PS as+< POLICY RESEARCH WORKING PAPER 2879 Externalities in Rural Development Evidence for China Martin Ravallion The World Bank Development Research Group m Poverty Team August 2002 I_POLICY RESEARCH WORKING PAPER 2879 Abstract Ravallion tests for external effects of local economic composition of local economic activity and private activity on consumption and income growth at the farm- returns to local human and physical infrastructure household level using panel data from four provinces of endowments. The results suggest an explanation for rural post-reform rural China. The tests allow for underdevelopment arising from underinvestment in nonstationary fixed effects in the consumption growth certain externality-generating activities, of which process. Evidence is found of geographic externalities, agricultural development emerges as the most important. stemming from spillover effects of the level and This paper-a product of the Poverty Team, Development Research Group-is part of a larger effort in the group to better understand the causes of poverty. Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Catalina Cunanan, room MC-3-542, telephone 202-473-2301, fax 202-522-1151, email address ccunanan@worldbank.org. Policy Research Working Papers are also posted on the Web at http:H/ econ.worldbank.org. The author may be contacted at mravallion@worldbank.org. August 2002. (35 pages) The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Produced by the Research Advisory Staff Externalities in Rural Development: Evidence for China Martin Ravallion' World Bank, 1818 H Street NW, Washington DC 20433, USA mravallion()worldbank.org Key words: Consumption growth, income growth, externalities, panel data, rural China JEL classification: D91, RI1, Q12 I The data used here were kindly provided by China's National Bureau of Statistics, and I am grateful for the assistance and advice provided by NBS staff in Beijing and at various provincial and county offices. For help with setting up the panel data set I am grateful to Shaohua Chen and Qinghua Zhao and for help with the calculations reported here I am grateful to Jyotsna Jalan. The support of the World Bank's Research Committee and a Dutch Trust Fund is also gratefully acknowledged. For their comments, I am grateful to Jyotsna Jalan, Peter Lanjouw, Forhad Shilpi, Dominique van de Walle and participants/discussants at the Cornell/LSE/WIDER Conference on Spatial Inequality and the World Bank's Economist's Forum 2002. 1. Introduction There is a long-standing view that externalities play an important causal role in economic development. Famously, Rosenstain-Rodan (1943) argued that the investment decisions made by one firm in a developing economy influenced the profitability of others, leading him to argue for international assistance for the industrialization of the lagging regions of Eastern and Southern Europe in the 1 940s. More recently, the hypothesis that there are extemalities through knowledge spillovers has been built into theoretical models of economic growth (notably Romer, 1986 and Lucas, 1993). In the context of rural development in poor countries, similar ideas have motivated policy arguments that getting one activity going locally stimulates others, in a "virtuous cycle" of growth; Mellor (1976) provided an influential statement of this hypothesis.2 Hazell and Haggblade (1990) tested the hypothesis using district and state level data for India, and reported seemingly strong effects of agricultural growth on rural nonfarn development.3 This paper explores the micro-empirical foundations of these arguments using household panel data for a developing rural economy. Some stylized facts about the setting will help motivate the subsequent analysis. One such fact is that in a poor rural economy, the income gains that are claimed to stem from linkage will be transmitted in large part through the farm- household economy, which accounts for the bulk of rural economic activity in most developing countries. No doubt, spillover effects will also involve rural-based firms. However, it is plausible in this setting that any external impacts of local economic activity on income growth would be evident at the farm-household level. A second stylized fact is that many farm- households engage in multiple activities simultaneously, including nonfarm activities. Casual 2 Building on Mellor and Lele (1972). Much earlier still, Clark (1940) had argued that higher agricultural productivity was a crucial precondition for industrialization. 3 Also see Haggblade et al. (1989) and Haggblade et al. (2002). Lanjouw and Lanjouw (2001) provide a useful review of the arguments and evidence on the rural nonfarm sector. 2 observations do not suggest that it is commonly the case that a rural household is fully specialized in either farm or nonfarm activities. Indeed, it has been argued that such income diversification is an important strategy by which rural households cope with uninsured risk (see, for example, Ellis, 1998). There is a large literature pointing to the problems of incomplete credit and risk markets in underdeveloped rural economies (for an overview see Besley, 1995). It is not implausible that there are extemalities in this setting. One way this happens is when farmers learn about new techniques of production from the experience of their neighbors; Feder and Slade (1985) provide survey evidence for northwest India that this is an important channel for knowledge diffusion amongst farmers. Foster and Rosenzweig (1995) find evidence of this type of leaming extemality in farm profitability from adopting new seed varieties in India. Network effects in the marketing of agricultural products can also generate externalities: a farmer can benefit from the infrastructure already in place locally. Another possible source of externalities is the presence of local nonfarm industries that encourage the acquisition of knowledge and skills that also benefit local farmers or non-farm enterprises at household level, possibly through knowledge sharing within households (Basu et al., 2002). In the case of China, it has been argued that higher output from the non-farm sector has brought extemal benefits to the traditional farm sector, through improved technologies and management (Sengupta and Lin, 1995). Or a higher density of commercial enterprises may enhance the local tax base, allowirig better local public goods, and so promoting higher growth for those not actually engaged in those enterprises. Alternatively, negative externalities might result when the expansion of one activity creates congestion, or otherwise crowds out, another activity. The most obvious way this could happen is though the existence of local-level fixed factors of production (including environmental assets) that are shared across activities. For example, with imperfect credit 3 markets leading to rationing of the available credit, an expansion in one activity may crowd out growth prospects in another. With restricted migration and wage stickiness, the same could happen with regard to labor. If the patterns found in aggregate data reflect such externalities this would provide an important insight into the causal processes creating rural underdevelopment. That depends crucially on whether markets exist for the externalities.4 That cannot be judged on a priori grounds. However, a complete set of such markets is not inherently plausible for the sorts of extemalities discussed above. Knowledge spillovers or network effects do not lend themselves to the excludability properties needed for a market. (It would clearly be difficult to define and enforce property rights for such externalities.) So there must be a reasonable presumption that private decision-makers will not typically take account of the extemal costs and benefits of their allocative decisions and so one will expect to see under-investment in the activities that generate positive externalities, and over-investment in those that have negative externalities. The externalities then impede or distort rural development. On the other hand, if the underlying linkage effects are purely intemal at the farm- household level then their welfare and policy significance is greatly diminished.5 Given the stylized facts summarized above, the averaging of purely intemal effects within diversified farm- household units could readily generate the appearance of extemalities in economic activity in aggregate data when in fact none exist at the micro level. For example, given capital market imperfections, higher farm income for a given household may create the resources needed to 4 On the economic theory of markets for externalities, see Dasgupta and Heal (1979, Chapter 3). s It is often argued that the same is true if the externalities are "pecuniary," meaning that they are transmitted through prices. However, it is known that with incomplete markets, pecuniary externalities can still be a source of inefficiency (Greenwald and Stiglitz, 1986; Hoff, 1998, 2000). The externality transmitted through prices could exacerbate the pre-existing inefficiency. 4 finance a new nonfarm activity. Farm and nonfarm incomes may then co-move in a process that one might identify as inter-sectoral linkage in aggregate data even though there is no genuine externality involved. The causal connection is of course unclear, nor is it obvious that there would be any believable identification strategy. The concern with geographic externalities goes beyond economic efficiency. It also raises concerns about horizontal equity. In particular, if the micro growth process involves such externalities then the economy will reward otherwise identical individuals differently depending on where they live. This may also help understand geographic dimensions of social unrest, as has been evident in China in the 1990s.6 Motivated by these observations, the central question addressed in this paper is whether the signs of linkage amongst economic activities found in geographic data stem from externalities. From what we know about the features of a developing rural economy it is clear that one cannot conclude from the existing literature on linkages in rural development that externalities are present to any significant extent. The signs of linkage in geographically aggregated data could easily stem from a process in which there is in fact no interdependence amongst individual farm-household units. Testing for externalities poses a problem, even with micro panel data. Correlations between individual outcomes and geographic variables have been widely reported in the literature. However, as is well-recognized, one cannot assume that the geographic placement of 6 For example, an article in the New York Times (Dec. 27, 1995, p.1) wrote that: "As China's economic miracle continues to leave millions behind, more and more Chinese are expressing anger over the economic disparities between the flourishing provinces of China's coastal plain and the impoverished inland, where 70 million to 80 million people cannot feed or clothe themselves and hundreds of millions of others are only spectators to China's economic transformation." 5 economic activity is exogenous at the micro level.7 Placement in a given locality cannot be expected to be independent of the characteristics of the households that live there - no doubt including characteristics that are unobserved by the analyst. Persistent spatial concentrations of individuals with personal attributes that inhibit growth in their living standards, and lead to a worse assignment of geographic assets, can readily entail that the cross-sectional correlations often found in the data are entirely non-causal, with little or no bearing on development policy. All one is really picking up in the data is the fact that households who are poor in terms of some latent characteristic tend to be grouped together spatially and are less able to attract infrastructure and other geographically assigned resources. To make this argument more concrete, consider t. ,..v,... - ... economy, the quality of farmland is likely to be important to the productivity of current and past investments and hence economic growth. Land quality tends to be spatially correlated; the quality of one farmer's land is positively correlated with the quality of his neighbor's. However, land quality is rarely captured well even in quite comprehensive surveys. At the same time, one can expect that the composition of economic activity and the placement of rural infrastructure (irrigation, roads and so on) will be influenced by land quality. In such seemingly plausible circumstances, one can expect to find correlations between one fanner's income growth rate over time and the attributes of the area in which he lives, even controlling for observable characteristics of the farmer, such as his capital stock. That correlation might look like an externality, but it may simply be picking up the geographically associated latent heterogeneity in land quality. For example, Foster and Rosenzweig (1996) report a significant coefficient on village placement of agricultural extension services in regressions for the adoption of high yielding varieties in micro data for India. As they point out, this cannot be considered a causal effect since the placement of extension services may depend on geographically-associated latent factors influencing adoption. 6 The paper presents results of a test for geographic externalities through the composition of economic activity that is robust to such latent heterogeneity. Both household panel data and geographic data are clearly called for to have any hope of identifying geographic extemalities in the growth process at the micro level. In modeling such data, one might turn to a standard panel data model with a time-invariant error component, as in (for example) the regressions for farm profits in Foster and Rosenzweig (1995). Allowing for latent heterogeneity in the household- level growth process will protect against spurious geographic effects due to time-invariant omitted variables. However, standard panel-data methods of eliminating the household-specific effect wipe out the time-invariant geographic variables of interest in this context, namely the initial composition of economic activity in the locality. Nor is it plausible that the latent heterogeneity in growth rates is time invariant; macroeconomic and geo-climatic conditions might well entail that the impact of these individual effects varies from year to year. However, by simply relaxing the assumption that the fixed effect has a time-invariant impact one can estimate the effect of geographic differences in the observed initial level of economic activity on the micro growth process robustly to the latent heterogeneity. In particular, .he analysis in this paper allows for nonstationary individual effects in the micro growth process, following Holtz-Eakin, Newey and Rosen (1988) and Jalan and Ravallion (2002). The analysis combines geographic data on the composition of economic activity and infrastructure endowments with longitudinal micro observations of consumption and income growth by sector. The growth rate of household consumption is decomposed by income source to explore the income effects of geographic differences in the composition of economic activity and other geographic characteristics. This allows a reasonably flexible description of the patterns of externalities within and between sectors of the economy, as they affect the growth process. 7 The following section outlines the econometric model. Section 3 describes the setting and data while section 4 presents the results. Section 5 summarizes the conclusions. 2. Econometric model The starting point is the following model of consumption growth for N households observed over Tperiods: A ln C1, = a +/, Xi, + ( Zi + i, (i=l,..,N; t2,..,7) (1) where C,1 is consumption by household i at date t, A In Ct is the growth rate of consumption, Xi, is a vector of time-varying explanatory variables, and Zi is a vector of exogenous time-invariant explanatory variables including measures of the initial economic activity in the locality in which household i lives. (The properties of the error term, ej,, are discussed below.) An economic model motivating equation (1) can be derived from a version of the Ramsey (1928) model of consumption growth with capital immobility (Jalan and Ravallion, 2002). In this model, output of the farm household is a concave function of the household's own-capital, but output also depends non-separably on characteristics of the area of residence, including the composition of economic activity. Given the constraints on access to credit, marginal products of own-capital are not equalized across farm-households. Households maximize the standard inter-temporally additive utility integral, with common preferences. The optimal rate of consumption growth is then directly proportional to the marginal product of own capital, which in turn depends on both the farm-household's capital stock and its geographic characteristics. The key feature of this model for the present purpose is that geographic externalities can influence consumption growth rates at the farm-household level, through their effects on the productivity of private investment, given capital market imperfections. (The extreme case in 8 which markets worked perfectly would imply that we had no power to explain the growth in consumption at the farm-household level.) Equation (1) is then obtained by assuming that the marginal product of own capital at the farm-household level is a linear function of Xi, and Zi. The assumptions made about the error term in (1) are of course critical. One naturally wants to include a fixed error component that may well be correlated with the regressors of interest, as discussed in the introduction. The potential endogeneity of the explanatory variables in (1) is assumed to be fully captured by non-zero correlations with this error component. However, it is not assumed that the impact of the heterogeneity is necessarily constant over time. For example, some farmers are more productive than others in ways that cannot be captured in the data and this matters more in a bad agricultural year than a good one. Following Holtz-Eakin et al., (1988), the specification of the error term allows for nonstationarity in the impacts of the individual effects: ui, = 0,1)i + pi/ (2) where aui, is the i.i.d. random variable, with zero mean and variance ao , and co is a time- invariant effect that is not orthogonal to the regressors, i.e., E(cojXj,) 0 and E(wiZi) X 0, while ,ui, is a white-noise innovation process, i.e., E(o)jpj,) = 0 and E(Zi,pi) = 0. The assurmed error structure in (2) facilitates quasi-differencing of the model in (1). Substituting equation (2) into (1) and lagging by one period one obtains: AlnCi,_1 =a+,8 Xi, + Zi +t10)i + il-l (3) Multiplying equation (3) by rt = 0, / 0,- and subtracting from equation (1), the quasi-differenced model for consumption growth: AInC1, = a(l - r,) + r, Aln Cj,X + 13(X1 - r, X, l) +(1- r,)Z + Pit - r,P,-,l (4) 9 It is evident from (4) that as long as r, 1 one can identify the impact of the time-invariant variables on the growth rate robustly to latent heterogeneity. The test described in Jalan and Ravallion (2002) (following Godfrey, 1988) is used to test the null hypothesis that r, = 1 for all t. In estimating equation (4) one must allow for the fact that InCi, -1is correlated with the error term, , t, - 4P,,i*. One can estimate equation (4) by Generalized Method of Moments (GMM) using differences and/or levels of log consumptions lagged twice (or higher) as instruments for lnCi,1. (So one loses two observations over time in estimating equation 1.) The essential condition to justify this choice of instruments is that the error term in (4) is second-order serially independent, as implied by serial independence of pi,.. The Arellano and Bond (1991) second- order serial correlation test is performed, given that the consistency of the estimator for the quasi-differenced model depends on the assumption that the composite error term is second- order serially independent.8 (Note that there is some first-order serial correlation introduced in the model due to the quasi-differencing. This means that consumption lagged once is not valid instruments.) Let us now see how the household-level impacts on consumption growth identified using the above model can be decomposed by income source. There are M-l income sources and let Yjit denote income from sourcej for household i at date t and (for notational convenience) let YMit denote savings. From the identity: M Ci, =ZYji, (5) j=l 8 To test if the instruments are valid, the Arellano and Bond (1991) over-identification test is also used. Lack of second-order serial correlation and the non-rejection of the over-identification test support our choice of instruments. For further discussion see Jalan and Ravallion (2002). 10 we have: A inCit _ c, = E AYjit (6) i,-I j=1 Ci, l This motivates a decomposition of equation (4) as follows: Ayjit / Ci, l - rtAYjt-, l Ci,_2 = aj (l - r, + /3j (Xit - rt Xi, l ) + {j (1- r,t)Z, + pji, - r,,uji-, l U = 1,..,M) 7 Summing equation (7) over allj yields equation (4), with a = Eaj, fi = ZJj and ¢ = Notice that for consistency with aggregation, the rt (t=1 ,..,T) parameters cannot vary by income source. To estimate (7), I replace the rt parameters by their estimates from the consumption growth model to give: AYji /Cit - P,AYj1, /l ICi-2 = aj (1- ;,).+,8 f(Xit -, Xi,-1) + Cj (1- r, zi + Ujit - ;,,uji,- l ( = II.-, M) 8 Thus, provided the individual effect has a time varying impact, one can identify geographic effects by income sources, which-are robust to latent (individual or geographic) heterogeneity. 3. Setting and data China experienced a surge in rural nonfarm activity in the 1 980s, in the wake of country- wide economic reforms (Byrd and Qingsong, 1990). An important element of this was the emergence and rapid growth of Township and Village Enterprises (TVE's). The fact that growth in the number of non-farm enterprises was preceded by more rapid agricultural growth (following de-collectivization starting in the late 1970s) is sometimes interpreted as evidence of a strong forward linkage from agriculture to non-farm rural development in the Chinese setting. For example, Jiacheng (1990) argues that agricultural growth provided the key pre-condition for the rapid expansion of nonfarm economic activities in the 1980s. However, there are other 11 interpretations in the literature; for example, Haiyan (1990) argues that, while the stimulus for nonfarm rural enterprise development came from agriculture, it was a negative stimulus, not positive - that the expansion of rural nonfarm enterprises was stimulated by low agricultural productivity in certain regions. Anti-poverty policy in China has emphasized poor-area development programs, which have traditionally emphasized the role of agriculture (Leading Group, 1988; World Bank, 1992). There has been debate in policy circles about this emphasis on agriculture, with some people arguing that non-farm enterprise development should be given priority instead. There has also been a debate about whether these programs are effective in longer-term poverty reduction, or are simply short-term palliatives (with out-migration from poor areas seen by some as the only long-term solution). In previous work using these data, evidence was found of dynamic income gains from the central and provincial poor-area development programs, implying quite reasonable economic rates of return (Jalan and Ravallion, 1998). The following analysis uses household level data from China's Rural Household Survey (RHS) done by the State Statistical Bureau (SSB). A panel of 5,600 farm households spanning 111 counties over the six-year period 1985-90 was formned for four contiguous provinces in southern China, namely Guangdong, Guangxi, Guizhou, and Yunnan. The latter three provinces form southwest China, widely regarded as one of the poorest regions in the country. Guangdong on the other hand is a relatively prosperous coastal region (surrounding Hong Kong). The RHS is a well-designed and executed survey of a random sample of households in rural China, with unusual effort made to reduce non-sampling errors (Chen and Ravallion, 1996). Sampled households fill in a daily diary on expenditures and are visited on average every two weeks by an interviewer to check the diaries and collect other data relevant to incomes. There is 12 also an elaborate system of cross-checking at the local level. The consumption and income data from such an intensive survey process are almost certainly more reliable than those obtained by the common cross-sectional surveys in which the data are based on recall at a single interview. For the six-year period 1985-90 the survey was also longitudinal, returning to the same households over time. While this was done for administrative convenience (since local SSB offices were set up in each sampled county), the panel can still be formed." The income aggregate includes imputed values of revenues from own production (net of costs) valued at actual local selling prices (rather than the planning prices used in the original data; see Chen and Ravallion, 1996). The consumption data include imputed values of the consumption streams from the inventory of consumer durables. Poverty lines designed to represent the cost at each year and in each province of a fixed standard of living were used as deflators. These were based on a normative food bundle set by SSB, which assures that average nutritional requirements are met with a diet that is consistent with Chinese tastes. This food bundle is then valued at province-specific prices. The food component of the poverty line is augmented with an allowance for non-food goods, consistent with the non-food spending of those households whose food spending is no more than adequate to afford the food component of the poverty line.'2 Income sources are broken down as follows: (i) Farm income: income from grain production and other farm crops. (ii) Nonfarm income type I: forestry, animal husbandry, fishery, gathering and hunting. Constructing the panel from the annual RHS survey data proved to be more difficult than expected since the identifiers could not be relied upon. Fortunately, virtually ideal matching variables were available in the financial records, which gave both beginning and end of year balances. The relatively few ties by these criteria could easily be broken using demographic (including age) data. 12 For further details on the poverty lines and the data set see Chen and Ravallion (1996). 13 (iii) Nonfarm income type nI: handicrafts, industry, material processing, construction, transportation, productive labor service, commerce, catering trade, services. (iv) Collective income: collective production, income from TVEs, collective welfare funds, collective prizes, other collective income. In adopting this classification, I wanted to distinguish the types of land-based nonfarm income sources that are often associated with farming (type I) from others (type II). My usage is not standard in this respect; it is more common in the literature to only refer to my "type II" as the "nonfarm sector" (see, for example, Lanjouw and Lanjouw, 2001). Of course, in a literal sense, my "type I" is not farming. And, as we will see, these three sectors behave differently, making their separation of interest. In 1985, these four income sources accounted for 58.4%, 24.5%, 15.0% and 2.1% (respectively) of aggregate household income in the sample. Multiple sources for one household are common. Indeed, every one of the sampled households who had income from farming also recorded at least some income from a nonfarm activity. Collective income is the most problematic of the four categories. Although income gains from non-household nonfarm enterprises are excluded from this analysis, the profits received from such enterprises by households are included under "collective income." However, the category accounts for only 2% of income. And it is likely that some of this comes from outside the county. One can be justifiably skeptical as to how well the following analysis will then be able to capture external effects on local non-household income growth. Echoing the empirical literature on linkages, one finds positive correlations across counties between farm income per capita and nonfarm income of type I above, though less so for type II. Table I gives the correlation coefficients in the time-means in the data set. There is very little correlation between the two types of nonfarm income. 14 In estimating equation (8), I shall use two distinct types of data on the geographic composition of,economic activity. The first uses the initial (1985) county mean of the income sources identified above. Initial values of the corresponding household variables are also included. This gives a conceptually clean representation of the four-by-four matrix of linkage effects. However, there is a potential concem that the explanatory variables are from the same survey-based data source. There are of course sampling errors in the county means, and possibly correlated measurement errors. For the second set of estimates, I draw instead on county administrative data. This has two advantages. Firstly, the data sources are then largely independent, relieving possible concerns about correlated measurement errors when using a common data source. Secondly, the county administrative data encompass the rural non-household sector, including TVEs. A disadvantage is that the available county data are less complete, which reduces the sample size to 4,800 (96 counties). From the county data, one can identify three obvious indicators of the extent of development of local agriculture, namely irrigated land area, fertilizer usage and agricultural machinery usage. For the rural nonfarm sector, I have used the county administrative data on the number of commercial enterprises in 1985 and the sector composition of gross product per capita at county level. The latter is broken down according to whether it is industry (distinguished according to whether the industrial enterprise is township, village or household-based), construction, transport or services. In this second model, controls are also added for geographic and household heterogeneity. The geographic variables at the county-level data base include population density, average education levels, road density, health indicators, and schooling indicators. Dummy variables for the province are also included. A composite measure of 15 household wealth can be constructed, comprising valuations of all fixed productive assets, cash, deposits, housing, grain stock, and consumer durables. Data are also used on agricultural inputs used, including landholding. These asset and farm input variables are time-varying, but are treated as endogenous, using lagged values as instruments. To allow for differences in the quality and ouantity of family labor (given that labor markets are thin in this setting), initial education attainments and demographic characteristics are also included. The Appendix provides descriptive statistics. 4. Results First the simpler model described above is estimated, in which consumption growth and its components by income source are regressed on the survey-based estimates of initial county mean income by source and initial own incomes. Table 2 gives the consumption growth regression (corresponding to equation 4), while Table 3 gives the decomposition by all four income sources (equation 8). (Saving is the residual, not estimated.) The diagnostic tests described in section 2 passed comfortably. (This was also true of the extended model, discussed later in this section.) The results in Tables 2 and 3 are for the full sample (n=5,600); the models were also estimated on the smaller sample for which county data are complete (as used in the extended specification below); the results were very similar. Consumption growth at the household level is significantly higher in counties with higher initial levels of farm income, nonfarm income type I and collective income. The size and significance of the effect of differences in county-mean farm income are notable; the regression coefficient in Table 2 implies that a 100 Yuan per month increase in mean farm income in the county of residence (equivalent to one standard deviation, or about 60% of mean farm income) increases the consumption growth rate by 0.0 195 - about two percentage points per annum. 16 In marked contrast to the county variables, higher own incomes tend to result in lower subsequent consumption growth. This pattern echoes the finding of Jalan and Ravallion (2002) that the micro consumption growth process tends to be convergent with respect to household characteristics (in that characteristics that tend to raise the current level of consumption lead to lower subsequent growth), but divergent with respect to geographic characteristics. Turning to the decomposition of consumption growth by income source, the results in Table 3 indicate a significant within-sector external effect in all cases except collective income. Higher initial mean incomes from farming in the county of residence entail higher subsequent income gains from farming. This is also the case for type II nonfarm incomes. For type I nonfarm incomes however, one finds a negative external effect within the sector, suggestive of a crowding-out effect. Looking at the cross-sectoral linkages in Table 3, one finds no significant effects of initial nonfarm income in the county on farm income gains at the household level. A significant positive effect of a higher initial level of farm incomes in the county on the growth of type I nonfarm incomes is found, but not for type II. Nonfarm incomes of type I in turn have positive effects on the growth of type II and collective incomes. However, higher collective incomes locally tend to attenuate growth in nonfarm incomes of type II. For each of the four income growth regressions in Table 3, one can convincingly reject the null hypothesis that the four coefficients on the county-mean income sources are equal.9 Thus the composition of economic activity matters. Summing the external effect of a given income component horizontally in Table 3, it is plain that farming is the largest generator of 9 Wald tests of the null hypothesis that the four coefficients on county-mean incomes in Table 3 are equal gave 44.9, 35.8, 71.7 and 19.5 respectively. For the consumption growth regression in Table 2 the Wald test gave 15.3. The test has a Chi-square distribution with four degrees of freedom, implying rejections of the null hypotheses at the 1% level or better. 17 of the other geographic controls are suggestive of positive extemalities from better local endowments of human and physical infrastructure; in particular, higher levels of literacy locally and higher road density promote higher consumption growth at household level. The indications of geographic extemalities are also evident in the decomposition by income source (Table 5). Echoing the results of Table 3, here too one finds strong indications that areas with more land and more developed agriculture tend to experience higher subsequent farm income gains; this effect is particularly strong for fertilizer usage, which is probably the best indicator in these data of the adoption of modern agricultural techniques. As in Table 2, the cross-effect of initial agricultural development on nonfarm incomes is also evident, for type I and type II nonfarm incomes. However, unlike Table 3, one now finds strong positive effects of the density of nonfarm commercial development and industrial output on farrn incomes. By allowing us to break up nonfarm incomes by sector (industry, construction, transport and services) the regressions using the county administrative data in Table 5 reveal that the more aggregate effects identified in Table 3 disguise some potentially important differences between sub-sectors. Indeed, while there are generally positive external effects of local industrial development, we see signs of negative external effects on farm and nonfarm income growth of greater local activity in the transport and service sub-sectors. (Notice that the transport income effect is probably not picking up an effect of transport infrastructure, since I am controlling for road density.) It appears that these sectors are competing with household-level farm and nonfarm activities for limited local resources that enhance the productivity of private investment and hence income growth at the farm-household level. Higher cultivated area per person in a county has a significant positive effect on the growth of nonfarm type I incomes, but the (positive) effect on type II is barely significant at the 19 10% level. These findings lead one to question the claims sometimes made (in the case of China, see Haiyan, 1990) that a shortage of cultivated land in an area was an inducement to nonfarm activities. One finds the opposite to be the case for nonfarm activities by the household, though there is a sign of this effect on collective income (which here includes income from enterprises). Higher fertilizer usage also has an external effect on both types of nonfarm income growth, though the dominant extemal effect is on farm incomes. The extended models in Table 5 also point to some diverse and in some cases surprising impacts across ixicome sources. The positive effect of higher population density on consumption growth (Table 4) appears to be transmitted entirely through nonfarm type I income growth. The effect of road density appears to be largely through higher farm incomes. Lower infant mortality (as an indicator of health care more generally) appears to have high returns to nonfarm (type II) income growth. Higher basic education appears to spillover more into farming. 5. Conclusions The literature on linkages in rural development has largely ignored what is surely the most relevant question for policy: do the signs of linkage found in geographic data reflect externalities at the level of the individual decision-maker? The data and methods used in past empirical work cannot distinguish externalities from other factors far more benign from a policy point of view. Yet the implications for understanding rural underdevelopment, and the implications for policy, depend crucially on whether the aggregate appearance of inter-sectoral linkage in rural development stems from externalities at the micro level. The paper has offered a test that can identify any genuine linkage externalities, and can also test for micro effects on the growth process of differing geographic endowments of human and physical infrastructure. The paper has implemented the test using data for rural China during 20 the post-reform period of farm and (particularly) nonfarm rural development. The aim has been to describe the patterns of linkage in a way that is robust to latent heterogeneity. Like any description, the results beg many questions. In particular, the analysis has thrown little light on the precise sources of external effects. Are we seeing the effects of knowledge spillovers, or something else such as network externalities or pecuniary externalities? The results do suggest that the level and composition of local economic activity has non- negligible impacts on consumption and income growth at the farm-household level. There are significant positive effects of the level of local economic activity in a given sector on income growth from that income source. And there are a number of significant sectoral cross-effects, notably from farming to those categories of nonfarm activities that tend naturally to be more linked to agriculture (forestry, animal husbandry, fishing), but also between the latter type of nonfarrn activity and other types (handicrafts, industry, processing, transportation etc.). Thus there is a direct link from the initial level of agricultural development to the first type of nonfarm activities and a more indirect link to the second. There is less sign of the reverse linkage - from initial level of nonfarm economic activity to growth in farm incomes. And there are indications of negative external effects from some nonfarm activities, notably involving non-industrial subsectors (construction and transport). While I do find significant cross-sector effects, they are dwarfed by the within-sector effects. The composition (as well as the level) of local economic activity matters, and the sector that clearly matters most quantitatively is agriculture. The results of this paper suggest that there are externalities at the farm-household level underlying the signs of linkage found in more geographically aggregated data. Under the paper's identifying assumptions, the linkages found can be interpreted as genuine externalities, suggesting that private agents in this economy are not going to take account of all the potential 21 income gains from their actions. Thus these results offer an explanation for rural under- development, arising from under-investment in externality generating activities, notably' agriculture and (to a lesser extent) certain nonfarm activities. By the same token; the results offer a micro-empirical foundation for the long-standing, but poorly validated, claims in the literature about the potential for "virtuous cycles" whereby a well-targeted external growth stimulus in a poor area can generate positive and more widely diffused income gains over time. Thus these results offer support for the types of poor-area development programs that have been pursued by the Government of China since the mid-1980s. The emphasis that these programs have given to agricultural development is consistent with this paper's findings that agriculture is the key externality-generating sector of the Chinese rural economy. Of course, the detailed design of such programs is crucial, and this is not something that the results of this paper can throw much light on. However, the present results also point to the importance of local endowments of human and physical infrastructure to the micro-growth process. When combined with data on the costs to the government's budget of alternative interventions, these empirical results will hopefully also help inform public choices on how best to balance agricultural development initiatives with infrastructure development, so as to assure maximum growth of living standards in poor areas. 22 References Ahn, S.C., Y.H. Lee and P. Schmidt (1994), "GMM Estimation of a Panel Data Regression Model with Time-varying Individual Effects", mimeo, Arizona State University. Arellano, M. and S. Bond (1991), "Some Tests of Specification for Panel Data: Monte-Carlo Evidence and An Application to Employment Equation", Review of Economic Studies, 58, 277-298. Basu, K., A. Narayan and M. Ravallion (2002), "Is Literacy Shared Within Households?", Labor Economics 8, 649-665. Besley, T. (1995), "Savings, credit and insurance", in Jere Behrman and T.N. Srinivasan (eds) Handbook of Development Economics Volume 3. Amsterdam: North-Holland. Byrd, W., and Lin Qingspong (1990), "China's Rural Industrialization," in W.A. Byrd and Lin Qingspong (eds) China's Rural Industry: Structure, Development and Reform, New York: Oxford Univserity Press. Chen, S. and M. Ravallion (1996), "Data in Transition: Assessing Rural Living Standards in Southern China", China Economic Review, 7, 23-56. Clarke, C., (1940), The Conditions of Economic Progress. London: Macmillan. Dasgupta, P.S., and G.M. Heal (1979), Economic Theory and Exhaustable Resources. Cambridge: Cambridge University Press. Ellis, F. (1998), "Household Strategies and Rural Livelihood Diversification," Joumal of Development Studies 35, 1-38. Feder G., and R. Slade, (1985), "The Role of Public Policy in the Diffusion of Improved Agricultural Technology," American Joumal of Agricultural Economics 67, 423-428. Foster, A.D. and M.R. Rosenzweig, (1995) "Leaming by Doing and Learning from 23 Others: Human Capital and Technical Change in Agriculture", Journal of Political Economv. 103(6), 1176-1209. Foster, A.D. and M.R. Rosenzweig, (1996) "Technical Change and Human-Capital Returns and Investments: Evidence from the Green Revolution," American Economic Review 86, 931-53. Godfrey, L.G. (1988), Misspecification Tests in Econometrics, Cambridge University Press, Cambridge. Greenwald, B., and J.E. Stiglitz (1986), "Externalities in Economies with Imperfect Information and Incomplete Markets," Quarterly Journal of Economics 229-264. Haggblade, S., P. Hazell and J. Brown (1989), "Farm-Nonfarm Linkages in Rural Sub-Saharan Africa," World Development 17(8), 1173-1201. Haggblade, S., P. Hazell and T. Reardon (2002), "Strategies for Stimulating Poverty-Alleviating Growth in the Rural Nonfarm Economy in Developing Countries,'' mimeo, International Food Policy Research Institute, Washington DC. Haiyan, Du (1990), "Causes of Rapid Rural Industrial Development," in W.A. Byrd and Lin Qingspong (eds) China's Rural Industry: Structure. Development and Reform, New York: Oxford University Press. Hazell, P. and S. Haggblade (1993), "Farm-Nonfarm Growth Linkages and the Welfare of the Poor," in M. Lipton and J. van der Gaag (eds.), Including the Poor. Washington DC., World Bank. Holtz-Eakin, D., W. Newey and H. Rosen (1988), "Estimating Vector Autoregressions with Panel Data", Econometrica, 56, 1371-1395. 24 Holtz-Eakin, D. (1988), " Testing for Individual Effects using Panel Data", Journal of Econometrics, 39, 297-307. Hoff, K. (1998), "Adverse Selection and Institutional Adaptation," mimeo, Development Research Group, World Bank. Hoff, K. (2000), "Beyond Rosenstain-Rodan: The Modem Theory of Underdevelopment Traps," Paper Presented at the Annual Bank Conference on Development Economics, World Bank, Washington DC. Jalan, J. and M. Ravallion (1998), "Are There Dynamic Gains from a Poor-Area Development Program?," Journal of Public Economics 67, 65-85. Jalan, J. and M. Ravallion (2002), "Geographic Poverty Traps? A Micro Model of Consumption Growth in Rural China," Journal of Applied Econometrics, in press. Jiacheng, He (1990), "Development Issues and Policy Choices," in W.A. Byrd and Lin Qingspong (eds) China's Rural Industry: Structure. Development and Reform, New York: Oxford University Press. Lanjouw, J.O., and P. Lanjouw (2001), "The Rural Nonfarm Sector: Issues and Evidence from Developing Countries," Agricultural Economics 26, 1-23. Leading Group (1988), Outlines of Economic Development in China's Poor Areas, Office of the Leading Group of Economic Development in Poor Areas Under the State Council, Agricultural Publishing House, Beijing. Lucas, R.E., (1993), "Making a Miracle," Econometrica 61, 251-72. Mellor, J.W. (1976), The New Economics of Growth: A Strategy for India and the Developing World, Ithaca, Cornell University Press. 25 Mellor, J.W., and U. Lele (1972), "Growth Linkages of the New Grain Technologies," Indian Journal of Agricultural Economics 18(1): 35-55. Ramsey, F. (1928), "A Mathematical Theory of Saving", Economic Journal 38, 543-559. Ravallion, M. (1998), "Poor Areas", in Handbook of Applied Economic Statistics, edited by David Giles and Aman Ullah, New York: Marcel Dekkar, 1998. Ravallion, M. and J. Jalan (1999), "China's Lagging Poor Areas," American Economic Review (Papers and Proceedings), 89(2), 301-305. Reardon, T., C. Delgado and P. Matlon, (1992). "Determinants and Effects of Income Diversification Amongst Farm Households in Burkina Faso," Journal of Development Studies 28: 264-96. Romer, P.M. (1986), "Increasing Returns and Long-Run Growth," Journal of Political Economy 94, 1002-37. Rosenstain-Rodan, P. (1943), "Problems of Industrialization of Eastern and Southeastern Europe," Economic Journal 53, 202-211. Sengupta, Jati L., and Bo Q. Lin (1995), "Recent Rural Growth in China: The Performance of the Rural Small-Scale Enterprises," mimeo, University of California, Santa Barbara, CA> World Bank (1992), China: Strategies for Reducing Poverty in the 1990s, World Bank, Washington DC. 26 Table 1: Correlation coefficients in sample mean incomes across 102 counties Farm Nonfarm Nonfarm Collective Income income I income II income Farm income 1.0000 Nonfarm income I 0.3240 1.0000 Nonfarm income II 0.1134 0.0027 1.0000 Collective income 0.4505 0.1125 0.2171 1.0000 Table 2: Consumption growth regressed on county-mean incomes and own incomes Consumption growth GMM estimates 1985-90 Coefficient t-ratio Constant -0.019034* -3.631332 Coefficients on lagged consumption 1987 -0.023637 -0.260700 1988 0.231193* 5.477698 1989 -0.036034 -0.974515 1990 0.192418* 4.036306 County mean household incomes by source, 1985 Farm income 0.000195* 7.029119 Nonfarm income I 6.77E-05 1.848970 Nonfarm income II 6.1OE-05 1.376225 Collective income 0.000148 1.925260 Household's own income by source, 1985 Farm income -5.07E-05* -3.616862 Nonfarm income I -7.25E-05* -4.473417 Nonfarm income II -7.47E-05* -5.055279 Collective income -2.25E-05 -0.795760 Note: * indicates significant at 1% level, two-tailed test; n=5,641 (111 counties). 27 Table 3: Decomposition of growth by income source Income change 1985-90, normalised by initial Farm income Nonfartn income I Nonfarm income II Collective income consumption Coefficient t- ratio Coefficient t- ratio Coefficient t- ratio Coefficient t- ratio Constant -1.042776 -0.606657 0.007246 1.581473 -0.001584 -0.304262 -0.006296* -2.776274 County mean household incomes by source, 1985 Farmn income 0.058360* 5.582170 9.02E-05* 3.587523 -1.43E-05 -0.527382 4.85E-06 0.405785 Nonfarm income 1 -0.019292 -1.864275 -7.72E-05* -2.922507 9.23E-05* 2.627234 7.64E-05* 4.199318 Nonfarm income 2 -0.012158 -0.836240 -2.46E-05 -0.722362 0.000358* 7.216553 1.95E-06 0.156793 Collective income 0.009052 0.365210 7.86E-06 0.104518 -0.000232 -2.364964 8.38E-05 1.705881 Household's own income by source, 1985 Farm income -0.065339* -7.964560 -2.23E-05 -2.032218 -2.72E-05 -2.032737 9.13E-07 0.225140 Nonfarm income 1 -0.009548 -2.082792 -8.46E-05* -5.212832 8.70E-07 0.069791 -1.86E-05* -3.288436 Nonfarm income 2 -0.005469 -1.357622 4.01E-06 0.394638 -4.41E-05 -1.707732 1.76E-06 0.414443 Collective income -0.024780* -2.654654 -1.04E-05 -0.393923 1.42E-05 0.485131 -0.000132* -5.666340 J statistic 0.073149 0.037610 0.020013 0.005590 Table 4: Consumption growth model using geographic data from county administrative records Coefficient t-Statistic Constant -0.328076* -3.938664 Coefficients on lagged consumption 1987 -0.563094* -5.580720 1988 0.226777* 6.313155 1989 -0.031837 -0.878866 1990 0.264715* 6.118230 Economic activity at county level (a) Farmn Cultivated area per 10,000 persons 0.003075* 3.424595 Fertilizer used per cultivated area 0.004131* 7.433107 Farm machinery used per cultivated area 0.000368* 2.651082 (b) Nonfarmn Number of commercial enterprises in county per 10,000 population 0.000220* 2.768617 Rural industry gross product per 10,000: township enterprises -6.63E-05 -1.759901 Rural industry gross product per 10,000 persons: village 0.000415* 3.729650 enterprises Rural industry gross product per 10,000 persons: household -1.77E-05 -0.173829 enterprises Rural construction gross product per 10,000 persons -0.000154 -2.063245 Rural transportation gross product per 10,000 persons -0.000509* -3.639974 Rural gross product from services per 10,000 persons 0.000169 0.715551 Other geographic controls Guangdong (dummy) 0.037373* 4.338988 Guangxi (dummy) 0.022666* 4.345667 Yunnan (dummy) -0.005237 -0.869316 Revolutionary base area (dummy) 0.050238* 3.248796 Border area (dummy) 0.002216 0.563537 Coastal area (dummy) -0.012471 -1.278915 Minority area (dummy). -0.012457* -3.714323 Mountainous area (dummy) -0.015838* -4.452355 Plains (dummy) 0.005659 1.459167 Population density (log) 0.021519 2.480439 Proportion of illiterates in 15+ population -0.000322 -1.866172 Infant mortality rate -0.000147 -1.296671 Medical personnel per capita 0.000584 1.988495 Kilometers of roads per capita 0.000455* 3.185796 Proportion of population living in urban areas -0.097467* -3.199404 Household variables Expenditure on agricultural inputs per cultivated area -0.001911 * -7.161740 Fixed productive assets per capita -1.27E-05 -0.883144 Cultivated land per capita -0.008748 -1.802922 Household size (log) 0.056994* 8.967627 Age of household head 0.002086* 2.617436 Age2 of household head -2.57E-05* -2.899381 Proportion of adults in the household who are illiterate 0.007032 1.125765 Proportion of adults with primary school education 7.77E-06 0.001468 Proportion of children 6-11 years 0.013395 1.377193 Proportion of children 12-14 years 0.032215' 2.502249 Proportion of children 15-17 years 0.002467 0.158605 Proportion of children with primary school education -0.002868 -0.736394 Proportion of children with secondary school education 0.020066 2.002172 Whether a household member works in the state sector (dummy) -0.001098 -0.147539 Proportion of 60+ members in the household 0.002312 0.187774 Notes: * indicates significant at 1% level, two-tailed test; n=4,778 (96 counties). 30 Table 5: Decomposition by income source Income change 1985-90, normalised by initial Farm income Nonfarm income I Nonfarm income II Collective income consumption Coefficient t-ratio Coefficient t- ratio Coefficient t- ratio Coefficient t- ratio Constant -0.037285 -0.551930 -0.317851 -5.899826 0.017037 0.312113 0.002492 0.213991 Economic activity at county level (a) Farm Cultivated area per 10,000 persons 0.001425 1.975995 0.004668* 7.661727 0.00096 1.694963 -5.92E-05 -0.494620 Fertilizer used per cultivated area (xlOO) 0.3298* 7.450607 0.1553* 4.623649 0.0775* 2.335067 0.000321 0.042174 Farm machinery used per cultivated area (xlOO) 0.0137 0.881606 0.00559 0.539991 -0.0262* -2.185012 0.00757* 2.829688 (b) Nonfarm Number of commercial enterprises per 10,000 -6.40E-05 -1.012804 1.02E-05 0.192262 0.000256* 4.728632 -1.06E-06 -0.084385 population Rural industry gross product per 10,000: township 1 .53E-05 0.472816 -7.68E-05* -3.695281 -5.09E-05 -2.126782 3.83E-05* 3.920372 enterprises Rural industry gross product per 10,000 persons: 0.000128 1.339266 -1.44E-05 -0.280215 0.000355* 4.522848 -2.83E-05 -1.738851 village enterprises Rural industry gross product per 10,000 persons: 0.000207* 2.549566 0.000188* 3.230548 -9.72E-05 -1.487319 -3.74E-05 -1.336554 enterprises owned by households Rural construction gross product per 10,000 -1.16E-06 -0.017589 5.40E-05 1.257481 -2.94E-05 -0.546224 1.46E-06 0.136096 persons Rural transportation gross product per 10,000 -0.000225 -1.893099 -0.000234* -2.760754 -0.000240 -2.109118 -3.1OE-05 -1.163832 persons Rural gross product from services per 10,000 -0.000870* -4.433674 0.000235 1.700295 3.25E-05 0.212639 -4.47E-05 -1.095567 persons Other geographic controls Guangdong (dummy) 0.040192* 5.578835 -0.039672* -7.215861 -0.001470 -0.261715 0.001851 1.382536 Guangxi (dummy) 0.007369 1.729346 0.006060 1.656530 0.000142 1.066986 0.000817 1.080041 Yunnan (dummy) -0.008800 -1.728590 -0.003452 -0.851059 -0.000270* -3.355940 0.000545 0.646179 Revolutionary base area (dummy) 0.051141* 3.953926 -0.005317 -0.896477 -0.000388* -1.856168 0.000267 0.172994 Border area (dummy) 0.015034* 4.753477 -0.006759 -2.516487 6.47E-05 0.512528 -0.000337 -0.645808 Coastal area (dummy) -0.050290* -5.715417 0.001306 0.172308 0.007192 0.300917 0.002023 0.869084 Minority area (dummy) (xIOO) -0.2865 -1.020337 -0.6569* -2.915940 -0.01381 -0.598987 0.0167 0.369802 Mountainous area (dummy) -0.019013* -6.629943 0.004989 2.179428 0.005246 2.187413 0.000247 0.478782 Plains (dummy) 0.003090 0.882888 0.009975* 3.684431 0.002272 0.751968 -0.000332 -0.527546 Population density (log) 0.003167 0.445301 0.026073* 4.853929 -0.001470 -0.261715 -0.000364 -0.297651 Prop of illiterates in 15+ population (xlOO) -0.0397* -2.810640 0.0183 1.371223 0.0142 1.066986 0.00253 0.982911 Infant mortality rate -3.89E-05 -0.437105 -0.000126 -1.605491 -0.000270* -3.355940 -3.62E-05 -2.187786 Medical personnel per capita (xlOO) 0.0368 1.435438 -0.00388 -0.148288 -0.0388* -1.856168 0.00424 1.070008 Kilometers of roads per capita (xlOO) 0.0678* 5.693957 -0.0100 -0.985167 0.00647 0.512528 0.00121 0.663914 Proportion of population living in urban areas -0.082497' -3.331008 0.039684 1.939799 0.007192 0.300917 -0.004783 -0.796981 Houselhold-level variables Expenditure on agricultural inputs per cultivated -0.1788* -9.474159 -0.00532 -0.578969 -0.0199 -2.058113 9.48E-06 0.435807 area (x I00) Fixed productive assets per capita (xlOO) -0.000515 -0.512903 8.22E-05 0.074993 0.00457* 2.566003 -0.000204 -0.867348 Cultivate land per capita -0.008281 -1.818687 -0.007585 -2.438985 -0.008547* -3.224610 -0.000337 -0.511164 Household size (log) 0.012321* 2.596482 0.014103* 2.919817 0.002724 0.615920 -0.000927 -1.012190 Age of household head 0.000470 0.718755 0.000379 0.635091 -0.000143 -0.251665 5.26E-05 0.476080 Age2 of household head (xlOO) -0.000731 -0.999952 -0.000355 -0.522368 2.45E-06 0.385523 -0.0001 -0.775087 Prop of adults in household who are illiterate -0.002009 -0.373594 0.000758 0.185174 -0.001015 -0.229459 0.002438 2.292417 Prop of adults in household with primary school -0.002942 -0.667418 0.004343 1.223840 -0.005701 -1.464877 0.001718 1.771110 education Prop of children in the household ages 6-1 1 years 0.005990 0.711307 0.009385 1.432443 -0.006400 -0.877534 0.001215 0.583170 Prop of children in household ages 12-14 years 0.004234 0.378313 0.012715 1.410109 0.012735 1.341365 0.003008 0.917733 Prop of children in household ages 15-17 years 0.002616 0.209421 -0.005501 -0.529853 0.018894 1.796201 0.005167 1.699518 Prop of children with prim school education (xlOO) -0.0409 -0.121831 -0.3429 -1.283922 0.001907 0.612289 0.0189 0.220488 Prop of children with secondary school education -0.001224 -0.149337 0.010930 2.011010 -0.007282 -1.247138 0.002381 1.411268 Household member works in state sector (dummy) -0.018599* -3.088751 -0.004086 -0.805931 -0.003461 -0.765217 -0.000913 -0.493164 Proportion of 60' members in the household 0.002762 0.261581 -0.005292 -0.646863 0.001195 0.151179 -0.000654 -0.363943 Notes: * indicates significant at 1% level, two-tailed test; n=4,778 (96 counties). 32 Appendix: Descriptive statistics Mean St. deviation Dependent variables Average growth rate of consumption, 1986-90 0.0042 0.0777 Farm income: mean change as a proportion of lagged -0.0065 0.0687 consumption, 1986-90 Nonfarm income I: mean change as a proportion of lagged 0.0027 0.0688 consumption, 1986-90 Nonfarm income II: mean change as a proportion of lagged 0.0157 0.0755 consumption, 1986-90 Collective income: mean change as a proportion of lagged -0.0005 0.0244 consumption, 1986-90 Economic activity at county level Farm income, 1985 (Yuan/person/month) 161.0459 100.846 Nonfarm income I, 1985 (Yuan/person/month) 92.9326 98.303 Nonfarm income II, 1985 (Yuan/person/month) 62.9430 100.405 Collective income, 1985 (Yuan/person/month) 9.4162 40.971 Fertilizers used per cultivated area (tones per sq.km) 11.5402 6.6497 Farm machinery used per capita (horsepower)a 151.7879 110.2427 Cultivated area per 10,000 persons (sq Iam) 13.0447 3.2518 Number of commercial enterprises per 10,000 population 52.5922 22.003 Rural industry gross product per 10,000: enterprises in 32.7465 132.874 townships (central administrative villages) Rural industry gross product per 10,000 persons: enterprises in 16.2585 45.475 villages Rural industry gross product per 10,000 persons: enterprises 27.5416 33.049 owned by households Rural construction gross product per 10,000 persons 32.5597 42.9291 Rural transportation gross product per 10,000 persons 13.3423 0.9594 Rural gross product from services per 10,000 persons 22.6664 23.121 Othergeographic variables Proportion of sample in Guangdong 0.1618 0.3683 Proportion of sample in Guangxi 0.3414 0.4742 Proportion of sample in Yunnan 0.2285 0.4199 Proportion living in a revolutionary base area 0.0191 0.1367 Proportion of counties sharing a border with a foreign country 0.1712 0.3767 Proportion of villages located on the coast 0.0316 0.1749 Proportion of villages in with an ethnic minority concentration 0.2978 0.4573 Proportion of villages that have a mountainous terrain 0.45563 0.498 Proportion of villages located in the plains 0.2292 0.4203 Population density (log) 8.20602 0.3929 Proportion of illiterates in the 15+ population (%) 36.9547 16.0225 Infant mortality rate (per 1,000 live births) 43.24006 23.8535 Medical personnel per 10,000 persons 7.816894 5.0388 Kilometers of roads per 10,000 persons 14.7122 10.9721 Proportion of population living in the urban areas 0.0907 0.0548 Household level variables Expenditure on agricultural inputs (fertilizers & pesticides) per 29.224 47.9954 cultivated area (Yuan per mu)8 Fixed productive assets per capita (Yuan per capita)' 129.8417 150.8919 Cultivated land per capita (mu per capita)8 1.2591 0.7802 Household size (log) 1.7086 0.3508 Age of the household head 41.8262 11.3887 Age2 of the household head 1879.114 1015.252 Proportion of adults in the household who are illiterate 0.33876 0.2932 Proportion of adults with primary school education 0.3787 0.3074 Proportion of children 6-11 years 0.1199 0.1415 Proportion of children 12-14 years 0.0845 0.1071 Proportion of children 15-17 years 0.06796 0.0988 Proportion of children with primary school education 0.2780 0.3689 34 Proportion of children with secondary school education 0.0484 0.1709 Proportion of a household members working in the state sector 0.0421 0.2008 Proportion of 60+ household members 0.06270 0.1222 Number of households: 4,778 Number of counties 96 Notes: I mu = 0.000667 km2; "a" indicates time-varying variables 35 Policy Research Working Paper Series Contact Title Author Date for paper WPS2854 Rich andPowerful? Subjective Michael Lokshin June 2002 C. Cunanan Power and Welfare in Russia Martin Ravallion 32301 WPS2855 Financial Crises, Financial Luc Laeven June 2002 R. Vo Dependence, and Industry Growth Daniela Klingebiel 33722 Randy Kroszner WPS2856 Banking Policy and Macroeconomic Gerard Caprio, Jr. June 2002 A Yaptenco Stability: An Exploration Patrick Honohan 31823 WPS2857 Markups, Returns to Scale, and Hiau Looi Kee June 2002 M. Kasilag Productivity: A Case Study of 39081 Singapore's Manufacturing Sector WPS2858 The State of Corporate Governance: Olivier Fremond June 2002 G. Gorospe Experience from Country Mierta Capaul 32623 Assessments WPS2859 Ethnic and Gender Wage Mohamed Ihsan Ajwad June 2002 Z. Jetha Disparities in Sri Lanka Pradeep Kurukulasuriya 84321 WPS2860 Privatization in Competitive Sectors: Sunita Kikeri June 2002 R. Bartolome The Record to Date John Nellis 35703 WPS2861 Trade-Related Technology Diffusion Maurice Schiff June 2002 M. Kasilag and the Dynamics of North-South Yanling Wang 39081 and South-South Integration Marcelo Olarreaga WPS2862 Tenure, Diversity, and Commitment: Somik V. Lall June 2002 Y. D'Souza Community Participation for Urban Uwe Deichmann 31449 Service Provision Mattias K. A. Lundberg Nazmul Chaudhury WPS2863 Getting Connected: Competition Frew Amare Gebreab June 2002 P. Sintim-Aboagye and Diffusion in African Mobile 38526 Telecommunications Markets WPS2864 Telecommunications Reform in Mary M. Shirley June 2002 P. Sintim-Aboagye Uganda F. F. Tusubira 38526 Frew Amare Gebreab Luke Haggarty WPS2865 Bankruptcy Around the World: Stijn Claessens July 2002 A. Yaptenco Explanations of its Relative Use Leora F. Klapper 31823 WPS2866 Transforming the Old into a David Ellerman July 2002 N. Jameson Foundation for the New: Lessons Vladimir Kreacic 30677 of the Moldova ARIA Project WPS2867 Cotton Sector Strategies in West Ousmane Badiane July 2002 A. Lodi and Central Africa Dhaneshwar Ghura 34478 Louis Goreux Paul Masson WPS2868 Universal(ly Bad) Service: George R. G. Clarke July 2002 P Sintim-Aboagye Providing Infrastructure Services Scott J. Wallsten 38526 to Rural and Poor Urban Consumers Policy Research Working Paper Series Contact Title Author Date for paper WPS2869 Stabilizing Intergovernmental Christian Y. Gonzalez July 2002 B. Mekuria Transfers in Latin America: David Rosenblatt 82756 A Complement to National/ Steven B. Webb Subnational Fiscal Rules? WPS2870 Electronic Security: Risk Mitigation Thomas Glaessner July 2002 E. Mekhova In Financial Transactions-Public Tom Kellermann 85984 Policy Issues Valerie McNevin WPS2871 Pricing of Deposit Insurance Luc Laeven July 2002 R. Vo 33722 WPS2872 Regional Cooperation, and the Role Maurice Schiff July 2002 P. Flewitt of International Organizations and L. Alan Winters 32724 Regional Integration WPS2873 A Little Engine that Could ... Liesbet Steer August 2002 H. Sutrisna Domestic Private Companies and Markus Taussig 88032 Vietnam's Pressing Need for Wage Employment WPS2874 The Risks and Macroeconomic David A. Robalino August 2002 C. Fall Impact of HIV/AIDS in the Middle Carol Jenkins 30632 East and North Africa: Why Karim El Maroufi Waiting to Intervene Can Be Costly WPS2875 Does Liberte=Egalite? A Survey Mark Gradstein August 2002 P. Sader of the Empirical Links between Branko Milanovic 33902 Democracy and Inequality with Some Evidence on the Transition Economies WPS2876 Can We Discern the Effect of Branko Milanovic August 2002 P. Sader Globalization on Income Distribution? 33902 Evidence from Household Budget Surveys WPS2877 Patterns of Industrial Development Raymond Fisman August 2002 K. Labrie Revisited: The Role of Finance Inessa Love 31001 WPS2878 On the Governance of Public Gregorio Impavido August 2002 P. Braxton Pension Fund Management 32720