POLICY RESEARCH WORKING PAPER 2263 When Is Growth Pro-Poor? Nonfarm economic growth in India had very different effects on poverty in different states. Evidence from the Diverse Nonfarm growth was least Experiences of India's States effective at reducing poverty in states where initial Martin Ravallion conditions were poor in terms of rural development and Gaurav Datt human resources. Among initial conditions conducive to pro-poor growth, literacy plays a notably positive role. The World Bank Development Research Group Poverty and Human Resources and South Asia Region Poverty Reduction and Economic Management Sector Unit U December 1999 F POLIcY RESEARCH WORKING PAPER 2263 Summary findings Ravallion and Datt use 20 household surveys for India's But the elasticities of poverty to (urban and rural) 15 major states, spanning 1960-94, to study how initial nonfarm output varied appreciably, and the differences conditions and the sectoral composition of economic were quantitatively important to the overall rate of growth interact to influence how much economic growth poverty reduction. reduced poverty. States with initially lowver farm productivity, lower The elasticities of measured poverty to farm yields and rural living standards relative to those in urban areas, development spending did not differ significantly across and lower literacy experienced a less pro-poor growth states. process. This paper-a joint product of Poverty and Human Resources, Development Research Group, and the Poverty Reduction and Economic Management Sector Unit, South Asia Region - is part of a larger effort in the Bank to better understand the conditions required for pro-poor growth. Copies of the paper are available free from the World Bank, 1818 H Street, NW, Washington, DC 20433. Please contact Joseph Israel, room MC8-174, telephone 202-458-5117, fax 202-522-1557, email address jisrael@worldbank.org. Policy Research Working Papers are also posted on the Web at www.worldbank.org/ research!workingpapers. The authors may be contacteclatmravallion@worldbank.org or gdatt@worldbank.org. December 1999. (33 pages) The Policy Research Working Paper Series dissemninates 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 Policy Research Dissemination Center 1. Introduction While cross-country comparisons indicate that measures of absolute poverty in developing countries tend to fall with growth in mean household income or consumption, there is considerable variance in the poverty reducing impact of a given rate of growth. For example, on studying spells constructed from successive household surveys for a set of developing countries, Ravallion and Chen (1 997, Table 6) obtain a point estimate of (minus) three for the elasticity of the proportion of the population living below $1/day (at 1985 Purchasing Power Parity) to change in the real survey mean. Their 95% confidence interval, however, goes from about one to five. Add to this the variance in how much GDP growth translates into growth in average household incomes and one can see that a rate of growth that brings rapid poverty reduction in one country might largely leave the poor behind in another. What explains why growth is more pro-poor in some economies than others? Addressing this question with cross-country data raises a number of measurement problems. There are concerns about the comparability of both the primary survey data used for poverty measurement and the likely explanatory variables such as indicators of inequality. This paper pursues an alternative route in which we compare the evolution of poverty measures over several decades in India, for which it is possible to construct a long time series of 20 comparable - highly comparable by the standards of international comparisons - surveys spanning a 35 year period. We test for inter-state differences in the poverty impact of various sources of growth in India, and we try to explain the differences. There are clearly country-wide factors that have influenced growth and poverty reduction in India, and comparisons of experiences over time in different states cannot throw much light on those factors. Here we focus on the extent to which differences in initial conditions at state level account for differing state experiences in the poverty-reducing impact of growth. 2 We build on our past work on poverty in India. Consistent with cross-country evidence for developing countries, measures of absolute consumption-poverty in India tend to fall with economic growth (Ravallion and Datt, 1996). But one also finds that: (i) the sectoral composition of growth, matters to the aggregate (country-wide) rate of poverty reduction: the aggregate time series data for India indicate that poverty measures have responded far more to rural economic growth than urban economic growth (Ravallion and Datt, 1996),2 and (ii) differences in initial conditions related to rural development and human resource development accounted for a sizable share of the long run differences between states in rates of rural poverty reduction (Datt and Ravallion, 1 998a). However, in Datt and Ravallion (1 998a), initial conditions entered additively with growth; favorable initial conditions meant a higher rate of poverty reduction at a given rate of growth, but a given growth rate was deemed to have the same impact on the rate of poverty reduction whatever the initial conditions. Clearly, this type of specification cannot tell us how initial conditions might influence the growth elasticity of poverty. This paper tests an encompassing mrodel of the evolution of the aggregate (urban plus rural) poverty measures at state level. Our model allows for multiplicative interactions of the sectoral composition of growth with initial conditions in determining the evolution of state-level poverty measures. In other words, we examine whether the impact of growth on poverty at given initial conditions also depends on those conditions. In addition to helping us understand what makes growth in a given sector of the economy more pro-poor, we will be able to test for 2 "Rural (urban) economic growth" refers to growth in mean consumption in rural (urban) areas; Ravallion and Datt (1996) also find that "primary" and "tertiary" sector growth had greater impact on poverty than "secondary" sector growth. Thorbecke and Jung (1996) come to a similar conclusion for Indonesia, using a different method based on simulations with a Social Accounting Matrix. There is a large literature, and much debate, on the role of agricultural growth in poverty reduction (Lipton and Ravallion, 1995, section 5.2; Datt and Ravallion, 1998b). 3 possible trade offs; for example, do certain conditions foster a more pro-poor agricultural growth process, but a less pro-poor non-farm growth process? The next section discusses the arguments that have been made as to why inequalities in various dimensions matter to reducing poverty. After reviewing the literature, we present a rudimentary theoretical model of how non-farm economic growth reduces poverty in a dualistic economy. The model assumes that any farm worker who wants to participate in the non-farm sector incurs a cost in doing so. This cost determines the equilibrium earnings differential between the poor farm sector and a non-poor non-farm sector. It is evident that such a cost lowers overall output. We argue that it also makes output gains less pro-poor. We show that the higher initial poverty rate, the less the rate of poverty reduction from a given rate of non-farm economic growth. The labor market dualism means that there will be less growth, and less of that growth will benefit the poor. Motivated by this model, we explore the determinants of poverty reduction in India. We first establish that there are significant differences between states in the extent to which the poor have benefited from (urban and rural) non-farm economic growth (section 3.1). Elasticities with respect to farm yields and state development spending have not, however, differed significantly between states. We then test for interaction effects between non-farm growth and initial conditions related to labor market dualism and human resource development (section 3.2). We find a number of significant interaction effects, broadly consistent with the predictions of our theoretical model. Section 4 concludes. 2. What makes growth pro-poor? It has been said often that economic growth is not sufficient for poverty reduction. Other things matter. For example, human resource development for poor people is now widely seen as a necessary component - alongside economic growth - of an effective strategy for fighting 4 poverty (World Bank, 1990). In the case of India, our previous work points to the importance of favorable initial conditions in human resource development and rural infrastructure in promoting higher rates of rural poverty reduction at a given rate of agricultural growth (Datt and Ravallion, 1998a). The issue we focus on here is whether the things that matter beyond the rate of economic growth enter additively or multiplicatively with growth. In other words, we ask here whether there a complementarity between these conditions and growth such that their combined effect is greater than the sum of their effects individually. Is it a matter of ticking off as many items as possible from a list of poverty-reducing actions, with a given additional impact of each action? Or are there important interaction effects, such that only certain combinations do the trick? And, if so, what are the key combinations? 2.1 Arguments from the literature It is evident that the distribution of consumption matters to the level of consumption poverty at a given level of mean consumption. But does initial distribution also matter to the subsequent rate of poverty reduction? One possible way it could matter is through the rate of growth in the mean. A long-standing view -- though for long periods a minority view amongst development economists it seems - has held that income inequality can be harmful to the pace of economic growth in poor countries. For example, in the 1920s and 1930s, Gunnar Myrdal believed that "..an equalization in favor of the low-income strata was also a productive investment in the quality of people ancl their productivity" (Myrdal, 1984, p. 154). A number of arguments have also been made as to why poverty can impede growth.3 One such argument points to the existence of credit market failures such that collateral is needed to 3 For surveys of recent arguments on this see B&nabou (1996) and Bruno et al (1998). 5 secure loans. Naturally the poor are less able to exploit opportunities for investment in (physical and human) capital; this lowers the growth rate, and more so the greater the proportion of poor (and hence investment constrained) people there are in the economy (see, for example, Binswanger et al., 1995). In addition to its potentially adverse implications for the rate of economic growth, high initial inequality is a likely candidate in explaining why the same rate of economic growth might be less effective in reducing poverty in one setting than another. Intuitively, in an economy where inequality is persistently low, the poor will tend to obtain a higher share of the gains from growth than in an economy in which inequality is high. There is supportive evidence from cross- country distributional data that higher initial income inequality entails a lower (absolute) elasticity of poverty to growth in average incomes (Ravallion, 1997; Timmer, 1997). For example, a country with a Gini index of 0.25 can expect a growth elasticity of the headcount index of around -3.3, while for a country with a Gini index of 0.60, the elasticity is -1.8 (Ravallion, 1997). But what specific determinants of "inequality" are likely to matter? The Gini index for incomes can be thought of as a product of various dimensions of inequality. Some inequalities may matter more than others to how much the poor share in growth. A potentially important factor in developing countries is the extent of the income disparities between urban and rural sectors. The existence of earnings and other income disparities between urban and rural sectors is clearly an important dimension of overall inequality in developing countries (Fields, 1980; Bourguignon and Morrison, 1998, who provide supporting evidence from cross-country comparisons.) Arguably, dualism also limits prospects for pro-poor growth. There is a long-standing view (though not a dominant one it would seem) that rural underdevelopment constrains prospects for industrialization; see for example Clarke 6 (1940). Again factor market distortions also entail that urban-rural inequality impedes poverty reduction through non-farm economic growth. Consider, for example, the classic model of Harris and Todaro (1970) in which wages in the non-farm sector are fixed above market clearing levels. While there is mobility between the urban and rural sectors, rural workers who move to the city will not all be able to get the new jobs, and so they will face unemployment, or turn to relatively low-paid urban informal sector activities. Then the extent of this distortion will influence both the wage rate in the rural economy and the extent to which the poor are able to gain from an expending non-farm sector. The initial population distribution between urban and rural sectors can also matter to the growth elasticity of poverty. In general, a change in the population distribution will shift the Lorenz curve in a dual economy. The direction of that effect is, however, theoretically ambiguous. For example, under the Kuznets Hypothesis, inequality will be low at both low and high levels of urbanization (Robinson, 1976; Fields, 1980; Anand and Kanbur, 1993). Urbanization is widely viewed as a positive factor in promoting rural non-farm economic growth, by expanding markets. Schultz (1953) noticed that rural non-farm activities tend to be more developed in the periphery of urban-industrial centers, and this has been confirmed in many countries. Enterprises are probably attracted to urban areas because of the larger local product markets, the availability of a skilled workforce, the wider variety of production inputs, the possibility of technological spillovers, and better infrastructure (Lanjouw and Lanjouw, 1997). It can be argued that these same factors will also matter to the growth elasticity of poverty. It is plausible that the poor will tend to be more constrained in their access to markets and infrastructure than the non-poor, and that the poor will tend to gain more from relaxing those constraints than do the non-poor. Assuming that the level of urbanization in an area reflects 7 these differences in access to markets and infrastructure, one can then expect that (other things being equal) the poor will be able to benefit more from non-farm growth when they live in a more urbanized area. Another factor influencing initial distribution and (hence) the elasticity of poverty to non- farm economic growth is the productivity of the main competing sector for workers, namely farming. For example, multiple cropping, irrigation and the spread of high-yielding varieties will probably entail a tighter labor market for the non-farm sector, and hence bring benefits to the new recruits into that sector through higher wages. Inequalities in ownership of physical and human assets are also likely to influence the prospects for poor people to participate in economic growth. Indeed, the credit-market failure argument as to why inequality matters to the rate of growth assumes that it will be the poor who are locked out of growth prospects. Low basic education attainments are often identified as a source of income inequality.4 Education will also influence how much the poor are equipped to participate in skill-demanding non-farm growth (relative to farming). This too is not a new observation. In the 1950s, Jacob Viner wrote that "The first requirements of high labor productivity under modem conditions are that the masses of the population shall be literate, healthy, and sufficiently well fed to be strong and energetic" (Viner, 1953, p. 100). More recently, the view that there are important synergies between human resource development and growth-oriented policy reforms has been a prominent theme in writings on development; examples include Dreze and Sen (1995) writing on India, and Thorbecke and Jung (1996) on Indonesia. The World Bank's approach to poverty reduction has also emphasized the importance 4 See, for example, the discussions in Tinbergen (1975) and Atkinson (1997). Evidence from cross-country comparisons of income-inequality reducing effects of average education attainments can be found in Li et al. (1998). 8 of combining human resource developmaent with economy-wide policies favorable to growth (particularly following World Bank, 1990). 2.2 Growth and poverty reduction in a simple dual economy model The last section sketched a number of arguments as to why economic growth might be more effective at reducing poverty in some developing economies than others. It would take a complex model to capture all these factors. Here we only offer a starkly simple model focusing on what is arguably the main way that the poor benefit from (urban and rural) non-farm economic growth, namely through its ability to absorb labor from the poor rural farm economy, and from a poor urban informal sector. The model divides the economy into "farm" and "non-farm" sectors, with higher wages in the latter. Obtaining a job in the non-farm sector requires a lumpy and risky investment (for which neither credit nor insurance are available). One possible interpretation is that the (rural) farm worker must finance migration to the (urban) non-farm economy, and must incur set-up costs at the destination including foregone earnings. Another interpretation is that the rural worker must invest in a minimal amount of schooling to qualify for a non-farm job. To cover the cost of this investment and the risks of moving, workers in the non-farm sector enjoy a wage advantage over those in the farm sector. This wage differential is taken to be exogenously fixed. In identifying "poverty" in this model, we assume that every worker in the farm sector is poor, but this is not true of any worker in the non-farm sector. So the headcount index of poverty, H, is simply the proportion of the population who fail to obtain work in the non-farm sector. Everyone else is in the non-farm sector, and none are poor. Firms in the non-farm sector can hire as many workers as they want at the prevailing wage rate, as can farms. Outputs are Ynf(1 -- H,Z) and Yf (H) for non-farm and farm sectors 9 respectively, where we allow for a vector of output enhancing variables in the non-farm sector, Z. The production functions have the usual properties; in particular, the marginal products of labor, MPf' (1 -H, Z) and MPf (H), are strictly positive and decreasing in the amount of labor employed, and approach infinity when no labor is employed, and approach zero when all the economy's labor is employed there. The marginal products of labor are equated with the wage rates in each sector, Wnf and Wf in obvious notation. Under these assumptions, there will be a unique equilibrium for the headcount index of poverty equating the relative marginal products of labor with the exogenous wage differential: MPf (I - H Z) W"f1 MPf(H) Wf (1) This equation shows how the headcount index varies with the productivity enhancing variables Z (interpretable as "growth effects" on poverty) and the wage disparity between sectors, W,fI /Wf (interpretable as "inequality effects" in this simple model). Given declining marginal products of labor in both sectors, the relative marginal product (the left hand side of 1) is strictly increasing in H given Z. So higher the inter-sectoral wage differential the higher will be the headcount index of poverty. The "growth effects" depend, however, on whether the output enhancing variables in Z are "labor augmenting" in that they increase the marginal product of labor. If so, then the only way to restore equilibrium after a productivity gain in the non-farm sector is to increase employment in that sector, so returning relative marginal products of labor to their previous level (as determined by the wage differential needed to cover the cost of switching sectors). Then the headcount index satisfying (1) will be strictly decreasing in Z. We can now return to our question: what determines how pro-poor growth will be? In terms of this model, we want to know what determines the elasticity of H to Z, as given by: 10 alnH = (1- H)i7z7q1 /;17<0 (2) where 71 and i7f are the wage elasticities of labor demand in the two sectors, 17 f + (1- H)77 < 0 (3) is the weighted mean elasticity of labor demand, and S is the vector of elasticities of the marginal product of non-farm labor to Z (i.e., ; = MPznfZ / Wnf ). The elasticities in (2) are implicitly functions of H. Thus (2) describes how the non-farm growth elasticity of poverty varies with the initial poverty incidence. From (1) we know that H is determined by the productivity paramneters of the non-farm sector (Z) and the wage differential. Thus the growth elasticity of poverty is a function of the same variables. How will the elasticities of poverty w.r.t. productivity gains in the non-farm sector vary with the initial poverty rate in the economy? One cannot predict on a priori grounds how the elasticities in (2) will vary with H. If we focus on the special case of constant elasticities then a particularly sharp result emerges, since (differentiating 2): a alnH H _ /77)2 >0 (4) aH aInZ Thus the (absolute) elasticity of poverty to labor augmenting productivity gains in the non-farm sector is decreasing in the initial poverty rate. Non-farm growth will reach the poor, but the poorer the setting, the lower the proportionate impact. The above model points to the imLportance to the poverty-reducing impact of non-farm growth of a specific aspect of initial distribution, namnely initial inter-sectoral income disparities, or "dualism" for short. This will motivate our empirical model (in section 3.2) of the determinants of the non-farm growth elasticity of poverty in India. 11 3. Econometric model and results We begin with a test for inter-state differences in the growth elasticities of poverty. We then try to explain the differences in terms of the variables identified in the last section. 3.1 Testing for inter-state differences in the elasticities ofpoverty There is a large literature studying the evolution of India's rural poverty measures over time, following (and debating) the seminal contribution of Ahluwalia (1978); Datt and Ravallion (1998b) survey this literature. In addition to updating and revising the data, our main point of departure is that we model the aggregate (urban and rural) state-wide poverty measures and we allow for state-specific growth elasticities of poverty.5 We also allow for state effects in the intercepts and for time varying variables that could well bias our results if they were omitted. We deliberately condition out the interstate differences in the level of poverty, by including state fixed effects. We also allow for state-specific time trends. From this specification we can then exploit the changes over time in aggregate farm and non-farm outputs, and see if their effects on the poverty measures vary across states. We have estimated various measures of absolute consumption-poverty for each of India's 15 main states using 20 rounds of the National Sample Survey (NSS) spanning the period 1960- 61 to 1993-94 at intervals of 0.9 to 5.5 years. We will be concerned with measures of absolute poverty, by which we mean that the poverty line is kept fixed in real terms (or in terms of the standard of living it commands) over the entire (spatial and temporal) domain of poverty measurement (Ravallion, 1994). We construct three different measures of poverty within the 5 It has been common in the literature to assume that growth responses are the same across all states. The only exception we know of is van de Walle (1985) who relaxes the pooling restrictions in the Ahluwalia (1978) model to allow the elasticities of rural poverty to agricultural output to vary between states. No attempt is made to explain the revealed differences. 12 Foster-Greer-Thorbecke (1984) class of measures: the headcount index (H), the poverty gap index (PG) and the squared poverty gap index (SPG). We have collated the surveys with data on farm yields, non-farm output, government spending, and variables describing initial conditions. The Appendix provides more detail. Figure 1 plots the trend rates of reduction in the headcount index by state against the trend rates of growth in non-farm product per capita (both estimated by OLS regressions of logarithms on time). There is a positive correlation (r-=0.30), though it is not significant (t=1.12). There is clearly enormous heterogeneity in the impact of non-farm economic growth on poverty; the rest of this paper will attempt to explain why. In testing for inter-state differences in the growth elasticities of poverty, the natural starting point is an econometric specification in which the log of the poverty measure is regressed on the log of mean income, allowing the regression coefficient to vary across states. We want to extend this specification to allow for differences in the elasticities with respect to different sectoral components of aggregate income, and for inflationary shocks. Since we are interested in describing (and later modeling) the growth elasticities rather than levels of poverty we also control for differences between states in the initial level of poverty. (So when we later try to explain the inter-state differences in elasticities, we will not be confusing this question with that of the effects of initial conditions on the initial level of poverty.) To allow for any state-specific omitted time trended variables we also include a state-specific trend. Combining these features, our test equation takes the form: In],, /J8NFP I NFPF + 1JYLD lnILD-t + 8jDEV InGO V, + 7,INFn, + 7r, t + V,i (5) where Pi, is the measure of absolute consumption poverty in state i at date t, NFP is real non- farm product per person, YLD is farm yield, GOV is real state development expenditure per capita, and INF is the inflation rate. (All these variables are defined more precisely in the 13 Appendix.) Consistently with Datt and Ravallion (1998a), we found that the fit of this model was improved if we used the two year moving averages of InYLD and lnNFP, and the lagged value of lnGOV. We initially estimated the model allowing e,, to be an ARI error term, allowing for the uneven spacing of the surveys (following the method in Datt and Ravallion, 1998a). However, the autoregression coefficient was not significantly different from zero so we set it to zero to simplify the estimation method. To be as flexible as possible, we initially write the f, 's as linear functions of a vector of state dummy variables. Since we also have state-specific time trends and differing effects of inflation, estimating (5) is equivalent to running a separate regression for each state. However, we found that a degree of pooling was consistent with the data. In particular, we could not reject the null hypothesis of constant coefficients at the 10% level for all variables except non-farm output per person and the state effects in the intercept.6 We could reject the null that the coefficients on NFP are the same across states at the 7% level or better for all poverty measures. Thus we impose a constant-coefficients restriction for YLD, GOV, INF and the time trend, leaving the coefficient on NFP free to vary between states and retaining the state fixed effects in the intercepts. An immediate implication of our finding that only the non-farm elasticities vary significantly across states is that we can reject the idea of significant trade-offs; it is not the case that states with higher elasticities with respect to non-farm growth tended to have lower elasticities to agricultural growth or development spending. 6 The failure to reject the null of constant coefficients for all except NFP was statistically convincing; probabilities for the tests of constant coefficients were no lower than 0.12, and most were above 0.25. Details are available from the authors. 14 Table I gives the estimated parameters of the restricted version of equation (5). Higher farm yields and higher development spending reduce all three poverty measures, and the coefficients are highly significant. Higher non-farm output per person lowers poverty in all states. Inflation is poverty increasing.7 Figure 2 gives the absolute values of the growth elasticities. (Notice that the elasticities are twice the ,B; estimate because lnNFP enters as the sum of the current and the lagged values.) For the headcount index, the elasticities vary from a low of 0.25 in Bihar to a high of 1.23 in both Kerala and West Bengal. For all states, the elasticities are higher (in absolute value) for the poverty gap index, and higher still for the squared poverty gap. This implies poverty-reducing gains below the poverty line. For the squared poverty gap index, the lowest elasticity is Jammu and Kashmir (0.49) and the highest is Kerala (2.47, though with West Bengal close behind at 2.40). To measure the quantitative importance to the overall rate of poverty reduction of the differences in elasticities we simulated rates of poverty reduction, in which we artificially set the non-farm growth elasticities of all states to a reference value (denoted /NFP ) set alternately at the lowest and highest elasticities across all states: dlnP7 dInPi +2(JNFP* -./3NAtFP) d t INFJ (6) dt dt dit where d ln Pi / dt and d ln NFIt / dt are the trend rates of poverty reduction and growth in non-farm output per capita respectively. This calculation assumes that the changes in elasticities leave other variables in the model unchanged. Consider the simulations when all states are given the highest elasticity of any state. If the more favorable initial conditions for the elasticities would 7 The significance of inflation confirms our earlier results (Datt and Ravallion, 1997, 1998a,b). In Datt and Ravallion (1998b) we argued that the main chiannel through which inflation mattered to India's poor was through its 15 have also led to higher (lower) growth rates then these simulations will have under (over) estimated the gains to rates of poverty reduction. Table 2 compares the national average trend rate of poverty reduction (obtained as OLS regression coefficients of log poverty measures on time) with the simulated rates of poverty reduction using equation (6). Recall from Table 1 that the lowest (absolute) elasticities are for Bihar (for the headcount index) and Jammu and Kashmir (for the other two poverty measures). If all states had Bihar's low elasticity then the mean rate of poverty reduction would have been only 0.3% per year for the headcount index (versus the observed mean of 1.3%). Under the Jammu and Kashmir elasticities, the average rate of decline in the headcount index would have been 0.8% and 1.0% for the poverty gap and squared poverty gap respectively (versus actual means of 2.2% and 2.8%). At the other extreme, if all states had Kerala's elasticity then the headcount index would have fallen at a trend rate of 3.5% instead of 1.3% per year. The trend rate of reduction in the squared poverty gap would have been 6.8% per year instead of 2.8%. Next we try to explain these inter-state differences in the growth elasticities of poverty. 3.2 Initial conditions and the non-farm output elasticities ofpoverty We now postulate that the elasticities of poverty to non-farm output are a function of initial conditions. Our model in section 2.2 pointed to the importance of labor market dualism between the urban (primarily non-farm) and rural (farm) sectors. So we look for indicators of urban-rural differences in incomes. We assume that 13NFP depends on the values (around 1960) of NFP, YLD, the initial population share in urban areas (URB), the initial ratio of urban to rural average consumption (CDIF), the share of the rural population that was landless at the beginning of the short-term adverse effect on the real wage rate for unskilled labor. 16 period (LLESS) and the initial irrigation rate in the state (IRR). (More precise definitions can be found in the Appendix.) Following the discussion in section 2, we also include the literacy rate; we use the female literacy rate (FLIT) following our previous work (Datt and Ravallion, 1 998a), though it makes little difference if one uses the male rate or the average. This can be interpreted as another determinant of the initial inter-sectoral earnings differential; a poorly educated rural population will presumably face higher costs of entry into the non-farm sector. We also allow for differences in the initial levels of government development spending, which might also influence initial distribution, including the extent of d.ualism. For example, low levels of development spending might be expected to favor the urban areas, with expansion into rural areas only occurring at higher levels of spending. Table 3 gives the data we will use on these initial conditions. Table 4 gives the results when we replace the state dummy variables in the sub-function for NFiP by the variables described above. All of these variables are entered in log form. We find that the non-farm elasticity of poverty is higher in states with higher initial farm yields, higher urbanization rates, lower urban-rwual disparities in consumption levels, and states with higher female literacy rates. Controlling for these variables, irrigation, landlessness, and initial non-farm output and development spending are not significant. There is of course considerable overlap in the explanatory powver of this set of initial conditions; for example, irrigation density is highly correlated with initial yield and the former becomes significant if the latter is dropped. Table 4 also gives a restricted form of the model dropping the jointly insignificant interaction effects with irrigation, landlessness, initial non-farm output per person, and initial development spending; the restrictions pass comfortably. On comparing the R2 values of Tables 1 and 4, it can also be seen that the variables we have used in explaining the inter-state differences in the non-farm growth elasticities of poverty 17 account for a large share of the variance. For example, with full state dummy variables, the value of R2 in the regression for the headcount index is 0.916; using our explanatory variables it drops to 0.887 indicating that only 3% of the variance in the poverty measures is accounted for by omitted variables (including measurement errors) influencing the measured non-farm growth elasticity. We know already that the differences in elasticities with respect to non-farm output were quantitatively important to the rates of poverty reduction. From the results in Tables 3 and 4, we can also say something about the relative importance of the various initial conditions we have identified in explaining the differences in elasticities. Consider the state with the lowest elasticity with respect to the headcount index, namely Bihar with an absolute elasticity of 0.25 (Table 1), well below Kerala's elasticity of 1.23. What growth elasticity of poverty would we have seen in Bihar if it had Kerala's initial conditions? The female literacy rate in 1960 was in Bihar was 6.9% (Table 3). The highest female literacy rate was (unsurprisingly) in Kerala, with 38.9%. If Bihar had Kerala's literacy rate then the parameter estimates in Table 4 (restricted forn) imply that the (absolute) elasticity of the headcount index to non-farm output per person in Bihar would have risen by 0.52, from 0.25 to 0.78.8 Performing the same calculation for the other initial conditions, we find that if Bihar had Kerala's urbanization rate then Bihar's elasticity would have risen by 0.17, while with Kerala's higher initial farm yields, Bihar's elasticity would have risen by a further 0.09. The urban-rural consumption disparity was, however, higher in Kerala; Bihar's elasticity would have fallen by 0.06 with Kerala's difference in initial urban-rural living standards. The combined effect of these four factors accounts for three-quarters of the difference between the non-farm output 8 The number of literate women per 1,000 adult women is logged, so 0.78=0.25+0.30x(5.96-4.23) (recalling that the regressor is the sum of current and lagged output). 18 elasticities of these two states. About two-thirds of the explained difference in elasticities is attributed to the initial differences in literacy alone. 4. Conclusions Using state-level data for India spaLnning 35 years, and allowing for both state-specific fixed effects and time trends, we find thaLt higher average farm yields, higher state development spending, higher (urban and rural) non-farm output and lower inflation were all poverty reducing. However, except for non-farm output, we could not reject the null that all these variables had the same elasticity across states for a given poverty measure. Farm yield growth, for example, was poverty reducing but with a similar elasticity in states with dissimilar initial conditions; thus it was the differences in the rate of agricultural growth that mattered to the poor. However, the elasticity of poverty to non-farm output varied significantly across states; growth in this sector brought much larger proportionate reductions in consumption-poverty measures in some states than others. Thus, we have been able to derive a state-specific measure of how pro-poor economic growth has been in. India over this period. To assess the importance of differing elasticities, we have simulated the rates of poverty reduction if all states had the non-farm growth elasticity of Kerala, which had the highest elasticity of any state. Then the annual trend rate of reduction in the headcount index over this 35-year period would have been more than two percentage points higher. For a state with the national average poverty rate0 in 1960 of4500this difference in trend rates of poverty reduction would have meant a difference in poverty rate by the mid-1990s of almost 15 percentage points; the poverty rate would have fallen to 13%No instead of 28%. Differing growth elasticities appear to have had a powerful longer term impact on the prospects of escaping absolute poverty in India through economic growth. 19 In attempting to understand the inter-state differences in the non-farm output elasticity of poverty, we have argued that certain inequalities can severely impede the prospects for poverty reduction through non-farm growth. The point can be illustrated by a simple model of dual labor markets, in which transition from the low wage farm sector to the high wage non-farm sector is necessary (and sufficient) for escaping poverty. But a successful transition comes at a cost. Then the initial inter-sectoral disparity in earnings, as needed to cover that cost, influences how much non-farm economic growth will reduce the incidence of poverty. Indeed, in its simplest version with constant labor demand elasticities, the initial poverty rate in the economy constrains the prospects for future poverty reduction through growth; the higher the poverty rate the less effective non-farm economic growth will be in reducing poverty. We have argued that the inter-state differences in the impact of a given rate of non-farm economic growth on consumption poverty reflect systematic differences in initial conditions. Broadly speaking, our results can be interpreted as indicating that non-farm economic growth was less effective in reducing poverty in states with "poor" initial conditions in terms of rural development (in both absolute terms and relative to urban areas) and human resources. Low farm productivity, low rural living standards relative to urban areas and poor basic education all inhibited the prospects of the poor participating in growth of the non-farm sector. Rural and human resource development appear to be strongly synergistic with an expanding non-farm economy in reducing poverty. Amongst the initial conditions we have found to matter significantly to prospects for pro-poor growth, the role played by initial literacy is particularly notable. For example, more than half of the difference between the elasticity of the headcount index of poverty to non-farm output for Bihar (the state with lowest elasticity) and Kerala (the highest) is attributable to the latter's substantially higher initial literacy rate. 20 Appendix This Appendix describes the main features of our data. A complete description of the data set assembled for this study (including sources of all variables) can be found in Ozler, Datt and Ravallion (1996) and the data and mantial can be found at the following web site: http://www.worldbank.orglpoverty/datalindiap aper.htm. We have used a consistent series of absolute poverty measures based on distributions of consumption per capita from 20 rounds of the National Sample Survey (NSS) spanning the period 1960-61 (round 16) to 1993-94 (round 50). All 20 rounds of the survey are covered for all 15 major states with the exception of Jammu & Kashmir, for which surveys were not held for the 48th and 50th rounds due to the prevailing political unrest. Punjab and Haryana had to be treated as a composite state because Haryana emerged as a separate state only in 1964. For NSS rounds since then, the poverty measures for the two states have been aggregated using rural population weights derived from the decennial censuses. Altogether, we use data from 298 consumption distributions to construct poverty measures. There is considerable variation in the sample sizes. For all states, the rural samples range from 3,762 households for the 16th round (July 1959-June 1960) to 99,766 households for the 32nd round (July 1977-June 1978), with a median sample size of 15,467 households for the 28th round (October 1973-June 1974). The s1mallest sample size for any state is 140 households for rural Gujarat for the 16th round. Assuming a simple random sample for the rural sector within the state, this implies a maximum standard en-or, for a headcount index of 50%, of 4.2 percentage points. The poverty lines we use are those defined by the Planning Commission (GOI, 1979). These lines were defined at the per capita monthly expenditure levels of Rs. 49 for rural areas and Rs. 57 for urban areas (rounded to the nearest rupee) at October 1973-June 1974 all-India prices. The Planning Commission followed the "food-energy method" in deriving the rural and urban lines; these poverty lines thus corresponded to levels of per capita total expenditure at which the caloric norms were typically attained in the rural and urban sectors. They correspond to a norm of per capita food energy intake of 2400 calories per day in rural areas and 2100 calories per day in urban areas. Poverty lines constructed this way have sometimes been found to not have the same purchasing power in urban and rural areas (Ravallion, 1994). However, 21 independent estimates of the urban-rural cost of living differential for 1973-74 (see Bhattacharya et al, 1980) confirmed the inter-sectoral cost of living differential of about 16% implicit in the Planning Commission poverty lines (see Datt, 1997, for further discussion). The nominal consumption distributions for each survey period were converted to constant prices using spatial (cross-state) price indices anchored to the consumption pattern of households in the neighborhood of the poverty line, and temporal consumer price indices for urban and rural sectors anchored to the consumption patterns of low-income workers. A substantial effort was invested into the construction of a consistent set of price indices across states and survey periods, using monthly data on consumer price indices from the Labour Bureau (disaggregated to the center level for the urban index). Our primary deflators were the Consumer Price Index for Industrial Workers (CPIIW) for the urban sector and the adjusted all- India Consumer Price Index for Agricultural Labourers (CPIAL) for the rural sector. The adjustment carried out to the CPIAL was for the price of firewood that has been held constant in the official CPIAL series since 1960-61. The nominal state-level distributions were further normalized for inter-state cost of living differentials estimated separately for urban and rural areas. For further details on the construction of the price indices, see Ozler, Datt and Ravallion (1996), Datt (1997), and Datt and Ravallion (1998a). The poverty measures are estimated from published grouped distributions of per capita expenditure using parameterized Lorenz curves; for details on the methodology see Datt and Ravallion (1992). As discussed above, the poverty measures are hypothesized to depend on both a set of time-dependent variables as well as a set of initial condition variables that determine how poverty-reducing the time-dependent variables are. Building on the empirical approach used in our earlier work (Datt and Ravallion 1 998a), we use time-dependent variables related to agricultural and non-agricultural growth, public spending on economic and social services and inflation. The specific variables used are as follows: (i) mean farm yield, given by real agricultural state domestic product (SDP) per hectare of net sown area in the state (denoted YLD), 9 (ii) non-farm output, measured by real non-agricultural state domestic product per person (NFP), 9 All real values were calculated using the (adjusted) state-specific CPIAL as the deflator. For further details on the State Domestic Product (SDP) data, see Datt and Ravallion (1 998a). 22 (iii) rate of inflation in the rural sector measured as the change per year in the natural log of the (adjusted) CPIAL,10 (iv) real state development expenditure per capita (GOV); development expenditure includes expenditure on economic and social services. The economic services include agriculture, rural development, special area programs, irrigation and flood control, energy, industry and minerals, transport and comm[unications, science, technology and environment. The social services include education, medical and public health, family welfare, water supply and sanitation, housing, urban development, labor and labor welfare, social security and welfare, nutrition, and relief for natural calamities. The data on SDP and state develo:pment expenditure are originally available on an annual basis, while the NSS surveys are not only not annual but they also do not always cover a full 12- month period. To match the annual data with the poverty data by NSS rounds, we have log- linearly interpolated the annual data to the mid-point of the survey period of each NSS round. We also identify a number of social and economic variables to describe initial conditions around 1960 (that we will later use in attempting to explain the growth elasticities of the poverty measures by state)."l The following variables (all measured in natural logs) describe these initial conditions: (i) the percentage of operated area which was irrigated in 1957-60 (IRR), (ii) the female literacy rate in 1961 defined as the number of literate females per thousand females in the total state population (FLIT), (iii) the percentage of landless rural households in 1961-62 (LLESS'J, as a measure of initial asset inequality in rural areas, (iv) the initial proportion of urban population (URB), (v) the ratio of the initial urban real mean consumption to that in the rural sector, where the initial real mean consumption in each sector is formed as an average over the first three NSS rounds available for that state (CDlF), as a measure of initial inter-sectoral disparity, (vi) the initial levels of the time-dependent variables, YLD, NFP and GOV. 10 This is state specific. However, the bulk of the effect is clearly through inter-temporal variation in the rate of inflation. We also tried adding the log of the ratio of the CPIIW to the (adjusted) CPIAL as an additional regressor, but that turned out to be insignificant. " The initial condition variables are assembled from a number of diverse data sources including the 1961 Census, the Statistical Abstract (Central Statistical Organization) for various years, and reports from a number of NSS surveys dealing with village statistics, land holdings and utilization, fertility, and infant mortality. 23 Note that we do not include the economic and human resource development indicators in time-varying form as additional explanatory variables in our model for two reasons. First, there are gaps in the available time series data on these variables over the period covered by our analysis. And second, even if a complete time series were available, these indicators in time- varying form would be arguably endogenous to the model. 24 References Ahluwalia, Montek S. (1978) "Rural Poverty and Agricultural Performance in India", Journal of Development Studies, 14(3): 298-323. Anand, Sudhir and Ravi Kanbur (1993) "The Kuznets Process and the Inequality- Development Relationship", Journal of Development Economics 40: 25-52. Atkinson, Anthony B. (1997) "Bringing Income Distribution in from the Cold", Economic Journal 107(March): 297-321. B9nabou, Roland (1996) "Inequality and Growth", in Ben Bernanke and Julio Rotemberg (eds) National Bureau of Economic Research Macroeconomics Annual, Cambridge: MIT Press, pp.1 1-74. Bhattacharya, S. S., Choudhury A. B. Roy, and Joshi P. D. (1980). Regional Consumer Price Indices Based on NSS Household Expenditure Data. Sarvekshana, 3:107-121. Binswanger, Hans, Klaus Deininger, and Gershon Feder (1995). "Power, Distortions, Revolt and Reform in Agricultural Land Relations." In J. Behrman and T.N. Srinivasan (eds) Handbook of Development Economics, Vol 3, Amsterdan: North Holland. Bourguignon, Fran,ois and Christian Morrison (1998) "Inequality and Development: the Role of Dualism," Journal of Development Economics 57(2): 233-257. Bruno, Michael, Martin Ravallion and ]Lyn Squire (1998) "Equity and Growth in Developing Countries: Old and New Perspectives on the Policy Issues", in Income Distribution and High-Quality Growth (edited by Vito Tanzi and Ke-young Chu), Cambridge, Mass: MIT Press. Clarke, Colin (1940) The Conditions of Economic Progress London: Macmillan. Datt, Gaurav (1997). Poverty in India 1951-1994: Trends and Decompositions, mimeo, World Bank and IFPRI, Washington D.C. Datt, Gaurav and Martin Ravallion (1992). "Growth and Redistribution Components of Changes in Poverty Measures: A Decomnposition with Applications to Brazil and India in the 1980s." Journal of Development Economics, 38: 275-295. and (1997). "Macroeconomic Crises and Poverty Monitoring: A Case Study for India." Review of Development Economics, 1(2): 135-152. -- and (1998a) "Why Have Some Indian States Done Better than Others at Reducing Rural Poverty?", Economica 65: 17-38. 25 and (1998b) "Farm Productivity and Rural Poverty in India", Journal of Development Studies 34: 62-85. Dreze, Jean and Amartya Sen (1995) India: Economic Development and Social Opportunity. Delhi: Oxford University Press. Fields, Gary (1980) Poverty, Inequality and Development. New York: Cambridge University Press. Foster, James, J. Greer, and Erik Thorbecke (1984). A Class of Decomposable Poverty Measures, Econometrica, 52: 761-765. Government of India (1979). Report of the Task Force on Projections of Minimum Needs and Effective Consumption. New Delhi: Planning Commission. Harris, John R., and Michael P. Todaro (1970). "Migration, unemployment and development: A two sector analysis", American Economic Review: 126-142. Lanjouw, Jean O., and Peter Lanjouw. 1997. "The Rural Nonfarm Sector: An Update." World Bank, Washington, D.C. Li, Hongyi, Lyn Squire and Heng-Fu Zou (1998) "Explaining International and Intertemporal Variations in Income Inequality", Economic Journal 108: 26- 43. Lipton, Michael and Martin Ravallion (1995). "Poverty and Policy", in Jere Behrman and T.N. Srinivasan (eds) Handbook of Development Economics Volume 3 Amsterdam: North- Holland. Myrdal, Gunnar (1988) in G. Meier and D. Seers (eds) Pioneers in Development, New York: Oxford University Press. Ozler, Berk, Gaurav Datt and Martin Ravallion (1996). A Database on Poverty and Growth in India, mimeo, Policy Research Department, World Bank. Ravallion, Martin (1994) Poverty Comparisons. Harwood Academic Press. (1997) "Can High Inequality Developing Countries Escape Absolute Poverty?" Economics Letters 56: 51-57. Ravallion, Martin and Shaohua Chen (1997) "What Can New Survey Data Tell Us About Recent Changes in Distribution and Poverty?", World Bank Economic Review 11(2): 357-382. Ravallion, Martin and Gaurav Datt (1996) "How Important to India's Poor is the Sectoral Composition of Economic Growth?", World Bank Economic Review 10(1): 1-25. 26 Robinson, Sherman (1976) "A Note on the IJ-Hypothesis Relating Income Inequality and Economic Development", American Economic Review 66: 437-440. Schultz, Theodore W. (1953) The Economic Organization ofAgriculture. New York: McGraw Hill. Thorbecke Erik, Jung Hong-Sang (1996) "A Multiplier Decomposition Method to Analyze Poverty Alleviation", Journal of Development Economics (48)2: 279-300. Timmer, Peter (1997) "How Well do the Poor Connect to the Growth Process?", mimeo, Harvard University. Tinbergen, Jan (1975) Income Distribution. Analysis and Policies, Amsterdam: North- Holland. van de Walle, Dominique (1985) "Population Growth and Poverty: Another Look at the Indian Time Series Evidence", Journal of Development Studies 34(4): 62-85. Viner, Jacob (1953) International Trade ana Economic Development, Oxford: Oxford University Press. World Bank (1990) World Development Report: Poverty. New York: Oxford University Press. 27 Table 1: Regressions for the state poverty measures allowing for inter-state differences in elasticities to non-farm output Variable Headcount index Poverty gap index Squared poverty (%) (%) gap index (xlOO) Real agricultural output per hectare of net -0.104 -0.191 -0.254 sown area: current + lagged (YLD) (4.44) (5.13) (5.01) Real per capita state development -0.140 -0.239 -0.336 expenditure: lagged (GOV) (2.55) (2.75) (2.83) Real non-agricultural output per person: current + lagged (NFP) Andhra Pradesh -0.288 -0.422 -0.521 (8.72) (8.09) (7.29) Assam -0.196 -0.254 -0.307 (4.92) (4.03) (3.56) Bihar -0.127 -0.332 -0.497 (2.50) (4.14) (4.53) Gujarat -0.282 -0.441 -0.546 (6.79) (6.72) (6.08) Karnataka -0.277 -0.406 -0.503 (7.13) (6.61) (5.98) Kerala -0.614 -0.976 -1.234 (14.66) (14.75) (13.62) Madhya Pradesh -0.147 -0.252 -0.332 (4.31) (4.71) (4.52) Maharashtra -0.203 -0.270 -0.296 (5.24) (4.42) (3.53) Orissa -0.333 -0.517 -0.671 (8.96) (8.83) (8.37) Punjab and Haryana -0.300 -0.407 -0.480 (9.17) (7.89) (6.79) Rajasthan -0.332 -0.488 -0.598 (7.23) (6.73) (6.03) Tamil Nadu -0.273 -0.391 -0.470 (7.77) (7.04) (6.20) Uttar Pradesh -0.249 -0.354 -0.437 (5.96) (5.36) (4.83) West Bengal -0.614 -0.932 -1.198 (11.39) (10.95) (10.27) Jammu & Kashmir -0.165 -0.209 -0.243 (4.99) (4.00) (3.41) Inflation rate (INF) 0.432 0.609 0.732 (5.29) (4.72) (4.15) Time trend 0.016 0.026 0.034 (6.18) (6.23) (5.94) Root mean square error 0.0950 0.1500 0.2054 R2 0.916 0.917 0.909 Test for equality of non-agricultural 15.29 15.19 14.84 growth elasticities across all states: (0.00) (0.00) (0.00) F(14,238) with p-value in ( ) Note: Absolute t-ratios in parentheses. All variables are measured in natural logarithms. A positive (negative) sign indicates that the variable contributes to a higher (lower) rate of increase in the poverty measure. The estimated model also included state-specific intercept effects, not reported in the Table. The number of observations used in the estimation is 272. 28 Figure 1: Trend rates of poverty reduction and non-farm output economic growth; states of India, 1960-94 3 0 Kerala P&H 0 = 0 AP g 2 -o0 West Bengal o0Gujarat -o: Tamil Nadu Ce ~~~~~~~~~~~~0 0 Orissa J&K 0~~~~~~~R ° -i Rajasthan KN oMaharashtra ° Q~~~~~~~~~ oUP -a D Madhya Pradesh 0Bihar o. 0 Assam 0 lTrend growth irk non-farm output per person (%/year) Note: Trend rates of growth estimiated by OLS regressions of the logarithms on time. Abbreviations: P&H=Punjab and Haryana; AP=Andhra Pradesh; J&K=Jammu and Kashmr; KN=Karrkataka; UP--I ttar Pradesh. 29 Figure 2: Elasticities of consumption poverty to non-farm output 2.5r 2.0 1.5 L 1.0- 0.0- 2 4 6 8 10 12 14 H PG _ SPGI Note: Absolute values of the elasticities by states of India implied by Table 2 for the headcount index (H), poverty gap index (PG) and squared poverty gap index (SPG). States ranked by elasticity for SPG. l=Kerala; 2=West Bengal; 3=Orissa; 4=Rajasthan; 5=Gujarat; 6=Andhra Pradesh; 7=Karnataka; 8=Bihar; 9=Punjab and Haryana; I O=Tamil Nadu; 11 =Uttar Pradesh; 12=Madhya Pradesh; 13=Assam; 14=Maharashtra; 15=Jammu and Kashmir 30 Table 2: Actual and simulated meani trend rates of poverty reduction across states Trend rates of change in Headcount index Poverty gap index Squared poverty gap poverty measure (%/year) index Actual mean across all states -1.315 -2.156 -2.816 Simulated mean with lowest -0.293 -0.763 -0.976 elasticity for all states Simulated mean with highest -3.493 -5.802 -6.827 elasticity for all states Note: Unweighted means across states of the silmulated rates of poverty reduction evaluated at state- specific trend growth rates of nonfarm output per person (based on regressions of log NFP on time). 31 Table 3: Initial conditions, around 1960 Female Urban- Urban-rural Irrigation Landlessness Agricultural Non-farm State literacy ization consumption rate (% rural output per product per development rate (%) rate (%) ratio (%) h'holds) hectare person spending Andhra Pradesh 12.0 17.4 1.24 23.8 6.8 19.3 1.06 42.7 Assam 16.0 7.2 1.25 4.4 27.8 52.0 1.60 59.0 Bihar 6.9 8.4 1.09 16.8 8.6 30.3 0.91 26.3 Gujarat 19.1 25.8 1.10 6.3 14.7 6.3 2.60 33.2 Karnataka 14.2 16.6 1.01 7.0 18.6 12.8 57.78 299.0 Kerala 38.9 22.3 1.19 12.4 30.9 96.4 0.52 36.0 Madhya Pradesh 6.8 15.1 1.14 4.2 9.1 8.0 4.03 72.3 Maharashtra 16.7 14.3 1.46 4.8 16.0 8.0 6.64 50.2 Orissa 8.6 28.2 1.01 15.0 7.8 18.3 0.22 19.6 Punjab and Haryana 14.1 6.3 0.97 41.0 12.3 16.7 1.71 44.1 Rajasthan 5.9 16.3 0.96 10.8 11.8 3.6 0.97 31.1 Tamil Nadu 18.2 26.7 1.47 38.3 24.2 51.0 1.80 43.2 Uttar Pradesh 7.1 12.9 0.94 34.8 2.8 41.9 1.19 27.0 WestBengal 16.9 24.4 1.46 18.8 12.6 76.0 4.86 39.2 Jammu & Kashmir 4.3 20.7 1.08 26.4 10.9 47.3 0.03 11.0 Notes: The units of initial farm yield are Rs.'000 per hectare at October 1973-June 1974 all-India rural prices; those of initial non-farm product Rs. '000 per person at October 1973-June 1974 all-India rural prices; those initial development spending are Rs. per person at October 1973-June 1974 all-India rural prices. Table 4: Explaining inter-state differences in the elasticity of poverty to non-farm output Headcount index Poverty gap index Squared poverty gap index Real agricultural output per hectare of net sown area: current + lagged -0.121 -0.126 -0.214 -0.222 -0.283 -0.293 (YLD) (4.60) (4.85) (5.12) (5.37) (4.97) (5.19) Real per capita state development expenditure: lagged (GOP) -0.043 -0.076 -0.125 -0.171 -0.206 -0.263 (0.72) (1.38) (1.33) (1.95) (1.60) (2.20) Real non-agricultural output per person : current + lagged (NFP) 0.758 0.822 1.317 1.265 1.835 1.638 (3.15) (4.67) (3.45) (4.52) (3.52) (4.29) NFP * initial female literacy rate (FLIT) -0.165 -0.151 -0.256 -0.231 -0.313 -0.282 (6.00) (8.15) (5.85) (7.85) (5.24) (7.02) NFP * initial urbanization (URB) -0.084 -0.089 -0.140 -0.148 -0.185 -0.193 (3.72) (4.18) (3.91) (4.36) (3.80) (4.17) NFP * initial urban-rural income disparity (CDIfl 0.293 0.325 0.511 0.527 0.692 0.691 (3 55) (4.46) (3.89) (4.55) (3.87) (4.37) NFP * initial irrigation rate (IRR) -0.010 - 0.009 - 0.028 (0.71) (0.40) (0.86) N1'FD * irnitial pet ceni of rural iandiess households (LLESS) -0.020 - 0.041 - 0.060 - (0.82) (1.05) (1.14) NFP * initial yield per hectare (YLD) -0.031 -0.039 -0.063 -0.060 -0.097 -0.081 (2.04) (3.70) (2.60) (3.56) (2.93) (3.54) NFP * initial per capita non-agricultural output (NFP) -0.011 -0.022 - -0.034 (0.54) (0.71) (0.79) NFP * initial per capita state development exp. (GOV) 0.033 - 0.067 - 0.093 - (0.83) (1.05) (1.07) Inflation rate (INF) 0.380 0.405 0.574 0.613 0.709 0.761 (4.10) (4.45) (3.90) (4.24) (3.52) (3.86) Time trend 0.011 0.012 0.018 0.020 0.024 0.026 (3.85) (4.53) (4.06) (4.56) (3.95) (4.36) Root mean squared error 0.1093 0.1094 0.1738 0.1738 0.2369 0.2370 R2 0.887 0.885 0.886 0.884 0.876 0.874 Test for joint significance of omitted variables: F(4,219) with p-value in () 1.08 1.00 1.06 (0.37) (0.41) (0.38) Note: Absolute t-ratios in parentheses; 272 observations. All variables are measured in natural logarithms. A positive (negative) sign indicates that the variable contributes to a higher (lower) rate of increase in the poverty measure. The regressions also included state-specific intercepts. 33 Policy Research Working Paper Series Contact Title Author Date for paper WPS2240 The Effects of Land Registration on Frank F. K. Byamugisha November 1999 E. Gelos Financial Development and Economic 37846 Growth: A Theoretical and Conceptual Framework WPS2241 How Land Registration Affects Frank F. K. Byamugisha November 1999 E. Gelos Financial Development and Economic 37846 Growth in Thailand WPS2242 Confronting Competition: investment Ritva Reinikka November 1999 H. Sladovich Response and Constraints in Uganda Jakob Svensson 37698 WPS2243 Designing Pro-Poor Water and Kristin Komives November 1999 M. Salehi Sewer Concessions: Early Lessons 37157 from Bolivia WPS2244 True World Income Distribution, 1988 Branko Milanovic November 1999 P. Sader and 1993: First Calculations, Based 33902 on Household Surveys Alone WPS2245 Opportunities for Improving Susmita Dasgupta November 1999 Y. D'Souza Environmental Compliance in Mexico 31449 WPS2246 Infrastructure's Contribution to David Canning November 1999 H. Sladovich Aggregate Output 37698 WPS2247 Does Deposit Insurance Increase Ashl Demirgu,g-Kunt November 1999 K. Labrie Banking System Stability? Enrica Detragiache 31001 An Empirical Investigation WPS2248 Privatization and Regulation of Antonio Estache November 1999 G. Chenet-Smith Transport Infrastructure in the 36370 1990s: Successes ... and Bugs to Fix for the Next Millennium WPS2249 Argentina's Transport Privatization Antonio Estache November 1999 G. Chenet-Smith and Re-Regulation: Ups and Downs Jose C. Carbajo 36370 of a Daring Decade-Long Experience Gines de Rus WPS2250 Universal Service Obligations ir, Omar Chisari November 1999 G. Chenet-Smith Utility Concession Contracts and Antonio Estache 36370 the Needs of the Poor in Argentina's Privatizations WPS2251 Will the Euro Create a Bonanza Daniel Cohen November 1999 H. Vargas for Africa? Nicolai Kristensen 37871 Dorte Verner Policy Reseairch 'Working Paper Series Contact Title Author Date for paper WPS2252 Productivity Growth, Capital Ejaz Gharii December 1999 N. Mensah Accumulation, and the Banking Vivek Suri 80546 Sector: Some Lessons from Malaysia WPS2253 Revenue Recycling and the Welfare Ian 'VV. H. Parry December 1999 R. Yazigi Effects of Road Pricing Antonio Miguel R. Bento 37176 WPS2254 Does "Grease Money" Speed Up Daniel Kaufmann December 1999 H. Sladovich the Wheels of Commerce? Shang-Jin Wei 37698 WPS2255 Risk and Efficiency in East Asian Luc Laeven December 1999 R. Vo Banks 33722 WPS2256 Geographical Disadvantage: Anthony J. Venables December 1999 L. Tabada A Heckscher-Ohlin-von Thunen Nuno Limcfo 36896 Model of International Specialization WPS2257 Infrastructure, Geographical Nuno Limnio December 1999 L. Tabada Disadvantage, and Transport Costs Anthony J. Venables 36896 WPS2258 Market Access Bargaining in the J. M/lichael Finger December 1999 L. Tabada Uruguay Round: Rigid or Relaxed Ulrich Reincke 36896 Reciprocity? Adriana Castro WPS2259 Predicting Currency Fluctuations Daniel Kajfmann December 1999 E. Khine and Crises: Do Resident Firms Gil lMehrez 37471 Have an Informational Advantage? Sergio Schmukler WPS2260 Regional Integration Agreements: Anthony J. Venables December 1999 L. Tabada A Force for Convergence or 36896 Divergence? WPS2261 Is Knowledge Shared within Kaushik Basu December 1999 M. Mason Households? Ambar Narayan 30809 Martin Ravallion WPS2262 How Inadequate Provision of Public Ritva Reinikka December 1999 H. Sladovich Infrastructure and Services Affects Jakob Svensson 37698 Private Investment