2.1. Theoretical background: choice experiment model

Choice experiments (CE) have their origins in conjoint analysis, a method that has been widely used to measure individual’s preferences in the areas of economics, transport, and geography (e.g. Green 1978; Ben-Akiva and Lerman 1985; Louviere 1988; Swait and Adamowicz 2001). Since the mid-1990s, CE have also been applied and discussed in environmental economics (e.g. Morrison et al. 1996; Hanley et al. 1998a). Valuation studies of non-market goods – for instance, the ecological value of a watershed – that employed CE are increasingly found in the economic literature (Hensher et al. 2006; Hanley et al. 2007; Peterson et al. 2007) and there has been a lively scholarly debate regarding the advantages and limitations of CEs in comparison to other environmental valuation methods, such as Contingent Valuation Methods (CVM) (Boxall et al. 1996; Adamowicz et al. 1997; Adamowicz et al. 1998; Hanley et al. 1998b; Swait and Adamowicz 2001; Bateman et al. 2002; DeShazo and Fermo 2002; Rolfe et al. 2002; Rosenberger et al. 2003). Proponents of CE claim that elicitation of the willingness-to-pay and/or willingness-to-accept is more subtle in choice experiments than in a CVM interview and, as a consequence, the danger of strategic answers and protest bids by respondents is reduced (Ahlheim and Neef 2006; Yabe and Yoshida 2006). The use of CE helps in eliciting trade-offs that individuals make when choosing among alternative options or attributes. Through determining the amount of money that respondents would be willing to pay to effectuate a proposed change, it is possible to compute the marginal value of changes in each attribute. In many situations, such an approach may provide more realistic outcomes than focusing on a single change in the provision of the whole good which is a characteristic of most CVM studies (Mogas et al. 2006). With its emphasis on the combination of various attributes and levels, CE can be used for designing multidimensional policies, for conducting a cost–benefit analysis of such policy measures and for supporting conflict resolution and negotiations over the protection and use of non-market goods (Bateman et al. 2002; Mogas et al. 2006; Hanley et al. 2007).

CE can be applied to generate estimates of compensating surpluses for different attribute levels. Choice experimentation is based on random utility theory in which an indirect utility function is comprised of a deterministic element (V_i) containing the attributes of the situation, and an error term (ε_i). The model can be represented by the equation (1):

where X_i is a vector of attributes of the option and β is a vector of coefficients that reflect the relative importance of the attributes. Selecting one situation over another implies that the utility gained from the first situation (U_i) is greater than the utility gained from an alternative (U_j). The probability of choosing alternative i over j is represented by the form:

The error term is assumed to be independently and identically distributed (IID) with a type I extreme value distribution, a so-called Gumbel distribution, then equation (3) applies. The scale parameter λ is typically assumed to equal one and is inversely proportional to the variance of the error term.

An estimated linear-in-parameters utility function for alternative i then comes in the form of equation (4):

To obtain welfare estimates or compensation variation (CV) in choice experiment studies, the following general formula described by Hanemann (1984) is employed:

where μ is the marginal utility of income, C is the choice set and V_i0 and V_i1 represent the indirectly observable utility before and after the optional changes. When the choice set includes only one ‘before’ and ‘after’ alternative, equation (6) applies:

From equation (6) it can be concluded that the marginal rate of substitution between two attributes is the ratio of their coefficients (Hensher and Johnson 1981). The marginal willingness to pay (MWTP) for a change in attribute is then provided by equation (7).

2.2. Study area

The Mae Sa watershed, Mae Rim district, Chiang Mai province covers an area of 142.2 km² and extends from 20 to 45 km northwest of the northern city Chiang Mai (Figure 1). The watershed is intensively used for market-oriented agriculture, mainly fruit, flower and vegetable production. 80 percent of the total agricultural area of 1086 ha is under irrigation (Schreinemachers et al. 2008). The population is composed of northern Thai (khon muang) and the Hmong ethnic minority group. Of the total 3046 households in the watershed, 1309 (43%) are engaged in agriculture on an average landholding of around 1 ha (Schreinemachers et al. 2008).

Major parts of the watershed are included in the Doi Suthep-Pui National Park, where agricultural activities are considered illegal under Thai environmental legislation. The watershed has been part of a pilot project of the Thai government to introduce river basin committees and sub-basin working groups to enhance public participation in water governance (Heyd and Neef 2006; Neef 2008). The Mae Sa Watershed Management Working Group is the first of its kind in Thailand. Recently, a sub-committee of the working group decided to establish the so-called “Mae Sa Watershed Conservation and Development Fund” with the objective of funding selected resource management projects that would improve watershed functions and services. The pilot project can thus serve as an example how such funds are established and how they can be further developed into viable compensation schemes for environmental services and – more specifically – for enhancing adoption of soil and water conservation practices by upland farmers.

2.3. CE design, sample selection, and model descriptions

This study applied the CM technique to examine upstream farmers’ preferences for different agricultural conservation measures, and downstream farmers’ preferences for various schemes of water resource improvement (Figure 2). The choice task required respondents to choose one outcome from a set of several possible alternatives (known as a ‘choice set’). Upstream respondents were presented with a series of 12 choice sets, while downstream farmers faced ten choice sets. Each choice set had three options, including a ‘no change’ option that characterized the status quo and two change options. The change options were generated by varying the levels of each attribute according to an experimental design.

Due to their long-term research collaboration with local stakeholders in the watershed, the research team had gained the trust of the local population which helped to minimize strategic responses and political correctness in stating their willingness to pay/to accept. Respondent fatigue was controlled through small compensation in kind for the participants’ opportunity costs of time.

The CE study involved a number of methodological steps. The first step was to select attributes and set their levels. Attributes in this CE study were based on both supply- and demand-driven approaches. Under the supply-driven approach, attributes were derived from what policy-makers and/or researchers perceive to be factors that can be influenced by policy measures. The demand-driven approach involved selecting the attributes that individual or groups of upstream farmers consider important when making choices. In this study, four attributes were selected including installing water saving technology, planting vetiver (Vetiveria zizanoides) grass strips for erosion control, applying bio-insecticides and compensation or subsidy. The levels of each attribute were designed based on the costs involved. In each upstream sample village focus group interviews with village leaders and key informants in the field of water resource management were conducted in two rounds. First, the selected attributes and levels were discussed as regards their likelihood to be put into practice. Among various types of compensation, cash compensation was the most preferred option. The second round of discussions provided the final sets of attributes and levels that were used for designing choice sets for upstream farm households’ decision making. Each attribute consists of four levels as shown in Table 1 below. Orthogonal design using SPSS provided choices which were then used for designing 12 choice sets. Each choice set contains three alternatives, two alternatives were obtained from the designed choices and the other – the so-called status quo – refers to the ‘no change’ alternative, i.e. maintaining the existing utility of the current practice and receiving no compensation (see example in Figure 3). The randomly selected 371 farm households from seven villages with a total population of 776 households located in the upstream part of the Mae Sa Watershed were asked to state their preferences on willingness to accept compensation according to the 12 choice sets.

In the case of the downstream area of the watershed, the same procedures as described for the upstream villages were applied to obtain attributes and levels of water resource improvement as shown in Table 2. Choice sets were constructed with three alternatives, namely two alternatives obtained from the designed choices and the status quo (see example in Figure 4). 151 farm households were randomly selected from the total 198 farm households that relied on water from the Mae Sa River for both irrigation purposes and for household consumption. Each respondent faced 10 choice sets.

This study used a simplified version of CE, which is constructed solely on the basis of non-price and price attributes of designed choices. Conditional logit models were employed to estimate implicit prices and willingness-to-pay/willingness-to-accept compensation as depicted in the appendix. In order to investigate the influence of socio-economic factors on the respondents’ preferences a logit model was developed. The marginal effect of each socio-economic factor could be estimated, providing the magnitude of the probability to choose improved alternatives over the status quo.

3.1. Upstream resource managers’ willingness to accept compensation for downstream-friendly practices

The major socio-economic characteristics of the upstream farm households in our sample are depicted in Table 3. The most important crops are vegetables, cut-flowers and fruit trees, mainly lychee. Farm sizes of northern Thai farmers are much smaller than those of their Hmong peers, but they also have less family members to feed. Average per capita income in Hmong communities is generally lower than that of the northern Thai communities, the latter being engaged in more intensive farming systems, often in greenhouses.

Table 4 shows the conditional logit model results with respect to upstream farmers’ stated willingness to accept compensation for switching to more environmentally friendly practices. The signs of the parameters are consistent with theory and a priori expectations. The negative coefficients for applying bio-insecticides and planting vetiver grass suggest that these attributes contribute negatively to utility and therefore need to be compensated for. Marginal willingness to accept was estimated by dividing the attribute coefficient with the price coefficient. The figures show that the upstream farmers would need to be compensated at the rate of about 100 Baht per ha and year for applying bio-insecticides on 1% of their agricultural area. For a 1% increase of area planted with vetiver grass strips upstream farmers would need a compensation of more than 400 Baht per ha and year. These results suggest that the willingness to adopt vetiver grass under a compensation scheme is lower than the willingness to adopt bio-insecticides, reflecting farmers’ scepticism vis-à-vis soil conservation measures that reduce land available for profitable crops.

The coefficients for installing micro-sprinkler systems on 50% of the area and for installing drip irrigation on 50% and 100% of the area respectively have a positive sign which means that farmers do not need any compensation for adopting these practices. These results indicate that upstream farmers are increasingly aware of the need to switch to water-saving irrigation technologies as competition among water users within the same community and upstream-downstream conflicts between neighbouring communities have dramatically increased in recent years (Neef et al. 2006; Becu et al. 2008). In focus group discussions, upstream farmers stated their willingness to invest in such water-saving technologies, if they were assured that they would have sufficient access to water around the year.

Table 5 depicts the regression results of the logit model employed to understand the socio-economic factors that influence upstream farmers’ willingness to engage in a compensation scheme supporting the adoption of environment-friendly agricultural practices.

Age and education played a significant and positive role in the willingness to adopt environmentally friendly practices under a compensation scheme. Households with a high number of family members (mainly belonging to the Hmong ethnic group, cf. Table 3 above) engaged in farming and with vegetable production as the main activity were also more likely to adopt such practices. Another positive influence on willingness to adopt stemmed from previous experiences with on-site soil erosion, water shortages in the household and drought. If farmers were member of an environmental conservation group they also had a higher probability to engage in such compensation schemes. On the other hand, household heads who perceived themselves as rich or medium wealthy were less willing to adopt environmentally friendly practices than their poorer peers.

The fact that poorer resource managers had a higher willingness to engage in a compensation scheme for providing better ecological services can be attributed to two reasons: first, poorer farmers usually suffer from low tenure security and adopting soil and water conservation measures improves their reputation among government officials, thus reducing the risk of losing their land rights (cf. Neef et al. 2006). Second, poorer households regard a viable compensation scheme as a way to generate a secure stream of cash income and to reduce their general livelihood risks, while more affluent farm households have accumulated their wealth through intensification of agriculture and are less willing to give up practices that have proved economically viable for them in recent years.

3.2. Downstream resource managers’ willingness to pay for upstream water resource improvement

Downstream resource managers’ agricultural systems are more homogeneous than the farming systems of their upstream counterparts. 31.2% of the agricultural area is under paddy rice, longan orchards constitute 28.7% and the area planted with soybean accounts for 13.6% of the total agricultural area during the rainy season (Schreinemachers et al. 2008). The major socio-economic characteristics of downstream resource managers are depicted in Table 6. The average farm size is just below one hectare. Many young people have moved to the nearby city of Chiang Mai to find non-farm income opportunities, reflected in a high age of the remaining farm population and a high share of non-agricultural income (78.1% of the total household income).

Downstream residents take water for non-food household use, e.g. for bathing and laundry, mainly from a communal water system fed by the Mae Sa river and its tributary creeks (Table 7). Only a minority of the households use water from ponds or groundwater (cf. Sangkapitux et al. 2007). Due to health concerns, most households buy their drinking water in bottles from nearby water companies. Only a quarter of the households use water from the community-based water system for their food and usually they would boil the water first before they drink it. For irrigation, most farmers can use water from the Mae Sa River only, while a small privileged group of farmers in the lowest part of the watershed have access to another water source, coming from the Mae Taeng irrigation dam.

Results of the CE’s conditional logit model for the downstream resource managers are presented in Table 8. Apart from the attributes of one-month water shortage for both cultivation and household use, all attributes are statistically significant at the 99 and 95% level. The positive coefficients of sufficient water for cultivation and drinking for the whole year and good water quality for drinking and household use (laundry, bathing, etc.) suggest that these attributes contribute positively to the respondents’ utility, while the attribute ‘water fee’ showed a negative coefficient. Hence, the signs of the parameters are consistent with theory and a priori expectations.

The implicit price or marginal willingness to pay could be estimated by dividing the attribute coefficient with the price coefficient. The willingness to pay of the downstream farmers to move from the status quo to improved water resources along the different attribute levels is calculated as indicated in equation (5) above. The results show that the downstream farmers’ willingness to pay was highest with 755.19 Baht per household per year for having the water quality improved from its current status (sufficient quality for irrigation) to drinking water quality. This is followed by the willingness to pay 737.42 Baht per household per year to avoid drought problem for agriculture, and 477.43 Baht per household per year to mitigate water shortage for household use. The lowest amount of money that the downstream resource managers are willing to pay is 178.68 Baht per household per year to obtain good water quality for household use (laundry, etc.). The results imply that downstream resource managers give the first priority to the aspect of water quality improvement as the welfare estimate is found highest, followed by the welfare improvement obtained from an elimination of water shortages for irrigation.

The welfare estimate indicates that the average downstream farm household is willing to pay 1492.61 (737.42+755.19) Baht per year for an improvement of water resources from the status quo (2-month shortage and insufficient water quality for drinking and other household use, such as washing) to sufficient water around the year and good water quality for drinking. This amounts to about 0.8% of the average household income of downstream communities.

Table 9 shows the regression results of the logit model determining the socio-economic factors that influence downstream farmers’ willingness to participate in a payment scheme for water resource improvement. Age and education of the household heads in the downstream communities were positively correlated with willingness to pay for an improvement of water resources. Farm households with high acreage and rice as the main crop were also more likely to participate in a compensation scheme. Those households that were characterized by a low water resource security – reflected in a low diversity of water sources for irrigation, past drought experiences and the need to buy bottled water for consumption – had a higher probability to get involved in such payment schemes (cf. Sangkapitux and Neef 2006). In accordance with the results from the upstream part, resource managers who had previously been engaged in environmental activities and supported an improvement of the ecological status of the watershed were more willing to contribute financially to an improvement of water resources.

On the other hand, a high share of non-farm income had a negative effect on willingness to pay. This result is particularly relevant for policy-makers and for the appropriate design of compensation schemes since rapid development in northern Thai cities and rural-urban migration may further reduce the share of farm revenues in rural people’s income portfolio.

3.3. Bringing upstream and downstream resource managers together in a payment scheme for environmental services

Based on the conditional logit model results depicted in Table 8 (Section 3.2) an estimate of the total willingness to pay of the 198 farm households in the downstream area for improved water resources could be calculated at 295,537 Baht (6157 Euro) per year. By contrast, the estimated total compensation needed for upstream farmers to adopt bio-insecticides on 30% and grass strips on 5% of their combined agricultural area¹ is 6,288,300 Baht (131,006 EUR) per year, according to the figures in Table 4. This means that around 4.7% of the amount needed to compensate upstream resource managers could be collected from downstream resource managers in the watershed, while the remainder would need to be contributed by other stakeholders (e.g. tourist resorts, drinking water companies, waterworks authority). For the implementation of a viable, long-term compensation scheme, an appropriate institutional framework would need to be set in place. It is also crucial that potential beneficiaries of improved water resources are certain that free-riders are excluded from the scheme and that the services they are asked to pay for are delivered in a reliable way. This principle of conditionality, however, poses considerable problems because it is unclear to what extent upstream resource managers would need to change their practices in order to achieve the standards requested by downstream water users.

Lack of trust between beneficiaries and providers of environmental services is probably one of the most constraining factors in setting up viable compensation schemes. Particularly in the Thai socio-political context, trust levels between potential beneficiaries/buyers of environmental services (e.g. downstream resource managers, water work authorities, drinking water companies, urban residents) and upstream providers/sellers of environmental services cannot be taken for granted and would need to be gradually built up in the course of designing appropriate compensation schemes. Experienced and trustworthy intermediary organizations can play a crucial role in (1) facilitating the institutional design of such schemes, (2) mediating in the negotiations that need to take place in their implementation, and (3) setting up appropriate monitoring and enforcement mechanisms.

In the Mae Sa watershed, such a role could be played by the “Mae Sa Watershed Conservation and Development Fund” (MSWCDF) established in 2007 under the Mae Sa Watershed Working Group, a pilot project of the Thai government towards greater stakeholder involvement in managing the kingdom’s water resources (Neef 2008). The primary objective of the MSWCDF is to fund various activities devoted to sustainable management of water resources and to minimize potential conflicts resulting from water use both in quantity and quality aspects. In principle, it has been agreed that the fund’s management committee would involve representatives from all stakeholder groups. Two elected chief executives from the four sub-district (tambon) administrative organizations (TAOs) in the watershed are entrusted as chairman and vice-chairman of the fund respectively. Yet, appointed representatives from government line agencies, such as the Watershed Management Unit (WMU) and the Water Resources Department’s regional office, are also involved in managing the fund’s budget, although they cannot contribute financially to the fund due to inhibiting ministerial regulations on budget expenditure. Given their history of not being very conducive to stakeholder participation in natural resource management, these organizations would need to build up a great deal of trust in order to gain the support of other local stakeholders.

Until October 2008, no concrete measures have been taken apart from designing a logo and setting up donation boxes at major tourist and business locations in the watershed. Voluntary monetary contributions are expected from all major stakeholder groups involved in water resource use. The agreement upon the regular contribution on an annual basis in order to sustain the funding scheme is still under negotiation. In its latest meeting in September 2008, the MSWCDF’s committee agreed that each of the four TAOs should contribute at least 50,000 Baht (1042 Euro) in the 2009 fiscal year for supporting various activities that would improve watershed services, such as removing sediments from the rivers, constructing check dams and reforestation.

Being one of the first of its kind in Thailand, the MSWCDF set up as a pilot project does not fully reflect the idea of a payment scheme for ecological services since all stakeholders in principle could benefit from the fund, not just the potential service providers, which are primarily the upstream communities. However, a compensation scheme could be easily incorporated in this type of fund. Hence, the creation of this fund can be seen as a positive sign and a first step towards supporting the principles of Payment for Environmental Services (PES). A proposed institutional framework for such a PES scheme is depicted in Figure 5. Such a framework could then serve as a model for other watershed areas where upstream and downstream stakeholders are willing to engage in negotiations and compensation schemes for improved agricultural and ecological practices in a watershed context.