of 8000 men and women, aged 18–75, randomly selected from the Swedish census registry. Two reminders were mailed at two-week intervals to those who did not reply. For the surveys about field elements of arable land, a total of 1163 respondents (29%) returned valid questionnaires. For the meadow and pastureland surveys, the response rate was 1474 respondents (37%). Footnote6

2.2 Reference points capturing observable respondent heterogeneity

In our econometric models, we separate the respondents who are accustomed to certain states of the agricultural landscape corresponding to attribute levels of the experiment, from those who are not, using dummy variable interactions. For example, some respondents may often visit grazing land with much brushwood. We then include this information into the econometric model and distinguish these respondents from other respondents. Their marginal utility of reducing brushwood is estimated separately.

This way of treating the respondents is partly influenced by the prospect theory of Kahneman and Tversky (Citation1979) that highlighted a body of evidence that strongly supported the existence of reference-dependent utilities. Footnote7 Reference dependency implies that the value that somebody assigns an attribute is not based on its absolute level, but rather on its deviation from a reference level. It leads to a sharp contrast between conventional and reference-dependent utility theory. The reference point might, for example, depend on past consumption, expectations or status quo. Unlike the conventional consumer theory, reference-dependent consumer theory also allows an individual to be loss averse, i.e. the value (utility) function is steeper for losses than gains. Moreover, a person may prefer attribute 1 to attribute 2 if she is endowed with 1, but may prefer 2 if endowed with 2. In another form, the value of a good to an individual increases when the good is in the individual's possession (endowment effect). This also explains features such as opt-out bias, i.e. a strong disposition for the status quo alternative. Footnote8 In a recent CE study, Hu et al. (Citation2006) found that reference dependency is important to understand consumers' choice of bread products, particularly with regard to understanding the effect of prices.

A challenge in modelling reference dependence is the identification of relevant reference points for each respondent. In our application, the reference point will be defined by the attribute and levels that are present in the agricultural landscape that the respondent most often visits, i.e. features with which the respondent is familiar. These attributes can be argued to serve as a natural point of comparison for the respondents (Samuelson and Zeckhauser Citation1988, Bruce et al. Citation1993). Footnote9 The use of a system of reference points based on familiarity with the subject to be valued should also mitigate the use of recognition heuristics, i.e. that respondents make inferences from patterns of missing knowledge (Goldstein and Gigerenzer Citation2002). We do not, of course, exclude other possibilities than that the respondents make their choices in the CE based on the attributes that they are accustomed to and value the alternatives based on deviation from their reference levels. They may visit different agricultural landscapes because of different preferences, although it has nothing to do with reference-dependency. Both of these interpretations are plausible.

To conclude, there are a number of reasons for including reference points in the econometric model. First, any way to decrease the magnitude of the noise term in the econometric model will in turn improve the statistical properties of the model. Second, including reference points could help to better understand the choices of the respondents, in terms of explanations given in the reference-dependent consumer theory supplementing conventional consumer theory literature.

2.3 Econometric model

In the econometric analysis we apply a random utility model. The utility consists of a systematic component (V_njt ) and a random component (ϵ _njt ):

The model is estimated with simulated maximum likelihood using Halton draws with 500 replications using Nlogit 4.0 (see Train Citation2003 for details on simulated maximum likelihood and Halton draws).

3.1 Values of field elements in the cultivated landscape

Table 2 presents the estimated models for linear elements and point elements, respectively. The alternative specific coefficients (ASC) are positive and significant in both versions, meaning that respondents are more likely to choose alternative 1. The ASC captures systematic but unobserved effects of choosing alternative 1; however, as this is a generic CE we do not expect this effect to be extensive. Since all the other attributes, besides one, are coded as dummy variables, we can directly relate the magnitude of the intercept to the dummy coded attributes. Although the ASC are statistically significant, they are small compared to the other coefficients; hence, they do not drive the model results.

The coefficients for the design specific attributes are significant, except for Headland, Field islet and Pollard, with expected signs. Furthermore, the value motive attributes (biodiversity, cultural heritage and visual characteristics) have significant coefficients of the expected positive signs. The cost coefficient is negative and significant, as expected. The interaction terms of the cost coefficient were rejected for our data, indicating that the marginal utility of income is constant across income groups. Most of the estimated standard deviations of the random parameters are significant. The significance is a sign of heterogeneity in preferences among the respondents.

A way to make the estimates of the two survey versions comparable is by estimating the WTPs of the attributes. The MWTP of an attribute is obtained by dividing its estimated coefficient by the cost coefficient (Hanemann Citation1984). Hence, the MWTP is simply the marginal rate of substitution between one of the attributes and the cost attribute. The reference point effect is included when calculating average MWTP according to the sample share of the respondents that had identified reference points. The MWTP for the attributes in the field elements surveys are presented in Table 3. Most of the standard deviations are sizeable, compared to the attribute coefficients. This results in rather wide distributions of the MWTPs, as demonstrated below.

The ranking is in accordance with experts' judgements (see e.g. SBA Citation2008), but this study also contributes with monetary estimates of the differences in value. Stone walls are the highest valued type of linear field elements. Their MWTP is 240 SEK per year, meaning that people on average are willing to pay this extra tax amount to get a landscape with this linear element. Field roads are MWTP-valued to 168 SEK/y and traditional wooden fences to 147 SEK/y. Lowest values are assigned to ditches, whose MWTP is 101 SEK/y, and headlands, whose MWTP is 54 SEK/y. Ditches may, besides giving variation to landscape scenery and providing biodiversity or historic qualities, be associated with wetlands draining and the replacement of natural rivulets, which may explain their relative low MWTP value. Headlands are mere strips of vegetation along the fields, and thus provide fewer environmental services than the other linear elements, which always have such strips along their other features. The ranking of linear element values is identical to the CVM preference ordering by Hasund (Citation1998), except that the respondents of that study valued ditches higher than field roads.

Linear elements with more biodiversity are worth 390 SEK per year more (in terms of MWTP) than elements with little biodiversity. That biodiversity is the policy objective most highly valued agrees with earlier landscape valuation studies (Drake Citation1992, Hasund Citation1998). ‘More visible’ linear elements are worth (MWTP) 317 SEK per year more than less visible elements. Swedish residents apparently value aesthetic and other visual characters. This latter result is striking and obviously contrasts with valuations made by influential ecologists and persons involved in the policy process, who have disregarded visual landscape characters in favour of biodiversity and to some extent cultural heritage qualities; see Dramstad et al. (Citation2006), SBA (Citation2006), MA (Citation2007a), Sandström and Klang (Citation2007). The MWTP for linear elements that have more cultural heritage qualities is 190 SEK per year, which is considerable compared to the present agri-environmental payments.

With regard to types of point field elements, there are fairly small differences of the estimates. The MWTP values range between 120–180 SEK/y for the four most common Swedish types, which is considerably higher than the present payments. The result should be interesting in a policy context. Notably, however, pollards have a low but negative and not significant MWTP value. This result may be either because most people are unfamiliar with pollards as they are now less common, or that a pollard is small and gives fewer environmental services. The MWTP for the quality attributes of point elements are similar to those of the linear elements.

In general, the respondents have a higher MWTP for the environmental quality attributes than for the element type attributes. This implies that people are willing to pay to preserve field elements, and value elements having more biodiversity, visual or cultural heritage interest considerably higher than those lacking these attributes.

The confidence intervals of the MWTPs presented in Table 3 are wide. They are calculated by the method outlined by Hensher and Greene (Citation2003), where both the mean and standard deviations from the RPL model are incorporated. Randomness across individuals as well the randomness introduced by the fact that both the central tendency and the dispersion of each parameter are themselves estimated quantities, and therefore subject to noise, are included. Substantial preference heterogeneity is revealed where some persons like and other dislike an attribute.

3.2 Values of permanent grassland

Table 4 presents the estimated models for the surveys related to permanent grasslands. The ASC are insignificant in both of the grassland survey versions. All the attributes have the expected signs and the estimates are all significant, except for the attributes ‘Shore meadow’ (in both survey versions), ‘Oak-wooded pasture’ (one survey version) and ‘Landscape context’ (one survey version). About half of the estimated standard deviations of the random parameters are significant, although there is less evidence of heterogeneity in preferences for these models than for the field element models. Furthermore, the standard deviations of many attributes are sizeable compared to the attribute coefficients. This may indicate considerable preference heterogeneity.

The cost coefficient is negative and significant as expected. With regard to the field elements model, the marginal utility of income seems to be constant across income groups in survey version 2. However, survey version 1 shows a decreased level of disutility with increased income, although the magnitude of this effect is quite small.

Table 5 displays the MWTP values obtained from the estimated grassland models. The marginal utility of the different types of grassland are estimated compared to the grazing land type ‘Dry meadow with junipers’, since it was the lowest valued in the surveys. The types of meadow or pasture are valued differently. Oak-wooded grassland is the most highly valued and has a mean MWTP in the range 120–340 SEK per year, depending on the valuation context. Grasslands with scattered oaks are symbols of a rich landscape, often with a high biodiversity and in many contexts recognised for their aesthetic or recreational qualities. The two survey versions consistently rank Oak-grown pasture at the top, followed by Open grazing land, Birch-grown pasture and Shore meadow lowest. Shore meadows have relatively low values – the MWTP ranges from 30–100 SEK/y. The result deviates from the general valuation of ecologists (see e.g. SBA Citation2008), particularly ornithologists, and the implicit valuation by policy measures (MA Citation2007a, Citation2007b), which assigns high values to many shore meadows.

The three land use types ‘Traditional mowed meadow’, ‘Semi-natural pasture’ and ‘Cultivated pasture’ are different with respect to biodiversity, historicity and aesthetic qualities, besides the differences following grassland type, discussed above. Cultivated grasslands have been fertilised or sprayed with pesticides at least one time, which reduce their biodiversity, historicity and the scenic amenities of flowers for many decades or more. The MWTP of semi-natural pastures and traditional mowed hay meadows is, respectively, 224 SEK/y and 89 SEK/y higher than that of cultivated grassland.

An environmental status attribute that is relatively highly valued is whether the grassland is maintained either by high grazing pressure or by mowing with removal of the cut grass. Much grazing or mowing gives a low grass sward and no accumulation of organic litter, which favours biodiversity, the site's historicity, recreational accessibility and aesthetic features. If the pasture is not grazed, the ground vegetation becomes dominated by a few trivial species of high grass and herbs, organic litter accumulates, so that biodiversity, cultural heritage and aesthetic qualities are affected and the object becomes less traversable for visitors. The MWTP of ‘Some grazing’ is estimated at 132 SEK/y, relative to the status of ‘Not grazed or mowed’). ‘Much grazed or mowed’ is valued by an additional 133 SEK/y in MWTP.

Valued even more highly is the grassland property of not being covered by brushwood or thicket. The MWTP for meadows or pastures with 10% of its area covered is 172 SEK/y relative to the WTP for grasslands with more than 50% of its area covered, and with no brushwood or thicket an additional 114 SEK/y. Brushwood and thickets are signs of abandonment or long periods of little management, which may be symbolic to persons who object to the decrease of agricultural land and landscapes. They reduce visibility, recreational accessibility, biodiversity, historicity and aesthetic qualities.

Grassland objects vary considerably with regard to biodiversity. One measure of the biodiversity is the number of red-listed species an object accommodates. Grassland with three red-listed species represents objects with high biodiversity, but not exceptionally high. Its MWTP value is 179 SEK/y, compared to the value of grassland without red-listed species. The MWTP is 194 SEK/y for grassland with ‘six red-listed species’, representing objects of biodiversity top class.

Of highest interest in the preservation and policy debate recent years in Sweden is the ‘landscape perspective’ issue; see e.g. MA (2003) or Miljömålsrådet (2008). This issue refers to the value of an object in the context of its size and surrounding landscape. Ecologists claim that many pastures have become too small or isolated, so that several of their species do not exist in long-term sustainable populations (Young and Clarke Citation2000). For other species, the ‘last’ remaining meadow or pasture may be a spreading point in the landscape (Öckinger and Smith Citation2007). For cultural heritage and recreation, large objects in ‘traditional’ landscapes with many similar objects are considered to be generally more interesting (SNHB Citation1999). The MWTP value of ‘large grassland objects surrounded by other similar grounds’ is 103 SEK/y higher than that of ‘small and isolated grassland objects’.

Another intensively debated policy issue concerns how to allocate the scarce public means. It has been questioned both to direct the support to remote meadow and pasture objects having a rich flora or fauna but hardly any visitors, as well as the support to much visited grasslands close to cities which are more or less void of biodiversity or historic interest. Relating the issue to TEV-components, it concerns use and non-use values, which are additive. To our knowledge, the MWTP of use values assigned to grassland objects have not been estimated in other studies. In this study, one of the designed choice attributes was ‘How much you experience the grassland object’. The MWTP for ‘Grassland that you will be in contact with only rarely’ is 96 SEK/y, whereas the alternative ‘Meadows or pastures that you often visit or see’ have a MWTP of 235 SEK/y/p. The baseline in either case was ‘Grasslands that you will probably never see’. If aggregating over the population, the high use-values do not necessarily mean that little visited grasslands with high biological or cultural heritage qualities have a lower total value, since the number of persons who assign non-values to it may be large.

In general, the wide confidence intervals in Table 5 of the MWTPs are striking and a sign of substantial preference heterogeneity.

3.3 Reference points – familiarity affects preferences

Parameter estimates for the interactions between attribute levels and reference points are reported in the last sections of Tables 2 and 4, respectively. In the permanent grassland survey, we distinguish between respondents that did answer the questions about reference points considering grazing and brushwood from those that did not (i.e. marked with No Ref. in Table 4), since not all of the respondents responded to these reference point questions. Our findings suggest that the reference-dependent utility theory appears to explain some of the heterogeneity in the preferences for agricultural landscapes, or, such explanations can at least not be rejected. In total, approximately half of all the reference point effects were significant in the econometric models. The results indicate that the preference for a specific attribute level is positively related to a prior reference with respect to a specific characteristic of the attribute, given that the estimate of the reference interaction is significant.

In the field element surveys, all of the coefficients of the interactions between the attributes and the reference points are positive; however, not all of these effects are significant. There is a significant reference point effect for the attributes Wood fence, Stone wall, Headland, Field road, Field islet and Pond. In short, people who are familiar with these elements are more likely to appreciate them than others.

A result from the grassland surveys is that the marginal disutility of brushwood is significantly lower for respondents who often visit pastures with the highest level of brushwood. Similarly, those who often visit pastures with high grass vegetation find them less inferior than other persons do, and will in general be willing to pay less for more intensively grazed pastures than other respondents. On the contrary, persons who are accustomed to meadows and pastures with low grass or without brushwood are not willing to pay more than other persons to get this environmental status instead of grassland with some high grass or 10% of the area covered by brushwood.

The marginal WTPs are for some of the grassland types driven by reference points, but for other types this is not the case. Respondents who often visit shore meadows or birch-wooded pastures, for example, are willing to pay for them, while the WTP of the remaining respondents is insignificant.

One interpretation of our findings is that persons who do not care about some landscape characters as much as other persons are more apt to visit such landscapes, for example, pastures having much brushwood. Even if this were true for some of the respondents, it should also be borne in mind that people's choice of what type of agricultural landscape to visit can be based on many other factors, including distance from home, accessibility, recommendations of friends etc. Consequently, an alternative interpretation would be that respondents find brushwood less bad as they become surrounded by and accustomed to it. It may then serve as a reference-point to them. While the former explanation can be interpreted in the light of conventional utility theory, the latter leans more to reference-dependent utility theory where individuals evaluate outcomes based on departures from a reference point. As it is not possible to distinguish reference-dependency from conventional preference heterogeneity in this case, we should be open to both of these interpretations.

3.4 Social values versus actual payment tariffs

Our CE-attributes have been designed to be applicable as criteria for agri-environmental payments. It might then be interesting to compare the estimated values, expressing the marginal values to society of the attributes, with the actual, present grants. It shows, for example, that traditional wooden fences are currently remunerated higher than headlands, by 60 and 6 SEK per 10 metres, respectively. The former is likewise more highly valued by the Swedish population, with an estimated MWTP 150 SEK compared to 54 SEK for headlands. The figures for the case of point field elements are listed in Table 6. Comparing the figures reveals that the actual payments are little correlated with the marginal WTPs of the point elements. Consider also that the payments are conditional Footnote11 and uniform for each element type. Contrastingly, the MWTP vary across the types of point elements (and a lot more across types of linear elements or grassland). Furthermore, their social values are considerably different also within each type, since a single object may have more or less biodiversity, cultural heritage or visibility qualities. If highly endowed in all these respects, a certain point element object is in the extreme valued 830 SEK more than another element of the same type.

The divergence between actual payments and the marginal WTPs estimated for grasslands is also striking if making similar comparisons. A pasture with, for example, 10% of its area covered by brushwood or thickets is worth 170 SEK less than if it had not, but will get 1100 SEK/ha less in the existing AEP-scheme, i.e. 0 SEK. On the other hand, a semi-natural pasture that is in the vicinity of some housing area and has a lot of visitors gets the same payment as a cultivated pasture in a remote area where just a few persons enjoy it, although our estimations indicate that its marginal value to society is about 560 SEK higher, ceteris paribus. It is evident that the differentiations of the actual Swedish AEP-schemes in some cases are too high and in others too low relative to the estimated MWTP.