1
Estimating Non-Market Environmental Benefits of the Conversion of
Cropland to Forest and Grassland Program: a choice modeling approach
Xuehong Wanga,Jeff Bennetta,[1],Chen Xieb, Zhitao Zhangb, Dan Liangb
aAsia Pacific School of Economics and Government, AustralianNationalUniversity, Canberra ACT 0200, Australia
bChina National Forest Economics Development and Research Centre, State Forestry Administration, Hepingli, Beijing 100714, China
Abstract
The non-market values of the environmental benefits derived from the Conversion of Cropland to Forest and Grassland Program (also known as the Grain for Green Program and the Sloped Land Conversion Program) in the Loess Plateau region of North West China were estimated using choice modeling both on-site in Xi’an and Ansai and off-site in Beijing. Separate choice models were estimated for the three sites and the results compared. Significant differences were found between the implicit price estimates derived from the multinomial logit (MNL) model and the random parameter logit (RPL) model for some environmental attributes. Based on theresults from the RPL models, the average willingness to pay per respondent household in Beijingwas CNY 882.56 (USD 109.44) each yearfor the environmental improvements on the Loess Plateau provided by the Program, a payment level significantly higher thanthe comparable estimates of CNY 342.56 (USD 42.48) in Xi’an and CNY 388.08 (USD 48.12) in Ansai.
Keywords: land use change, choice modelling, environmental benefits, China
- Introduction
The growing intensity of environmental deterioration and consequential economic losses, caused mainly by land and water degradation in North West China, triggered the adoption of a new land use management strategy by the Chinese Government in 1999. Formally launched in 2002, the Conversion of Cropland to Forest and Grassland Program (CCFGP), also known as the Grain for Green Program and the Sloped Land Conversion Program, involves the world’s largest expenditure on a single environmental services program after the US Conservation Reserve Program (CCICED, 2002; Xu et al., 2004). Under the Program, participating farmers are supported by thegovernment to grow trees and grass on land previously used for annual cropping. With government funding for the CCFGP drawing to an end in 2007 (PRC, 2002), increasing attention amongst the Chinese policy makers,as well as the wider Chinese and international communities,has been devoted to the long-term sustainability of the land use changes triggered by the CCFGP.
Few attempts have been made to devise innovative ways to sustain the supply of fundsrequired for the CCFGP from when the government funding stops until alternative livelihood sources for participating farmers become financially viable. For instance, the potential for the Chinese communitiesthatbenefit from the environmental improvements provided by the CCFGP to pay directly for their gains remains largely unexplored.This is, in part, because of a lack of information about the monetary value of the environmental benefits so derived. While some of these values can be inferred through market transactions – such as more fish production due to water quality improvement – others, especially the non-use values, must be estimated using stated preference techniques.The research reported here aims to fill this information gap by estimating the array of non-market values that the wider Chinese community place on the environmental benefits derived from the CCFGP usingchoice modeling (CM).This technique allowsenvironmental changes to be disaggregated into relevant ‘attributes’, such as landscape aesthetics and biodiversity, and marginal values for each attributeto be estimated. This allows increased flexibility in analysis and facilitates benefit transfer(Rolfe and Bennett, 2006), and offers the potential toreduce problems of framing bias because substitution effects can be incorporated within the questionnaire design(Rolfe and Bennett, 2000).
This paper is structured as follows. In the next section, details of the CCFGP are presented. The CM method isoutlined in Section 3. This is followed by a description of the application in Section 4. Model specifications are detailed in Section 5. Results derived from the selected models, including people’s willingness to pay for improvements in specific environmental attributes and the compensating surplus for aggregate changes under the CCFGP are set out in Section 6. The paper concludes with a discussion of the policy implications of the results.
- Case Study
The focus of this paper is the Loess Plateau, where the pilot phase of the CCFGP was implemented. Located in the north and central west of China, in the middle reaches of the Yellow River,the Plateau has an arid to semi-arid climate and its 630,000 sq km includes parts of Gansu, Shaanxi, Shanxi, Qinghai and HenanProvinces, Ningxia Autonomous Region and Inner Mongolia Autonomous Region (McVicar et al., 2002; Sun and Zhu, 1995).
The Loess Plateau is considered to be the largest highly erodable area on earth (Peng, 2001). With reduced vegetation cover, harsh winters and intense summer monsoon rainfall, a total of 2.2 billion tons of topsoil are washed away every year from the Loess Plateau(Huang, 2000).About 70 per cent of this flowsinto the Yellow River, contributing 90 per cent of all the river’s sediment (Huang, 2000; Hessel, 2002). Severe soil erosion and sedimentation has led to increased sandstorm frequencies in the past two decades and deteriorating water quality has affected drinking water supplies both within the Loess Plateau and in off-site areas to the east (PTFDSSD, 2003). Land degradation on the Plateau has also caused deteriorating landscape aesthetics and biodiversity loss. Experts from the Chinese Academy of Sciences and the Chinese Academy of Forestry predict that,without the CCFGP implemented on the Loess Plateau, by the year 2020,average sandstorm occurrence in the region will increase from the current level of 20 days to 22 days per year, vegetation cover will drop from 14 per cent to 10 per cent, sediment discharge into the Yellow River will increase from 1.6 billion tons to 1.9 billion tons, and the number of plant species present will decrease from the current 2000 species to 1600 species (Wang et al., 2005; Q. Lu, Y. Wang, and Q. Zhu, personal communication, 2005). With the CCFGP in place, this trend of deteriorating environmental conditions is predicted to be reversed.
Among the potential beneficiaries of the predicted environmental improvements arising from the CCFGP are urban households both on-site and off-site. The values enjoyed may be influenced by their proximity tothe Loess Plateau and differences in their socioeconomic characteristics.An aim of this study is to test if these influences are significant.
- Choice Modeling
A CM application involves respondents to a survey selecting their preferred option from a range of potential, hypothetical environmental management strategies, including ‘business as usual’.CM is consistent with Lancaster’s theory in which consumption choices are determined by the utility or value that is derived from the attributes of a particular good or situation (Lancaster, 1966). It is also based on the behavioural framework of random utility theory (RUT), which describes discrete choices in a utility maximising framework. Utility (Ui) is assumed to be comprised of two parts:
Ui = Vi + i (1)
where Vi is the systematic and observable component of the latent utility for option i; and iis the random or “unexplained” component(Louviere et al., 2000).
Because of the random component, the researcher can never expect to predict choices perfectly. This leads to the expression for the probability of choice:
P(i) = P[(Vi + i) (Vj + j)] (2)
for all available j options. Different assumptions of the distribution of the random error terms yield different models. Assuming that the random error terms are distributed independently and identically (IID) and follow the Gumbel distribution with scale parameter μ(which is typically assumed to equal one), the probability of choosing option i can be estimated using the multinomial logit (MNL) model:
P(i) = (3)
The individual’s indirect utility function (Vi) in equation (3) for a choice option can be modeled by way of various specifications. The simplest functional form involves an additive structure which only includes the attributes from the choice sets:
Vi = C + Σ βk.Xk where i = 1, …., K (4)
where C is an alternative specific constant (ASC), β is a parameter vector, and X is a vector of K attributes from a choice set. The attributes enter the utility functions at various levels as specified by an experimental design (Bennett and Blamey, 2001).
The IID error term assumption underpinning the MNL model implies a number of restrictions, in particular, the property of independence of irrelevant alternatives (IIA). The nested logit (NL) model represents a partial relaxation of the IID assumption of the MNL model. It recognizes the possibility of different variances across the alternatives and correlation among partitions of alternatives (Louviere et al., 2000; Hensher et al.,2005; Champ et al., 2003). The mixed logit model, or random parameter logit (RPL) model represents a full relaxation of the IID assumption. It allows model parameters to vary randomly over individuals(Revelt and Train,1998; Revelt and Train, 1999; Brownstone and Train,1999).Similar to equation (1), the utility that a person obtains from alternative i among j options in a choice set is:
Ui = (4)
where Xi is a vector of observed variables, vector varies over individuals with density f (β*) where * represents the parameters of the pooled distribution, and i is an unobserved random term that is IID extreme value, independent of and Xi. The vector can be expressed as the population mean (b) and the individual specific deviation from that mean . Hence the above utility function can be rewritten as:
Ui = b’Xi + ’Xi + i (5)
Because the choice probability cannot be calculated exactly –a closed form solution does not exist –it is estimated as a “mixture” of logits with the density f as a mixing distribution.
These model forms can be used to generate probabilities of choice and hence estimates of marginal values for each attribute and compensating surpluses (CS) for changes between different choice profiles (Bennett and Blamey, 2001).
4. Application Details
4.1 Survey and questionnaire design
To test if people living proximate to the environmental impacts of the CCFGP have different valuesto those who live further away (following Champ et al., 2003), the same CM questionnaire[2] was used in surveying residents in Ansai on the Loess Plateau, Xi’an which is the capital of Shaanxi Province and located at the edge of the Plateau, and Beijing, the national capital which is located to the east.
A draftCM questionnaire was initially developed using the results of two focus groups held in Beijing in December 2004. Attribute levels were specified based on scientific research conducted in the region and projections were made by experts from the ChineseAcademy of Sciences and ChineseAcademy of Forestry (Table 1). Four more focus groups were held in July 2005 in Beijing, Xi’an and Ansai to refine the draft questionnaire.
(Table 1 here)
An orthogonal fractional factorial experimental design was used to allow the estimation of all main effects of the attributes. It has been argued that the main effects typically account for 70 to 90 per cent of explained variance (Louviere et al., 2000). Dominant and implausible alternatives were removed from the design but orthogonality was preserved. The resulting 25 choice sets were allocated to five blocks of five choice sets. Hence there were five versions of the survey questionnaire differing only in the attribute levels in the choice questions.One-on-one interviews were conducted to pre-test the revised survey questionnaire.
In the questionnaire, respondents were told that there were two broad options available for land use management on the Loess Plateau: to revert to the pre-CCFGP land use practices; and to continue the land use practices initiated under the CCFGP. The scenario presented to respondents was that the financial support from the government in implementing the CCFGP will cease in 2007. This would mean that farmers may revert to previous land use practices because of their subsequent income losses. To keep farmers growing trees and grass on the Loess Plateau, financial support for the farmers involved in the Program was stated to berequired for 10 years beyond 2007. Specifically, a compulsory annual payment from urban households across China for a 10-year period would be required under this scenario. It was stated in the questionnaire that “the money from the payment would go into a special fund used only for the continuation of the CCFGP in the Loess Plateau region. For the collection, use and administration of the fund, a complete system would be put in place and would be monitored by the State Auditing Administration”. The size of the payment was stated to depend on the extent of the environmental improvements that the CCFGP aimed to achieve on the Loess Plateau.
Respondents were then presented with five choice sets showing various options for land use management in the Loess Plateau region (see Figure 1 for an example). Respondents were asked for their preferred choice from each of the five choice sets. They were also reminded of their available income and all other expenditure.
Figure 1 Example of a choice set from the Land Use Management on the Loess Plateau questionnaire
Option 1（R (Reversion to land use practices pre-CCFGP)
By 2020: / Continuation of CCFGP
Option 2
By 2020: / Option 3
By 2020：
Compulsory Payment per annum (CNY) / 0 / 100 / 100
Sandstorm days per year / 22 days / 20 days / 18 days
Landscape (vegetation cover) / 10% / 30% / 20%
Water quality (annual sediment discharge) / 1.9 billion tons / 15% less than Option 1 / 10% less than Option 1
Plant species present / 1600species / 1900 species / 2200 species
I would choose
Tick one box only / 1 / 2 / 3
4.2 Survey method and logistics
The population of interest is urban residents in North China. The definition of urban resident needs to be carefully drawn as the former distinction between urban and rural residents based on their “hu kou” (household register) no longer holds. Many rural people now live and work in urban cities without an urban “hu kou”. According to the occupation-based social stratification completed by the Chinese Academy of Social Sciences (CASS) (Lu, 2004), rural people who have been living and working in urban cities for four to five years no longer belong to the social class of farmers but instead fall into other categories based on their new occupations. This principle is applied here. Furthermore, household rather than individual values are elicited in this survey.
According to CASS (Lu, 2004), in the past 20 years, social stratification in China is more and more reflected by occupation stratification. Ideally, stratified random sampling based on occupation should be adopted. However, due to the lack of sample frame for some occupation groups, a mixture of sampling methods was used for this research.
A target of 200 households was set to be interviewed in each of the Beijing, Xi’an and Ansai sub-samples. Because the occupation-based social structure is homogenous across different geographical districts in the urban areas in Beijing and Xi’an, the geographical boundaries of the Third Ring Road in Beijing and the Second Ring Road in Xi’an were set as the limits of the survey area. This enabled survey cost savings for the in-person survey and also ensured that rural residents were excluded. Systematic sampling was adopted to select five blocks each from the 15 blocks within the Third Ring Road in Beijing and from the 16 blocks within the Second Ring Road in Xi’an. Quota sampling against occupation and systematic sampling in terms of the households approached within each block were then applied in combination. About 40 households were drawn randomly from each geographical block, with the specific number of households drawn from each occupation group in proportion to the percentage of the population as a whole in the respective cities (Table 2). This sampling strategy ensured that each of the five versions of the questionnaire was taken by 40 respondents in each of the five geographical blocks with respondents from different occupation groups being involved evenly. Altogether 200 valid questionnaires were collected in Beijing and 203 questionnaires in Xi’an.
(Table 2 here)
Quota sampling based on occupations was not used in Ansai given its smaller urban population (around 17,000) and its greater population homogeneity. All the residential buildings in the urban area of Ansai are within walking distance from the town centre, so four geographical blocks were first selected at random and systematic sampling was then applied in approaching the households living in each block. Because the urban area of Ansai is home for both the urban population and rural population, a screening question was asked to identify urban respondents. Altogether 203 valid survey questionnaires were collected in Ansai.
4.3 Socio-demographics of respondents
The socio-demographics of the respondents and the population averages are shown in Table 3. Across the three sub-samples, the age structure of the samples is close to the city/ town average. The sex structure of the sample in Beijing is close to the city average, but is not representative in Xi’an and Ansai. This is because the survey was mainly conducted at people’s homes, and in most cases the male is the decision-maker in the households. Both the education and income characteristics of the sub-samples are different from the populations in each of the three places. The difference in education can be explained by the inclusion of rural residents who generally have lower educational levels than urban residents in the population average. The lower income level of the sample does not present a problem, as the sample stratification based on occupation ensures income representation.
(Table 3 here)
- Model Specification
5.1The Multinomial Logit (MNL) Models
The choice data were analysed using LIMDEP. With the sub-samples separated, three data sets emerge. Variables used in these models and their coding are specified in Table 4. Mean values for the socio-demographic and attitudinal variables used in these models are shown in Table 5.