An Estimation of the Effects of China’s Priority Forestry Programs on Farmers’ Income

Abstract

In the late 1990s, the Chinese government initiated some new programs and consolidated other existing ones of ecological restoration and resource development in its forest sector, and renamed them as “Priority Forestry Programs,” or PFPs. They include the Natural Forest Protection Program (NFPP), the Sloping Land Conversion Program (SLCP), the Desertification Combating Program around Beijing and Tianjin (DCBT), the Shelterbelt Development Program (SBDP), and the Wildlife Conservation and Nature Reserve Development Program (WCNR). In addition to improving the environmental and resource conditions, a frequently reiterated goal of these PFPs is to increase rural households’ income, therefore discussing why looking at rural household income impacts might be an important part of forest program evaluation. Thus, an interesting and important question is: How has implementing the PFPs affected the farmers’ income and poverty status? This article addresses this question using a fixed-effects model and a panel dataset that covers 1968 households in four provinces for ten consecutive years (1995–2004). The empirical evidence indicates that their effects are mixed. The SLCP, the SBDP, and the NFPP have made positive impact and, by far, the SLCP has the largest effect. But the WCNR and the DCBT still have not had a pronounced overall effect due to their short time span of execution, even though they may have exerted certain influence at the margin. Notably, the impact of the WCNR, if any, is negative.

Introduction

With an eightfold increase of per capita gross domestic product (GDP) since 1978, China has witnessed tremendous economic growth over the past three decades (Zheng and others 2008). Accordingly, Chinese population living in absolute poverty (below a per capita annual income of 637 Yuan in 1995 real price) was over 250 million in 1978, but it reduced to 21.5 million by the end of 2006 (China National Bureau of Statistics or CNSB 2007). Despite these great achievements, poverty, especially in the rural areas of western China, remains a troublesome phenomenon. Residents living below the World Bank poverty line of $1 per capita per day amounted to 135 million in 2004 (Chen and Ravillion 2004). In its “Human Development Report 2005,” the United Nations Development Program (UNDP) noted that the pace of poverty alleviation in China has slowed down markedly over the past decade (UNDP 2005). Obviously, how to further reduce rural poverty and increase farmer’s income still is a top policy priority to the Chinese government.

Forests often play a crucial role in the lives of many poor people. Worldwide, almost 70 million people—many indigenous—live in remote areas of closed tropical forests and another 735 million rural people live in or near tropical forests and savannas (FAO 2006; World Bank 2000). In China, most of its 592 poverty counties are found in areas that are far away from urban centers and have poor traffic access; meanwhile, they tend to possess relatively plentiful forests (State Forestry Administration or SFA 2003). In numerous impoverished places, forestry has indeed been a main source of income for farmers (Liu and Lü 2008). To enhance the income of its rural residents as well as to improve its environmental and resource conditions, the Chinese government initiated some new programs and consolidated other existing ones of ecological restoration and resource development in its forest sector in the late 1990s, and renamed them as “Priority Forestry Programs” or PFPs (SFA 2002). So far, implementing these PFPs has substantially altered the land use patterns in many upland regions, where both a significant portion of the country’s primary forest ecosystems and a high rate of poverty incidence are found. As a result, a large amount of sloping cropland has been converted to forest and grass coverage and many existing forests, including quite some managed ones, have been subject to strict regulation for commercial use. Thus, a question of broad interest and major relevance is: How has implementing the PFPs affected the farmers’ income and poverty status? The objective of this article is to address this question empirically.

At a time when ecological restoration has become a common cause and payment for ecosystem services has been widely promoted in pursuant of sustainable development (FAO 2009; Millennium Ecosystem Assessment 2005), scrutinizing China’s recent experience in general and its implementation of the PFPs in particular is interesting. This is because evaluating the program impacts on participating households’ welfare is essential to determine the directions toward which public funding and policy should be mobilized. Lessons learned from China can thus benefit many other countries, especially developing counties that face challenges of both environmental protection and poverty reduction.

Natural disasters in the late 1990s intensified an environmental debate in China and triggered the government to initiate the Natural Forest Protection Program (NFPP) in 1998 and the Sloping Land Conversion Program (SLCP, also known as the ‘Grain for Green’ program) in 1999 (Yin and others 2005). Following successful experimenting during 1998–1999, the NFPP was formally launched in 2000 with an initial investment of 96.4 billion Yuan or equivalent to roughly US$14.1 billion given the current exchange of $1 = 6.85 yuan for the decade (Yin and Yin 2009). A key component of the NFPP was logging bans over 30 million hectares (ha) of natural forests in the upper reaches of the Yangtze River and the upper/middle reaches of the Yellow River. In other areas, harvest restrictions were tightly imposed. The SLCP was piloted in Sichuan, Shaanxi, and Gansu provinces in 1999 and 2000. Its primary goal was to convert 14.6 million ha of sloping and desertified farmland into forest and grass coverage from 2001 to 2010. When it was formally launched, the SLCP covered 25 provinces, with a budget of 225 billion Yuan (Yin and Yin 2009).

In addition to the above two mega programs, a number of other efforts of ecological restoration and forest expansion have been consolidated into the following four programs: the Desertification Combating Program around Beijing and Tianjin (DCBT), the Shelterbelt Development Program in the Three-Norths (referred to the northwestern, north-central, and northeastern regions of China) and the Yangtze River Basin (SBDP), the Wildlife Conservation and Nature Reserve Program (WCNR), and the Industrial Timber Plantation Program (ITPP). Together with the NFPP and the SLCP, these programs comprise the six PFPs (SFA 2004, 2005). The various policy measures taken in implementing them are summarized in Table 1.

Table 1 Key policy measures of the PFPs

From the perspective of rural households, the direct effects of these PFPs are reflected mainly in: (1) the government subsidies they receive for retiring and converting the sloping and desertified cropland or rehabilitating grassland (under the SLCP or the DPBT); (2) the government restrictions imposed on their logging, collecting, and even managing practices in case their forests are put under protection for providing more important ecosystems services (under the NFPP or the WCNR); and (3) the government incentives offered for them to engage in plantation and shelterbelt establishment and other related activities (under the ITPP or the SBDP). As shown in the table, there exist numerous tradeoffs, most of which can result in changed patterns of land use and production. Induced by the land reallocation and production shift, farmers have to intensify farming and commercial forestry activities on their remaining lands, switch animal husbandry from open gazing to pen raising, or search for off-farm jobs in order to sustain their income growth. Therefore, it is expected that following their participation in the PFPs, farmers’ income sources and employment structure will undergo major transformation. To be sure, in addition to farmers’ own initiatives, efforts, and inputs, the extent and trend of their income and employment changes depend critically on the availability and effectiveness of technical, financial, and personnel assistances provided by the local public agencies (Yin and Yin 2009; Yao and others 2009). Finally, after the implementation of the PFPs, the local ecological conditions have changed, which would or might benefit for local production.

The analytic focus of this article is the impacts of the PFPs on farmers’ income. As noted, while some of the impacts are direct, others are indirect; some are positive, but others are insignificant or even negative. Moreover, the PFPs have a high degree of overlapping both spatially and temporally, confounding the employment and income impacts. This situation implies that examining the effect of a single program in isolation may lead to biased or incomplete findings. Additionally, it is more feasible to disentangle the confounded effects with a large panel dataset containing sufficiently long time series. We hope that our study will be able to tackle these issues and thus capture most, if not all, of these impacts in an unbiased manner.

The impacts of the SLCP on farmers’ income and livelihood have received a great deal of attention in the literature. In addition to examining its cost effectiveness and sustainability, Uchida and others (2005) and Uchida and others (2007) analyze its influence on eradicating poverty in the countryside. They find that the program has been successful in poverty alleviation, even though poor households may not have benefited the most. Further, their evidence suggests that households participating in the program have already begun transferring their labor to non-farming sectors more rapidly than those not participating in the program. In contrast, using data collected from Sichuan, Shaanxi, and Gansu for the first few years of the program, Xu and Qin (2004) show that it made little difference in affecting farmers’ income. Their conclusion thus implies that the SLCP’s role in relieving poverty is limited; the reduction of poverty is more likely driven by the overall economic development, which provides greater opportunities to farmers, rather than the direct subsidies of the land conversion.

Observers also point out that in some cases, the goal of the program is not well understood by farmers; and it may even be inconsistent with their aspirations, which have affected their enthusiasm for participation (Du 2004). According to Xu (2003), the main reasons that some farmers lack interest in the program are partly because the subsidies are not delivered on time and in full, and partly because no appropriate remedies were put in place to address the restrictions on intercropping in the forested fields and gathering fuelwood. These factors have led to an adverse effect on the livelihoods of the farmers, who not only rely heavily on forest resources but also tend to be among the poorest rural population. Based on case studies in the upper watersheds of the Mekong and Salween Rivers in northwestern Yunnan, Weyerhaeuser and others (2005) also reveal certain negative impacts of the SLCP and the NFPP on the livelihoods of highland communities.

On the other hand, the study by Zhi and Shao (2001) claims that the income of farm households would be improved significantly during the time period of government compensation. If the subsidies are terminated after the program expires in 5–8 years and farmers are not allowed to utilize their retired lands for economic purposes, they could suffer a loss. In their opinion, the policy that mandates that the proportion of economic forest be no more than 20% has failed to consider the regional disparity and the basic fact that the country has a large rural population but a relatively small amount of cropland. Thus, the government must take steps to improve farmland quality and increase farming productivity in order to address the issue of food supply following the land conversion.

Compared to the SLCP, there have been fewer studies of the socioeconomic impacts of the NFPP and other programs. Using household data, Liu and others (2005) and Ni and others (2002) demonstrate that the NFPP had a negative effect on the income of farmers living close to the protected natural forests. With data from 18 counties in Shanxi, Inner Mongolia, and Hebei, Liu and Zhang (2006) show that the DCBT had a positive effect on farmers’ income, but the effect varied from county to county, due in part to uneven program investments. However, they did not consider all of the relevant factors affecting farmers’ income, including production inputs and household characteristics. In addition, little work has been conducted so far to make a comparative assessment of the impacts of the PFPs. In fact, most of the existing studies have focused on a single PFP, with few studies dealing with two. This is unfortunate given the fact that these overlapping PFPs have all somehow affected farmers’ income and their impacts may interact (Yin and Yin 2009).

Here, we use a fixed-effects model to quantify the income impacts of the PFPs, with a panel dataset of 1968 households in ten counties of four provinces. Broader cross sections and longer time series are two unique features of this dataset. That is, every sample county has at least two PFPs and covers a period of 10 years from 1995, before they were initiated or consolidated, to 2004, when their implementation was well underway. Based on such a large and comprehensive dataset, it is more likely for us to remove the influences of covariates, including production inputs and household and village characteristics, and thus enable us to identify the specific effect(s) of each program on farmers’ income in a rigorous and appropriate way.

Another feature of our study is a simultaneous assessment of the income impacts by adopting two different approaches—one based on dummy variables of whether a household participates in a program and the other on the area that has been enrolled in or occupied by the program. We hope that these two approaches will generate a set of complementary findings, which will allow us to draw clearer and stronger conclusions regarding the income effects of the PFPs. Our results are indeed corroborated by the two approaches, and they indicate that different programs have had quite different impacts. The impacts of the SLCP, the SBDP, and the NFPP on household income are positive, whereas the impacts of the WCNR and the DCBT are either insignificant or negative for the former but positive for the latter. By far, the SLCP has the largest positive income impact.

The article is organized as follows: The next section is devoted to methods and data, the third section reports our empirical results, and we present our conclusions and discussion in the final section.

Methods and Data

Generally speaking, farmers’ income is determined by their production inputs and other biophysical and socioeconomic factors. Production inputs constitute labor, capital, and land. Included in land are farmland, forestland, and other land for growing vegetables and fruits. In addition, land-based production activities entail cash outlays for commercial seeds, fertilizers, plastic sheets, and the like. Moreover, household/village characteristics affect farmers’ income. For example, as part of the human capital, educational attainment is commonly viewed as an important household feature (Schultz 1964). And biophysical and socioeconomic variables at the village level, like rainfall and plot size of farmland, are also relevant to income determination. Of course, because the policy measures taken, while providing variable levels of subsidy, have caused substantial land use changes to the sample households, implementing the PFPs has affected farmers’ income (see Table 1 and discussion below). Certainly, the direction and magnitude of each program’s impact may well vary.

The PFPs effects on sample households’ income can be examined in two ways—one based on whether a household participates in a program and the other on how much of its land is enrolled in or occupied by the program. Therefore, the household income (R) can be defined as a function of production inputs (X ₁,…,X _j ,…,X _J ), characteristics of households, natural, and village conditions (Z ₁,…,Z _m ,…,Z _M ), and families’ engagements in the PFPs (Y ₁,…,Y _k …,Y _K )—areas enrolled in or occupied by the specific PFPs or dummy variables (if yes = 1; otherwise 0) indicating the participation status of households. That is,

$$ R_{it} = e^{{\alpha_{0} }} \prod_{j = 1}^{J} X_{it}^{{\alpha_{j} }} \prod_{k = 1}^{K} Y_{k}^{{\beta_{k} }} \prod_{m = 1}^{M} Z_{mt}^{{\gamma_{m} }} \varphi_{it} $$

(1)

where i the ith household (i = 1, 2,…, I), t the time period (t = 1, 2,…,T); α _j , β _k and γ _m are coefficients to be estimated; and φ _it is the error term that is assumed to be independent and identically distributed.

According to the distributions of the farmers’ income and the PFPs as well as our discussions with officials of provincial forestry and other departments and local experts, we first selected ten counties for our surveys. They are: Zhangbei and Pingquan in Hebei; Xiushui, Xingguo, and Suichuan in Jiangxi; Zhen’an and Yanchang in Shaanxi; and Nanbu, Nanjiang, and Mabian in Sichuan (see Table 2). Each of these counties has participated in at least two PFPs. For instance, Zhangbei County has participated in the DCBT and the SBDP, and Nanbu county has participated in the NFPP, the SLCP, and the SBDP.

Table 2 Participation of sample counties in the PFPs

Households and villages in a county were then chosen randomly. That is, we chose the villages from the village list of a county and households from the household list of a village. Except for Zhangbei where three townships were selected, six townships were chosen in each county. Altogether, we interviewed over 2000 households in 171 villages of 57 townships. And our initial survey was carried in 2004 as part of our program monitoring and assessment efforts supported by the Asian Development Bank and China’s Ministry of Finance. To understand the microeconomic shifts over time, we asked interviewees to recall their production activities and other relevant information back to 1995. Then, in 2005, we repeated our surveys. As such, we were able to assemble a panel dataset covering 10 years (1995–2004), which has a longer and more continuous coverage than almost any other datasets used by others. In order to help interviewees to recall their production and consumption behaviors, we designed the questionnaires in term of each production and consumption activity, and asked as many as possible family members to recall their household activities in each year, and cross-checked with knowledgeable villagers and statistical data and other information of the case study counties, townships and villages. All these steps taken ensure high quality of data collected.

However, our surveys did not get complete information from some households. This is because a few of them moved to places other than the sample villages, errors occurred to some interviews, or certain families failed to clearly recall what had happened to them in the previous year(s). These factors led to the outcome of a slightly unbalanced panel over time. To avoid complications in the estimation based on such an unbalanced panel, we decided to remove those observations with incomplete information and/or incomplete interviews, resulting in a balanced panel of 1968 households for this study. We hope that doing so will have minimal consequences on the estimated results.

It can be seen from Table 3 that over the period of study, more and more households were involved in the PFPs as these programs were undertaken. While some participated in multiple PFPs, others did not participate in any of them. More specifically, a large number of households took part in the NFPP and the SLCP, but only a few took part in the SBDP and fewer in the ITPP. It seems odd that only a few sample households were involved in the SBDP, which is widespread graphically. This is due in large part to the fact that protective shelterbelts are mostly established on public lands, instead of lands devolved to households. Similarly, as a consequence of the limited extent of the ITPP, in conjunction with our random drawing, the ITPP was not captured by any of the sample households. Therefore, it will be excluded from the following empirical estimation.

Table 3 The evolution of sample households’ participation in the PFPs

Included in the dataset are the following variables: (1) household demographics (household size, educational achievement of the household head, and the like); (2) monetary outputs (total income, off-farm income, and income from land-based enterprises) and inputs (labor, farmland, cash expenditure, etc.) for land-based and off-farm activities; (3) the statuses of the PFP participation—effective areas enrolled in or occupied by each of the programs (except for the WNRP, for which the relevant variable is defined as the inverse of the distance to the nearby nature reserve), or dummy variables; and (4) natural and socioeconomic conditions (annual precipitation and average plot sizes of forestland and farmland). The average plot sizes of forestland and farmland of sample households partially reflect the production efficiency of agriculture and forestry. Total income and cash outlay of sample households were deflated and converted to the 1994 constant Yuan, using the rural consumer price index and rural industrial product price index from the Chinese Statistical Yearbooks, published by the CNSB (http://www.stats.gov.cn).

Undoubtedly, household income and other activities are the key variables of this study and reporting income for households can be prone to errors. To ensure the quality of that variable, we compared the household data (including data and information of household incomes, production and consumption behaviors) we collected with statistics compiled by the county and provincial government agencies, and also we double checked with local villagers and other stakeholders (such as local banks, trust foundations and extension agencies) And we found that our data are generally reliable. In addition, we plotted the income distribution of the sample households and its shift over time (Fig. 1). The distribution appears normal and its shift sensible.

Fig. 1

Sample household income distribution

Preliminary statistics of the household data are listed in Table 4. Due to sample dispersals in biophysical and socioeconomic domains, the data have large variances. Nonetheless, our description below will feature the mean values and their changes over time, given their greater relevance. Agriculture in these sample counties remains a major source of income, accounting for more than one-third of their GDP. Implementing the PFPs, especially the SLCP, has caused a sharp land use change as reflected in farmland reduction and forestland increase. Average area enrolled in the SLCP is 1.14 mu (15 mu = 1 hectare) and 2.90 mu per household in 2000 and 2004, respectively. In 2004, the average area enrolled in the SLCP in the Yellow River basin reached 14.90 mu per family, while the figure in Yangtze River basin was only 2.06 mu. The SLCP dummy also shows that households participating in the program increased from 24% in 2000 to 42% in 2004. For the DCBT, participation grew by 9.10 percentages from 2000 to 2004, and Table 2 indicates that it is located in Zhangbei and Pingquan counties. Due to the nature of their top–down, unilateral imposition, households involved in the NFPP and the WCNR did not increase much even since their beginnings.

Table 4 Summary statistics of the household data in 1995, 2000, and 2004

The average farmland per household was 8.72 mu, 7.84 mu, and 5.70 mu respectively, in 1995, 2000 and 2004; and the average forestland per household was 11.20 mu, 12.25 mu, and 15.08 mu, respectively. Clearly, much of the lost farmland was converted into forestland. The mean plot size of farmland dipped from 1.50 mu in 1995 to 1.49 mu in 2004, while the mean plot size of forestland increased from 3.92 mu to 4.36 mu during the same period. Labor input for land-based activities declined slightly over time, i.e., 236.12 person-days in 1995, 231.44 person-days in 2000, and 220.52 person-days in 2004. On the other hand, off-farm employment went up a lot, from 104.72 person-days in 1995 to 159.37 person-days in 2000 and 200.36 person-days in 2004. Cash expenditure for fertilizers, seeds, and other inputs increased significantly as well, from 490.50 Yuan in 1995 to 595.35 Yuan in 2000 and 615.50 Yuan in 2004.

Following log transformation and adding a time trend variable to capture the potential effect of technical and institutional changes, our empirical model becomes:

$$ \begin{aligned} \ln R_{it} = \alpha_{0} & + \alpha_{1} \ln X_{1it} + \alpha_{2} \ln X_{2it} + \cdots + \alpha_{5} \ln X_{5it} + \beta_{1} \ln Y_{1it} + \beta_{2} \ln Y_{2it} + \cdots + \beta_{5} \ln Y_{5it} \\ & + \gamma_{1} \ln Z_{1it} + \gamma_{2} \ln Z_{2it} + \gamma_{3} \ln Z_{3it} + \gamma_{4} \ln Z_{4it} + \theta T + \varphi_{it} \\ \end{aligned} $$

(2)

where T = 1, 2,…,10 is the time trend variable for 1995–2004. A detailed definition of the variables is also given in Table 4.

Empirical Results

A couple of technical issues—choice of random-effects vs. fixed-effects estimation technique and potential endogeneity bias—must be resolved before any formal empirical attempt. First, note that Eq. 2 can be estimated as a fixed-effects or random-effects model. Whether we adopt the random-effects or fixed-effects estimation technique hinges on the outcome of a Hausman test (Wooldridge 1999). To that end, we ran the corresponding regressions of the gross income of sample households against the PFP areas or dummies. It is found that in both cases the χ² values (73.30 for the PFP area model and 1162.38 for the PFP dummy model) are greater than the critical values at the 99% confidence level. These results indicate that we should estimate a fixed-effects model, rather than a random-effects one. One advantage of the fixed-effects estimation is its control over unobserved fixed factors that could confound the estimation (Pender 2005).

Another issue is the potential endogeneity bias—whether or not households’ participation in any of the PFPs is endogenously decided. If households have the freedom to select for participation, then their participation becomes endogenous and an assessment of the program impact must be made accordingly. Otherwise, ignoring the endogenous choice by households will lead to biased estimates (Wooldridge 1999; Uchida and others 2007). In the current context, endogeneity is most pertinent to the SLCP, given that the other programs were largely imposed on the rural households, who had little power to determine whether and how they should participate.

As to the SLCP, it is true that any household had certain autonomy of participation in principle. In practice, however, if its cropland plots fell into the planned project area and were eligible, it had to and indeed desired to participate, because of the very attractive compensation (Uchida and others 2005). On the other hand, if a household’s plots did not fall into the planned project area, it could not participate even if it wanted to. Therefore, we do not expect that endogeneity would be a major issue.

To formally test whether there exists an endougeneity bias, we first estimated a model, in which the likelihood of participation was determined by a set of exogenous variables (Table 5), from which we derived the predicted probabilities of participation by individual households. Once we had obtained these predicted values, we then used them in identifying the income effect of the program participation (Table 6). The Hasusman test indicates that χ² = 9.54, which is much lower than the critical value. So, we reject the hypothesis that there is a significant endougeneity bias in household’s participation in the SLCP. Moreover, comparing the coefficients of the SLCP dummy and the predicted probabilities of participation, we found that both are 0.13 and significant at 99% level. Again, it suggests little effect of endogenous choice. Therefore, it seems that voluntarism of the SLCP participation might be a questionable thesis. That is, farmers can choose to participate in the “take-it-or-leave-it” program only when their croplands are eligible for it. They will not have the option if their land is considered “ineligible.” A similar result was supported by Uchida and others (2007) and Liu and Zhang (2006).

Table 5 Regression results of the SLCP participation determinants

Table 6 Testing the endougeneity bias of the SLCP participation

Next, we present our regression results in Table 7. Four alternative scenarios for the PFP income effects were executed. Recall that we used both the PFP dummy and area (or distance inverse) variables in estimating the income effects. Given that the SLCP adopted different standards of subsidy for the Yangtze River basin and the Yellow River basin, two additional scenarios were construed for the SLCP compensation variation.

Table 7 Estimated impact of the PFPs on household income

Results of the four scenarios are remarkably consistent and they show that using the balanced panel, while conducive to estimation, has not caused any major alteration to the coefficients of our primary interest. While the R ² values are somewhat low—about 0.44, they are acceptable. This is because of the heterogeneity of geography and economy on the one hand and the fact that the PFPs, the other key independent variables, the straight geography (dummy variables), and the fixed-effects term will not be able to explain the variations in income within and across counties particularly well (Wooldridge 1999). Some covariates, such as labor for land-based activities, off-farm employment, and time trends, have significantly affected household income. But our result presentation will concentrate on the impacts of the PFPs, given that are the central question of this study.

When dummies were used to capture the effects of the PFPs on household income, the coefficients of the WCNC and the DCBT are insignificant. The coefficients of the SLCP are 0.13 for the whole sample, its effect in Yellow River basin (0.27) more than doubles that in Yangtze River basin (0.11), and both are significant. The coefficients of the SBDP dummy are 0.14 and 0.16 at the 99% confidence level, and the NFPP has a positive coefficient of 0.02 that is also significant at the 99% level.

If areas of the PFPs were used, the WCNC has a negative effect (−0.312 and −0.317) on household incomes at the 99% confidence level. As expected, the closer a household to a natural reserve, the lower its income. In contrast, the coefficients of the DCBT, ranging from 0.05 to 0.06, are significant at the 99% level. The contribution of the SLCP to household income is, again, the largest. The breakdown of the SLCP further shows that it has a larger positive impact (0.18) in the Yellow River basin, compared to that in the Yangtze River basin (0.10). The coefficient of the NFPP area is positive and significant at the 1% and 5% confidence level. While this effect is inconsistent with the negative impact reported in the literature (Xu and others 2004; Liu and Zhang 2006; Du 2004), it is very small. Finally, the coefficient of the SBDP is 0.04 and 0.05 and significant at 99% level.

Conclusions and Discussion

This study was motivated by our observation that while it is of broad interest and great relevance to exam the impacts of forest priority programs on incomes, little has been done along this direction of research. So, the first thing we did was to build a large panel dataset based on an appropriate sampling and enumerating approach. Ultimately, we were able to gather survey information from 1968 households in ten counties and for a period of ten years (1995–2004). In addition to household income, included in our data are an extensive number of variables of production inputs, program engagements, and household and village features. Then, we estimated a fixed-effects model of multiple specifications to investigate the income determinants, including the PFPs. These steps have allowed us to derive a rich and interesting set of empirical findings.

It is found that the ecological restoration and resource development programs have affected household income in different ways. The SLCP, the SBDP, and the NFPP—have all made positive contributions, regardless of how the program participation variables are defined. Among them, the SLCP has the largest income effect. It is easy to understand the positive impacts of the SLCP, because the Chinese government has provided handsome subsidies for the cropland conversion to forest and grass coverage, which in many cases are higher than that revenues generated from their farming on the sloping cropland. Also, these highly significant effects are consistent with the findings by others (Uchida and others 2005; Yao and others 2009). Although the subsidy is higher in Yangtze River basin (3450 Yuan per ha) than that in the Yellow River basin (2400 Yuan per ha), the contribution of the SLCP to household income in Yangtze River basin is much lower than that in Yellow River basin. This is a result of the much larger SLCP area per household in Yellow River basin than that in the Yangtze River basin. And it indicates that farmers in the former region have benefitted much more than those in the latter, even though grain subsidy per unit cropland in the latter is a bit higher. This conclusion is further reinforced by the spatial heterogeneity in our sample—the income of a household located in the Yangtze River basin is 24–28% higher than that in the Yellow River basin.

The effects of the SBDP on household income are significantly positive as well. This is because the primary objective of the program is to improve ecological conditions for farming and other economic activities and it has taken place for quite long. With the program being implemented, the local ecological conditions for farm and other economic activities have improved, leading to a positive effect. Nonetheless, it should be reiterated that in our sample only a few households have been involved in the SBDP.

In contract, implementing the DCBT and the WCNR has either no impact on household income (with the dummy variables approach) or significantly positive impact for the former but significantly negative impact for the latter (with the area variables approach). It seems that the significance indicates that the programs are making a difference on the margin. Given their short durations (2–3 years covered by the data), however, their overall impacts have not necessarily become pronounced yet. Further, as shown in Table 1, part of the DCBT is to reduce overgrazing by mandating households to raise their cattle and sheep in pens; their income from animal husbandry has been adversely affected. Thus, the insignificance of the DCBT’s income effect is not out of the anticipation.

The establishment of a natural reserve implies that using the forestland for commercial or self-consumptive purposes by surrounding households is forbidden or strictly controlled. Therefore, the effect of the WCNR on household income can be negative on the margin; and if a household lives closer to the reserve, it may suffer a greater income loss. Because the share of forestry income is not necessarily high in its gross income, however, the overall impact of the WCNR remains insignificant. In this regard, a key question is how the government can more effectively integrate nature conservation with economic development to mitigate the negative impact of excluding or restraining local farmers’ access to forests in the nature reserve. Certainly, it is our view that the effectiveness of the SLCP, the SBDP, and the NFPP can and should be improved as well.

The positive effect of the NFPP contradicts much of what has been reported in the literature (Xu and others 2004; Wu and Liu 2002) and appears counterintuitive. It is true that at the beginning, sudden logging bans and forestry activity contractions inflicted harms on the local farmers, including their foreland put under the NFPP being not considered for any compensation. But farmers have responded to the restrictions by altering their production and employment, especially when they have had sufficient time. For one thing, they have used their farmland more intensively. For another, they have transferred more of their labor from land-based to off-farm activities. Also, the government has modified its policy by hiring local farmers for forest protection and management and relaxing timber harvest restrictions on non-state forests that were subject to the NFPP coverage (Liu and Lü 2008). Essentially, these considerations have been confirmed by a more recent study by Mullen and others (2009), who find that the NFPP has had a negative impact on income from timber harvesting but has actually had a positive impact on total household income from all sources, and that off-farm labour supply has increased more rapidly in NFPP areas than non-NFPP areas.

To trace out the temporal effect of the program and farmers’ response, we decided to take the following three options: (1) re-estimation our model in a slightly different manner, by including annual dummies or areas for the NFPP; (2) exploring the variation of farming intensity changes between affected and non-affected counties; and (3) examining the potentially reduced timber production induced by the logging bans. The re-estimated results of our model are presented in Table 8. In general, the effects of the NFPP on land-based income and off-farm income experienced a process of change from insignificantly negative to significantly positive. Based on the dummy variables approach, for instance, the effects on total income and land-based income were insignificant during 1998 and 1999, but it became significant between 2000 and 2004 and the coefficient got larger accordingly. The effect on off-farm income experienced a similar trend, except for 2003, as well. The regression results based on the area variables approach tell virtually the same story, with the only difference in that the significant income effect appeared later and not as strong.

Table 8 Estimated result of the impact of NFPP on total household incomes and marginal incomes

To find the farming intensity variation, we calculated the indexes of production expenditure and labor input for land-based production (see Table 9). It can be seen that since the NFPP was launched in 1998, the production expenditure index for farming in the affected counties has become higher than that in the non-affected counties. It was 1.44 in 2000 and 1.87 in 2004 for the affected but 1.20 and 1.46 for the non-affected, respectively. The labor input index witnessed similar paths of change. Consequently, we postulate that more intensive farming has resulted as a response to NFPP explains part of the positive income response to the program., and more off-farm employment has been associated with higher household income responded to the NFPP and other PFPs. These observations lead us to an important conjecture. That is, the direction and magnitude of a program’s impact has to do with both the direct and indirect effects of the policy measures, as well as the agent’s response and the time frame for analyzing the response.

Table 9 Production expenditure and labor input indexes for farming

Finally, it is useful to look at the age structure of forests in the affected counties in assessing the potential reduced timber production caused by implementing the NFPP. Figure 2 indicates that the proportion of matured and over-matured forest is quietly low in the affected counties—below 5% in Nanbu, Mabian and Yanchang, and between 5 and 10% in Zhen’an and Nanjiang. Even without implementing the NFPP, log production would have been limited and income generated from timber would have been small. Therefore, the negative impact of the NFPP on household forest income could not have been very large.

Fig. 2

The age structure of natural forests in relevant sample counties in 1998

Overall, the contributions of various production inputs and other control variables are in line with our expectations. Notably, labor and cash expenditure for land-based activities and off-farm employment have contributed to income growth positively. The significant effect of the time trend variable shows that technical and institutional changes have contributed to household income growth as well. It is unsurprising to find that the average sizes of farmland and forestland plots are somehow negatively correlated with household income. Similar evidence is also reported by Chen and Brown (2001). This suggests a strong need to address the question of how to improve land use efficiency and productivity. While the Chinese government has been making efforts along this direction as part of its strategy of rural development, our results accentuate their importance.

To be noted, because of production factor changes, we found that rural households’ production behaviors and choices have been changed since these the PFPs have been launched, such as intensive farming and more off-farming activities, in order to maximum their income.

In closing, this article has shown that some of the PFPs—the SLCP, the SBDP, and the NFPP—have made positive impact on household income. By far, the SLCP has the largest impact. Other programs—the WCNR and the DCBT—still have not had a pronounced overall effect on household income due to their short time span of execution, even though they may have exerted certain influence at the margin. Notably, the impact of the WCNR, if any, is negative. It should be pointed out that issues related the income inequality and mobility have not been investigated. This should be one fertile area of future research. Also, as we have emphasized, time is a crucial factor for any program to take effect. What we have looked at here is primarily the short-term effect. It is thus necessary to continue the monitoring and data accumulation efforts, so that the long-run effects of the PFP can be analyzed.