Oxford Economic Papers

Oxford Economic Papers Advance Access published online on July 8, 2009
Oxford Economic Papers, doi:10.1093/oep/gpp021

This Article Abstract Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Alert me to new issues of the journal Add to My Personal Archive Download to citation manager Request Permissions Google Scholar Articles by Groom, B. Articles by Zhang, S. Social Bookmarking          What's this?

© Oxford University Press 2009 All rights reserved

Relaxing rural constraints: a ‘win-win’ policy for poverty and environment in China?

Ben Groom^*, Pauline Grosjean, Andreas Kontoleon, Timothy Swanson^¶, and Shiqiu Zhang

*Department of Economics, School of Oriental and African Studies, Russell Square, London, WC1H 0XG; e-mail: bg3{at}soas.ac.uk
University of California Berkeley, Department of Agricultural and Natural Resource Economics
Department of Land Economy, University of Cambridge
¶Department of Economics and Faculty of Law, University College London
College of Environmental Sciences and Engineering, Peking University

JEL classifications: C33, J22, O22

	Abstract

TOP Abstract 1. Introduction 2. The Sloping Lands... 3. Theoretical model of... 4. Econometric approach 5. Results 6. Conclusion Supplementary material Appendix description of... Acknowledgements References

 
The link between institutional and market failures, rural poverty and environmental degradation suggests a ‘win-win’ policy intervention: relax local ‘constraints’ and achieve poverty alleviation and environmental goals. We evaluate the ability of the Sloping Lands Conversion Programme (SLCP) in China, a reforestation payments programme, to relax constraints on off-farm labour markets and achieve these dual objectives. Our model of the agricultural household allows for heterogeneous exposure to constraints and impacts. The model predicts that the impact of the SLCP on off-farm labour supply will be larger for constrained households if constraints are relaxed. To test the predictions we combine a switching regression with difference in differences. Applied to panel data, this technique allows identification of the heterogeneous impact of the SLCP on constrained and unconstrained households. Our results identify some support for the ‘win-win’ hypothesis in the case of the SLCP, and how the targeting of the programme can be improved.

	1. Introduction

TOP Abstract 1. Introduction 2. The Sloping Lands... 3. Theoretical model of... 4. Econometric approach 5. Results 6. Conclusion Supplementary material Appendix description of... Acknowledgements References

 
An expansive literature on household behaviour in developing countries points to the near ubiquity of missing markets, imperfect institutions, and high transactions costs (Singh et al., 1986; Jacoby, 1993; Key et al., 2000).¹ As a result, rural households are constrained in their choice of production patterns and occupations and are precluded from many income enhancing opportunities. This economic environment conspires to limit development and even trap households in poverty (de Janvry et al., 2005; Dumas, 2007). At the same time, these localized failures are frequently the root cause of wider environmental externalities. For instance, failures in the off-farm labour market have been shown to underpin slash-and-burn agriculture, deforestation, and other apparently inefficient household land-use practices in many developing countries (Bluffstone, 1995; Shively and Pagiola, 2004). Likewise, externalities at the river-basin level, such as flooding, often arise as a direct consequence of inefficient labour and land allocations (Yin and Li, 2001). Thus, developing countries can be doubly immizerated since, not only do these failures leave rural households impoverished, but the coping strategies that households employ frequently impose externalities upon the wider population.

It is difficult to imagine a silver lining to this cloud, and yet a closer examination of this account suggests some grounds for optimism. Since the cause of poverty lies in local institutional and market constraints, a ‘win-win’ intervention immediately presents itself: by relaxing local constraints both poverty and negative externalities can be reduced. Of course, this realization is not new (e.g. Ellis, 1998; Carter and Olinto, 2003),² but ensuring local interventions are successful is not without its difficulties. Among these, the heterogeneity of rural households' exposure to constraints presents a particularly important stumbling block, since this means that the impact of local interventions will also tend to be heterogeneous (Carter and Yao, 2002; Vakis et al., 2005). Combined with the fact that exposure to constraints is frequently unobserved, the design and targeting of interventions, predicting household responses, and undertaking retrospective evaluations, all become fraught with difficulties.³

The primary aim of this paper is to develop a general theoretical and empirical method which makes sense of this complicated environment and allows careful examination of the impact of interventions when households are heterogeneous in their exposure to constraints. To this end, we first develop a model of the household consumer-producer in the presence of market and institutional failures. The model distinguishes households according to the nature of the institutional and market failures they face and gives rise to several testable hypotheses concerning the impact of relaxing binding constraints on off-farm labour supply: arguably the most important potential source of higher income in rural areas (e.g. Rozelle et al., 1999). We then develop a novel empirical approach which combines an exogenous switching regression with ‘difference-in-differences’. The first feature allows identification of constrained and unconstrained households when constraints are unobservable and heterogeneous (Carter and Yao, 2002; Vakis et al., 2005). The second feature is a programme evaluation technique which accounts for non-random selection and identifies an intervention's treatment effect, here, on off-farm labour supply. The combination of these methods allows identification of heterogeneous treatment effects and the most important constraints.

The method is used to assess the impact of a large environmental and poverty alleviation programme implemented in rural China: the Sloping Lands Conversion Programme (SLCP). The model is developed in the context of households participating in the SLCP and the empirical method uses panel data from SLCP participant and non-participant households from two provinces: Ningxia and Guizhou. The evaluation of the impact of the SLCP on off-farm labour supply is an extremely appropriate application for several reasons. Firstly, our focus on off-farm labour supply is highly relevant to rural China. Evidence suggests that off-farm opportunities offer the greatest potential for increasing rural household income in China (Rozelle et al., 1999; Knight and Song, 2005, ch.8).⁴ Also, previous analyses have shown that off-farm incomes represent the predominant substitute for crop incomes for participants in the SLCP (Xu et al., 2004; Groom, 2005). Secondly, China is especially prone to market and institutional failures given its historically centralized economy and currently only nascent markets in rural areas (de Brauw et al., 2002). In particular, land use and exchange rights restrictions and tenure insecurity limit participation in the off-farm labour market (Carter and Yao, 2002), which itself is highly segmented and discriminatory against rural migrants (Knight and Song, 2005, ch.8). These failures result in rationing of off-farm labour for rural households, with the adverse consequences on rural incomes dependent upon exposure to constraints (Carter and Yao, 2002).⁵

The link between institutional and market failures and environmental degradation is also clear. Failures in the market for land have caused inefficient agricultural practices in a number of provinces of China (Deininger and Jin, 2003), while limited off-farm opportunities and grain shortages have induced the cultivation of marginal, highly sloped lands (Feng et al., 2004). Inefficient land and labour allocations, e.g. high land-labour ratios (Wang et al., 2005) and low levels of agricultural and land saving investments are also observed (Jacoby et al., 2001). In turn, these household responses have caused large scale externalities at the river-basin level. For instance, cultivation of forested sloping lands in the upper reaches of the Yangtze River is widely blamed for the serious flooding and loss of life downstream in 1998 (Yin and Li, 2001).

The SLCP was implemented in this context as a response to the environmental degradation of 1998. It provides compensation to households for reforesting cultivated sloped land in the upper reaches of the major river basins. Its stated aims speak directly to the notion of a ‘win-win’ solution: to curb land degradation and its consequences and reduce rural poverty (Xu et al., 2004). However, the SLCP compensation, which is the main policy instrument, is temporary. Following the argument above, the long-term success of the SLCP rests upon its ability to address pivotal local market and institutional failures and invoke dynamic mechanisms which ‘push’ participants from constrained to unconstrained equilibria (Dumas, 2007).⁶ Failure to do so will mean that the impact of the SLCP will be at most transitory.

A secondary aim of this paper is, therefore, to examine whether the SLCP has achieved its dual objectives via the relaxation of the constraints that bind on the off-farm labour decisions. The theoretical model predicts that participation in the SLCP relaxes ‘production’ constraints thereby increasing off-farm labour supply and household income, and alleviating pressure to cultivate and degrade marginal sloped land. However, due to heterogeneity in exposure to constraints the impact is predicted to be larger for the ‘off-farm constrained’ than for unconstrained households. The predictions are tested using household panel data from Guizhou and Ningxia provinces. The results show that the impact of the SLCP on household off-farm labour supply differs radically across households depending upon exposure to constraints and is only positive for particular ‘constrained’ households. The results also reveal the nature of household constraints and discuss the effectiveness of the SLCP to act upon them.

The following section describes the SLCP. Section 3 presents the household model and the hypotheses regarding the impact of the SLCP. Section 4 describes the empirical approach and data. Section 5 shows the results and Section 6 concludes.

	2. The Sloping Lands Conversion programme (SLCP)

TOP Abstract 1. Introduction 2. The Sloping Lands... 3. Theoretical model of... 4. Econometric approach 5. Results 6. Conclusion Supplementary material Appendix description of... Acknowledgements References

 
The SLCP is effectively an Payments for Ecological Services scheme (PES) and is by far the largest of its kind in the developing world. The objective is to encourage reforestation of previously converted, mainly sloping land via compensation for changes in land-use practices.⁷ The aim is to convert 15 million hectares of cropland, approximately a third of which will be on land which has a slope of at least 25 degrees (Xu and Cao, 2002).⁸ The principle motivation for this intervention was to address the basin level externalities, including the severe flooding of the Yangtze river basin in 1998, and local externalities, including loss of fertile topsoil and siltation of streams, which have inhibited the productivity of agriculture and the availability of water resources. Deforestation in the upper reaches of the river basins was seen as the cause of these externalities.

The main instrument of the SLCP is the direct compensation of farmers in cash, grain, or seedlings for trees provided by the local forest agencies. Depending upon particular circumstances, SLCP participants receive approximately 100 to 150 kgs of grain per mu per year and an additional Y20 per mu per year in cash.⁹ This means that grain is the major component of the compensation package.¹⁰ Secondly, compensation varies from region to region reflecting local conditions. In the Guizhou and Ningxia, the regions studied in this paper, the level of the annual cash compensation was Y20 and Y12.5 per mu respectively, reflecting distinct opportunity costs of land in each region. Importantly, the SLCP has tended to over-compensate participants (Xu and Cao, 2002). Finally, compensation varies in its duration depending on whether sloped land is converted to ‘ecological’ forest (eight years) or to ‘productive’ forest (five years).¹¹ The rules of the SLCP state that a minimum of 80% of the reforested area in any region must be ecological forest and in our study area the rate is almost 95%.

Participation in the SLCP is in principle voluntary from the perspective of the farmers. However, the SLCP is implemented by local governments and local SLCP implementation agencies which gauge households' suitability and/or make participation involuntary (Xu and Cao, 2002). Our interviews with village leaders show that selection of participants takes on two distinct types in the study areas: either participation is compulsory (60%) or households can volunteer for selection (40%), that is, the implementing agency chooses participants from a self selected pool. Given this selection into the programme is unlikely to be random. This gives rise to some empirical issues which are discussed in Section 4.

	3. Theoretical model of household responses to the SLCP

TOP Abstract 1. Introduction 2. The Sloping Lands... 3. Theoretical model of... 4. Econometric approach 5. Results 6. Conclusion Supplementary material Appendix description of... Acknowledgements References

 
We now develop a household consumer-producer model in the presence of institutional constraints and market failure. The model shows how heterogeneous households can be distinguished according to the severity of the constraints they face. The model is quite general, but the constraints are tailored to the case of the SLCP.

3.1 Basic model
The farm household has preferences defined over income, y, leisure time l^l, and a vector of consumption shifters z^c. The household is endowed with a total amount of time T which is allocated between leisure l^l, on farm work lⁱ and off-farm work l^o, which is remunerated at a wage w^o. The land endowment is assumed to be distributed in parcels of an increasing productivity Agricultural output is produced with a technology: , where represents the amount of land under cultivation. That is, households cultivate their highest productivity lands first and decide on the productivity at the extensive margin, . We make usual assumptions on the production technology,¹² and assume labour and land are complements: q₁₂(.) > 0.¹³

The constraints of the model are now tailored to the conditions in rural China. We allow for the possibility that households can sell some of their production on agricultural markets, at a price p. However, there is evidence in China of large imperfections in the agricultural markets. Park et al. (2002) find that during the reform period between 1988 and 1995, grain markets have not substantially developed as a result of an erratic reform path, with several retrenchments from the liberalization policy,¹⁴ but also because of transaction costs and infrastructure bottlenecks. In interior provinces in particular, market deepening has lagged and autarky rates have actually increased over the period. Burgess (2001) describes how Chinese rural households respond to these adverse market conditions by relying on own production to meet their food requirement.¹⁵ The presence of quotas and agricultural taxes, which have to be paid in kind also impose agricultural production requirements.¹⁶ In addition, uncultivated land faces a high risk of being confiscated and redistributed by the village authorities, a practice known as the ‘use it or lose it’ rule (Deininger and Jin, 2002).¹⁷ Uncultivated land is easily observable to village authorities and other households who may have interest in denouncing perpetrators so as to benefit from land redistribution. This confiscation risk induces households to display that land is useful and potentially produce more than optimal levels of output. In combination, these constraints mean that households must meet a minimum level of production. In addition, since land rental markets are non-existent or poorly developed (Carter and Yao, 2002; Bowlus and Sicular, 2003), households can only cultivate land allocated to them by local government or village leaders.¹⁸

These rigidities are captured in the model by two pivotal constraints: (i) a ‘farm output constraint’: (ii) a ‘land constraint’: ₀. With utility represented by the twice differentiable, concave function: U (y, l_l, z^c), with grain as the numeraire (p = 1), hence , and l^l = T–lⁱ – l^o, the household maximization problem can be represented as follows:

(1)

s.t.

(2)

(3)

(4)

(5)

where µ^c and µ are the Lagrange multiplier associated with constraints (2) and (3) respectively.

We consider only households who work both on- and off-farm. Appendix 2 shows how households' behaviour in the off-farm labour market differs according to whether constraints are binding on their optimizing behaviour. When no constraint is binding on household behaviour, or when only the land constraint is binding, households' production and consumption choices are separable (Singh et al., 1986). We label these household ‘off-farm unconstrained’. For these households the decision wage for labour supply is the market wage.

Alternatively, there are households for whom both the output and, by extension, land constraints are binding. We label these households ‘off-farm constrained’. Here, the decision price in the off-farm labour market is no longer the market wage, but a shadow wage, which depends on the farm output constraint and which is lower than the off-farm market wage. The marginal productivity of on-farm labour of such households is lower than the off-farm market wage and leads to excessive labour allocated on-farm compared to the optimal situation. Surplus on-farm labour is characteristic of rural China (Knight and Song, 2005, ch.8).

Appendix 2 presents reduced form equations of the off-farm labour supply of these two types of household. Importantly, the ‘off-farm constrained’ households supply is affected by the output constraint due to the idiosyncratic shadow decision wage. However, for the separable ‘unconstrained’ households supply only depends on production characteristics and household preferences. This distinction provides the basis for the empirical identification of these heterogeneous labour supply functions (see Section 4).

3.2 The impact of the SLCP
The model illustrates that participation in the SLCP has two effects. Firstly, the land constraint (3) is tightened. The SLCP targets highly sloped lands of the lowest productivity, which were previously cultivated, and imposes with ¹⁹ Secondly, the programme provides subsidies, which are largely distributed in grain, in order to compensate for lost production. Let A be the unit subsidy distributed for every piece of land set aside. The total subsidy is thus: Where over-compensation occurs, grain subsidies act to relax the ‘output’ constraint (2) for those households that are off-farm constrained.²⁰ The household's optimization problem can now be represented as before only replacing (1) and (2) respectively with:

(6)

s.t.

(7)

Comparative static analysis of the solution leads to the following proposition:

Proposition 1
(a) The SLCP generates a positive substitution effect from on-farm to off-farm labour by reducing the amount of cultivated land; (b) the revenue effect of the SLCP subsidies on the off-farm labour supply is either negative, nil, or positive, according to whether households are over, exactly, or under compensated for their loss in agricultural production, respectively; (c) in addition to the substitution and revenue effects, the SLCP grain subsidies relax the ‘output constraint’, which increases the off-farm labour supply of ‘off-farm constrained’ households.

Proof
See Appendix 2.

The intuition for Proposition 1(a) is that participation in the SLCP reduces cultivated land and, since labour and land are complements in the agricultural production function, induces a positive substitution effect from on-farm labour to off-farm labour.²¹ Proposition 1(b) is established from the fact that, if the subsidies exactly compensate households for their loss of agricultural income, leisure does not vary, and the revenue effect is nil. Conversely, if households are over (under) compensated, there is a positive (negative) revenue effect and leisure will increase (decrease) assuming normality. The revenue effect is dominated by the substitution effect if the marginal subsidy, A, is lower than the (real) marginal productivity of land under cultivation.²²

Proposition 1(c) captures the heterogeneity in household responses to the SLCP. In addition to the substitution and revenue effects, the SLCP compensation relaxes the ‘output constraint’. This makes the shadow off-farm wage of constrained households increase towards the market wage, w^o. This generates an increase in their off-farm labour supply. For this reason, the impact of the SLCP will differ for off-farm constrained and unconstrained households in one important dimension: the relaxation of the ‘output constraint’.

	4. Econometric approach

TOP Abstract 1. Introduction 2. The Sloping Lands... 3. Theoretical model of... 4. Econometric approach 5. Results 6. Conclusion Supplementary material Appendix description of... Acknowledgements References

 
The model presented in Section 3 together with Proposition 1 (a)–(c) point to a number of distinct hypotheses for those households that participate in the off-farm labour market:

Hypothesis 1
The off-farm labour supply of unconstrained households depends only on production side characteristics and on the household's preferences, while that of the constrained households depends also on the presence of the ‘farm output constraint’.

Hypothesis 2
If the revenue effect does not offset the substitution effect, participation in the programme should induce an increase in the off-farm labour supply of all households.

Hypothesis 3
The increase in off-farm labour supply should be larger for off-farm constrained than for off-farm unconstrained households.

Hypothesis 4
Participation in the SLCP reduces the probability of being off-farm constrained, since participation in the SLCP may fully relax the production constraint, so that households which were formerly ‘off-farm constrained’ households become ‘off-farm unconstrained’.

An empirical test of these must overcome two important challenges. Firstly, it must accommodate the heterogeneous exposure to constraints identified by the theoretical model. This is particularly difficult since the constraints faced by rural households are numerous and unobservable (Vakis et al., 2005). Secondly, the problem of selection bias must be addressed to identify the treatment effects (e.g. Abadie, 2005). We propose a novel solution to these empirical issues which combines a switching regression with unobserved sample separation with difference in differences (DID). Firstly, the switching regression model accommodates heterogeneity among households by separating the sample into two distinct labour supply equations or ‘regimes’, consistent with the theoretical model. The regimes cannot be labeled constrained or unconstrained a priori. To identify the regimes we exploit the theoretical differences in the labour supply equations for constrained and unconstrained households. Statistical significance of constraint variables in one regime's labour supply equation would indicate the constrained group. This constitutes a test of Hypothesis 1.

Secondly, panel data allows us to control for selection into the SLCP using DID (Abadie, 2005). Combined with the switching regression model we can identify the treatment effect of the SLCP on each regime and hence test whether this effect is heterogeneous. In this way we can test Hypotheses 2 and 3. This empirical approach also allows a direct test of Hypothesis 4.

4.1 Difference in differences (DID)
We use DID to estimate the ‘treatment’ effect or impact of participation in the SLCP on off-farm labour supply.²³ Under suitable assumptions DID controls for endogeneity of the SLCP participation decision running through an unobservable fixed effect, and consequently provides a consistent estimate of the impact of the SLCP for the population supplying labour off farm (Abadie, 2005). Our approach follows work by Gary Chamberlain.²⁴

We specify a reduced form labour supply function as a components of variance model. For each household i (i = 1,...,N) labour supply at time t (t = 1,..., T),l_it, is modelled as a linear function of K – 1 household and village level characteristics (X_it and Z_it respectively), and unobservable components of variance: _i, _t and u_it. The latter represent permanent household specific, time specific, and individual transitory effects respectively, and we make the usual assumptions about them.²⁵ Finally, the dependence of labour supply on participation in the SLCP is captured by the dummy variable D_it, and the scalar represents the impact of the SLCP on the outcome variable l_it:

(8)

where x_it = [X_it, Z_it] is a (K–1) xNT matrix and β' is a 1x (K–1) vector of parameters. For simplicity of exposition, the treatment effect in this model () is assumed identical for all households. This assumption can easily be relaxed by including interactions with D_it. To account for endogeneity and obtain consistent estimates of the parameters of this model we follow Hsiao (1986, p.57) and recast (8) in terms of a T–variate regression for a household i. The method employed is described in Appendix 3 and yields the following model of Difference in Differences:

(9)

From this , as well as the parameters in β and _t, can be consistently estimated using OLS. We use this approach to identify , and the coefficients on interactions with D_it and the time varying explanatory variables we expect to be correlated with _i.²⁶ One assumption underpinning DID is the equality of the trend in the error terms over time between treatment and control groups. One benefit of the approach used here is that time varying parameters are estimated, which capture some of the variation over time that would usually appear in the error terms (Abadie, 2005).

4.2 Switching regression using DID
As discussed above, households face idiosyncratic institutional and market failures. Consequently, it would be unusual for all households to be constrained in any given region and, if they are, for all constraints to be equally important for every household. In addition to this, the cause of constraints is not always observable (Carter and Yao, 2002; Vakis et al., 2005). In this setting, and given the theoretical model of Section 3 the econometric model of labour supply decisions accommodates the following features: (i) a reduced form approach; (ii) the source of separability is not confined to a single market failure; (iii) recognition of heterogeneity across households with respect to the nature and extent of constraints; (iv) unknown sample separation between constrained and unconstrained households. These features of the problem point to a switching regression with unobserved sample separation as the most suitable econometric framework (Hartley, 1978).

The model belongs to the family of mixture-distribution models that aim to ‘unmix’ the sample by simultaneously identifying the stochastic structures governing the separation of the sample into two latent regimes while explaining the behavioural decisions of each observation in the regimes (Hartley, 1978). However, what distinguishes our approach from previous applications (e.g. Vakis et al., 2005) is that the regime regression equations are of the form shown in eq. (9). This innovation means that not only are points (i)-(iv) above accommodated, but we can also account for endogeneity of the variables of interest by exploiting the panel nature of the data. In particular, under the assumptions of the DID model, we can consistently estimate treatment effects in each equation of the switching regression.

The model uses the following system of equations to define household behaviour:

(10)

(11)

where j = 1, 2. l represents the latent off-farm labour supply of two heterogenous groups or regimes of households and _it is a latent variable that determines sample separation. The error terms are assumed to be normal i.i.d disturbances with zero means and variance (j =1, 2) with = 1 for identification. For each household i in each time period t we only have data on the observable counterpart of l such that:

(12)

Given that we cannot observe the regime classification, each randomly selected household i will have a probability 1– = (–x β) of belonging to the first regime and probability of belonging to the second. The probability density function of each observation is hence given by the mixture of two distributions. The resulting likelihood function is maximized with respect to the parameters using the E-M method as articulated by Hartley (1978).

To interpret the empirically defined regimes we refer to the theoretical model. Equations (24) and (30) in Appendix 2 show: where for the ‘off-farm unconstrained’ households, and according to (38): for the ‘off-farm constrained’ households. The difference between these groups is the presence of the constraint variables, C, in the latter. The constrained regime could then be identified if the variables associated with the production constraint are significant in one regime and not the other. This would be an indication that our sample represents observations drawn from two distinct samples and would be a test of Hypothesis 1. One feature common to each equation is the treatment effect, ^j and . If there is support for Hypothesis 1 then Hypotheses 2 and 3 can be tested by inspection of the estimates of ¹ and ², while Hypothesis 4 can be tested by inspection of the estimate of in the switching equation, which also yields information about the determinants of being in one regime or another.

In order to identify the parameters of interest in this model it is assumed that the error terms are independent across equations. This assumption means that the disturbance term in (11), which affects the probability of falling into the first regime, is independent of that affecting the continuous labour supply decision.²⁷

4.3 Data
The data set contains information on 286 households in Ningxia (155) and Guizhou (131) collected in the summer of 2004.²⁸ Forty village leader questionnaires were also undertaken. Two years of data were collected for the pre- and post-programme periods, respectively 1999 and 2003, for participants and non-participants. Table A1 in Appendix 1 presents descriptive statistics of the variables included in the switching regression. Our interest is in the constraints on households participating in the labour market, and for this reason we estimate the model on a sub-sample of households that supply off-farm labour both pre- and post-programme.²⁹ This reduces the number of households to 159, 30% of which are non-participants in the SLCP.

Two main constraints have been identified in the literature on rural China: the incompleteness of local labour and agricultural markets, and institutional constraints (Burgess, 2001; Carter and Yao, 2002; Bowlus and Sicular, 2003). Our theoretical model suggests that household size and composition will influence off-farm labour supply, not only through the labour endowment effect, but also through its influence on the quantity of labour that needs to be allocated to cultivation in order to meet quotas and family food requirements in the context of imperfect agricultural markets (Burgess, 2001). The presence of young children and the elderly, who do not contribute to the labour endowment but represent extra mouths to feed, are included in the analysis to capture the latter effects. As an indicator of transaction costs we use distance to a main road to capture the ease of access to agricultural and labour markets. Institutional constraints are captured by the development of land rentals within the village (Deininger and Minten, 1999) and by the distance from the village to the nearest credit agency (Key et al., 2000), while the level of tenure security is proxied by soil quality at the village level. Soil quality is found to be highly correlated with tenure security in the literature. Frequently this is because the disincentive effect of land reallocations on farmers' investment incentives results in land degradation (Jacoby et al., 2001; Rozelle et al., 2004). Unfortunately, the approach described in Section 4.1 fails to deal with the endogeneity problem inherent in household level institutional constraints, since these data are time invariant.³⁰ Consequently, we follow Carter and Yao (2002) and Deininger and Minten (1999) and include village level variables to proxy for such institutional constraints and mitigate the endogeneity problem that arises from using household level variables.

Estimation of off-farm labour supply requires reasonably complete and accurate data on wages. However, our wage data is unable for three reasons. Firstly, our wage data does not vary over time, which may indicate that people are mis-reporting wages. Secondly, employment in Township and Villages Enterprises (TVEs) often provides in-kind payments, so that the comparison of self-employed earnings with TVE wages may hide significant differences. Thirdly, many households do not provide any information on wages. Since the type of off-farm labour opportunities in the region is likely to drive the main differences in off-farm earnings, using a regional dummy mitigates difficulties arising from a complex wage structure. Education is used as a household level proxy for off-farm wages.³¹ The panel structure of our data allows to control for household specific differences in labour quality that may influence off-farm earnings and are constant over time.

4.4 Treatment variables
We include three treatment variables: the treatment dummy D_it which we label treatment, along with two interactions: (i) treatment * outdum which is an interaction of treatment with labour supplied outside of the village as opposed to within, measured by the indicator outdum (a dummy variable equal to one if labour is supplied outside the village, and zero otherwise), and; (ii) treatment * region, an interaction effect of treatment with a region dummy (region). The former is included to capture differing impacts of the SLCP on labour supplied outside the village compared to within, since evidence suggests that different constraints are likely to affect each (Guang and Zheng, 2005).³² The latter is included to account for regional differences, including the fact that the SLCP's compensation differs across regions.

The coefficients on each of these variables capture heterogeneity in the treatment effect. As discussed above, selection into SLCP is likely to be non-random. In addition to this, the decision to work outside of the village and any interaction with this is also likely to be endogenous. Identification of the parameters associated with these variables is obtained using the approach described in Section 4 and Appendix 3. Just as the DID approach allows any type of selection into the programme based on the household fixed effect, _i, endogeneity through the fixed effect of outdum and treatment* outdum is also dealt with using this approach.

	5. Results

TOP Abstract 1. Introduction 2. The Sloping Lands... 3. Theoretical model of... 4. Econometric approach 5. Results 6. Conclusion Supplementary material Appendix description of... Acknowledgements References

 
5.1 Household heterogeneity
Table 1 shows the results of the switching regression for the separate regimes of the off-farm labour market (columns 3 and 4), as well as the initial pooled regression for comparison purposes (column 2).³³ The switching regression separates the sample into two distinct regimes. Sample separation is unobserved and hence the interpretation of the regimes can only be made by reference to the results of the component regressions in Table 1 and the ‘switching’ equation shown in Table 2. Table 1 shows that the two regimes have an interpretation that is consistent with our theoretical model.

View this table: [in this window] [in a new window]   Table 1 Switching Regression: Regime Regressions

 

View this table: [in this window] [in a new window]   Table 2 Switching equation

 
In Regime 1 the variables that were included to reflect market failures and institutional constraints, the so-called ‘constraint’ variables, are found to be highly significant. The model predicted that only the off-farm labour supply of constrained households is influenced by these variables. For such constrained households, the presence of children under 16 years of age (child) and elderly household members (elderly) reduce off-farm labour supply, as expected. The remoteness from credit agencies (credit constrained) significantly reduces off-farm labour supply. The development of the land rental market (rentease) significantly increases off-farm labour supply, by over 127 days per year in our sample. Ease of land rental is, indeed, the most economically important constraint in the determination of off-farm labour supply, followed by the presence of elderly family members. These findings are coherent with previous evidence of a strong impact of land transfers possibilities and tenure security on the development of off-farm labour in rural China, as well as the influence of credit constraints on production decisions of Chinese rural households.³⁴

Conversely, in Regime 2 the only significant variables are those which reflect the structure of the local labour market and the household factor endowments. Indeed, only the regional dummy (region), the dummy indicating the destination of off-farm labour (outdum), land area (land), and household size are significant. Importantly the constraint variables are insignificant. The theoretical model showed that this is a feature of off-farm unconstrained households. We thus reject the null hypothesis of homogeneous household behaviour in favour of Hypothesis 1 and conclude that households are heterogeneously exposed to the presence of market failure and institutional constraints. The separation into regimes is also seen to be robust when we compare the outcome under the two regimes to the pooled regression in column 2 of Table 1. Many of the explanatory variables in the pooled regression are insignificant and yet the unobserved sample separation shows that ignoring household heterogeneity in our sample masks important behavioural differences.

Descriptive statistics for each of the two groups in Table A2 in Appendix 1 confirm that sample separation is consistent with our hypotheses. Households that are subject to constraints supply labour off-farm at a shadow wage, which is lower than the market wage. Their off-farm labour supply should consequently be lower than that of unconstrained households. Indeed, the average off-farm labour supply of households belonging to the constrained group is 302 days per year, well below the average of 433 days per year for unconstrained households.³⁵

5.2 The treatment effect of the SLCP
Estimates of the average impact on the treatment group, or Average Treatment on the Treated (ATT) , of the SLCP are contained in Tables 1 and 2. In Table 1, all three treatment effects (coefficients on treatment, treatment * region, and treatment * outdum) are significant at the 1% level in the constrained regime, while none is significant in the unconstrained regime. For the constrained households, the impact of the SLCP is large, positive and significant at the 1% level, for the base group, inducing an extra 194 days per household per annum on average. This represents an increase of 56% with respect to pre-programme off-farm labour supply of SLCP participants. For the unconstrained regime, the treatment effect is negative but not statistically significant. However, for constrained households, programme participation has a negative, significant, and large impact on the off-farm labour supply of those who supply their labour outside of the village. This negative impact offsets the treatment effect of the programme, so that the overall effect of the programme on such households is not statistically different from zero. The regional treatment effect (treatment* region) is also significant and negative, indicating that the treatment effect on households in the region of Guizhou is negative. However, the overall treatment effect of the programme on households in Guizhou is not statistically different from zero. For those households in Guizhou that supply their labour outside the village the treatment effect is negative, but not statistically different from zero even at a 10% level.

The impacts are clearly heterogeneous. In some cases we reject the null hypothesis in favour of Hypothesis 2: the substitution effect of programme participation offsets the revenue effect and the overall effect of the SLCP is to induce participating households to reallocate their labour towards off-farm employment. Similarly, we reject the null in favour of Hypothesis 3: the impact of the programme on off-farm labour supply is larger for constrained than for unconstrained households. This implies that part of the effect of programme participation operates through the constraints that impede household behaviour. However, the labour reallocation which results from programme participation is realized at the village level, the programme having a negative impact on labour which is supplied outside the village, despite off-farm activities being more lucrative outside the village than within.³⁶ We can thus conclude that the SLCP does appear to relax local constraints that were forcing households to supply excess labour on land, but does not address, and may even sharpen, the constraints on the off-farm labour market itself, namely those which restrict migration outside of the locality. Anecdotal evidence suggests that one reason for this may be that the presence of the household is required by program participation, for example because participating households have to monitor reforested areas. We now turn to the nature of local constraints, and the impact of the programme upon them.

5.3 The nature of the constraints
Table 2 provides some indication of the nature of the constraints generating the division between households.³⁷ The ‘switching’ equation corresponds to eq. (1), indicating the impact of the explanatory variables on the probability of being in Regime 1 and therefore the probability of being constrained.³⁸ First of all, we reject the null hypothesis in favour of Hypothesis 4: participation in the programme significantly reduces the probability of being constrained. The results suggest that in addition to participation in the SLCP, household education (education), family size ( familysize), as well as development of land transfers (rent) and better soil quality (indicating more secure tenure) reduce the probability of being constrained, as does living in the Guizhou region. Conversely, arable land endowment (land), distance to the nearest road (distroad) and to the nearest credit agency (credit dist), presence of children under 16, and elderly members of the household (child and elderly respectively) increase the probability of being in the constrained regime. The results show that there are a number of important market failures and transactions costs underpinning the constrained households.³⁹

As seen elsewhere in China, inadequate access to credit, difficulties in land rental and insecure tenure are prominent market and institutional failures in the areas in question.⁴⁰ The effect of the area of cultivated land (land) further points to a link between off-farm labour supply and the constraints imposed on households behaviour by imperfect land rights. Indeed, the rather surprising fact that more land increases the probability of being constrained (land has a positive coefficient in the switcher equation) confirms that having more land increases the burden of constraints and compels households to attach more labour to cultivation. As a consequence, the availability of labour for more lucrative off-farm opportunities is reduced, as illustrated by the impact of land rental rights on such constrained households' off-farm labour supply. Here, more developed land rights decrease the probability of being constrained, hence contributing to allow households to supply freely their labour off-farm and enhance the efficiency of their labour allocation choices (Carter and Yao, 2002). The impact of the presence of young children or elders, which increases the probability of being in the constrained regime, reflects the fact that it is traditional in China (and imposed by the lack of a social security system) for care for the elderly to provided by younger family members.⁴¹

Predictions from the ‘switcher’ equation yield the probability that a household belongs to a particular regime and allows us to identify constrained and unconstrained households in the sample.⁴² Comparing the characteristics of constrained versus unconstrained households points to policy recommendations related to programme targeting, and allows to check that no single source of constraints determines sample separation. Confirmation of the last point validates the use of a switching regression, one of the main advantages of which is that it does not confine the source of constraint to any specific institutional or market failure.⁴³

As we can see from Table A2 in Appendix 1 constrained households differ significantly from unconstrained households in many respects. This indicates that no single variable drives the partition of the sample, and this confirms our motives for using the switching regression. Constrained households supply significantly less off-farm labour, are more likely to live in Ningxia, have a larger land endowment, exhibit lower levels of education and are further from main roads and formal credit agencies. These households are also more likely to live with young children and elderly family members than unconstrained households. The institutional environment of these two groups of households also differs significantly. Rental rights are more likely to be prohibited and soil quality to be lower (indicating less tenure security), for those households that are identified as being constrained. However, the propensity of participating in the SLCP does not significantly vary across groups. This may signal that the targeting of the SLCP was inadequate, as is further discussed in the conclusion.

	6. Conclusion

TOP Abstract 1. Introduction 2. The Sloping Lands... 3. Theoretical model of... 4. Econometric approach 5. Results 6. Conclusion Supplementary material Appendix description of... Acknowledgements References

 
Several conclusions and policy implications can be derived from this analysis. Firstly, our methodology and results highlight the importance of accommodating heterogeneity in the evaluation of interventions in a context where households face idiosyncratic institutional and market failures. Applying traditional policy evaluation methods, which treat all households as a homogeneous group, would yield erroneous policy implications. This is demonstrated in column 2 of Table 2, which indicates that a global policy approach confounds the impact of the programme and would lead us to conclude that the impact of the SLCP on off-farm labour supply was insignificant. It is widely acknowledged that the ability of the programme to produce a sustainable ‘win-win’ outcome rests upon its ability to enable households to access alternative employment opportunities (Xu et al., 2004; Uchida et al., 2007). Yet, our evaluation of the SLCP is the first to show a significant impact of the SLCP on off-farm labour supply, albeit one limited to local activities in Ningxia. Nevertheless, this shows that accommodating heterogeneity in exposure to constraints can reveal heterogeneous impacts.

Secondly, our analysis brings important policy recommendations with respect to programme targeting. The results imply that in order to improve the cost effectiveness of the programme subsidies should target constrained households. This is a common problem for poverty alleviation programmes (Besley and Coate, 1992). The analysis of the characteristics of the constrained and unconstrained households, as identified by the switching equation in Table A2, leads us to conclude that the programme should focus not only on households with large land endowments, as is currently the case, but should also consider the education level, household structure, and the institutional environment of recipient households. The compensation package offered by the SLCP at present is too uniform: it only accounts for two different levels of grain subsidy on the basis of gross regional averages, not on the basis of households characteristics. However, a more flexible design on the basis of household and land characteristics may give way to rent seeking by households or by local authorities, which are countered to some extent by current arrangements (Xu et al., 2004). Targeting the programme on household characteristics may, thus, be complicated and costly, but the results here indicate two potential improvements to the current situation. Firstly, policy design should embody adequate screening mechanisms to counter issues of adverse selection. For example, introducing work requirements for programme participants could induce constrained households to self-select into the programme (Besley and Coate, 1992). The SLCP arranges for such requirements on tree planting and maintenance, but this clause was often misused by local authorities: ad hoc teams were appointed which diverted programme subsidies away from SLCP participants (Xu et al., 2004). Secondly, our analysis suggests that the objectives of the programme could be achieved more efficiently, by alleviating institutional constraints on the land exchange market, on tenure security or on the credit market. These constraints remain the major impediments to labour reallocation and drive important behavioural differences between constrained and unconstrained households. The last implication of our analysis is that accompanying policies should focus on such constraints.

In this regard, Chinese central authorities appear to be moving in the right direction. Indeed, the National Peoples Congress announced in March 2005 that all agricultural taxes were to be removed progressively.⁴⁴ Also, the household registration system, or Hukou, will be lifted in 11 regions, which will facilitate migration to cities.⁴⁵ Concerning institutional constraints, in 2002, the National People's Congress adopted the Rural Land Contracting Law, which reaffirmed rural households' land use rights and their rights to transfer, exchange, and assign their land use rights to other households (Wang et al., 2005). Nevertheless, how these changes and reforms will affect the rural poor depends greatly on the implementation will and capacity of local authorities.

	Supplementary material

TOP Abstract 1. Introduction 2. The Sloping Lands... 3. Theoretical model of... 4. Econometric approach 5. Results 6. Conclusion Supplementary material Appendix description of... Acknowledgements References

 
Supplementary material (the Appendix) is available online at the OUP website.

	Appendix description of explanatory variables and regime characteristics

TOP Abstract 1. Introduction 2. The Sloping Lands... 3. Theoretical model of... 4. Econometric approach 5. Results 6. Conclusion Supplementary material Appendix description of... Acknowledgements References

 

Table A1 Descriptive statistics

Variable Survey source Mean Std. Dev. Off farm labour Hshld off-farm labour supply (hrs) Household 380.1 284.6 Off farm labour (t = 0) Hshld off-farm labour supply (hrs) pre-SLCP Household 335.5 271.9 Off farm labour (t = 1) Hshld off-farm labour supply (hrs) post-SLCP Household 402.1 287.0 Region =1 if Guizhou Household 0.44 0.50 Outdum =1 if work outside village Household 0.68 0.46 OutPRE (t = 0) =1 if work outside village pre-SLCP Household 0.70 0.48 OutPOST (t = 1) =1 if work outside village post-SLCP Household 0.65 0.45 Treat Treatment Household 0.38 0.49 TreatReg Treatment * Region Household 0.17 0.37 TreatOut Treatment * Work out of village Household 0.26 0.44 Land Arable land (mu) Household 15.8 15.5 Education Hshld education level per capita (yrs) Household 2.29 0.93 HH size Household Size Household 5.29 1.59 (HH size)2 Household Size Squared Household 30.5 20.3 Elderly =1 if hshld has elderly members Household 0.11 0.32 Child =1 if hshld has children Household 0.46 0.50 Rentease =1 if land rental unrestricted Village 0.80 0.40 Credit constrained Dist to formal credit institution (km) Village 5.45 4.32 Soil quality Degree of soil erosion Village 0.34 0.47 Distance Dist to nearest town in (km) Village 4.87 3.95

Table A2 Comparison of constrained and unconstrained households

Constrained Unconstrained Unconstrained-constrained Variables Mean s.d. Mean s.d. Difference t-stat p-value OffLabour 302.3 18.7 433.4 22.7 131.1 4.6 0.0 Region 0.2 0.03 0.6 0.04 0.4 8.5 0.0 Treat 0.3 0.04 0.4 0.04 0.09 1.7 0.09 TreatReg 0.05 0.02 0.2 0.03 0.2 5.5 0.0 TreatOut 0.3 0.04 0.3 0.03 0.01 0.09 0.9 Outdum 0.63 0.46 0.73 0.55 0.01 1.9 0.05 Land 23.6 1.8 10.4 0.6 –13.2 –7.2 0.0 Education 1.9 0.048 2.5 0.08 0.6 6.6 0.0 HH size 5.3 0.1 5.2 0.1 –0.1 –0.5 0.6 Elderly 0.2 0.03 0.07 0.02 –0.1 –2.8 0.01 Child 0.6 0.04 0.4 0.04 –0.2 –2.7 0.01 Rentease 0.7 0.04 0.9 0.02 0.2 4.1 0.0 Credit constrained 7.0 0.4 4.5 0.3 –2.6 –4.9 0.0 Soil quality 0.2 0.04 0.4 0.04 0.2 3.5 0.0 Distance 5.9 0.4 4.2 0.3 –1.7 –3.6 0.0

	Acknowledgements

TOP Abstract 1. Introduction 2. The Sloping Lands... 3. Theoretical model of... 4. Econometric approach 5. Results 6. Conclusion Supplementary material Appendix description of... Acknowledgements References

 
The authors would like to thank the China Council for International Co-operation on Economic and Development and the UK's Department for International Development for financial assistance. Pauline Grosjean would like to thank the Ciriacy Wantrup foundation for financial support. We would also like to thank Emmanuelle Auriol, Jean Paul Azam, Tim Besley, Phoebe Koundouri, the late David Pearce, Elisabeth Sadoulet, Paul Seabright, Alban Thomas, Renos Vakis, Jeremy Warford, the Toulouse development economics seminar, and two anonymous referees for their many helpful comments. The usual disclaimer applies.

	Notes

 
¹Other examples include missing local credit and labour markets (Ellis, 1998), weak property rights (e.g. Carter and Olinto, 2003) or the absence of local public goods.

²Ellis (1998) states: ‘removal of constraints to, and expansion of opportunities for, diversification are desirable policy objectives because they give individuals and households more capabilities to improve livelihood security and to raise living standards’.

³The assumption of global rather, than local separability can mask the impact of interventions on particular types of household (Singh et al., 1986; Carter and Yao, 2002).

⁴Poverty reduction in China over the last two decades was largely achieved through increases in rural incomes including off-farm income (Park et al., 2002; Bowlus and Sicular, 2003; Xu et al., 2004).

⁵Knight and Song (2005, ch.8) estimate that, on average, returns to off-farm employment are still 50% higher than those of on-farm employment. Our data show that these earning differentials can reach 250%.

⁶If risk aversion underpins failure to exploit on and off-farm wage differentials, those ‘pushed’ by the SLCP may not return.

⁷The environmental costs of cultivation also extend to airsheds. The increased incidence of dust-storms in the Northern plains, and the associated loss of topsoil, has been attributed to the extensive cultivation of natural grasslands. The SLCP has also targeted these flatter areas to return the land to its natural grassland state (Xu and Cao, 2002).

⁸Since the policy commenced in 1999, approximately 15 million farmers have become participants in 20 provinces and over 27,000 villages (Uchida et al., 2007).

⁹1 hectare = 15 mu.

¹⁰At the government price in 1999 of RMB1.4/mu, the grain subsidy is equivalent to 140–210 RMB/mu per year. At this price, the proportion of the grain versus cash subsidy is thus 7–10.5 (Xu et al., 2004).

¹¹With ‘ecological’ forest farmers have no rights to forest products. For ‘productive’ forests participants have rights to collect non-timber forest products (e.g. fruits, nuts, mushrooms, and limited quantities of timber) hence compensation lasts for a shorter period of up to five years (Xu and Cao, 2002).

¹² q₁(.) > 0, q₂(.) > 0, q₁₁(.) < 0, q₂₂(.) < 0. Where: and

¹³Where, This assumption is motivated below.

¹⁴The government continued to intervene strongly in grain markets, namely in periods of sharp grain price inflation in 1988–99 and 1994–95. Despite the presence of some private grain traders, state-owned grain enterprises (SOEs) have dominated the market at least until 2004, when new regulations on grain circulation have been adopted by the state council (China Daily, Xinhua, 2004-06-08).

¹⁵Key et al. (2000) describe how some households may opt out of the market and remain in autarky in presence of a wedge between buying and selling prices driven by transaction costs.

¹⁶The plan for most agricultural taxes is that from 2006 is for incremental reductions. They were still in force at the time of our survey however. Only few villages allow for cash payments.

¹⁷Indeed, although land readjustments have been either circumscribed or completely prohibited by the 1999 revised Land Management Law for a 30 year period, legal provisions have not always translated into effective tenure security.

¹⁸Again, this is reflected in our data in which only 12% of households rent land in or out. Only 7% of all cultivated land is involved in such transfers. Furthermore, the majority of such rental exchanges is informal and generally does not give rise to any monetary compensation (Bowlus and Sicular, 2003).

¹⁹It is assumed that at this stage, once the household has been chosen to participate in the program, the amount of land enrolled is not a decision variable for the household. In addition, we assume that implying that the programme is supposed to target only land which was previously cultivated, and this is observable so that is enforceable. This is a plausible assumption given that in our sample all land is used in cultivation. Lastly, given that the SLCP targets highly sloped land it is plausible to assume that the aim is for low productivity land to be reforested.

²⁰As described above, the SLCP tends to over-compensate participants.

²¹Given that this assumption is pivotal to the predictions we make in the following sections, we have estimated the agricultural production function using a multi-output distance function approach. We estimated a Trans-log production function in the two main outputs (wheat and potatoes) in land, household labour, and fertilizer. The cross partial of land and labour was positive (0.12) and significant at the 5% level.

²²The implicit function theorem yields:

²³We refrain from using the term Average Treatment on the Treated (ATT) since this implies the average impact on the population of potential SLCP participants. While similar in spirit, we estimate the treatment effect for the subsample of households who supply positive amounts of labour off farm in both periods. We thank an anonymous referee for this point of clarification.

²⁴Abadie (2005, p.2–5) provides a succinct explanation of the DID method which effectively draws on Chamberlain's work which is discussed in detail in Hsiao (1986, p.57).

²⁵These are (Hsiao, 1986):

²⁶We only include in the projection those variables considered to be endogenous. This allows us to identify parameters on exogenous but time invariant variables (e.g. Abadie, 2005, p.5).

²⁷This assumption is an integral part of the switching regression model with unobserved sample separation of Hartley (1978). The assumption is routinely employed in the applied labour economics literature (e.g. Dickens and Lang, 1985; Vakis et al., 2005).

²⁸The SLCP survey was administered by moderators from Beijing University.

²⁹Consequently, we do not deal with censored data. This approach has also been taken by Vakis et al. (2004). Groom (2005) analyses the decision to participate in the off-farm labour market and finds that the average treatment on the treated is positive and significant.

³⁰Household level perceptions of tenure security, rental rights or access to credit can be endogenous to the off-farm labour supply decision.

³¹For the same reasons, a similar approach is employed and found to be robust in other studies of the rural labour market in China (Lohmar et al., 2000).

³²Guang and Zheng (2005) show that there a numerous factors, including the relationship to local party cadres and hence power, which can determine whether a household migrates outside of the village or works in off-farm in local enterprises.

³³In both tables, *indicates significance at 10%, ^**significance at 5%, and ^***significance at 1%. Standard errors are robust.

³⁴See Carter and Yao (2002); Deininger and Jin (2003); and Bowlus and Sicular (2003).

³⁵A t-test shows that this difference is statistically significant.

³⁶Our, albeit imperfect, data on wages indicates that average earnings outside the village are 50% higher than within the village.

³⁷The table reports robust standard errors.

³⁸Note, the coefficients do not represent pure marginal effects on the probability.

³⁹The dependent variable in the switcher equation begins with a guess of the classification. In our case this was a binary variable reflecting incomes higher than (1) or lower than (0) averagea income. In subsequent iterations the dependent variable becomes the probability of being in Regime 1. The results are robust to the initial guess.

⁴⁰E.g. Carter and Yao (2002), Deininger and Jin (2003), Bowlus and Sicular (2003).

⁴¹Many households in our survey cited care for the elderly as an important constraint to finding off-farm labour.

⁴²We define constrained households whose probability of being in the constrained regime is greater that the mean probability in the sample: 0.66.

⁴³For instance, Carter and Yao (2002) focus on the land exchange rights, Bowlus and Sicular (2003) consider only the differences in land endowments or the development of the off-farm labour market, in exclusion of each other.

⁴⁴Source: People's Daily Online, March, 05, 2005; China Daily, March, 06, 2005.

⁴⁵Source: China Daily, 25/11/05; BBC 10/11/2005.

	References

TOP Abstract 1. Introduction 2. The Sloping Lands... 3. Theoretical model of... 4. Econometric approach 5. Results 6. Conclusion Supplementary material Appendix description of... Acknowledgements References

 

Abadie A. Semi-parametric difference in differences. Review of Economic Studies (2005) 72:1–19.[CrossRef][Web of Science]

Besley T, Coate S. Workfare versus welfare; incentive arguments for work requirements in poverty-alleviation programs. The American Economic Review (1992) 82:249–61.[Web of Science]

Bluffstone RA. The effect of labour market performance on deforestation in developing countries under open access: an example from rural Nepal. Journal of Environmental Economics and Management (1995) 29:42–63.[CrossRef][Web of Science]

Bowlus AJ, Sicular T. Moving towards markets? Labour allocation in rural China. Journal of Development Economics (2003) 71:561–83.[CrossRef][Web of Science]

Burgess R. Land and welfare: theory and evidence from China, working paper. (2001) London: London School of Economics.

Carter M, Yao Y. Local versus global separability in agricultural household models: The factor price equalization effect of land transfer rights in China. American Journal of Agricultural Economics (2002) 84:702–15.[CrossRef][Web of Science]

De Brauw A, Huang J, Rozelle S, Zhang L, Zhang Y. The evolution of China's rural labour markets during the reforms. Journal of Comparative Economics (2002) 30:329–53.[CrossRef][Web of Science]

De Janvry A, Sadoulet E, Zhu N. The role of non-farm incomes in reducing rural poverty and inequality in China. (2005) Working Paper Series, Department of Agricultural and Resource Economics, University of California, Berkeley.

Deininger KW, Minten B. Poverty, policies, and deforestation: the case of Mexico. Economic Development and Cultural Change (1999) 47:313–44.[CrossRef][Web of Science]

Deininger KW, Jin S. The impact of property rights on households' investment incentives, risk coping and policy preferences. Economic Development and Cultural Change (2003) 51:851–82.[CrossRef][Web of Science]

Dickens WT, Lang K. A test of dual labour market theory. American Economic Review (1985) 75:792–805.[Web of Science]

Dumas C. Why do parents make their children work? A test of the poverty hypothesis in rural areas of Burkina Faso. Oxford Economic Papers (2007) 59:301–29.[Abstract/Free Full Text]

Ellis F. Household strategies and rural livelihood diversification. Journal of Development Studies (1998) 35:1–38.

Feng Z, Yang Y, Zhang Y, Zhang P, Li Y. Grain for green policy and its impact on grain supply in West China. Land Use Policy (2004) 22:301–12.[CrossRef][Web of Science]

Groom B. The Sloping Lands Conversion Programme: impacts on income, income distribution and poverty alleviation, Applied Economics of Resource Conservation. (2005) PhD dissertation (Ch. 5), Department of Economics, University College London.

Guang L, Zheng L. Migration as the second best option: local power and off farm employment. The China Quarterly (2005) 181:22–45.[CrossRef]

Hartley MJ. Estimating mixtures of normal distributions and switching regressions: comment. Journal of the American Statistical Association (1978) 73:738–41.[CrossRef][Web of Science]

Hsiao C. Analysis of Panel Data (1986) Cambridge: Cambridge University Press.

Jacoby HJ, Li G, Rozelle S. Hazards of expropriation, tenure insecurity and investment in rural China. The American Economic Review (2001) 92:1420–47.

Key N, Sadoulet E, de Janvry A. Transaction costs and agricultural household supply response. American Journal of Agricultural Economics (2000) 82:245–59.[CrossRef][Web of Science]

Knight JB, Song L. Towards a labour market in China (2005) Oxford University Press. Studies on Contemporary China.

Liu S, Carter M, Yao Y. Dimensions and diversity of property rights in rural China: dilemmas on the road to further reform. World Development (1998) 26:1789–806.[CrossRef][Web of Science]

Lohmar B, Rozelle S, Zhao C. The rise of rural labour markets in China. (2000) Working paper No. 1029, Department of Agricultural and Resource Economics, University of California, Davis.

Park A, Wang S, Wu G. Regional poverty targeting in China. Journal of Public Economics (2002) 86:123–53.[CrossRef][Web of Science]

Rozelle SJ, Taylor E, de Brauw A. Migration, remittances, and agricultural productivity in China. American Economic Review (1999) 89:287–91.[Web of Science]

Shively G, Pagiola S. Agricultural intensification, local labour markets, and deforestation in the Philippines. Environment and Development Economics (2004) 9:241–66.[CrossRef][Web of Science]

Singh I, Squire L, Strauss J. Agricultural Household Models: Extensions, Applications and Policy (1986) Baltimore: John Hopkins Press for the World Bank. 153–82.

Uchida E, Xu J, Xu Z, Rozelle S. Are the poor benefiting from China's land conservation program? Environment and Development Economics (2007) 12:593–620.[CrossRef][Web of Science]

Vakis R, Sadoulet E, de Janvry A, Cafiero C. Testing for separability in household models with heterogeneous behaviour: a mixture model approach. (2005) CUDARE Working Paper No. 990, Department of Agricultural and Resource Economics, University of California, Berkeley.

Wang X, Han H, Bennett J. China's land use management. (2005) Australian Center for International Agricultural Research Project: ADP/2002/021.

Xu J, Cao Y. The sustainability of converting the land for forestry and pasture. International Economic Review (2002) 3–4:56–60.

Xu Z, Bennett M, Tao R, Xu J. China's sloping land conversion programme four years on: current situation, pending issues. International Forestry Review (2004) 6:317–26.[CrossRef][Web of Science]

Yin H, Li C. Human impacts on floods and flood disasters on the Yangtze River. Geomorphology (2001) 41:105–9.[CrossRef][Web of Science]

CiteULike    Connotea    Del.icio.us    What's this?

This Article Abstract Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Alert me to new issues of the journal Add to My Personal Archive Download to citation manager Request Permissions Google Scholar Articles by Groom, B. Articles by Zhang, S. Social Bookmarking          What's this?

Online ISSN 1464-3812 - Print ISSN 0030-7653

Copyright © 2009 Oxford University Press

Oxford Journals Oxford University Press

• Site Map
• Privacy Policy
• Frequently Asked Questions

Other Oxford University Press sites: Oxford University Press Oxford Journals China Oxford Journals Japan American National Biography Booksellers' Information Service Children's Fiction and Poetry Children's Reference Corporate & Special Sales Dictionaries Dictionary of National Biography Digital Reference English Language Teaching Higher Education Textbooks Humanities International Education Unit Law Medicine Music Online Products Oxford English Dictionary Reference Rights and Permissions Science School Books Social Sciences Very Short Introductions World's Classics