Payments for Ecological Restoration and Rural Labor Migration in China: The Sloping Land Conversion Program in Ningxia

Transcription

1 Payments for Ecological Restoration and Rural Labor Migration in China: The Sloping Land Conversion Program in Ningxia Sylvie Démurger Université de Lyon, Lyon, F-69003, France CNRS, GATE Lyon St Etienne, 93 Chemin des Mouilles, Ecully, F-69130, France & CNRS, CEFC, USR 3331 Asie Orientale, Hong Kong WAN Haiyuan School of Economics and Business Administration, Beijing Normal University, China & Université de Lyon, Lyon, F-69003, France; Université Lyon 2, Lyon, F-69007, France; CNRS, GATE Lyon St Etienne, Ecully, F-69130, France February 2011 Abstract: This paper employs difference in difference (DID) method with recent data to study the impact of the Sloping Land Conversion Program (SLCP) on labor migration in China. The endogenous problem during the policy implementation is alleviated with the year when program was first introduced to the village instrumental variable. We find that the policy does exert a significant effect on labor migration, though a great number of variables are controlled. In the perspective of migration decision, the policy promotes more farmers to migrate out to find an off-farm job. Migration probabilities increases due to the SLCP policy are 23.3% and 17.4% in 2008 and 2007, respectively. These results are stable and robust in three sub-panel datasets, which implies a strong and significant labor migration effect of the SLCP policy. Key words: Sloping Land Conversion Program; Labor migration; Difference in Difference method. JEL Classification: J22; O13; Q23

2 Introduction After a decade of implementation, the Sloping Land Conversion Program (SLCP) 1 in China came to a new stage in This large-scale program, which was part of a national ecological restoration plan, aimed at reducing soil erosion by converting more than 14 million hectare of cultivated land, of which 4.4 millions of land with a slope over 25 degrees (Xu et al., 2004). The program also entailed the objective of restructuring agricultural production and reducing poverty through a compensation scheme, which granted participating households the payment of in-kind and cash subsidies for a period of 5 to 8 years. A systematic evaluation of this program from capital using efficiency, environment improvement, and income increasing and so on is very necessary. So far, most of the economic literature analyzed and evaluated the effects of the policy on income and poverty and very few papers have been devoted to measuring its impact on labor migration (see notably Uchida et al., 2009). As a matter of fact, the study of the impact of SLCP program on labor allocation in China is part of a wider set of studies examining the question of how this policy affect peasants off-farm-labor decisions, a subject of long-time interest to labor economists in China. Furthermore, the existing literature on the effect of SLCP program on migration provides mixed evidence. On the one hand, some papers highlight the attractiveness of the program to households because of the subsidy mechanism, and find that it induces a rapid growth of labor migration. As a consequence, they conclude that the policy has achieved the target of environmental protection, poverty alleviation, and labor migration (Loucks et al., 2001; Xu et al., 2004; Xu, 2006; Chen et al., 2009; Yin, 2009). On the other hand, some papers argue that the peasants participation to the program was intervened by the government, and that actually most peasants were not willing to participate at all. At the same time, the subsidy and free job training could not apparently change the cost-benefit to the peasants. These papers conclude that the policy had little effect on peasants participation and labor migration (Zhao, 2004; Wunder, 2007; Wang et al., 2007). At present, how much is the effect of the policy on labor migration remains unclear. This paper uses a dataset collected in 2009 to focus on evaluating whether and to what extent the SLCP policy 1 The English direct translation of the Chinese official program title (tuigeng huanlin huancao) is Conversion of Cropland to Forests and Grassland. The program is also known as Grain-for-Green Program (Uchida et al., 2009). 2

3 influences labor migration in China. We focus on three questions: what makes peasants decide whether to participate to the program? Does the policy exert an impact on labor migration? If yes, to what extent? Compared to the existing literature, we use the updated dataset that it allows us to cover the full implementation period of the program. On August, 2007, the General Office of the State Council issued a Circular on improvement of the Sloping Land Conversion Program ", which is regarded as a critical turning point of the SLCP program. This new policy adjusts the main planning of SLCP and the work priorities is shifted from converting large-scale new land to consolidating the existing achievement, shifted from quantity extention to forest quality improvement. Most important of all, considering the peasants low income, the central government extends a circle of compensation payments to participates with a sharply decreased compensate criterion, which may reduce the motivation of the policy on labor migration (as can be seen from the tables, our estimation results are much smaller than theirs, which in turn justify this conclusion.). So far, none of the existing paper analyzes the impact of the SLCP policy in time though the State Council adjust the main direction of the SLCP from The advantage of using 2009 data is that it allows us to cover the full implementation period of the program. Meanwhile, our samples also are much larger and more representive than previous. It is two times more than that in Uchida et al (2009). More importantly, we carefully deal with the endogenous problem before and after of the policy implement. The endogenous problem during the policy implementation is alleviated with the year when program was first introduced to the village instrumental variable. In perspective of peasant migration possibility with instrument variable, the coefficient of policy s net effects on labor migration increases a lot compared to the simple DID models. Most important of all, some of the coefficients in simply DID regression is not significant, but all of them present clearly trends to pass the significance test with the IV-DID approach, which demonstrates a serious endogenous problem during the implement. Meanwhile, we have two sub-panel datasets , This extravagant condition allows us to compare the cultivation effect of program with the time after taking endogenity problem into consideration. The rest of the paper is organized as follows. Section 1 reviews the literature on the policy effects on migration in developed and developing countries. Section 2 gives an introduction to the SLCP policy, and develops the household model that illustrates how SLCP may affect farmer s decisions to 3

4 migrate when faced with land and income constraints. Section 3 provides data description. Section 4 gives an overview of the empirical approach, discusses the identification strategy and displays regression results with the difference in difference method. Section 5 takes endogeneity issues into consideration and estimates further the net effect of SLCP on migration. Section 6 concludes and summarizes the results. 1. Literature review The Sloping Land Conversion Program (SLCP) implemented in China falls into the category of programs implemented worldwide under a scheme of Payments for Environmental Services (PES). The United States started such programs in 1986 with the implementation of the Conservation Reserve Program (CRP), a long-term reforesting program founded by the government. Similarly, Canada implemented a Permanent Cover Project (PCP) in 1989, and the European Union implemented its Conservation Protection Project (CPP) in Payment for Environmental Services programs have also been implemented in a series of developing countries including Costa Rica, Bolivia (Ecological Compensation System program), Mexico (Payments for Hydrological Environmental Services program), Colombia (Environment Protection Project), and El Salvador (Environment Conservation Program under which the government pays upstream farmers to safeguard the forest). Similarly, many other countries have also launched a wide range of programs to protect local environment and ecosystem. Include reference to general papers: Clement et al (Ecological Economics, 2010) / Engel et al. (Ecological Economics, 2008). As for China, refer to Uchida et al. (Land economics 2005; Environment and Development Economics 2007). Generally speaking, these programs have been found to have a positive effect on environmental protection and on the reduction of soil erosion (Dehejia, 1999; McMaster and Davis, 2001). In the US, Szentandrasin et al. (1995) and Mapemba et al. (2007) found an increase in species diversity after the implementation of the CRP program. In Mexico, O Ribaudo et al (2001) found a positive impact of the PSAH program on water quality. Moreover, programs like the CRP in the US and the PES in Costa Rica have been found to be effective in preventing extensive deforestation (Dirzo and Garcı, 1992; Mas et al, 2002; Ricker et al, 2007). 4

5 Regarding the economic impacts of reforestation programs, especially for household activity choice or labor migration, evidence is rather mixed. [On the US part: Reduce as much as possible (one paragraph could be enough)] In the US, Loucks et al. (2001) argued that the implementation of reforestation projects may have some positive externality, which means that it increases the households income and improves the labor migration as well. Cooper (1998), who used discrete selection model and random utility model to analyze the impact of returning farmland to forest policy on farmers income and on labor migration, revealed that this policy has made a more than 10% increase in the American s labor migration. Beck (1999) used an input-output model to evaluate both the advantages and disadvantages that CRP brings to the rural economy in America and finds that the implementation of this policy activates the labor market and improves labor migration. On the contrary, most of the literature found that government payments to farmers decrease off-farm labor participation (Saltiel, 1994; El-Osta and Ahearn, 1996; Mishra and Goodwin, 1997). Ahearn et al. (2005) found that the payments from the Conservation Reserve Program (CRP) decreased the likelihood of a farm operator to work off the farm. Cash payments from CRP actually create an important substitution effect, which makes peasants unlikely to migrate out for off-farm work. Williams et al. (2004) used micro survey data from Tennessee and a probit model about whether to participate in reforestation program to illustrate that the policy of reforestation has no effect on farmers participation and the labor migration as well. Hamdar (1999) also employed the concept of linear programming and sensitivity analysis to the process of evaluating returning farmland to forest policy. By calculating the benefits and costs of participating in reforestation projects and the amount of appropriate household subsidy for improving labor migration, he argued that it is not rational for farmers to participate in reforestation program and labor migration. As for other developing countries, there is a large body of literature that shows that reforestation policies have a significant positive impact on labor migration as well as on household income (Alix-Garcia et al. in Mexico, 2003; Pagiola et al. in Costa Rica, 2002; Russman in Costa Rica, 2006; Asquith et al. in Bolivia, 2008). Resosudarmo and Thorbecke (1996) analyzed the impact of environmental policies on household incomes and migration for different socio-economic classes in Indonesia. The results showed that the policies increase labor migration and improve income distribution. According to Pagiola et al. (2005) s research, the PES program in Latin America actually 5

6 increases the migration possibilities of rural farmers about 10%. In China, researches on the effect of the implementation of the Sloping Land Conversion Program (SLCP) also exhibit contrasting results. A small number of studies highlight some positive aspects. For instance, Zhao (2004) analyzed the threshold point of the farmers returning to farmland and participating in labor migration in the perspective of compatibility of investment incentives. By doing this he found that the SLCP policy decreases the threshold point, thus facilitating the outward flow of labor. Guo (2005) pointed out the difference in outside work due to the surplus labor and the surplus time after returning their farmland to forest. The percentage of families with migrant workers has increased from 26% to 66% after the implementation of the SLCP, and the number of people who leave their home to seek a job elsewhere in TIANQUAN village of Sichuan province has grown from 30 to 70, which is 1.6 times more than before. In addition, there was a trend that the households that have already had some migrant workers are increasing in the number of working out as well as the time of doing that. Wei et al. (2006) formulated a decision model on the basis of a research on SLCP conducted in Chongqing and in Sichuan province. They argued that the SLCP policy has a positive impact on labor migration because it changes the benefits and the costs of the farmers. Xu (2006) held that the SLCP itself influences the decision made by the farmers, therefore, contributes to increasing of the labor migration. Zhang (2008) analyzed and reviewed the compensation policy to the SLCP itself as well as to the ecology from the perspective of externalities to conclude that the SLCP does have a significant impact on labor migration. At the same time, the financial compensation policy, the optimal forest scale and the establishment of the compensation standard were the main factors that affect the households migration. According to Xu et al (2004), the SLCP project has made a 20% growth in labor migration compared with last year. Chen (2009) held that the SLCP project in China makes a substantial increase in household income and there was a 15% increase in labor force flow as well. Loucks et al. (2001) and Liu (2008) argued that the situation of labor transfer has improved significantly as the continuous implementation of the program. Groom et al. (2006) and Yi (2006) used household survey data and found a positive effect of SLCP on off-farm labor migration. Xu et al (2004) first introduced DID methodology to evaluate the effect of SLCP on households income and showed that this policy reduced poverty incidence, improved the income of the household and also contributed to labor migration. However, Du (2003) pointed out the existence of a selection problem when evaluating the policy by DID approach and suggested a two-stage estimation method 6

7 (2SLS) to improve it. Uchida et al. (2009) used a panel data from household surveys implemented in 2003 and in 2005 in rural China and found that on average the SLCP program has a positive effect on off-farm labor participation. However, they found that the program impact varies across groups of individuals in the sample. More researches on labor migration showed that the effect of SLCP on labor migration was not obvious and may be negative. For instance, Brockett et al (2003), did some research on the change of farmers economic situation after attending reforestation project by starting from the attitude of the farmers behavior and it turned out that this policy only has positive effect on the stock of household income but does not have a significant effect on the decision of labor migration itself. Wang et al. (2007) argued that the SLCP policy plays an important role in environmental protection and reduction in soil erosion, however, at the same time, it reduced the economic interests of farmers and had a low efficiency in labor migration. Ke and Zhao (2008) analyzed the factors that affect labor migration by mainly using the research data from Asian Development Bank (ADB) project, they showed that the households who have a smaller stock of farmland are more likely to take part in labor migration. Nevertheless, the influence was not that significant overall. Wu et al. (2008) estimated the impact of the SLCP on households income and the time they spend working out on a panel survey from Zhejiang province in China. There was no clear impact of SLCP program on labor migration, however, family s agriculture income and family s energy consumption were the key factors that determine labor migration. Weyerhaeuser et al. (2005) held that the reforestation policy has both positive and negative impacts on labor migration and the overall effect was ambiguous. Xu et al. (2004) used the data collected two years after the program began and found that the program had no impact on off-farm labor participation. Wunder (2007) found that the impact of reforestation projects on households income and on labor migration is still uncertain. Overall, the existing literature in China mostly focuses on the qualitative levels to investigate the impact that reforestation policy imposes on labor migration. Furthermore, they did not come to an unanimous conclusion. This paper will empirically look into the impact that SLCP program has on labor migration, using research data collected by joint investigation from Hitotsubashi University and Beijing Normal University in

8 2. Policy background and conceptual model 2.1 The background of the SLCP policy The past decades have witnessed a deterioration of natural environment in western China, especially water shortages, soil erosion, the worsening of ecological environment and desertification. Improving the ecological environment has become a prime task and reforestation projects are part of the agenda. The background of the implementation of such policies can be traced back to May 1957 when the State Council passed the People s Republic of China Soil and Water Conservation Program in the twenty-fourth plenary meeting, which states that the original terraced fields on steep slopes, which exceed the required slope should be stopped gradually and returned to forests or pastures. Then, in January 1985, the CPC Central Committee and the State Council launched ten policies on activating the rural economy, which explicitly provisioned that mountain areas with a slope of more than 25 percent should be converted into forestry and pasture use. In June 1991, the People's Republic of China Law on Water and Soil Conservation forbad the cultivation of land with a slope more than 25 percent. Cultivated land on steep slopes before the implementation of the law was supposed to be returned to forestry and pasture use. In August 1998, the revised Article 39 of the Law on Land Administration of the People's Republic of China prohibited farmers from destroying forests and grasslands for cultivation use, and stated that the lands that had already been reclaimed had to be converted into forest, husbandry and lakes. Moreover the Implementation Regulations of the People s Republic of China Forestry Law issued in January 2000 clearly states that land with a slope of more than 25 percent should be used to tree planting and grass growing. In January 2000, the SLCP has been set as an important issue in the Central Committee's Document No. 2 and the meeting on the development of China s west region. In March, the project was officially launched, and it involved 25 provinces (autonomous regions and municipalities) and 1,877 counties. The total project area covers billion mu, accounting for 73.9 percent of the total land area. The total population in the area is 712 million, of which the agricultural population is 533 million, taking 77.6 percent of the total population. According to the plan, 100 million mu were to be returned into forest from 2001 to 2005 and 220 million mu from 2001 to The SLCP aims at stopping land cultivation on steep plots that easily lead to soil erosion and on 8

9 sand land that easily get desertification with carefully designed plans and steps from the perspective of protecting and improving ecological environment. In this process, a principle is strictly carried out, that is planting grass, forests and pastures according to the local conditions. Basically, the purpose of the reforestation projects is to protect ecology and maintain the natural environment. Therefore, the natural condition of the land is the criteria when determining the land to convert, its amount and which farmers participate in the project. In some pivotal and fragile areas, lands that deteriorate the ecology and that are not beneficial for environmental protection should be enforced to return back to forests, no matter how high the economic benefits are. Those policies and implementation methods indicate that the SLCP policy can be considered as an exogenous enforcement for farmers decided by the government. The provision of this policy guarantees the enforcement and the exogeneity of its implementation. The policy design and implementation consist in two parts: government enforcement and follow-up service. On one hand, the evaluation of land and forests is made by local governments. Land plots that are not suitable for cultivation are set to be returned to forests, and farmers are given a reasonable compensation calculated by acres. On the other hand, farmers are provided with free tree seed, grass seed, free vocational training, and free technical guidance after returning farmland to forests. Apart from that, the government has established some public employment service agencies to provide skill training for farmers who participate in this policy, and guidance and help for those choosing to work and live in outside cities. By adapting this kind of free vocational training, job search information and help service, peasants psychological costs of labor migration and the costs of job search can be greatly reduced, which would make them are more involved in labor migration. As for the channel, farmers who participate in the program can increase their income if the payments they receive exceed the opportunity cost associated with retiring their land. In addition, farmers can use the compensation to finance other productive activities, both on and off the farm. It is through this way that the reforestation policy reallocates farmer s time and labor endowment, and therefore can indirectly exert a crucial influence on labor migration. As mentioned above, the government actually issued many related supporting policies that may affect peasants labor migration activity. These policies include a food subsidy (of 300 Jin per mu converted per year in the upper reaches of the Yangtze and 200 Jin in the Yellow River basins before 2007, but it decreases sharply to a half after 2007 according to State Council s new policy), a cash 9

10 subsidy (of 20 Yuan per mu converted per year), free seeds and seedlings, and free labor training, employment information and necessary help (State Environmental Protection Administration, 2005; State Forestry Administration, 2007). On the one hand, the provision of free job information and basic living guidance could reduce the peasants information collection and psychological costs for labor migration, as well as increase their expected wage in the future. On the other hand, the increased leisure time provided to peasants who participate in the program may give them an incentive to seek to migrate out (Becker, 1965; Gronau, 1973; Evenson, 1978). At the same time, the food subsidy and the cash subsidy relax the peasants budget constraint, which in turn may increase their probability to migrate out (Harris and Todaro, 1970; Stark and Taylor, 1991; Zhao, 1999; Lundborg and Segerstrom, 2000; Uchida et al, 2005). Taken together, the follow-up policies associated to the SLCP could change the peasants cost-revenue comparison and promote migration. If it is the case, it is also very reasonable to expect that the longer of the duration the peasants participate, the larger the program effect on migration. 2.2 Theoretical model of SLCP policy affects labor migration Given the interactions between factors that influence how a SLCP policy affects a farmer s time allocation, we construct a conceptual model to understand how land and labor allocations are interlinked with income and other constraints that a farmer might face. We extend the literature on off-farm labor allocation in a household production framework by including land constraints as well as the choice to allocate land to the SLCP program. While analyzing the peasant time allocation, we consider the peasant as a consumption unit as well as a production unit. The peasant time allocation should meet the utility maximization rule if the peasant is completely rational while making a decision. His time is divided into leisure, farm working, and non-farm working. We make the assumption that peasants can get the utility from the consumption of leisure and a composite commodity: l U = U( T, g; X p ) where g is the composite commodity, T l is leisure time, and X p is a series of characteristics that affect the peasant s utility. The derivation of income to leisure time (W l ) is the marginal evaluation 10

11 value of leisure. We assume that it can be calculated in currency unit, which is the same as regular commodities. Peasants land can be used in two ways: one for on farm production, another for participation in SLCP program. There are three kinds of income sources for each peasant: on-farm profit, off-farm income, and net transfer income and government subsidy through the SLCP participation. Regarding on-farm profit, we assume that the representative peasant is a producer. He produces and sells agricultural products to acquire profits, with the price of factors and products given. The production function is assumed to be quasi-concave technology, so that the marginal income from on-farm time is diminishing. The agricultural production is a function of on-farm time and input and output prices and quantities. The exogenous factors for peasants that affect agricultural income include: variable input and output s price P v, fixed agriculture input and output K, laborer s human capital H f. f f f I = I ( T, P, K, H ). v f The marginal income from on-farm working time is given by: δ I δt f f f f = I ( T, P, K, H ) v f Given the assumption of concave production function, it is negative. As for off-farm income, it comes from wages from off-farm work. It is an increasing function of off-farm labor time (T o ), labor experience and technology (H m ), labor market conditions (R m ) and job characteristics (Z m ). The marginal net income of off-farm time is: δ I δt o o o = I ( R, Z, H ). m m m The reforested plots income comes from the government subsidy. According to the design and the implementation of the SLCP, the government will provide cash and food subsidy. We assume the subsidy rate τ to be constant per acre. With the budget constraint of time and income, T l and g present a substitution relation. So, the marginal value of leisure reflects the marginal substitution rate of T l and g, that is: l U T U g l = W( T, g, X ) l p The time allocation can be induced from the maximization of utility subject to the budget constraint (I = I f + I o + I SLCF + Ī), and time (T= T l +T f +T o ) and land constraints (L= L f +L SLCP ). The 11

12 first-order conditions are derived through the Lagrangian: U T U T U T = = U g U g U g l f o (W l = W f = W o ). The time allocated to off-farm activities T o can thus be expressed as follows: T o = T o (L SLCP ; X p, Ī, P v, K, H f, L m, Z m, H m ),where X p, Ī, P v, K, H f, L m, Z m, H m are all of the exogenous variables that affect W l, W m, W f. 3. Data description 3.1 Dataset This paper uses a dataset collected by Hitotsubashi University and Beijing Normal University in The investigation was implemented from March to April 2009, and was especially designed for the research on Sloping Land Conversion Program (SLCP). A stratified random sampling method was used to collect the data, and the survey investigated the townships, the villages and the peasants. The collected information includes family and individuals basic information, labor migration, employment, income and expenditure, consumption and the SLCP participation information. It requests the enumerators to come into the household and ask them to recall the related information. As a result, we have data for the year 2002, as well as for the years 2007 and The survey collected 812 questionnaires. Out of the 800 valid questionnaires, 500 households participated in the program and 300 households were not participating. For the participating group, the investigation covered 9 counties, 44 townships and 52 villages, and for the non-participating group, it covered 9 counties, 31 townships and 31 villages. On the whole, one village was randomly selected in each township, and within each village 10 respondents on average were randomly chosen. As for the year information of participation, all the peasants within participating group did not participate this program all the time from 2000 to Finally, we have 300 non participating households in our analysis. But as for the participating group, it is necessary to present some more detailed explanations. In our sample, the first year when the program was first implemented in the village was And 6 12

13 households firstly were enrolled in this program in After that, there were 23, 52, 128, 187, 104 households began to attend this program in 2001, 2002, 2003, 2004 and 2005, respectively. In our sample, we have information in 2007 and 2008, which can be treated as after-policy information, and the only previous data in 2002 just can be regarded as pre-policy information. In order to strictly use the difference in difference method, it is very necessary to request peasants should not participate the policy before So, we have to delete 81 households in our sample, because those 81 households (6 households in 2000, 23households in 2001, and 52 households in 2002) had began to participate the SLCP program before As a result, we have 419 households in the participating group, who did not attend the program before 2002 and finally all of them attend it after In other words, the participation duration in the program for participated peasants range from 2 to 6 years, and we can test the impact of the duration in the program on labor migration. Furthermore, our research topic is about peasants migration, so we then go on to delete the individuals whose ages are under 15 in two groups. 2 Finally, we have 300 households (1435 individuals) in non participating group, and 419 households (2460 individuals) in participating group. 3.2 The exogeneity of the SLCP policy The compulsory and exogenous nature of the policy clearly appears from a set of questions. To the question Were you free to decide to participate in the program before the implementation of the policy?, 96 percent of the peasants answered No. Moreover, to the question Were you free to choose the land to return to forest before the implementation of the policy?, 95 percent of the participating peasants chose No, it has been compulsorily arranged by the village/government. These two questions show that the policy was compulsory, and that it can be considered as exogenous to rural households. Tao et al (2004) points out that most peasants did not have the option to choose whether they can participate in the program or not, how much land they had to return, and what kind of the trees they would plant on the returned farmland. Actually, 85 percent of the respondents in the participating sample and 72.2 percent in the non-participating sample declared that they did not have any option at 2 Modern labor economics theory and international statistical convention usually believe labor is the people who are above 16. So we choose the 16 as a threshold to define the labor in our sample. 13

14 all in our sample. The two questions listed above also indicate that the policy itself (whether to participate or not) was compulsory. As demonstrated in Zuo (2001) and Bennett (2008), the targeting of areas to retire has generally been conducted via a top-down approach, starting with retirement quotas that were distributed from the central government to the provinces, followed by subsequent distribution down through counties, townships and finally to participating villages, and then the local government divided different households into different categories by a land criterion. 3.3 The SLCP selection criterion In the view of the policy design and implementation, the only standard to decide on a peasant s participation to the program is the land condition, which has nothing to do with labor migration. Impoverished lands on steep hills and far away from the village are designed to return to forest. Table 1 reports statistics on the five biggest plots per household, which accounts for more than 95% of total land for 90% of farmers in our sample. There are great differences between the participating and the non-participating groups in the slope of plots, the distance of land to home and land quality. Regarding land slope, only 14.9 percent of the land is flat in the participating group, while it is 56 percent in the non-participating group. This illustrates the fact that higher quality and flat land is still cultivated. In the participating group, more than 59.2 percent of the land has a slope higher than 25 percent, while it is only 22 percent for the non-participating group. It reflects that most of the high slope lands on the hill are returned to forest. Regarding land quality, more than 45 percent of the land is of a bad or very bad quality in the participating group, while the corresponding figure for the non-participating group is less than 4 percent. On the other hand, the good and very good land amounts to about 20 percent in the participating group, and to nearly 70 percent in the non-participating group. Table 1 also presents great disparities between the participating and non-participating group in land quality, geographical position and land slope, which are the criteria chosen by the government to decide whether a peasant take part in the program or not. To be specific, flat and high quality land will continue to be cultivated, but the land with bad geographical characteristics, far away from home, with a steep slope and of bad quality may be arranged by the government to return to forest. This kind of policy design and rule guaranty the exogeneity and the compulsory nature of the policy. 14

15 Besides the policy aspect, migration decision also depends on a set of various factors that can be summarized as follows: individual and household characteristics variables, opportunity cost, and family assets variables. In theory, explanatory variables should be the factors that affect the farmers time allocation decision. We expect that the larger the household size, the higher the possibility moving out for off-farm work (Stark and Bloom, 1985). Moreover, younger and more educated people are more likely to migrate out because of the lower psychological costs and higher revenue streams, and they are more likely to acquire skills to get a better paid job (Becker, 1993; Ehrenberg and Smith, 2005; McConnell et al., 2007). Meanwhile, the migration networks also exert some effect on peasants migration decision, because it can reduce the information collection cost and psychological cost. As can be seen from table 2, there are indeed some differences between the two groups at 5% level with the two-sample Kolmogorov-Smirnov test (a non-paramatic test, which is used to test the distribution differences between two samples.). Take education variable for example, the average education in participating group is higher and the KS value is 0.000, which we can conclude that participating group peasants carry out more of education compared to that of non participating group. The number of children at school within a family also presents an apparently difference. The KS test value is 0.075, which demonstrates the participating peasants have much less number of children at school within their families. The percentage of dry land to total land variable also passes the KS test at 5% level, peasants high dry land percentage determines their agriculture s low productivity, which then make peasants opportunity cost of moving is lower as compared to peasants who own low percentage of dry land. Generally speaking, individuals in the participating group are really more educated, with less children at school and higher percentage of dry land. The dependent variable is measured by a dummy variable to distinguish between individuals who migrate and those who do not migrate. Table 2 shows that the migration pattern is very different between the participating and the non-participating groups. The migration possibilities are significantly higher in the participating group compared to that of non participating group, because the KS test is apparently different with zero at 1% level. 4. The impact of the policy on labor migration 15

16 4.1 Empirical approach Our objective is to test whether there are significant difference in migration patterns between the participating and the non-participating group. We construct some natural experiment according to the policy that affects labor migration, so as to control for self-selection problem. We use the difference-in-differences (DID) method with controlled variables to measure the net effect of the SLCP policy on labor migration. We assume that the participating peasants are the treatment group ( 1 for short), and the non-participating peasants are the control group ( 0 for short). The time before the policy implementation is t=0, and the time after is t=1. Our target is to examine the difference before and after the policy for the treatment and the control groups. The difference-in-differences estimator is given by: DD = E( Y Y T = 1) E( Y Y T = 0) = E( G T = 1) T T C C T where Y is the outcome of interest (migration in our case) for individual i at time t (with t=0, 1). The 1 first term can eliminate separate changes trend for themselves. And the second term not only can control the time series trend and some other time-varying factors, but also can get the policy s net impacts (Bertrand, 2004). We use the 2 2 quadrat method and the first order difference method to measure the DID. Let us define A, B, C and D as the number of migration days in the following situations: A stands for the non-participating group before the implementation of the policy, B stands for the participating group before the implementation of the policy, C stands for the non-participating group after the implementation of the policy, and D stands for the participating group after the implementation of the policy. We use C-A to eliminate the non-participating group s time-varying trend and D-B to eliminate the participating group s time-varying trend. As a result, (D-B)-(C-A) can eliminate the whole time series trend, and the remaining effect is the net impact of the policy on labor migration. We can translate this into a basic model: Y= β + β * year+ β * decision + β *( year _ decision) + ε Year is a dummy variable (0 before the policy, 1 after the policy); and decision is a dummy variable equals to 0 for a SLCP non-participating peasant and to 1 for a SLCP participating peasant. It 16

17 is straightforward to show that the coefficient β 3 of the interaction term is (D-B)-(C-A), that is the net impact of the policy on labor migration in our model Another assumption is that during the experiment, all the factors should stay the same or be randomly changed, except program participation changes. Since it is difficult to hold such assumption in social science, we control as many variables as possible to fill up this kind of deficiency (Lalonde, 1986; Sun, 2009). On the basis of the controlled variables, the new DID model can be written as follows: Y= β + β * year+ β * decision + β *(year _ decision) + X ʹ γ + ε X is a vector of control variables, including time invariant or time-varying observable variables, such as sex, ethnicity, family scale, education, age, the number of family labor, land square per person and the period dummy. ε is a random error, which reflects the unobservable factors that affect labor migration. Unobservable effects and time-invariant factors could be eliminated by difference. In conclusion, β i, β 3 and γ in the model are the coefficients to be estimated. Moreover, we also add a variable of duration in the program to test whether the duration of participation in the program make some differences with number of years participating the program. Thus, heterogeneity in treatment effects can be studied by including interactions between duration and the treatment dummy variable. 4.2 Estimation results from a simple DID method Following the theoretical analysis above, we use the simple DID method with controlled variables to regress migration on SLCP policy. Considering the correlation within households, we will regress the model using the cluster option to account for clustered errors at the household level. The estimation results are reported in Table 3. Indeed, in the columns (1), (4) and (7), the coefficients of the interaction term are positive, and all of them pass the 5% level significance test. In the perspective of the estimation result, the coefficient is rather robust for the three controlled group. Table 3 shows that the sloping land conversion program does exert a significant impact on labor migration. On average, if other conditions remain unchanged, the policy increases the migration possibilities for 5.6% in 2007, and 5.9% in 2008 respectively. 3 3 We calculate the partial influence using %ŷ = 100(e b 1). For details, see Wooldridge (2003). The marginal effect for a dummy variable is the difference in probability of migration relative to the reference group; for 17

18 In addition, table 3 also demonstrates that the migrants usually are young, educated male individuals; peasants who have those characteristics are more likely to migrate. Take the case of group, female s migration possibility is 5.1% lower than that of male while other factors are unchanged. Likewise, age is identified as negatively affecting the probability of migration, 1 year older actually decreases 2.53% migration possibility at 1% level, this kind of negative effect probably comes from less number of benefits stream for the older people. Moreover, education is found to have an apparently effect on raising the probability of migration as well, because it passes the significance test at 1% level, which is rather robust in three controlled groups. As can be seen from Table 3, the number of children at school in a household was an important factor determining the migration decision. Take group for example, holding land size and the rest of the variables constant, decreasing the number of children at school by one raised the probability of migration by 5 percent. Household land number significantly affects labor migration, peasants have less land are more likely to migrate. Decreasing the size of household land per adult by one mu increased the probability of migration by 0.9 percent. Because household land is the important factor that affect peasants agriculture income, the effect of less land on migration is via less expected agriculture income. From the data, we find that this effect is not very large, which probable reflects low percentage rate of agriculture income to other income sources (non-agriculture income or migration income) (Sjaastad, 1962). Besides the treatment dummy in the DID model, we also consider the intensity of participation and the duration in the program. continuous variables, the effect is evaluated at the mean. It is the same below. 18

19 5. Correcting for endogeneity 5.1 The source of the endogeneity problem Simple difference in difference (DID) method can reasonably estimate the net effect of an exogenous policy on an outcome variable. But in empirical research, there are many reasons why there might be endogeneity, including selection bias from the non-random sample (Xu, 2004). For example, there may still exist non-randomly extracted samples, though we use a stratified random sampling method during the design of the questionnaire. During the policy implementation process, there might be an endogenous bias for four reasons. First, there is a fact that the program has been implemented throughout the whole period of the 2000s. The implementation experience accumulated over the years may tell the local government which peasants are more likely to voluntary join this program and which are not. So, when the land quality and the geographical position are neither very good, nor very bad, the government may potentially make a decision on characteristics related to the willingness of the peasant to participate. The peasants who were thought to be voluntary to participate in the program will actually be designated by the government officials. Doing in this way, the government achieves a Pareto efficiency point, in which the peasants intention and the policy can match with each other very well. On the contrary, if officials just try to achieve their own goal, regardless of the peasants will, the local government may meet with a good deal of opposition or resistance during the implementation of the policy. So, the government may take peasants will into consideration while deciding who will be part of the program. This may induce serious endogenous bias and make the policy not exactly a natural experiment. Second, the area where we did our investigation belongs to a national poverty-covered area. A majority of peasants over there live below the poverty line. Usually, the aim of the SLCP is not simply to renew the environment, but it also intends to increase income, rectify the agricultural structure, etc. (Pagiola and Platais, 2005). Actually, the local government takes migration as an effective way to alleviate poverty (there is a local government document entitled Promoting the non-farm employment 19

20 and labor migration is a basic way to reduce poverty in the next 10 years 4 ). Given the poverty reduction target, a local government may tend to select a poor family to participate in the program, because it can help them to migrate out and probably get out of the poverty trap. If it is the case, the government can not only greatly alleviate the poverty in that area, but also achieve the goal of protecting regional environment. As a result, the local government s poverty alleviate endeavor may also lead to endogenous problem. Third, an opposite endogenous problem may also come from local government s financial pressure. One of the supporting policies of the SLCP requests local governments to subsidy every participating peasant, which may represent a heavy burden for local governments. In order to reduce the burden as much as possible, local governments may arrange fewer peasants to participate the program though some of them should participate by land quality standard. Fourth, there might be a problem related to the fact that we use the recall data. Since people are usually more familiar with the situation they are contemporary facing, it is possible for farmers to confuse the present and the past, which means less accuracy of the recall data. 5.2 Instrument variable- DID method For the reasons stated above, the Sloping Land Conversion Program and the labor migration decision may affect each other because of the endogenous problem during the policy design and implementation. So, it is necessary to find a proper method to deal with this kind of problem. In this paper, we will introduce an instrumental variable for the Sloping Land Conversion Program. As a good instrument in theory, it not only can explain the policy variance to different people, but also it must meet the exogenous requirement. In other words, this instrument variable must only exert some indirect impact on labor migration through the Sloping Land Conversion Program. We will introduce a village level instrument variable into our model. The year when the SLCP program was first introduced to the village is decided by the higher-level government, which is exogenous for individuals. On the one hand, the earlier the program being introduced into the village, the earlier the farmers in this village are possible to participate in this policy. There is no doubt that the 4 Baseline of poverty alleviation development in rural Ningxia from 2001 to

21 peasants would not be involved in the project if the program has not been introduced into the village. Thus, the year when the SLCP program was first introduced to the village is highly and directly related to involvement of the peasants in the program. On the other hand, the year when the SLCP program was first introduced to the village variable has no direct relationship with personal labor migration, because whether the program is introduced to the village is decided by the higher-level government rather than the individual peasant in the village. As a result, the individual peasant will not take the year when the SLCP program was first introduced to the village into consideration when they make their migration decisions. In conclusion, we believe that the year when the SLCP program was first introduced to the village variable and individual migration decision cannot directly connect with each other. From the analysis above, we believe the year when the SLCP program was first introduced to the village has a decisive impact on the SLCP participation, and at the same time, it has nothing directly to do with individual s labor migration. Especially, this variable is an objective, exogenous variable for peasants, which cannot be altered with the change of policy or other forces. So, we expect that the year when the SLCP program was first introduced to the village is a good instrument variable for the Sloping Land Conversion Program in our model. In the empirical analysis, we use the 2SLS regression method to test whether this variable is a good instrument. First, we regress labor migration on the SLCP participation and on the year when the SLCP program was first introduced to the village separately, and then we run a regression which takes both of them into one model. If the instrument does affect labor migration through the SLCP participation, we may expect that both independent variables coefficients are significantly different with zero in model 1 and model 2 (see table 4). As expected, in model 1, in the sub-panel dataset, the policy variable coefficient is , and is significantly different from zero at 1% level. And in model 2, the instrument variable coefficient is 0.042, and is significant at 10% level. But when we put policy and instrument together into one model, the instrument variable doesn t significantly affect labor migration any more. It demonstrates that the instrument variable actually affects migration just through SLCP policy. In order to test and verify our judgment, we regress migration decision on the instrument variable directly; the result in model 10 (see Table 4) demonstrates that the instrument variable significantly affects migration decision at 1% level. Furthermore, Table 5 also shows the first stage regression results for 21