当前位置:首页 >> 英语 >>

Education and allocative efficiency household income growth during rural reforms in China

Journal of Development Economics 74 (2004) 137 – 162 www.elsevier.com/locate/econbase

Education and allocative efficiency: household income growth during rural reforms in China
Dennis Tao Yang *
Department of Economics, Virginia Polytechnic Institute and State University, 3033 Pamplin Hall, Blacksburg, VA 24061, USA

Abstract This paper studies the contribution of schooling to rural income in China during factor market liberalization between 1986 and 1995. The relaxation of controls permitted farm households to reallocate productive inputs from agriculture to nonagricultural activities. It is hypothesized that education facilitates this adjustment. Panel data from Sichuan province suggest that schooling enhanced the ability of farmers to devote labor and capital to nonfarm production given the evidence that less-than-optimum levels of these inputs were allocated to nonagricultural uses. During the transition, the expansion of nonfarm activities contributed significantly to household income growth. D 2004 Elsevier B.V. All rights reserved.
JEL classification: J43; O15; O12 Keywords: Education; Market liberalization; Input allocation; Rural income; China

1. Introduction Policy reforms in China have revitalized the rural economy. Since the inception of reforms, real rural per capita income has increased more than fourfold; earnings rose sharply between 1978 and 1985, followed by a period of continued growth (SSBa, 1998). Several factors have contributed to this remarkable performance. The adoption of the household responsibility system (HRS) and increases in state procurement prices were identified as the major sources of income growth prior to 1985, creating a profound onetime effect on earnings through increased labor effort and price incentives (McMillan and

* Tel.: +1-540-231-7474; fax: +1-540-231-5097. E-mail address: deyang@vt.edu (D. Tao Yang). 0304-3878/$ - see front matter D 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.jdeveco.2003.12.007


D. Tao Yang / Journal of Development Economics 74 (2004) 137–162

Zhu, 1988; Lin, 1992). Agricultural research and technological change are also found to have significantly raised crop yields (Huang and Rozelle, 1996; Fan and Pardey, 1997). While their coverage extends to the collectivization period, these studies are primarily concerned with productivity gains within agriculture, especially during the early period of economic reforms. The main purpose of this paper is to examine the sources of sustained income growth between 1986 and 1995 in a broader context of the rural economy that includes nonagricultural development. The focus is on farmers’ responses to factor market liberalization as they expanded nonfarm production and on the role of their education in facilitating resource allocation decisions. Starting in 1983, the government announced a series of policies that loosened restrictions on labor mobility out of agriculture. The regulatory changes, including permission for long-distance transport, marketing of commodities, and employment in small towns, encouraged farmers to establish nonfarm businesses and seek off-farm jobs with better pay. At the same time, farm households also diverted funds and capital equipment to industrial and service activities for higher returns. During this 10-year period, the percentage of rural labor force employed in township and village enterprises (TVEs) increased from 12.8% to 22.2% (SSBb, 1996). In 1986, the gross output value of TVEs was about 88% of the gross value of agriculture, but in 1995, the former was more than three times the latter. These industrial developments are analyzed as a major force behind rapid income growth in rural China. The human capital approach to production efficiency has long postulated that education may enhance the ability of farmers to perceive and interpret market information so that they can better respond to economic disequilibria (Schultz, 1975). In particular, investment in education may improve farm allocative decisions as well as workers’ production skills (Welch, 1970). Returns to education are high when productive learning opportunities can be exploited; these opportunities are often associated with technical innovation or changes in the market and political regimes (Rosenzweig, 1995). While these views are supported by a large body of empirical literature that finds education to have positive effects in modern, dynamic environments,1 few attempts have been undertaken to assess returns from schooling during the transition of market and political systems.2 Hence, the recent policy reforms in China present an unusual opportunity for examining the role of education in production during the transition from a planning to a market economy.

1 For instance, schooling is found to have higher returns during periods of technical change because it facilitates the use of fertilizer and other modern inputs (Huffman, 1977; Jamison and Lau, 1982). In addition, during the Green Revolution, more schooled farmers were able to achieve higher profits because of more effective adoption of high-yield seed varieties (e.g., Foster and Rosenzweig, 1996) and of better execution of production (e.g., Pitt and Sumodiningrat, 1991). Schooling also enhances information skills of the rural population (e.g., Strauss and Thomas, 1995), which in part improve their opportunities to participate in rural to urban migration (e.g., Schultz, 1988). 2 One exception is Orazem and Vodopivec (1995) who find a large increase in compensation to the more educated in Slovenia as it moved from the centralized economy to the market economy. Another is Li and Zhang (1998) who find that education was not rewarded during the commune period in rural China, whereas, under the postreform household farming system, there were positive returns to farmers’ education. Significant returns to education are also reported for China’s paper industry in the postreform era (Fleisher et al., 1996). However, these studies do not investigate the mechanisms through which education realizes its returns, the focus of this paper.

D. Tao Yang / Journal of Development Economics 74 (2004) 137–162


To clarify the productive value of education, this paper sets up a simple model of profit maximization in which the farm household engages in agricultural and nonagricultural activities.3 Under central planning that emphasized local grain self-sufficiency, factors of production were devoted excessively to agriculture, resulting in resource misallocations with higher input returns in the nonagricultural sector. Therefore, as restrictions on factor mobility were relaxed during reforms, rural families increasingly reallocated inputs towards nonfarm production. The central hypothesis proposed in the paper is that better educated families would adjust to these changes more quickly than less educated families. The empirical analysis of this paper uses household-level panel data between 1986 and 1995 from the Sichuan province to analyze how schooling affects intersectoral input allocations and how factor utilization in turn determines farm profits. The panel data are constructed from the Rural Household Survey collected by China’s State Statistical Bureau. The rich structure of the data enables control for household fixed effects, region and time specific factors, and endogeneity associated with idiosyncratic shocks to individual households. These factors are not properly treated in previous studies on the selection of rural income activities due to limitations of cross-sectional data (Taylor and Yunez-Naude, 2000; Yang and An, 2002). The empirical findings of this paper indicate that, during the period of transition, less-than-optimum levels of labor and capital were allocated to nonagricultural uses. More importantly, the findings suggest that schooling played a critical role in allocating more of these inputs to the factor-scarce sector. Thus, schooling was an important factor behind the rapid expansion of rural industries, which was a major source of sustained income growth. The rest of the paper is organized as follows. Section 2 provides an overview of policy reforms in rural China and especially of regulatory changes governing factor mobility. Section 3 presents a simple household model in which farmers make allocative decisions regarding inputs across agricultural and nonagricultural production. The model illustrates the distortion in factor utilization due to policy interventions, expected responses of farmers to market liberalization, as well as the role of education during market transition. Section 4 discusses the data set, the econometric specifications, and reports the estimation results. Section 5 presents concluding remarks.

2. Policy reforms in China Prior to the start of reforms in 1978, massive distortions in the allocation of resources existed in China’s centrally planned system. The cumulative effects from pursuing a heavy-industry-oriented development strategy since the 1950s resulted in excessive allocation of capital assets in urban areas, and a high percentage of the labor force being concentrated in the countryside.4 Within the rural sector, this national
This framework is closely related to Taylor and Yunez-Naude (2000) and Yang and An (2002) who analyze the effect of education on the choice of income activities. 4 The main enforcement mechanisms include the state control of agricultural production and procurement, and restrictions on rural-to-urban migration via a household registration system. See Yang and Zhou (1999) for discussions on resource allocation across rural and urban sectors.


D. Tao Yang / Journal of Development Economics 74 (2004) 137–162

policy stressed agricultural production and local grain self-sufficiency, a strategy rigorously pursued by prominent leaders ever since the tragic experience of the Great Leap famine between 1959 and 1961. Before the reforms, rural industrial activities concentrated on a narrow range of products, emphasizing ‘‘five small’’ industries: iron and steel, cement, chemical fertilizer, hydroelectric power, and farm implements. Enterprises in the countryside were not oriented towards market and consumer products and remained subsidiary to agriculture (Findlay et al., 1994; Naughton, 1996). In 1978, only about 7% of the rural labor force nationwide was in nonagricultural employment, generating approximately 7% of rural household earnings (SSBa, 1988), a level far below that of other comparable developing countries (Anderson and Leiserson, 1980). Due to restrictions on nonfarm production, capital and labor were scarce and their returns were high in that sector, creating opportunities for rapid expansion along with policy reforms. Market-oriented development in rural China started with a package of three reforms: the replacement of production teams with households as units of basic production (HRS), official increases in agricultural product prices, and the liberalization of markets for rural products. These reforms provided the necessary conditions for the boom in rural industrial development starting in the mid-1980s. The change from communes to a household-based farm system began in 1979 and was essentially completed by the end of 1983. This institutional change induced strong family work effort, thus reducing the demand for workers on small Chinese farms. More importantly, the household responsibility system enabled individuals to have increased command over their productive resources. During the same period, the government also implemented reforms in production planning in which the state reduced the number of production planning targets (or categories). Of the remaining targets, few were mandatory, and many were guided by complementary prices and incentive schemes (Sicular, 1988). Therefore, farmers not only had incentives, but also certain freedom in relocating labor and capital to nonfarm uses. In 1979, the government also implemented large increases in state procurement prices for agricultural products. Quota prices for grain, oil crops, cotton, sugar crops, and pork were increased by an average of 17.1%. In addition, the premium paid for above-quota sale of grain and oil crops was raised from 30% to 50% of the quota prices. The weighted average increase was 22.1% for all agricultural products.5 In effect, these price adjustments injected a large amount of funds into the rural economy, which created a demand for industrial products and supplied the flow funds for capital investment, especially nonfarm production. Finally, the liberalization of rural markets not only accommodated the sales of nonfarm products, but also facilitated the purchase of inputs for nonagricultural activities. It is evident that these three reforms were interrelated; each helped reinforce the impact of the other. Consequently, by the mid-1980s, the economic basis for accelerated growth in rural industries was already embedded in China’s rural economy. Input and output markets had

5 For details of these price changes and agricultural price adjustments in the following years of reforms, see Sicular (1988).

D. Tao Yang / Journal of Development Economics 74 (2004) 137–162


emerged; households were conscious of their alternative opportunities; and they had incentives to quickly allocate resources to nonagricultural activities that would generate higher returns than those from farming. This view is supported by the empirical findings of Putterman (1993), who analyzed intersectoral factor allocation in five production teams of Dahe Township in Hebei province. The study suggests that, in 1985, the marginal productivity of capital and labor in the noncrop sector exceeded the levels in the cropping sector, indicating ‘‘overallocation’’ of resources in agriculture. The catalyst for the rapid expansion of nonfarm production was a series of policies that loosened restrictions on labor mobility and the operation of rural enterprises. The policy’s evolution can be briefly described as follows. In 1983, Document No.1 of the Central Committee of the Chinese Communist Party (CCP) provided general guidelines that encouraged the emergence of specialized households and praised their effectiveness in making the best use of limited funds and labor. Skillful workers and craftsmen were permitted to leave farming and engage in a variety of nonagricultural activities, including long-distance transport and the marketing of commodities. In addition, the document allowed cooperative ventures, as well as rural industrial and commercial households, to employ labor (Ash, 1988). In accordance with these liberalization measures, the state continued to narrow the range of products for compulsory procurement. It was after the inception of this document that some farmers began to quit farming to take up jobs in product transport, goods retail, or business and handicraft. In March 1984, the Central Committee of the CCP and the State Council issued the ‘Report on Creating a New Situation in Commune and Brigade-run Enterprises,’ which outlined a new development strategy that targeted industry as the focus of future rural development. Industrial development was expected to provide inputs for agriculture, absorb rural labor, and help raise rural earnings (Findlay et al., 1994). This strategy sharply contrasted with the old policy of local grain self-sufficiency in which rural industries had only a subsidiary role to agriculture. In 1985, Document No.1 of the CCP permitted farmers to seek employment and establish businesses in nearby towns, if they could provide their own food grain and were financially capable to run a business. This landmark deregulation officially relaxed the controls on labor mobility within rural regimes; in the past, farmers had to live and work in villages where they held household registration. In addition to the relaxation of controls on labor mobility, a major reform in agricultural production and procurement helped trigger the rapid growth of rural industries. At the beginning of 1985, after consecutive years of good crop harvests, the state announced that it would no longer set any mandatory production plans in agriculture and that obligatory procurement quotas were to be replaced by purchasing contracts negotiated between the state and farmers (Lin, 1992). The loosening of farming constraints, together with the increased freedom in allocative decisions, prompted farmers to adjust their productive activities in accordance with profit margins.6 In

However, it should be acknowledged that in some parts of China, implementation of contract negotiations fell short of the contract rhetoric, in which mandatory quotas continued after 1985. A significant price liberalization occurred again in 1992 – 1993, but came to a halt in 1994 when China experienced high inflation.



D. Tao Yang / Journal of Development Economics 74 (2004) 137–162

1985, the grain-sown area at the national level fell by 4%, output by 7%, cotton-sown area by 26%, and cotton output by 34% (Sicular, 1988). In contrast, the number of TVEs more than doubled in the same year, and their total labor force increased by more than 30%, following a year of strong growth in 1984 (SSBa, 1988). These dramatic changes in policies and in farmers’ responses marked the beginning of sustained expansion in nonagricultural activities.7

3. An illustrative model To better understand the mechanisms through which state interventions lead to distortions in factor allocations and the rational responses of farmers to policy changes, I set up a farm household model with two activities. For simplicity, the analysis focuses on labor mobility regulations and the consequences of policy changes, taking other aspects of the reforms, such as the adoption of HRS and the emergence of rural markets, as given. Consequently, the model accords with the economic reality of rural China in the early 1980s and the subsequent changes after factor market liberalization. Moreover, the model provides a framework that also accommodates the role of education in affecting the profitability of farm business through resource allocation decisions. Consider a static profit maximization problem in which the household engages in both agricultural (a) and nonagricultural (n) activities, ja{a,n}. Assume that the activityspecific production function takes the following form: y j ? f j ?x j ; k j ; l j ?; ?1?

where y j is the output; {x j,k j,l j} are variable inputs and quasifixed factors of capital and labor used in the jth activity.8 I make the usual assumptions about the neoclassical production function f j in that it has diminishing returns in each of the inputs and the three factors are complementary in production. Analytically, Bf j > 0; BX B2 f j B2 f j < 0; and >0 2 BXm BXmV BX for m p mV ; ?2?

} where X={x j,k j,l j} and {m,mV index the inputs. The allocative decisions of the household involve the purchase of the variable input x, the utilization of this input, as well as the use of family labor and capital across the two lines of production. Following conventional treatment, total family labor (l) and capital
7 It should be noted that obstacles to rural labor mobility still exist today despite continued improvements since the early years of reforms. For instance, a rural worker currently employed in the enterprise of another village does not receive the allocation of homestead or other housing arrangements, even if the job is permanent, thus incurring high costs to the migrants. As reported in Yao (1999), local protectionism is also a significant issue, in which village workers often earn much higher wages than outsiders. In addition, in some regions, local government has continued to implement voluntary production contracts with a certain degree of coercion. Clearly, much has been improved regarding labor mobility, but further reforms are needed. 8 Note that land is omitted in the model, although it could be added to the agricultural production without affecting any of the following analytical results. This omission simplifies notation, and it will be relaxed in the empirical analysis.

D. Tao Yang / Journal of Development Economics 74 (2004) 137–162


assets (k) are assumed to be fixed inputs; their accumulation decisions are only affected by long-term considerations. Because the three inputs are used in both activities, they satisfy the following resource constraints: l ? la ? ln ; k ? k a ? k n; x ? xa ? xn : ?3?

The equalities in the equations are consistent with full-capacity usage and zero cost of adapting the inputs to the alternative activities. The variable values associated with each sector reflect the extent of activity participation. Let the aggregate profit P be the sum of profits from the two activities, P = Pa + Pn. The maximization problem of the household can be written as:
l a ;k a ; x; xa

max P ? pa ya ? pn yn ? wx x;

subject to conditions(1) – (3); { pa, pn, wx} represent the prices for the agricultural product, the nonagricultural product, and the price of the variable input, respectively. The optimal solutions to the problems can be denoted as {la*, ka*, x*, xa*}. These derived demand for inputs are functions of household endowments and input and output prices, satisfying the conditions that the marginal return of x is equal to the marginal cost of purchase, and that the marginal values for each of the three inputs are equalized across the two sectors. These standard optimal choices under the competitive situation can be used as a reference for studying farm household behavior in China. By the early 1980s, Chinese rural households still could not adopt optimal production plans despite the fact that markets had emerged and they had reasonable command over their resources within agriculture. This is because officially they were still not permitted to engage in nonagricultural production, reflecting the old development strategy that emphasized agriculture and local grain self-sufficiency. The set of labor market distorting policy interventions ( P) that prevented optimal labor flow from agriculture to nonagriculture can be represented by a binding constraint l aP ? l a * > 0; or l n * ? l nP > 0 where laP corresponds to overutilization of labor in agriculture under the policy interventions. Denoting knP as farmers’ choice of capital for nonagricultural production with policy controls, it is easy to show that under the conditions given above k n * ? k nP > 0; a result suggesting systematic underutilization of capital in the nonfarm sector. Therefore, prior to factor market liberalization during the period of 1983 – 1985, Chinese farmers operated under a second-best scenario. Their key productive resources, labor and capital, were restrained from being allocated to nonfarm activities despite


D. Tao Yang / Journal of Development Economics 74 (2004) 137–162

households’ incentives for intersectoral reallocation. Hence, when the series of policies lifted labor mobility controls, farmers would respond by closing the inefficiency gaps of factor allocation, (kn*?knP) and (ln*?lnP). While the above analysis focuses on the allocation of inputs across the two sectors, the framework can be extended to address capital accumulation decisions. If the household is permitted to purchase capital equipment, rather than treating it as a quasifixed factor, the effect of policy interventions on capital accumulation can also be analyzed. The total capital stock with policy restrictions, kP = knP + kaP, would depend on cumulative capital investments in the two sectors over time, and it would fall short of market optimal level when capital has stronger complementarity with labor in nonagricultural sector, i.e., kP < k*, if B2f n/Bk nBl n>B2f a/Bk aBla. In China, land per farm is limited, forming a constraining factor in agriculture, which suggests that this complementarity condition is likely to hold. Consequently, the set of labor market policy interventions would cause underinvestment in capital. In a dynamic framework, because newly purchased equipment does not fully depreciate in the current period, reduced capital spending in each period would result in a less-than-optimal level of total capital stock for the household. In this paper, I do not systematically explore capital accumulation decisions over time because of the following two reasons. First, capital assets owned by rural families in China in the mid-1980s were largely determined by factors going beyond the exogenous variables in the above, simple model. As documented by Wen (2000), assets previously owned by collective units were divided and distributed to individual families during the disbandment of communes between 1982 and 1984 based primarily on the size of the family population, and, in some regions, on the size of family labor force or other criteria. The empirical analysis of the paper covers 1986– 1995. Data limitations on local regulations regarding division of assets do not allow proper investigations into past asset accumulation. Second, from a theoretical perspective, capital accumulation decisions are closely related to occupational choices of family members. For instance, self-employment in nonfarm activities would usually require more capital investment relative to having wage jobs in local labor markets. Formally incorporating these considerations into the current framework would require a different modelling strategy, which would go beyond the scope of the current paper. Therefore, throughout the paper, I focus on intersectoral input allocation decisions, taking the capital assets of the household as given. The above static model, while highlighting factor misallocation due to policy interventions, has not yet incorporated the role of education in the adjustment process with market liberalization. The idea that schooling may enhance the efficiency of farmers to respond to economic disequilibria and to explore learning opportunities is well known (e.g., Schultz, 1975; Rosenzweig, 1995). In particular, education may contribute to profitability through a ‘‘worker productivity effect’’ and an ‘‘allocative effect’’ (Welch, 1970). The former refers to education’s effect on technical efficiency, and the latter refers to the skills of obtaining and using information for managerial decisions, including the allocation of inputs to alternative uses. Applying these theses to rural reforms in China, we expect that, once policy restrictions were loosened, more educated farmers would have better information about resource misallocations and their

D. Tao Yang / Journal of Development Economics 74 (2004) 137–162


schooling would facilitate quicker adjustments in allocating more capital and labor to nonagricultural production. Two issues arise regarding the specification of education in production. The first pertains to the measures of education to approximate allocative skills within the household. The approach taken incorporates the idea that family members have incentives to share information with each other so that the member with the highest schooling level is most likely to contribute to managerial decisions. The fact that family members often live together and jointly engage in production certainly facilitates information sharing and collective decision making. Hence, this paper distinguishes the highest level of schooling in the household (sh) from the average schooling of other family workers (sa) as proxies for allocative skills.9 In the context of this paper, the allocative effect of schooling is modelled through input allocations across agricultural and nonagricultural sectors. Therefore, if there is centralized decision making, we would expect that sh plays a more important role than sa in affecting resource allocations. Later empirical analysis will also investigate whether family schooling enhances farm efficiency through other channels including the worker productivity effect. The second issue pertains to unobserved characteristics of the household. For instance, unobserved ability and managerial skills specific to the ith farm (mi), which is likely correlated with the schooling attainment of family members, may have an effect on farm profits independent of schooling. Omitting these unobserved variables are omitted in empirical analysis may cause bias to the schooling estimates. Therefore, careful and explicit specifications for family unobserved variables are called for in both the analytical model and empirical analysis. It should be noted that the adjustment process towards a new equilibrium after policy reforms may last for a long period of time. The establishment of new nonagricultural business may require months or even years of preparation. The scale of business often grows gradually towards optimum. Employment in the labor market also involves collection of information and, perhaps, long-term planning in accordance with the ongoing farming activities. Therefore, taking into account the allocative role of education during the period of adjustment, the optimization problem for household i for year t leads to the following input demand functions:
nP kit ? k nP ?lit ; kit ; pt ; sit ; mi ; Fit ?;


nP lit ? l nP ?lit ; kit ; pt ; sit ; mi ; Fit ?:


The superscript ‘‘P,’’ as noted earlier, refers to choices of inputs during the adjustment period in which the influence of policy interventions is still not fully corrected. The
9 Hypotheses concerning the effects of worker schooling composition on farm efficiency are proposed and tested in Yang (1997). Empirical findings based on small-scale farming in China indicate that educational returns come primarily from the highest farm schooling through allocative effects. This result is consistent with the findings by Foster and Rosenzweig (1996) that whether anyone in the household has a primary education is a good predictor for the adoption of high-yield seed varieties during the early stages of the Green Revolution in India. Also, see Basu et al. (2001) for evidence of intrahousehold externality of literacy for Bangladesh.


D. Tao Yang / Journal of Development Economics 74 (2004) 137–162

explanatory variables that determine the allocation of capital and labor to nonagricultural activities in Eqs. (4) and (5) include aggregate labor and capital endowment {lit, kit}, price vector pt, a vector of family schooling sit=(shit, sait), unobserved managerial ability and other time-invariant, family-fixed factors mi, and a vector of household/local characteristics Fit, including local weather conditions, family land endowment, worker experience, geographic features and agricultural tax burdens. These household/local variables may or may not change with time. Note that the size of family labor force, capital stock, and the schooling variables may differ over time, although their changes are assumed to be determined by long-term considerations and are exogenous to period-specific, intersectoral input allocations. To clarify the allocative role of education, note that if the sectoral n nP n nP allocations of inputs are already at the optimum, i.e. kit* = kit and lit* = lit , sit would have n* nP n nP no role in affecting input demand. But, when inefficiency gaps (kit -kit )>0 and (lit*-lit ) >0 exist, we would expect that sit is positively associated with input allocations to the nonfarm sector. In other words, better educated families would adapt to optimal input uses more quickly than less educated families. The effect of input adjustments on earnings is revealed in the household net profit function, which is defined as total sales minus the expenditures on variable inputs xt. Net profit for household i in year t, denoted as Vit, can be represented as follows: Vit ? V ?lit ; dnP ; kit ; dnP ; pt ; sit ; mi ; Fit ?; lit kit ?6?

nP nP where {dlit , dkit } are shares of labor and capital devoted to nonagricultural production. This profit function, together with Eqs. (4) and (5), encompass the ideas that education may affect input allocations and that those allocations in turn may affect farm profits. More specifically, if schooling facilitates input adjustment and less-than-optimal levels of capital and labor are devoted to nonfarm uses, we would expect a positive association between sit nP nP nP nP and {dlit ,dkit }, and between {dlit ,dkit } and Vit.10 Note that to capture the effects of schooling on profits other than sectoral input uses, I have also specified sit in the profit function conditional on factor allocation decisions. A positive association between sit and Vit would imply additional contribution of education to efficiency on top of the allocative nP nP effects already embedded in the factor shares. Because {dlit ,dkit } are choice variables, proper treatment for endogeneity is called for in empirical implementation. In summary, the central hypotheses concerning the adjustment process and schooling effects on farm efficiency can be stated as follows. In a specific year during the period of factor market liberalization, we would expect:

(1) that less-than-optimal levels of capital and labor are allocated to nonfarm activities (misallocation of resources);
An alternative empirical strategy to demonstrate that policy interventions indeed resulted in binding constraints on input allocations is to estimate the marginal returns of capital and labor in the two sectors. Evidence on higher returns in nonagriculture would confirm the binding constraints. However, it can be shown that the binding constraints in factor allocations, (kn*?knP)>0 and (ln*?lnP)>0, are necessary and sufficient it it it it conditions for a positive association between factor shares in nonagriculture and overall household profit. Thus, nP nP empirical evidence that higher {dlit ,dkit } actually raised Vit would imply that the resource constraints were actually binding.

D. Tao Yang / Journal of Development Economics 74 (2004) 137–162


(2) that given household endowments, the allocations of capital and labor to nonagricultural activities are positively related to the schooling of household workers, in particular the highest level of schooling attainment (allocative effects); and (3) that given input allocations, the household profit is positively related to the schooling attainment of family workers (worker productivity and other allocative effects). Hypotheses (2) and (3) can be readily expressed in analytical forms. Taking partial derivatives of Eq. (6) with respective to the schooling variables, the mechanisms and sources of schooling returns are derived as: BVit BVit BdnP BVit BdnP BVit kit lit ? nP ? nP ? ; Bsit Bdkit Bsit Bdlit Bsit Bsit ?7?

where the first two terms are associated with allocative effects of sectoral capital and labor uses, and the last term with all other effects of schooling. I will use a panel data set covering 1986 –1995, a period immediately after labor market policy reforms, to describe the changes in farm performance and to test the above hypotheses.

4. Empirical analysis 4.1. Data The data used for this study, which were collected by China’s State Statistical Bureau (SSB), are from the Rural Household Survey (RHS) for the Sichuan province for 1986 – 1995.11 Sichuan, the most populous province in the inland of China, is historically praised as the ‘‘land of fish and rice’’ because of its favorable climatic conditions for farming. The survey rotates a fraction of the sampled households each year. The data consist of two panels, one for 1986– 1989 and the other for 1991 – 1995, that are constructed from the original, complete sample.12 A number of adjustments were required in order to make the data suitable for this study. The Data Appendix provides detailed information on sources and adjustments. Here, I report only a summary description of the data set. In this study, agricultural activities include cropping, animal husbandry, forestry, fishery, and sideline production, a breakdown that is consistent with the standard definition
This national survey, started in 1952, consists of large random samples and records of detailed diary information on production, incomes, and expenditures. Data since 1986 are in computer-usable form, but they have not been released to the public. I have limited access to the Sichuan data through a collaborative project with researchers at the SSB. 12 The two separate panels reflect the facts that SSB started a complete new sample in 1991, and that the 1990 data I received from SSB was corrupt. A second attempt to restore the data by a different means of data transfer was also not successful.


D. Tao Yang / Journal of Development Economics 74 (2004) 137–162

Table 1 Summary statistics of real per capita income and major inputs Year Real per capita income (yuan) (1) 501 507 509 486 (187) (198) (224) (186) – 653 (288) 674 (305) 620 (337) 661 (343) 710 (368) Land per farm (mu) (2) 7.7 6.9 6.5 6.6 6.2 6.1 6.2 6.1 5.9 (8.9) (5.6) (4.9) (5.4) – (4.8) (4.7) (4.9) (5.0) (4.9) Labor force per family (3) 2.6 2.6 2.6 2.7 2.6 2.6 2.6 2.6 2.6 (1.1) (1.0) (1.1) (1.1) – (0.9) (1.0) (1.0) (1.0) (1.0) Capital stock per family (yuan) (4) (643) (697) (1058) (1181) – 1046 (1359) 1101 (1682) 1337 (2124) 1554 (2466) 1613 (2520) 502 551 590 646 Number of observations (5) 809 798 770 788 – 1519 1516 1511 1516 1518

1986 1987 1988 1989 1990 1991 1992 1993 1994 1995

(1) The figures in column 1 are in 1986 prices. The deflator used is the consumer price index of rural residents (SSBb, 1996). Yuan=$0.125. (2) In column 2, mu = 0.165 acre. (3) Figures in column 4 are in nominal prices because of no appropriate price deflator. (4) Figures in parentheses are standard deviations; the same applies to Table 2 – 4.

of ‘‘agriculture’’ in Chinese statistics. Nonagricultural activities consist of a variety of production, ranging from industry to handicrafts. The profit from each line of production is equal to revenue minus variable costs. Wage employment in nonagricultural activities is special because it does not incur variable costs. To our advantage, the RHS records the utilization of capital and labor by industry. Therefore, we can aggregate the factor allocations into agricultural and nonagricultural activities. Table 1 reports summary statistics of income and major inputs owned by the farm households. Real per capita income for the sample rose from 501 yuan in 1986 to 710 yuan in 1995, indicating sustained growth after the initial burst in earnings between 1978 and 1985.13 Despite the small scale of Chinese agriculture, land per farm declined during the period, while the number of workers per family stayed constant. In contrast, the value of capital equipment increased over time, a fact consistent with the national trend. The two panel data sets have, respectively, approximately 8 and 15 hundred households for statistical analysis. Due to missing information for some families in certain years, the number of observations is not exactly the same for each of the years because the year-specific observations with missing information are deleted. The number of observations reported in [column (5) of Table 1] is the sample used for the following empirical analysis. The figures reported in Table 2 on activity-specific allocations of inputs reveal that, in general, the households devoted increasingly more labor and capital to nonagricultural activities over time. During the 10-year period, the share of labor with nonagricultural work as its main occupation increased by about 10% points. Accordingly, the share of
13 For Sichuan, rural real per capita income rose by 115.7% between 1978 and 1985; and it accomplished 40.8% of growth between 1986 and 1995 (SSBb, 1996). The increase of 41.7% in real income for the sample households between 1986 and 1995 is representative of the provincial income growth, despite the fact that there is noticeable discontinuity across the two panels.

D. Tao Yang / Journal of Development Economics 74 (2004) 137–162 Table 2 Allocation of labor and capital in agricultural and nonagricultural activities (household average) Year No. of workers in agriculture (1) 2.35 (1.04) 2.25 (0.96) 2.27 (1.01) 2.35 (1.07) – 2.23 (0.94) 2.21 (0.96) 2.19 (1.01) 2.09 (1.01) 2.10 (0.98) No. of workers in nonagriculture (2) 0.28 (0.44) 0.34 (0.49) 0.35 (0.49) 0.32 (0.46) – 0.34 (0.49) 0.37 (0.53) 0.44 (0.60) 0.54 (0.67) 0.52 (0.65) Labor share in nonagriculture (3) 10.8 13.0 13.3 12.0 – 13.1 14.2 16.8 20.6 19.8 Capital in agriculture (4) 178.6 (360.0) 181.8 (214.0) 171.7 (254.8) 184.7 (254.3) – 286.4 (454.7) 289.5 (483.6) 306.3 (461.0) 367.1 (650.0) 423.0 (1089.2) Capital in nonagriculture (5) (501.8) (630.6) (997.7) (1121.5) – 770.6 (1156.5) 819.4 (1513.0) 1031.6 (2015.8) 1187.2 (2282.0) 1192.0 (2225.6) 323.0 368.8 419.2 461.5


Capital share in nonagriculture (6) 64.4 67.0 71.0 71.4 – 72.9 73.9 77.1 76.4 73.8

1986 1987 1988 1989 1990 1991 1992 1993 1994 1995

capital equipment for nonfarm uses also increased by close to 10%. It appears that nonagricultural production is much more capital-intensive than farming.14 Table 3 contains information on schooling and labor force experience. The RHS reports the level of schooling completion for the rural workers instead of years of schooling attained. I organize the completion levels into four categories: illiterate and semiilliterate, elementary school, secondary school, and high school and plus (see the Data Appendix for details). It is evident from [columns (1) – (4) of Table 3] that the educational attainment of the labor force improved over the 10-year period. The percentage of workers who were illiterate or semiilliterate fell from 31.4 in 1986 to 14.9 in 1995. This fall in the illiteracy rate and the rise in the percentage of workers with middle and high school degrees are mainly due to the entrance into the labor force of more educated workers, the exit of older workers with less schooling, and other changes in family demographics, such as deaths and migration. Within each subperiod, there were some changes in the highest and average levels of schooling within households. For the 1986– 1989 period, 21% of the households reported changes in the highest level of schooling, while 57% of the households had changes in average schooling. For the 1991 –1995 period, the corresponding percentages were 24 and 59. Also note that the average schooling, reported in the level of completion, is significantly below the average highest level of education of the households, indicating schooling variability within families. The average experience of the labor force, defined as (ageschooling-7), is stable over time. For empirical analysis, I will also examine whether geographic environments influence factor allocations and household income growth. Table 4 contains information on the geographic features of the sample (plain, hilly areas vs. mountainous regions) and
14 The statement is subject to one caveat that the definition adopted for occupation may under report the extent of labor participation in nonagricultural activities. A worker is classified as in nonagriculture if, according to the survey, his or her ‘‘main occupation’’ is a nonfarm industry. This definition systematically underrepresents part-time participation in nonfarm activities for some farmers. Unfortunately, there is no other information in RHS that could remedy this data limitation.


D. Tao Yang / Journal of Development Economics 74 (2004) 137–162

Table 3 Schooling and experience of the labor force Year Education of the labor force (%) Illiterate and Primary semiilliterate (1) school (2) 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 31.4 (30.7) 30.9 (30.3) 29.0 (29.4) 28.8 (29.9) – 16.7 (26.0) 16.7 (26.3) 15.2 (24.6) 14.7 (24.5) 14.9 (25.1) 43.0 (33.3) 43.5 (32.9) 44.0 (32.3) 43.5 (32.6) – 45.9 (33.0) 45.4 (33.1) 45.5 (33.0) 44.7 (32.9) 43.8 (32.9) Middle school (3) 22.5 (27.8) 22.6 (27.5) 24.0 (28.4) 24.6 (28.6) – 32.4 (32.3) 32.8 (32.1) 33.6 (31.9) 34.5 (32.2) 35.5 (32.6) Level of Level of Average work average highest experience (7) High school+ (4) education (5) education (6) 3.0 3.0 2.9 2.9 4.9 5.0 5.6 5.9 5.6 (11.3) (10.9) (10.9) (11.0) – (15.5) (15.6) (16.3) (16.7) (16.0) 1.9 (0.5) 1.9 (0.5) 2.0 (0.5) 2.0 (0.5) – 2.2 (0.5) 2.2 (0.5) 2.3 (0.5) 2.3 (0.5) 2.3 (0.5) 2.5 (0.7) 2.5 (0.7) 2.5 (0.7) 2.5 (0.7) – 2.7 (0.7) 2.7 (0.7) 2.7 (0.7) 2.8 (0.7) 2.8 (0.7) 23.1 (6.5) 23.3 (6.4) 23.3 (6.7) 23.4 (6.5) – 21.2 (6.7) 21.6 (6.7) 21.9 (6.5) 22.5 (6.8) 23.5 (7.2)

(1) Schooling attainment reported in columns 5 – 6 is an arithmetic average of schooling completion levels, where illiterate and semiilliterate is represented as level 1, primary school as level 2, middle school as level 3, and high school and plus as level 4. (2) Experience is approximated as (age-schooling years-7).

compulsory levies on agriculture, which reflect policy environments for the households. It should be noted that for each subperiod, the small variations in the percentage of households belonging to a geographic type reflect the fact that the number of observations is not exactly the same for each year. Some families with missing yearspecific information are deleted from the sample for the corresponding years. Moreover, note that the second panel draws a higher percentage of households from plain areas, which in part explains the differences in sample characteristics across the two time periods as revealed in Tables 1 –3. Therefore, caution is needed when making acrosspanel comparisons. 4.2. Econometric specification To estimate the input allocation functions for labor and capital to nonagricultural activities, as in Eqs. (4) and (5), I use the following parametric specification: lnXit ? aX lndit ? aX lnlit ? aX lnkit ? d l k ? X
a a cXa git ? cX fit ? g f


bXc sc ? bX sait ? bX xait sh hit sa x ?8?


cXr Dr ? mi ? eit ; d t

nP nP where Xita{lit , kit }, and Xa{lnP, knP} which denotes input-specific parameters. To simplify notation, I remove the superscript ‘‘np’’ from{lnP, knP} in the following analysis, given the fact that the corresponding parameters are for input demand functions for the nonfarm sector. In Eq. (8), {dit, lit, kit} are aggregate land, labor and capital endowments, c respectively, for household i in year t; shit are dummy variables relating to the completion level for the most educated family worker: c={elementary, secondary, high school and

D. Tao Yang / Journal of Development Economics 74 (2004) 137–162 Table 4 Summary statistics of geographic and policy variables Year Geographic features of sample households (%) Plain areas (1) 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 6.3 5.7 5.9 5.8 – 16.5 15.8 16.4 16.4 16.4 Hilly areas (2) 60.1 61.4 62.0 61.6 – 55.8 56.1 55.1 54.3 53.8 Mountainous areas (3) 33.4 32.8 31.9 32.4 – 27.6 27.9 28.4 29.2 29.7


Compulsory levies on farming (yuan) (4) 31.9 (27.8) 40.8 (33.2) 47.7 (60.2) 63.9 (56.3) – 103.6 (103.4) 107.6 (117.6) 103.6 (144.8) 139.9 (179.3) 155.1 (173.7)

plus}, with elementary school as the reference group;15 sait is the average completion level of schooling for other family workers; and xait is the average work experience for all a family workers. The dummy variables git represent geographic areas: a={plain, hilly, mountainous regions}, with plain areas as the reference group. The variable fit stands for compulsory fees imposed by the government on farming. This estimation equation also includes household fixed effects (mi) relating to factors such as unobserved managerial ability and land quality, as well as region-and time-specific effects (Dtr ) relating to factors such as time-varying weather shocks at the region level and likely price variations both over time and space.16 An additional advantage of having time-regional dummies is that they can control for changes in factor demand in the nonfarm sector. The superscript r in Dtr indexes 23 prefectures from which the sample households were drawn. The error term eit is assumed to represent the effects of remaining omitted variables, which are independent of the explanatory variables and are independently and identically distributed. X Xc X Xa Xr {ad , aX, aX, bsh , bsa , bX, cg , cX, cd } are parameters to be estimated.17 l k x f The two activity-specific factor demand functions will be estimated conditional on total factor endowments for the household in each period, namely land, labor and capital. Accordingly, the parameter estimates would yield meaningful economic interpretations. For instance, the parameter ak would represent the propensity to invest in nonfarm k production: ak = 1 is associated with a neutral strategy, while ak>1 implies a high k k propensity of investment. That is, for a given percentage increase in total capital, a higher
15 The illiterate and semiilliterate group is merged with the group with elementary education because the percentage of households belonging to the former category is low, around 7% for the 1986 – 1989 panel, and dropping below 3% for the 1991 – 1995 panel. Therefore, the following analysis focuses on the efficiency of the households with elementary education relative to the secondary and high school and plus groups. 16 This treatment for price variations is necessary because the RHS does not contain price information for various agricultural and nonagricultural products. The approach taken is consistent with the fact that there likely exist competitive markets in local regions, but not necessarily across regions. 17 For this and the following empirical functions, the superscript to a parameter indicates association with a particular dependent variable. School completion level c, geographic type indicator a, and region index r are also included in the superscript. The subscripts refer to the independent variables associated with the parameters.


D. Tao Yang / Journal of Development Economics 74 (2004) 137–162

percentage will be allocated to the nonfarm sector. Similar interpretations also apply to all. Moreover, because land is only used in agriculture, and if land is a complementary factor l k to capital and labor, we would expect {ad ,ad} < 0, implying that abundant land endowment would have the effect of retaining the other two inputs in agriculture. More importantly, the estimation results will shed light on the effect of schooling attainment on factor allocations to nonagricultural activities. If the highest level of c schooling in the household (shit) is a reasonable proxy for allocative skills, and schooling facilitates a narrowing of the inefficiency gaps during the adjustment period, we would lc kc expect {bsh,bsh }>0. That is, families with middle and high school education systematically to make better input decisions than primary schooled families. The profit function in Eq. (6) that incorporates the effects of input allocation decisions will take the following empirical form: lnVit ? av lndit ? av lnlit ? hv lndnP ? av lnkit ? hv lndnP ? l lit k kit d l k ? bv xait x ? X
a a cva git g


bvc sc ? bv sait sh hit sa ?9?


cv fit f



cdjr Dtr

? mi ? lit :

The parameters for the education variables and factor shares are of particular interest. nP kit vc v First, because factor allocation decisions are already embedded in {dlit ,dnP }, {bsh , bsa}>0 would imply productive value of education on top of its allocative role in capital and labor uses across the two sectors. This additional value may include the use of modern crop varieties, fertilizer, other allocative decisions, as well as worker productivity effects. Second, when allocations of labor and capital are already at the optimum across the two sectors, marginal changes in the shares of labor and capital around their sample means would have no implications for profits, i.e., hlv = hv = 0. However, we would expect {hlv,hv}>0, if the k k sample households were still in the process of making adjustments towards the new equilibrium. When these two parameters are positive, the role of education in reallocating resources to nonagricultural activities would contribute to the farm household earnings. While {hlv,hv} will indicate systematic misallocations of labor and capital to nonfarm k sector, it would be interesting to further examine whether the extent of inefficiency declines with the deepening of economic reforms. We expect that, ceteris paribus, the estimated parameters will become smaller over time as the gaps of distortion are gradually closed up with adjustments. However, if nonfarm opportunities improve continuously with rapid growth and structural transformation, the need for input relocation towards that sector may persist for an extended period. To investigate the efficiency of factor allocations over time, I will add to the basic model in Eq. (9) interaction terms of the nP nP time periods (1986 –1989 and 1991– 1995) with lndlit and lndkit , thus allowing different parameter estimates across the two panels. If the estimated parameters are positive but smaller for the latter panel, they would imply improvements in intersectoral factor utilization with market liberalization. Given the structure of the model, the two factor shares are endogenous variables, e.g., idiosyncratic shocks to individual households (such as being lucky in landing a nonfarm job), would affect both factor shares and household profit, violating the orthogonality

D. Tao Yang / Journal of Development Economics 74 (2004) 137–162


condition of the error term. Therefore, it is necessary to estimate Eq. (9) using instrumental variables. From the model specifications, it is also evident that lagged factor shares may affect input allocation decisions for the current period, but not the current profit level of the household. Therefore, I use factor shares lagged by one period as instruments. Consequently, in addition to controlling for family-fixed effects (mi) and region/time specific effects (Dtr ), the predicted factor shares from the first-stage equations will be used in estimating Eq. (9). 4.3. Estimation results Table 5 reports fixed effect estimates of input demand functions for the allocations of capital and labor to nonagricultural activities based on pooled data of the 1986– 1989 and 1991 –1995 panels. For reference of comparison, I first fit the two input demand functions with a random effects procedure based on the assumption that the effects of unobserved characteristics, such as household managerial skills and time/region fluctuations in weather and prices, are independent of the included explanatory variables. Then, to be consistent with the analytical model presented earlier, I estimate the functions with a fixed effect procedure, controlling for both household (mi) and time/region (Dtr ) heterogeneity. These two procedures yield quite different estimates. To select appropriate specifications, the results of the Hausman test strongly reject the random effect models, suggesting that the unobserved household and time/region effects are dependent on the explanatory variables. As such, the following discussions will concentrate on the fixed effect results. The negative coefficients estimated for the land variable are consistent with the view that land endowment raises the productivity of capital and labor in agriculture, thus

Table 5 Estimates of input demand functions for capital and labor in nonagricultural activities Explanatory variables Dependent variables ln(capital in nonagriculture) Fixed effects estimates (1) ln(land) ln(total labor) ln(total capital) Middle school, highest High school +, highest Average education Average experience Average experience2 (? 1000) Hilly areas Mountainous areas Levies on agriculture (? 1000) R2 ? 0.190** (0.869) ? 0.306** (0.073) 1.383** (0.024) 0.100* (0.055) 0.177* (0.099) 0.017 (0.039) ? 0.005 (0.013) ? 0.018 (0.243) 0.373** (0.169) 0.879** (0.198) 0.117 (0.111) 0.747 ln(labor in nonagriculture) Fixed effects estimates (2) ? 0.058* 0.523** 0.048** ? 0.025 0.131* ? 0.003 ? 0.029** 0.239 ? 0.269** ? 0.457** 0.299** 0.721 (0.032) (0.056) (0.018) (0.043) (0.077) (0.031) (0.010) (0.190) (0.132) (0.155) (0.086)

Both regressions include household and region/time dummies. Sample size is 8480. Standard errors are given in parentheses. * Significant at 10% level. ** Significant at 5% level.


D. Tao Yang / Journal of Development Economics 74 (2004) 137–162

reducing the outflow of these two inputs from farming. The elasticity of capital investment in nonfarm activities, as represented by ak = BlnknP/Blnk, indicates that a 1% increase in k total capital is associated with a 1.38% increase in nonagricultural uses. Therefore, new capital is increasingly being directed towards rural industries and services. Agriculture still appears to be the main sector for labor employment because the elasticity of nonagricultural labor allocation, all = BlnlnP/Blnl, is 0.52. With regard to education, the findings support the hypothesis that allocations of capital and labor to the nonfarm sector are positively related to the schooling attainment of the household. Using the highest level of education as a proxy for household allocative skills and controlling for household aggregate capital assets, the middle-schooled families devote 10% more capital to nonfarm uses than the reference group, the primary-schooled households. Households having members with a high school or college education invest even more capital in nonfarm uses 17.7% higher than the primary-schooled households and 7.7% higher than the middle-schooled households. Moreover, the data yield evidence that households having high school and college-educated members allocate 13.1% more labor to nonfarm activities relative to the reference group, although middle-schooled families do not appear to allocate more workers away from agriculture. There is also evidence of centralized decision making on the farms supported by the fact that the average education of family workers excluding the highest attainment does not contribute significantly to either capital or labor allocations. Moreover, the results indicate that, while the experience of workers does not significantly influence the intersectoral distribution of capital, older farmers with more general work experience are less likely to participate in nonfarm work. This finding is consistent with the standard implications of human capital theory.18 With regard to geographic and policy variables, the estimates indicate that levies on agriculture discourage farming activities, thus having the effect of encouraging factor allocations to nonfarm uses, although the coefficient for capital does not reach the conventional level of significance. Geographic location also has a significant impact on sectoral input allocation. Being in hilly and mountainous regions increases capital investment in nonfarm uses, a result that is consistent with the view that adverse geographic characteristics may have relatively less negative effect on returns for capital in the nonfarm sector. In contrast, being in hilly and mountainous regions is associated with less labor utilization in nonagriculture, which may reflect the fact that geographically disadvantaged locations present less nonfarm opportunities. Table 6 reports estimation results for the profit function in Eq. (9) and its two variant specifications. The model in column (1) is a baseline case excluding labor and capital shares, and the model in column (3) includes interaction terms of the two time periods with factor shares in order to investigate changes in resource misallocation over time. Since the factor shares are choice variables, their lagged values are used as instruments for models
18 Because the experience measure is highly correlated with the age of laborers, the finding is also consistent with a cohort preference effect. That is, the newcomers to the labor force prefer leaving the farm independent of their work experience. Unfortunately, I am not able to disentangle the preference and human capital effects due in part to the lack of data on farm vs. nonfarm experience. I thank a referee for pointing out the issue regarding cohort preference.

D. Tao Yang / Journal of Development Economics 74 (2004) 137–162 Table 6 Estimates of profit functions Explanatory variables Dependent variable = ln(household net profit) Fixed effects (1) ln(total land) ln(total labor) ln(labor share in nonagriculture) ln(labor share in nonagriculture) ? first period ln(labor share in nonagriculture) ? second period ln(total capital) ln(capital share in nonagriculture) ln(capital share in nonagriculture) ? first period ln(capital share in nonagriculture) ? second period Middle school, highest High school and +, highest Average education Average experience Average experience2 ( ? 1000) Hilly areas Mountainous areas Levies on agriculture ( ? 1000) R2 0.105** (0.017) 0.185** (0.031) – – – 0.084** (0.029) – – – 0.030 0.014 ? 0.007 0.016** ? 0.339** 0.062 0.137* ? 0.197** 0.719 (0.023) (0.041) (0.017) (0.006) (0.103) (0.070) (0.082) (0.047) IV Fixed effects (2) 0.137** (0.018) 0.138** (0.037) 0.178** (0.038) – – 0.027 (0.030) 0.215** (0.048) – – 0.019 ? 0.043 ? 0.000 0.022** ? 0.352** 0.103 0.152 ? 0.269** 0.722 (0.023) (0.042) (0.017) (0.006) (0.103) (0.073) (0.091) (0.049)


IV Fixed effects (3) 0.135** (0.019) 0.136** (0.037) – 0.182** (0.044) 0.179** (0.038) 0.035 (0.043) – 0.309** (0.071) 0.193** (0.049) 0.018 ? 0.040 0.003 0.022** ? 0.344** 0.107 0.149 ? 0.265** 0.722 (0.023) (0.042) (0.016) (0.005) (0.102) (0.072) (0.091) (0.048)

All regressions include household and region/time dummies. Sample size is 8480. Standard errors are given in parentheses. * Significant at 10% level. ** Significant at 5% level.

(2) and (3). Table 7 reports the first-stage regressions in which the lagged factor shares appear to have strong explanatory power for the current period factor allocations.19 Specification tests are also performed in which the Hausman test statistics strongly reject the random effect model in favor of the alternative fixed effect specification. Therefore, the following discussions will concentrate on the IV fixed effect estimates. In column (2) of Table 6, family-owned assets—land, labor and capital—all contribute to household earnings, although the coefficient for capital is not statistically significant. Controlling for the level of these quasifixed factors, if labor and capital are already optimally allocated across the two uses, any intersectoral redistribution, as reflected in {dlnP, dnP}, would have no effect on profits. The positive and statistically significant k coefficients estimated for these parameters indicate that, during the period of adjustment in rural China, the allocations of the two inputs have not yet reached the optimum. In particular, ceteris paribus, a 10% increase in the share of labor in nonfarm activities would
19 The F-values are F(2,6190) = 18.13 in the capital share function and F(2,6190) = 11.11 in the labor share function, both rejecting the null hypotheses at 1% significance level that the two coefficients for the instruments are jointly 0.


D. Tao Yang / Journal of Development Economics 74 (2004) 137–162

Table 7 First-stage instrumental variable estimates for capital and labor shares in nonagricultural activities Explanatory variables ln(capital share in nonagriculture) ln(labor share in nonagriculture) Fixed effects estimates (1) ln(lagged labor share in nonagriculture) ln(lagged capital share in nonagriculture) ln(land) ln(total labor) ln(total capital) Middle school, highest High school +, highest Average education Average experience Average experience2 (? 1000) Hilly areas Mountainous areas Levies on agriculture (? 1000) R2 0.010* (0.006) 0.085** (0.014) ? 0.103** (0.024) ? 0.164** (0.041) 0.241** (0.014) 0.058* (0.032) 0.097* (0.057) 0.003 (0.022) ? 0.006 (0.007) 0.043 (0.140) 0.139 (0.097) 0.435** (0.114) 0.047 (0.063) 0.615 Fixed effects estimates (2) 0.104** (0.012) 0.110** (0.030) ? 0.072 (0.050) 0.473** (0.087) 0.071** (0.029) 0.001 (0.067) 0.239** (0.120) ? 0.011 (0.005) ? 0.023 (0.016) ? 0.015 (0.294) ? 0.419** (0.204) ? 0.642** (0.239) 0.395** (0.133) 0.711

Both regressions include household and region/time dummies. Sample size is 8,480. Standard errors are given in parentheses. * Significant at 10% level. ** Significant at 5% level.

raise household incomes by 1.8%, whereas a 10% increase in the share of capital in nonfarm activities would raise incomes by 2.2%. These findings provide support for the hypothesis that less-than-optimal levels of capital and labor were allocated to rural industries and services during the period of transition. Combining these results with the findings in Table 5, we see evidence that more schooled households allocated more inputs to nonagriculture, and that these adjustments in turn raised family earnings. In addition to the allocative role of education embedded in the factor shares, schooling variables are also included directly in the profit function to capture other aspects of their effects on earnings. However, those coefficients are not statistically significant. We also find that experience contributes to income through a concave schedule. Controlling average experience at 0, the marginal value of 1 year of experience is 2.2% of annual household profit; at sample mean level of experience, the marginal value is 0.6% of the profit. Conditional on family endowments and resource allocations, households who live in hilly and mountainous regions are no more disadvantaged than households living in plain areas, according to the data. This finding could be sample specific, however, because a large percentage of farm households in the Sichuan province (see Table 4) live in hilly and mountainous areas where the climate and soil conditions are generally good. Consistent with expectations, compulsory taxation on farming unambiguously reduces overall household earnings. [Column (3) in Table 6] presents the results of interacting factor shares in nonagricultural activities with the two time periods, thus allowing changes in the extent of resource misallocation over time. The estimated coefficients for all variables remain stable, and the estimates for the factor shares are both positive and significant, confirming the existence of resource misallocation. More specifically, the estimates show that the elasticity of profit

D. Tao Yang / Journal of Development Economics 74 (2004) 137–162


with respect to capital shares in nonagricultural activities is reduced considerably. In the first period (1986 – 1989), ceteris paribus, a 10% increase in the share of capital in nonfarm production would raise total income by 3.1%, but it lowered to 1.9% in the second period (1991 –1995). This result implies that policy reforms had reduced the severity of capital misallocation over time. For the labor shares, the coefficients for the two periods are not statistically different. This result is consistent with the possibility that rapid economic structural changes taking place in rural China during transition may have created continued demand for labor in the nonfarm sector, and that the adjustments had not fully responded to the changes. It would be interesting to investigate the changes over a longer period of time. Using estimates reported in Tables (5) and (6), decomposition analysis as outlined in Eq. (7) can be conducted to shed light on the sources and mechanisms of schooling returns to farm profits. Because BlnknP/Bs = bskc as in Eq. (8) and BlndnP = BlnknP when aggregate k capital stock is held fixed, it follows BlndnP/Bs = BlnknP/Bs = bskc, where s=(sh, sa). k Similarly, BlndlnP/Bs = Blnl nP/Bs = blc. Therefore, applying Eq. (7) to the empirical s estimation with logarithmic specifications, the marginal value of schooling can be computed as: BV ? ? ?v ? ? ? hv ? bkc ? h1 ? blc ? bv ; k s s s Bs ?k ? ?s where BlnV/BlndnP = h v , BlnV/BlndlnP = h vl, and BlnV/Bs = b v from Eq. (9); the parameters k with hats are estimated values. The first term captures the rate of schooling returns from intersectoral capital allocation; the second term relates to the rate of schooling returns associated with intersectoral labor allocation; and the last expression is the rate of returns associated with all other sources of schooling’s effects on profitability. Empirical results reported in Table 5 and [column (2) of Table 6] suggest that for this sample of Chinese farms, the effects of education on profitability come primarily from optimally allocating capital and labor across the two sectors. If statistically insignificant estimates are ignored, when the highest level of education of the household is raised from primary to middle school, the return would be 2.15% (= 0.10 ? 0.215) of the total household profit. This return is attributable to more efficient capital allocations. If the highest level of education is raised from primary to high school, the family earnings would increase by 6.13%; this effect is attributable to more efficient capital (3.8% = 0.177 ? 0.215) and labor (2.33% = 0.178 ? 0.131) allocations. These results corroborate earlier empirical findings reported by Yang (1997) based on cross-sectional data from rural China that schooling returns come primarily from allocative decisions and that the highest level of schooling contributes the most to farm efficiency. Finally, data reported in Table 2 indicate that labor shares in nonfarm activities increased by 83.8% from 10.8 to 19.8 between 1986 and 1995, and capital shares increased by 14.5% from 64.4 to 73.8. Based on estimates presented in [column (2) of Table 6], these changes in intersectoral input allocations would result in an 18% increase in farm household earnings. Therefore, the rapid expansions in nonfarm activities would account for approximately 43.6% of the total farm income growth because real earnings grew by 41.2% in this 10-year period (see Table 1). While caution must be given to these


D. Tao Yang / Journal of Development Economics 74 (2004) 137–162

estimates, due to the fact noted earlier that there is an issue of data comparability of the two panels, these sample estimates are nevertheless broadly consistent with provincial level information.20 Although the model and results presented in the paper are not suited to trace out many dynamic effects of the changes, the mechanisms concerning the effects of education on input allocation and profits are clearly revealed. Schooling significantly reduces the inefficiency of intersectoral input allocations; the increase in efficiency in turn contributes to farm household income growth.

5. Concluding remarks This paper uses panel data with rich economic and demographic information to investigate the sources and determinants of sustained income growth in rural China between 1986 and 1995. This was a period of market adjustment when the relaxation of labor mobility controls induced rural families to reallocate their productive inputs from agriculture to nonagricultural activities. The findings suggest that schooling plays a critical role in raising the efficiency of farmers to respond to changing market conditions. Households with better-educated members acted more quickly in devoting more capital and labor to nonagricultural activities that yielded higher returns. As a result of these allocative effects, education contributed significantly to the sustained rural income growth. The above findings have wider implications than simply improving the understanding of a special period of income growth in rural China. The centrally planned system created massive misallocation of resources both within the rural sector and across rural – urban regions. While past reforms have greatly improved allocative efficiency within the rural economy, China is still facing long-term, arduous structural adjustments across the sectors.21 Mobility of resources will be a key aspect of this process; consequently rural schooling will have a high value during the transition.22 In view of the fact that the conditions of many rural schools have deteriorated during recent organizational changes, public attention and investment in infrastructure is imperative. To a large extent, this study of China also mirrors the experience of other developing countries, in which rural people must face the selection of income activities and the prospect of leaving agriculture. Schooling plays a critical role in these allocative decisions for raising incomes in the current period as well as for the future.

20 At the provincial level, rural labor shares in nonfarm activities increased from 12.8% in 1986 to 22.2% in 1995, an increase of 73.18% (SSBb, 1996). In addition, the growth in real per capita income of the sample is comparable with the provincial level growth (see Footnote 13). Unfortunately, aggregate statistics do not contain sufficient information for computing capital utilization devoted to agriculture vs. nonagriculture. 21 For instance, Johnson (2000) forecasts that the agricultural labor force in China will likely to fall by as much as 63% between 1997 and 2030. Therefore, the rural nonfarm sector and urban regions will face a long-term challenge of absorbing large number of workers released from agriculture. 22 This view is corroborated by Zhao (1997) who finds that senior high school education raises the accessibility of rural people to urban formal employment.

D. Tao Yang / Journal of Development Economics 74 (2004) 137–162


Acknowledgements I would like to thank Barry Naughton, Tiejun Wen, Shukai Zhao, Xiaodong Zhu, conference participants in New Research on Education in Developing Countries at Stanford University, and especially Albert Park and two anonymous referees for valuable discussions and suggestions. I am also grateful to Xian Zude, Sheng Laiyun, Wang Pingping and other researchers at Rural Survey Organization of China’s State Statistical Bureau for data support, and to the Center for Research on Economic Development and Policy Reform at Stanford University for financial support.

Appendix A The Rural Household Survey (RHS) organized and collected by China’s State Statistical Bureau (SSB) has been the major and continued source of statistical information for rural China. Chen and Ravallion (1996) provide detailed descriptions about the survey design, sampling issues, administration, and contents of the Survey. This paper uses the data from the Sichuan province covering 1986– 1995. Note that starting in 1993, SSB changed the questionnaire for RHS. Variables were extended from 640 to 1412, covering additional information. This paper uses variables that can be found in both sets of questionnaires. The following paragraphs describe the method adopted for constructing the two panel data sets, variable definitions, and adjustments that are necessary to make the data suitable for econometric analysis. A.1 . Construction of panel data In 1986, RHS in Sichuan started with about four thousand rural households. In each of the following years, SSB officials replaced some households with new samples. The general guidelines suggested the rotation of about 20% of the sample every year, but in practice, the rate of rotation may differ significantly from the general rule. A key feature of the rotation is that the replaced household and the replacing household always share the same household code, which is a 10-digit number that distinguishes the province, the county, the village, and the household number within the village. However, no information is available regarding sample rotation. In 1991, SSB completely dropped the old sample and started with an entirely new sample. Over the years, the size of the sample grew, reaching approximately 6000 households in 1995. My goal is to construct two panels: the 1986 –1989 panel consisting of households that were surveyed for four consecutive years during the period (see Footnote 12 for information about 1991 data) and the 1991– 1995 panel consisting of households that stayed in the survey for five consecutive years. I use three sets of information to sort out the panels: the household code, the age of the household head, and the age of the head’s spouse. On a year-to-year basis, for two households with the same family code, I match the age of the household head in the first year sample with the age of the head in the second year sample. If the age of the head’s spouse also matches with the age of the spouse in the second year sample, I designate these two families as the same households. Otherwise, if


D. Tao Yang / Journal of Development Economics 74 (2004) 137–162

the age of one pair does not match, I consequently drop the household. This method is applied to the entire sample. Because the probability of matching the age of two different couples is close to 0, given the additional information on household code, the sorted panel is therefore very reliable. Table 1 reports the number of households available for statistical analysis for the two subperiods 1986 –1989 and 1991– 1995. A.2 . Income and variable costs in agriculture Consistent with the standard definition in Chinese statistics, agricultural activities include cropping, animal husbandry, forestry, fishery, and sideline production. RHS reports information on the sales and the variable production costs for each of these activities. A.3 . Income and variable costs in nonagricultural production Nonagricultural activities include handicraft, industry, construction, transportation, commerce, food retail, services and other nonagricultural activities. RHS also reports information on the sales and the variable production costs for each of these nonfarm activities. RHS reports wage information separately from the above self-employed activities. For the 1986 –1992 survey, wage earnings include incomes ‘‘from collectives, economic unions, and labor employment.’’ For the 1993 –1995 survey, labor earnings is the sum of labor incomes from ‘‘collective organizations, enterprises, and other work units.’’ A.4 . Household profit This variable is defined as the sum of net incomes from agricultural and nonagricultural sources. A.5 . Land RHS records three types of land: cultivated land, mountainous land, and water areas for fish ponds. The land area used in this paper is the sum of all three types. A.6 . Capital The fixed assets are reported in their ‘‘original value.’’ Agricultural capital includes draft animals, large and middle-size farming tools, and machinery for cropping, forestry, fishery and animal husbandry. Nonagricultural capital includes industrial machinery, transport equipment, households for production and other nonagricultural fixed assets. A.7 . Labor Labor is recorded according to their capacity as ‘‘whole’’ or ‘‘half’’ workers. The paper uses the sum of these capacity measures to approximate the household labor force.

D. Tao Yang / Journal of Development Economics 74 (2004) 137–162


A.8 . Schooling and experience Education is reported in completion levels rather than years of schooling. The six levels are illiterate/semiilliterate, primary school, middle school, high school, technical school, and specialized colleges. Because there are only few workers belonging to the two highest categories, I group them with the high school graduates as ‘‘high school and plus.’’ To approximate work experience, the six completion levels are first converted into years of schooling with the designation of 3, 5, 8, 10, 11, and 13 years of education for each category. Then, experience is computed as (age-schooling-7). A.9 . Levies on agriculture This item includes government extractions through compulsory procurement and other types of agricultural levies.

Anderson, D., Leiserson, M.W., 1980. Rural nonfarm employment in developing counties. Economic Development and Cultural Change 28 (2), 227 – 248. Ash, R.F., 1988. The evolution of agricultural policy. China Quarterly 116, 529 – 555. Basu, K., Narayan, A., Ravallion, M., 2001. Is literacy shared within households? Theory and evidence for Bangladesh. Labour Economics 8 (6), 649 – 665. Chen, S.H., Ravallion, M., 1996. Data in transition: assessing rural living standard in southern China. China Economic Review 7 (1), 23 – 56. Fan, S.G., Pardey, P.G., 1997. Research, productivity, and output growth in Chinese agriculture. Journal of Development Economics 53, 115 – 137. Findlay, C., Watson, A., Wu, H.X., 1994. Rural Enterprises in China St. Martin’s Press, New York. Fleisher, B., Dong, K.Y., Liu, Y.H., 1996. Education, enterprise organization, and productivity in the Chinese paper industry. Economic Development and Cultural Change 44 (2), 571 – 587. Foster, A., Rosenzweig, M., 1996. Technical change and human capital returns and investments: evidence from the Green Revolution. American Economic Review 86 (4), 931 – 953. Huang, J.K., Rozelle, S., 1996. Technological change: rediscovering the engine of productivity growth in China’s rural economy. Journal of Development Economics 49, 337 – 369. Huffman, W.E., 1977. Allocative efficiency: the role of human capital. Quarterly Journal of Economics 91, 59 – 77. Jamison, D.T., Lau, L.J., 1982. Farmer Education and Farm Efficiency. Johns Hopkins Univ. Press, Baltimore. Johnson, D.G., 2000. Agricultural adjustment in China: problems and prospects. Population and Development Review 26 (2), 319 – 334. Li, T.Y., Zhang, J.S., 1998. Returns to education under collective and household farming in China. Journal of Development Economics 56, 307 – 335. Lin, J.Y., 1992. Rural reforms and agricultural growth in China. American Economic Review 82, 34 – 51. McMillan, J.W., Zhu, L.J., 1988. The impact of China’s economic reforms on agricultural productivity growth. Journal of Political Economy 97, 781 – 807. Naughton, B., 1996. Growing Out of the Plan: Chinese Economic Reform 1978 – 1993. Cambridge Univ. Press, Cambridge, UK. Orazem, P.F., Vodopivec, M., 1995. Winner and losers in transition: returns to education, experience, and gender in Slovenia. World Bank Development Review 9 (2), 201 – 230.


D. Tao Yang / Journal of Development Economics 74 (2004) 137–162

Pitt, M.M., Sumodiningrat, G., 1991. Risk, schooling and the choice of seed technology in developing countries: a meta-profit function approach. International Economic Review 32, 457 – 473. Putterman, L., 1993. Continuity and Change in China’s Rural Development: Collective and Reform Eras in Perspective. Oxford Univ. Press, New York. Rosenzweig, M.R., 1995. Why are there returns to education? American Economic Review 85, 153 – 158. Schultz, T.W., 1975. The value of ability to deal with disequilibria. Journal of Economic Literature 13, 827 – 846. Schultz, T.P., 1988. Education investment and returns. In: Chenery, H., Srinivasan, T.N. (Eds.), Handbook of Development Economics, vol. 1. North-Holland, New York. Sicular, T., 1988. Agricultural planning and pricing in the post-Mao period. China Quarterly 116, 671 – 705. SSBa, State Statistical Bureau of China, 1988 – 1998. Statistical Yearbook of China. China Statistical Publisher, Beijing. SSBb, State Statistical Bureau of China, 1996. China Regional Economy: A Profile of Years of Reforms and Opening-Up. China Statistical Publisher, Beijing. Strauss, J., Thomas, D., 1995. Human resources: empirical modeling of household and family decisions. In: Srinivasan, T.N., Behrman, J.R. (Eds.), Handbook of Development Economics, vol. 3. North-Holland, Amsterdam, pp. 1883 – 2005. Taylor, J.E., Yunez-Naude, A., 2000 (May). The returns from schooling in a diversified rural economy. American Journal of Agricultural Economics 82, 287 – 297. Welch, F., 1970. Education in production. Journal of Political Economy 38, 35 – 59. Wen, T.J., 2000. Rural China: A Study of the Basic Economic System (in Chinese). China Economic Press, Beijing. Yang, D.T., 1997. Education in production: measuring labor quality and management. American Journal of Agricultural Economics 79, 764 – 772. Yang, D.T., An, M.Y., 2002. Human capital, entrepreneurship, and farm household earnings. Journal of Development Economics 68 (1), 65 – 88. Yang, D.T., Zhou, H., 1999. Rural-urban disparity and sectoral labor allocation in China. Journal of Development Studies 35 (3), 105 – 133. Yao, Y., 1999. Rural industry and labor market integration in eastern China. Journal of Development Economics 59, 463 – 496. Zhao, Y.H., 1997. Labor migration and returns to rural education in China. American Journal of Agricultural Economics 79, 1278 – 1287.