33259 SOUTH ASIA REGION PREM WORKING PAPER SERIES Poverty in Sri Lanka: the Impact of Growth with Rising Inequality Ambar Narayan and Nobuo Yoshida July 2005 Report No. SASPR-8 A WORLD BANK DOCUMENT The findings, interpretations, and conclusions expressed in this paper are entirely those of the author(s) and should not be attributed in any manner to the A WORLD BANK DOCUMENT The findings, interpretations, and conclusions expressed in this paper are entirely those of the author(s) and should not be attributed in any manner to the World Bank, to its affiliated organizations, or to members of its Board of Executive Directors or the countries they represent. World Bank, to its affiliated organizations, or to members of its Board of Executive Directors or the countries they represent. About the SASPR Working Paper The purpose of the SASPR Working Paper Series is to provide a quick outlet for sharing more broadly research/analysis of issues related to development in South Asia. Although the primary source of such research/analysis is SASPR staff, other contributors are most welcome to use this outlet for rapid publication of their research that is relevant to South Asia's development. The papers are informal in nature and basically represent views/analysis of the concerned author(s). All papers submitted for publication are sent for an outside review to assure quality. I provide only a very light editorial touch. For enquiries about submission of papers for publication in the series or for copies of published papers, please contact Naomi Dass (telephone number 202-458-0335). Sadiq Ahmed Sector Director South Asia Poverty Reduction and Economic Management World Bank, Washington D.C. Acknowledgements This paper has been prepared as one of the background papers for the Poverty Assessment for Sri Lanka, which is due to be completed in FY2006, and reflects work in progress being conducted in collaboration with the Department of Census and Statistics (DCS), Sri Lanka. The paper was also presented at an Inequality Workshop in Delhi (December, 2004) organized by South Asia Region to inform the World Development Report (WDR, 2006) on Equity and Development. We gratefully acknowledge DCS for household survey data and active collaboration on poverty analysis and poverty mapping, Yoko Kijima for her work with DCS on poverty line for Sri Lanka, Uwe Deichmann (World Bank) for constructing maps based on GIS information, Peter Lanjouw (World Bank) for advice on poverty mapping, and Tara Vishwanath (World Bank) for advice and guidance. We are also grateful for the thoughtful comments and suggestions received from Shankar Acharya, discussant for this paper, the WDR 2006 team and other distinguished participants at the Inequality Workshop, which will greatly benefit our ongoing work for the Poverty Assessment report. All poverty maps included in this paper are provisional maps, based on 5 percent Census data. For final versions of these maps, see World Bank (2005, draft) ­ a policy note that fully documents the poverty mapping exercise that was undertaken jointly between the DCS and World Bank, as part of technical assistance program by the World Bank for poverty monitoring in Sri Lanka. Table of Contents Introduction......................................................................................................................... 1 I. Consumption Poverty In Sri Lanka: Trends And Patterns......................................... 1 II. Poverty, Growth And Inequality................................................................................ 4 III. What Explains The Pattern Of Poverty And Inequality In Sri Lanka........................ 8 Conclusion ........................................................................................................................ 16 Reference .......................................................................................................................... 17 Figures And Appendix...................................................................................................... 18 Introduction 1. Poverty reduction in Sri Lanka has been a story of mixed success. On the one hand, Sri Lanka has made impressive gains in providing access to basic social services that has resulted in significant human development. At the same time, reduction in income poverty has been modest and uneven, with the gains being largely limited to Colombo and neighboring districts. The gap between average urban and rural incomes has widened, as have the differences across provinces and districts. A key challenge facing Sri Lanka today is not only how to improve growth to increase the pace of poverty reduction, but also to ensure that the benefits of growth spread to the lagging regions and sectors of the economy. 2. The last decade has seen a marginal reduction of poverty in Sri Lanka of about 3 percentage points, in spite of a relatively healthy growth performance ­ annual per capita GDP growth averaged around 3.5 percent between 1991 and 2002. Poverty reduction was also uneven across different parts of the country. This paper ­ also intended to be a background paper for the upcoming Poverty Assessment for Sri Lanka ­ will attempt to lay out the evidence on evolution of poverty in Sri Lanka over the last decade or so. We will particularly focus on understanding how the pattern of poverty reduction is related to growth in consumption and incomes, as well as the change in distribution of these measures. The evidence will also look at how poverty reduction in Sri Lanka over the last decade is characterized by widely divergent experiences across sectors (urban, rural and estates) and geographical regions, and the factors that appear to be strongly correlated with the geographical dispersion of income/consumption poverty. While analyzing the full range of factors that explain the uneven pattern of poverty reduction in Sri Lanka is beyond the limited scope of this paper, the evidence presented here can serve as a useful starting point for such analysis, by hinting at the nature of constraints that hinder equitable growth and poverty reduction in the country. 3. Section I will present evidence from household data on the trends and patterns of consumption poverty in Sri Lanka during the last decade, focusing on widening inequality between regions and provinces. Section II will attempt to draw links between poverty trends and movements in growth and inequality measures, including the sectoral pattern of growth during recent years. Section III will examine the correlates of poverty, to understand which underlying factors explain the wide geographical differences in income and poverty outcomes. I. Consumption Poverty in Sri Lanka: Trends and Patterns 4. National poverty headcount for Sri Lanka Table 1: Poverty headcounts for Sri Lanka1 showed a modest decline over the decade ­ from 26.1 percent in 1990-91 to 22.7 percent in 2002 90-91 95-96 2002 (Table 1) with uneven poverty reduction across National 26.1 28.8 22.7 urban, rural and estate sectors.2 In the Urbana 16.3 14.0 7.9 intervening period, national poverty increased by Rurala 29.4 30.9 24.7 almost 3 percentage points from 1990-91 to Estate 20.5 38.4b 30.0 1995-96, followed by a decline of more than 6 Source: HIES, using official poverty lines (DCS) Notes: a: The classification of urban and rural areas is different percentage points from 95-96 to 2002. Other between HIES 90-91 and HIES 95-96 onward as discussed measures of poverty, namely depth and severity below. The changes do not alter the arguments on the poverty of poverty, show similar trends (Table 2). trend, although the levels of headcount ratios (urban and rural) need to be cautiously interpreted. Evidence presented later will show that even b: Comparability of estate headcount for 95-96 with that for during the period of poverty reduction between other years may be affected by the fact that HIES in 95-96 was sampled differently for the estate sector 1Based on official poverty lines (Rs. 1423, Rs. 833 and Rs. 475 for 2002, 95-96 and 90-91) respectively. The official poverty line is derived using "cost of basic needs" method on 2002 HIES data, and deflated by Colombo CPI to obtain nominal lines for other years. 2Excludes the Northern and Eastern provinces. 1 1995-96 and 2002, the benefits of growth continued to be unevenly distributed, resulting in increasing inequality co-existing with gains in average levels of consumption. 5. During the decade from 1990- 91 to 2002, the gap in poverty Table 2: Depth and severity of poverty incidence between the urban Poverty gap Severity of poverty/Sqd. pov gap sector and the rest of the country 90-91 95-96 2002 90-91 95-96 2002 widened. Urban poverty halved National 0.056 0.066 0.051 0.018 0.022 0.016 during the period, while rural Urban* 0.037 0.029 0.017 0.013 0.009 0.005 poverty declined by less than 5 Rural* 0.063 0.072 0.056 0.020 0.025 0.018 percentage points. Poverty in the Estate 0.033 0.079 0.060 0.009 0.025 0.018 estate sector, on the other hand, Source: HIES for relevant years, using official poverty lines (DCS) increased by about 50 percent Note: * The classification of urban rural areas is different between HIES 90-91 and over the decade, making this HIES 95-96 onward as discussed below. sector now the poorest.3 The trend in rural poverty is consistent with positive trend in per-capita agricultural production during the later part of the decade (as the sector recovered from the severe drought of 1996), while per-capita growth in the preceding years was negative. 6. In interpreting the urban and rural poverty estimates above, it is important to note one caveat: the classification of urban and rural areas changed between HIES 90-91 and HIES 95-96, with the Town Council areas considered as urban in HIES 90-91 being re-classified as rural in the later two rounds.4 As a result, the estimated proportion of urban population reduced by around 7 percentage points from HIES 90-91 to HIES 95-96 and stayed almost constant from 95-96 to 2001-02. While this should suggest caution in interpreting the absolute magnitude of changes in rural and urban poverty from 90-91 to the later years, the implication of this change on the main conclusion above ­ that urban poverty has been reduced dramatically while rural poverty has been stagnant ­ is likely to be minimal. The estimates reported below (Tables 3 and 4) also support this conclusion: poverty in Colombo district (and in Western Province where it is situated), which constitutes a large part of the urban sector, more than halved between 90-91 and 2002, while poverty reduction in predominantly rural districts has been minimal during the same period of time. Regional Differences in Poverty Incidence 7. The sharp differences in poverty incidence Table 3: Share in total and poor population of across sectors are also mirrored by geographical the country by province (2002) differences. As Figure 1 shows, poverty incidence % of headcount % of was by far the lowest in Western Province in 2002, population (%) poor with a poverty headcount of 11 percent, with Western 33 11 16 North Central having the next lowest poverty North-Central 7 21 6 incidence of 21 percent. In contrast, the poorest Central 15 25 16 provinces of Sabaragamuwa and Uva have Northwest 13 27 16 headcount poverty rates of around 35 percent. The Southern 14 28 18 low poverty incidence in the Western Province is Sabaragamuwa 11 34 17 largely due to the location of Colombo where most Uva 7 37 12 of Sri Lanka's economic activity is concentrated. Source: WB staff calculations using HIES (2002) The high incidence of poverty in some of the other provinces is consistent with the evidence on wide disparity in provincial incomes and share of 3Note that because the estate sector comprises of a relatively small part of the population, the HIES sample (about 4 and 7 percent of the weighted sample in 95-6 and 90-1 respectively) yields poverty estimates with a higher degree of "error." 4This implies that the definition of "urban" in 90-91 HIES included Urban, Municipal and Town Councils, whereas that in 95-96 and 2002 HIES included Urban and Municipal Councils only. 2 GDP, with Western Province accounting for almost half of the country's GDP. The low poverty incidence in Western Province implies that it accounts for about 33 percent of the population, but for only 16 percent of the poor (Table 3). In comparison, Uva and Samaragamuwa together are home to 18 percent of the population and 29 percent of the poor.5 Table 4: The headcount indices by districts (%) 8. Disaggregating poverty headcounts by Province District 90-91 95-96 2002 districts shows even wider differences in Colombo 16 12 6 poverty incidence (Table 4). In 2002, while Western Gampaha 15 14 11 Colombo district has a headcount rate of only 6 Kalutara 32 29 20 percent, 37 percent of the population in Badulla Kandy 36 37 25 and Monaragala are poor. Another interesting Central Matale 29 42 30 fact is that in 2002, poverty headcount in only 5 Nuwara Eliya 20 32 23 out of the 17 districts for which data are Galle 30 32 26 available (covering 41 percent of the population) Southern Matara 29 35 27 happened to be at or below the national Hambantota 32 31 32 headcount, indicating that the national figures North- Kurunegala 27 26 25 masks the high incidence of poverty in large West Puttalam 22 31 31 parts of the country. North- Anuradhapura 24 27 20 9. Thus by all accounts, poverty is Central Polonnaruwa 24 20 24 concentrated geographically in Sri Lanka. Uva Badulla 31 41 37 Evidence also suggests that over the past Monaragala 34 56 37 decade, there has been a tendency towards wider Sabara- Ratnapura 31 46 34 regional disparity (Table 3 and Fig. 2). Between gamuwa Kegalle 31 36 32 90-91 and 2002, poverty reduction has been Source: HIES 90-91, 95-96, and 2002 (DCS) substantial for the three districts in Western Province including Colombo (ranging between 27 and 63 percent) ­ as well as in other districts like Kandy, Anuradhapura and Galle (between 13 and 31 percent) that include key urban centers. In comparison, poverty headcount has declined by about 7 percent in Kurunegala and Matara, and remained unchanged or increased in the remaining nine districts, four of which (Ratnapura, Nuwara Fig. 1: % change in headcount between 90-91 and2002 by Eliya, Badulla and Puttalam) district registered increases of 10 60 percent or above. District 40 poverty trends also show the seu 20 0 vulnerability of certain alv -20 districts to shocks such as %-40 droughts. From 90-91 to 95- -60 96, there have been large -80 increases in poverty incidence ob ar el al ara ta elat a ur mal for districts (like Monaragala, mol aratul ydn ahap pu Ga ga alud Ka m Mat ar Ma Ratnapura, Matale and Co Ka ha neu Ga adr urK nuA tonabmaH awu elag Ke alagar ayilEar Ba ttauP wa Puttalam) that experienced nnoloP na Mo aptnaR Nu severe drought (Table 3). 5The share of total population by province remained fairly stable over the decade, with the share of Western Province increasing by about 2 percentage points, while those for Southern, Northwest and Sabaragamuwa declined slightly (by 1 percentage point or less) 3 Table 5: Share of GDP by Province 10. The trend towards wider regional disparity in incomes is also apparent from the figures on Province 1990 1996 2002 provincial shares in national GDP. Western Western 40.2 43.7 48.1 Province's share in national GDP increased from 40 North-Central 4.8 4.6 3.9 percent in 1990 to 48 percent in 2002, while that of Central 12.1 10.0 9.4 all the other provinces ­ with the sole exception of Northwest 11.1 11.3 10.1 Southern province ­ declined by between 1 and 4 Southern 9.5 9.0 9.7 percentage points (Table 5). The share of Uva, the Sabaragamuwa 8.1 9.0 6.9 poorest province, in national GDP halved during this Uva 8.1 5.1 4.3 period (from 8 percent to 4 percent) while that of Sabaragamuwa fell from 8 percent to 7 percent. Source: Dept. of National Planning Note: the shares do not add up to 100%, since Northeast This meant that Uva and Samaragamuwa together province is excluded from this table accounted for only 11 percent of the GDP in 2002, while being home to 18 percent of the population (see table 1.9 of the World Bank 2004 in detail). II. Poverty, growth and inequality 11. The observed pattern of poverty reduction during the Table 6: Mean real per capita consn. last decade ­ with large variation across sectors and (at 2002 prices) by quintiles regions ­ occurred during a period of significant growth in Consn. quintiles 90-91 2002 the national average consumption level, along with an Q1 1045 1068 equally significant skewing of the distribution of consumption.6 Q2 1499 1596 Q3 1909 2168 12. Mean per capita consumption for the country Q4 2489 3117 increased by about 29 percent in real terms between 90-91 Q5 4871 7325 and 2002. However, as Table 6 shows, this increase was Aggregate 2363 3055 unevenly distributed ­ ranging from around 50 percent for Source: HIES 90-91, 95-96, 2002 and CCPI the top quintile and 25 percent for the 4th quintile, to 2 and 6 percent for the 1st and 2nd quintiles respectively. Table 7: Inequality: Gini coefficient Increasing inequality over the decade is also reflected by of per capita expenditure* the Gini coefficients (Table 7). The Gini increased 90-91 2002 between 1990-91 and 2002 ­ by 25 percent for the country Urban* 0.37 0.42 as a whole, including an increase of about 15 percent for Rural* 0.29 0.39 the urban sector and about 35 percent for the rural sector. Estate 0.22 0.26 National 0.32 0.40 13. An analysis of how growth in per capita consumption Source: HIES 90-91, 95-96, 2002 and CCPI expenditure was distributed among the population, using Note: * The classification of urban and rural areas is different between HIES 90-91 and HIES 95-96 the well-known device of Growth-Incidence Curves onward as discussed above. (GICs) provides more insights into the pattern of that occurred during the period (Figure 2).7 The picture that emerges nationally is that while growth in levels of consumption had a poverty-reducing effect during the last decade, the benefits 6This paper focuses on per capita consumption expenditure rather than per capita income for the following reasons. First, the per capita income data in HIES 1990-91 is not comparable with that for other years due to different treatment of income of self-employment. Second, income data is more vulnerable to measurement errors than consumption data. (see Deaton (1997) for details). However, looking at income estimates for comparable years (i.e., HIES 1995-96 and 2002), the trends and patterns in inequality are found to be almost identical as that for per capita consumption expenditure (see Appendix 3 for details). 7The GIC maps the average annual rate of growth of real per capita consumption between the relevant years for all centiles (1 percent quantile) of the consumption distribution (see Ravallion and Chen, 2003 for details). 4 accrued disproportionately among the better-off. This pattern is also unchanged whether one looks at the entire period of 90-91 to 2002, or the most recent sub-period of 95-96 to 2002.8 Figure 2: Growth Incidence Curves for Per Capita Consumption Expenditure National (90-91 to 01-02) National (95-96 to 01-02) 12 12 11 11 02 02 01-ot 10 01- 10 9 ot 9 91 8 96 0-9 8 pxe 7 95- p 7 6 ex pcniht 6 pc 5 ni 5 ht ow 4 ow 4 gr 3 gr 3 ual 2 2 ann 1 annual 1 0 0 1 10 20 30 40 50 60 70 80 90 100 1 10 20 30 40 50 60 70 80 90 100 % of population ranked by pcexp % of population ranked by pcexp Rural (90-91 to 01-02) Rural (95-96 to 01-02) 12 12 11 11 02 02 10 01-ot 9 01-ot 10 9 91 8 96 8 90-p 7 95-p 7 ex 6 ex 6 pcnih 5 pcnih 5 owt 4 owt 4 grl 3 grl 3 2 2 annua annua 1 1 0 0 1 10 20 30 40 50 60 70 80 90 100 1 10 20 30 40 50 60 70 80 90 100 % of population ranked by pcexp % of population ranked by pcexp Urban (90-91 to 01-02) Urban (95-96 to 01-02) 12 12 11 11 02 02 1-0 10 1-0 10 ot 9 ot 9 91 8 96 8 90-p 7 95-p 7 ex 6 ex 6 pcni 5 pcni 5 h h owtrg 4 3 owtrg 4 3 2 2 annual annual 1 1 0 0 1 10 20 30 40 50 60 70 80 90 100 1 10 20 30 40 50 60 70 80 90 100 % of population ranked by pcexp % of population ranked by pcexp 14. Between 90-91 and 2002, while growth in per capita consumption was positive for the entire distribution of rural population, the gains were negligible (below 1 percent) for the bottom 40 percent and sizeable for the top 20 percent. A similar pattern is seen for the period between 95- 96 and 2002, although the gains in absolute terms were higher for all groups ­ consistent with a relatively larger reduction in rural poverty during this period, after the spike in poverty in 95-96.9 The GICs for the urban sector show similar skewed growth in per capita consumption over the 8It is important to recall here the problems associated with the change in definition of urban and rural areas between the surveys of 90-91 and 95-96, as mentioned above. Note however that in Figure 2, this problem can affect the results of only the GICs drawn separately for urban and rural areas for the period 1990-91 to 2002; it does not affect national GICs for all periods, neither does it affect urban/rural GICs for the period 95-96 to 2002. 9 Between 1990-1 and 1995-6, consumption growth was in fact negative for the lower 50 percent of the rural consumption distribution ­ which explains the increase in rural poverty during this period. 5 decade and during the last five years. For the decade, the shape of the urban GIC closely resembles that of the rural GIC for the same period, albeit with somewhat higher levels of per capita consumption growth for the entire distribution ­ consistent with the fact that urban poverty declined more than rural poverty during this period. A notable difference between urban and rural areas is seen only for 95-96 to 2002: the gains for those near the top of the urban distribution appear to be especially large, in comparison to the rest of the distribution of urban as well as rural population. 15. The GICs thus tell a story of highly skewed growth in per capita consumption over the decade, for urban and rural areas alike ­ a pattern that was even more pronounced when one looks at the more recent sub-period of 95-96 to 2002. At the same time, poverty incidence declined by 6 percentage points for urban and rural areas alike during this period after it had increased during the previous five years ­ attributable entirely due to an upward shift in the distribution, rather than any redistribution towards the less well off. The results from an exercise in growth- redistribution decomposition further demonstrates this fact, by quantifying the relative "contribution" of growth in mean consumption and change in distribution of consumption to the change in poverty headcount over the relevant periods. 16. The growth-inequality decomposition of changes in poverty headcount is a useful device to quantify the links between poverty reduction, growth and inequality in consumption. The so- called growth effect between selected years measures the simulated impact of the increase in mean per capita consumption expenditure on poverty headcount (keeping the distribution unchanged at that for the initial year); while the redistribution effect measures the simulated impact on headcount of the change in the distribution of per capita consumption (keeping the mean unchanged from the initial year).10 17. This exercise done for Sri Lanka for the period 90-91 to 2002 shows Fig. 1.14: Growth-inequality decomposition between 90-91 Fig. 3: Growth-inequality decomposition between 90-91 and 2002 and 2002 that if inequality had not increased, a significantly greater reduction in 20.0 14.2 13.9 poverty would have been achieved as 10.0 8.4 a result of the observed growth in mean per capita consumption (Fig. 3). 0.0 Actual change With no change in distribution from -0.6 Growth -4.7 -4.8 -3.3 -1.8 that in 90-91, the rise in mean -10.0 Redistribution -8.3 consumption (of about 29 percent) -12.0 Residual -20.0 -15.5 would have been enough to reduce -18.4 poverty by more than 15 percent -30.0 nationally between 90-91 and 2002 Urban Rural National (by 12 and 18 percent in urban and rural areas respectively), instead of the observed reduction of only 3 percent (8 and 5 percent in urban and rural areas respectively). How poverty trends are related to sectoral patterns of growth 18. Macroeconomic data are consistent with the growth in average levels of consumption observed in household survey data. Per capita GDP for Sri Lanka grew in real terms by 21 percent during the period 1991-96, and by 16 percent during 1996-2002, which translated to an annual average growth rate of 3.9 percent and 2.5 percent respectively for the two periods respectively. Cumulatively over the period 1991-2002, real per capita GDP increased by 40.6 percent, compared to the growth in per capita real mean consumption of 29 percent over the same 10See Datt and Ravallion (1992) for details. 6 period.11 Thus it is clear that whether one looks at macroeconomic data or household surveys, growth performance over the last decade in Sri Lanka was healthy enough to have had a larger impact on poverty that it actually did. 19. Increase in inequality, as shown above, no doubt explains why the impact of growth on poverty was tempered, with much of the benefits of growth being concentrated in urban areas and for the upper income groups. Useful insights on uneven poverty reduction, particularly on why rural areas lagged significantly behind urban areas, are also gained from looking at the distribution of output/income across sectors of economic activity. Figure 4 shows that even as real GDP increased substantially between 1990 and 2002, the share of the agricultural sector declined, while that of industry and services increased. Agriculture accounted for 28 percent of GDP in 1991 and 20 percent in 2002; the share of industry increased from 24 percent to 27 percent, while that of services increased from 50 percent to 54 percent over the same period. Fig. 4:Contribution to GDP (at constant 1996 prices) by Fig. 5: Real per capita GDP and sectoral outputs Sector 1000 5 800 4.75 ion)illb(.sR 600 .sR 4.5 of 4.25 400 gol 4 200 3.75 0 3.5 1 2 3 4 5 6 7 8 9 0 1 2 9911 9921 9931 9941 9951 9961 9971 9981 9991 0002 0012 0022 199 199 199 199 199 199 199 199 199 200 200 200 Agriculture Industry Services GDP Agriculture Industry Services Source: Central Bank of Sri Lanka Annual Report (multiple issues) 20. This phenomenon by itself would not necessarily explain the observed Table 8: Average annual % growth in per capita terms pattern of unequal poverty reduction, GDP Agriculture Industry Services since a natural process of development 1991-96 3.9 -0.2 6.2 4.2 can be expected to lead to declining 1996-02 2.5 0.4 2.6 3.3 importance of the agricultural sector. Source: Central Bank of Sri Lanka Annual Reports However, what is seen for Sri Lanka is also stagnation in the sector itself over an entire decade. Figure 5, which traces the log of per capita output from each sector over time clearly shows that agricultural output per capita remained almost unchanged, and in fact registered Fig. 6: Employment by Sector negative growth during certain years 7 ) (1991-92, 1995-96 and 2000-01). While 6 industry output in per capita terms grew illions 5 by an annual average rate of 6 percent m( 4 nte during 1991-96 and 3 percent during m 3 1996-2002, and services grew by 4 2 ploy percent and 3 percent respectively over mE 1 the same periods, output from agriculture 0 per capita fell by an annual average rate 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 of 0.2 percent during 91-96, and grew by Agriculture Industry Services an anemic annual average of 0.4 percent Source: Central Bank of Sri Lanka Annual Report (multiple issues) during 1996-2002 (Table 8). Note: Excludes North and East provinces 11This difference is explained by the fact that per capita GDP estimates from national accounts and household consumption from household surveys do not match precisely for most countries for which such data are available 7 21. Agriculture's share in total employment also fell, from 43 percent in 1991 to 34 percent in 2002, in keeping with the trend of its stagnation and declining importance in the economy (Fig. 6). The decline in agriculture's share was largely compensated by an expansion on the services sector: while the share of industry in total employment remained stable at around 21-23 percent, that of services increased from 37 percent in 1991 to 45 percent in 2002. Notably however, the absolute number of people employed by agriculture actually increased by 3.5 percent from 1991 to 2002 ­ the decline in agriculture's share in total employment was a result of total employment increasing by 29 percent over the same period. 22. A few important facts thus emerge from the evidence. The growth that occurred in the Sri Lankan economy largely bypassed the agricultural sector, resulting in its declining importance in terms of share of national income ­ which is also consistent with falling share of the agricultural sector in employment. At the same time, there appears to be no evidence of a structural shift in the labor force away from the agricultural sector. Even in 2002, agriculture employed more than 1/3rd of the total workforce (and this share is likely to be far higher in rural areas), and had in fact registered a modest increase in the number of people employed compared to 1991. This suggests that a large part of the population of Sri Lanka still rely on agriculture for livelihood, and the stagnation in agriculture thus appears to be an important contributing factor to the persistence of poverty in rural/estate areas. To make this connection with greater clarity, it will be necessary to look at evidence from different sources, including household surveys, to explore how lack of growth in agriculture translated into wages and household incomes in rural areas, and how these impacts are distributed geographically. III. What Explains the Pattern of Poverty and Inequality in Sri Lanka? 23. In this section, the potential factors associated with the pattern of inequality and uneven incidence of poverty are examined, with a particular focus on those that appear to best explain the sharp geographical differences in income and poverty levels that are observed. Evidence from various studies seem to suggest that lack of access to market opportunities, infrastructure and employment opportunities constrain growth and poverty reduction in lagging areas of Sri Lanka; and it is important to see whether HIES data provides information that support these hypotheses. In the context of the discussion above on a possible relationship between stagnation in agriculture and poverty, household data also offers the opportunity to understand further the nature of this relationship, and whether low agricultural productivity and incomes disproportionately affect particular groups within farm households. 24. With these broad objectives in mind, we first explore what characteristics appear to explain ­ at least partially ­ the widely divergent poverty rates across provinces. However, since a large degree of heterogeneity exists within provinces, provincial comparisons can be useful only to a limited extent, and one would need to go down to a much lower level of geographical disaggregation to draw useful inferences about what factors explain the occurrence of high levels of poverty in certain areas of Sri Lanka. We make a beginning in this direction, by looking at some of the patterns that emerge when poverty incidence is measured (using small-area estimation techniques for poverty mapping) at a lower level of disaggregation (Divisional Secretariat or DS Division level). When such a poverty map is juxtaposed against other maps depicting potential poverty correlates like access to infrastructure and rainfall patterns, certain patterns of association emerge that are instructive ­ as useful information to policymakers about the constraints to poverty reduction in specific areas and the kind of policy interventions that may be necessary, as well as a starting point for more rigorous analysis of the underlying processes that explain these patterns. 8 Comparison among provinces 25. Accessibility to business opportunities and poverty: Table 9 indicates the relationship between poverty indices and various measures of access to business opportunities.12 Clearly, Western Province, which has performed outstandingly in terms of economic growth among provinces, has significant advantages in terms of access to business opportunity in any measure. On the other hand, many of these accessibility measures do not necessarily explain the extent of deprivation in other provinces. For example, the province with the highest poverty headcount ratio, Uva, has almost the same level of telephone connection as the Western province and even better access to a bank, while it has poor access to the nearest market, nearest city, and electricity. Importantly, among these indicators of accessibility, the indicators of geographical isolation such as the distance to nearest market or nearest city seem to be most closely correlated with poverty. We will investigate the relationship of such geographical isolation with poverty indices in more detail later in this paper. Table 9: Poverty indices and access to infrastructure by province Average Average enterprises that Share Share of Share of Poverty of enterprises enterprises Headcount distance to distance to with a land line or located in a Ratio (%) nearest market nearest city use electricity (km) (km) (%) mobile phone (%) community with a bank (%) Western 11 5 5 79 24 70 Central 25 5 10 80 7 47 Southern 28 13 3 68 18 62 North Western 27 6 6 61 15 70 North Central 21 7 8 61 8 75 Uva 37 11 15 62 23 78 Sabaragamuwa 35 18 4 76 15 70 Total 23 8 9 69 15 67 Source HIES 2002 WB ICS ICS ICS ICS ICS Notes: "HIES 2002 WB" denotes that the world bank staff calculated these figures using HIES 2002; "ICS" refers to "Sri Lanka: Improving the Rural and Urban Investment Climate (2004)". 26. The share of the rural and estate sectors in the economy: As discussed earlier, population in the rural sector and especially the estate sectors are worse off than those in the urban sector at the national level. Table 10 indicates this pattern to be largely true, albeit with some significant differences, like the case of Sabaragamuwa. Mean per capita consumption expenditure in the urban areas of the Sabaragamuwa province is significantly lower than that of other provinces, and is in fact comparable to the mean per capita consumption in rural Western Province. This suggests that poverty in Sabaragamuwa is not only a rural problem ­ economic opportunities appear to be limited even in the urban areas of Sabaragamuwa. 27. Table 10 also indicates that a large share of estate sector in a province contributes to a high poverty incidence in the province. The two poorest provinces, Uva and Sabaragamuwa, have relatively high shares of estate population. A comparison between Central and North-Central provinces is also instructive: although Central province has the richest urban sector and the second richest rural sector, its poverty headcount ratio is worse than that in the North Central province, partly because Central province has the highest share population in the estate sector. These facts are consistent what was shown in Section I above, that the estate sector is the poorest in Sri Lanka, and contributes disproportionately to the poor population in the country. 12These measures are taken from "Sri Lanka: Improving the Rural and Urban Investment Climate (2004)" 9 28. Finally, Table 10 illustrates how inequality within a province is also important in explaining its relative rank in terms of poverty incidence. The case of North Central province, which has the second lowest among provinces, is instructive. In terms of business opportunities shown in Table 9, North Central is not in a particularly good position. Also, the majority of population in North Central province resides in the rural sector (more than 95 per cent), whose mean expenditure is not high in comparison with other provinces. One potential explanation for the relatively low poverty incidence in North-Central is that the province has the lowest inequality as measured by GINI coefficient of per capita consumption expenditure. Another potential explanation could be that the North Central province has a very low share of estate population that usually consist of many poor households, and furthermore, the estate population appear to be less poor than that in other provinces in terms of average per capita consumption. Table 10: Poverty headcount and inequality within province Mean per capita monthly consumption GINI of per Poverty Share of population (%) expenditure (Rs) capita Headcount consumption Ratio (%) RURAL ESTATE URBAN RURAL ESTATE expenditure (%) Western 11 69.7 0.9 5447 3800 1911 40 North Central 21 95.2 1.0 5001 2574 4833 33 Central 25 70.2 21.0 5644 2630 2067 38 North Western 27 95.7 0.5 5113 2582 1908 37 Southern 28 90.2 2.0 4918 2488 1979 37 Sabaragamuwa 35 87.4 8.8 3864 2317 1817 35 Uva 37 80.8 15.6 5282 2342 1765 39 Total 23 80.6 6.0 5285 2865 1985 40 Source: World Bank staff calculations using HIES 2002 29. Poverty and educational attainment: Educational attainment of household heads is usually highly correlated with the welfare level of their households in developing countries. In fact, Western province, which is the richest province in Sri Lanka, also records the Table 11: Poverty and educational attainment by highest share of household heads with province tertiary education, while Uva has the Share of Share of lowest share of the same indicator (Table Poverty household household 11). However, for other provinces, Headcount heads with at heads with educational attainments are not Ratio (%) least primary tertiary particularly good indicators of education (%) education (%) consumption poverty. For example, North Western 11 97 35 Western province records the second North Central 21 93 17 highest educational attainment in the Central 25 92 19 shares of household heads with primary North Western 27 95 20 education and tertiary education, but its Southern 28 92 19 poverty headcount ratio is the fourth best Sabaragamuwa 35 92 17 in Sri Lanka. The tenuous link between Uva 37 87 15 educational attainment and differences in poverty rates across provinces is mainly Total 23 94 23 due to the long-lasting initiative of the Sri Source: World Bank staff calculations using HIES 2002 Lanka government to provide uniform educational opportunities, which is manifested in high rates of primary education attainment even in the poorest provinces. Having said that, access to higher levels of education is not universal, and (as Table 11 shows) the substantial advantage 10 enjoyed by Western Province in this regard may in part explain its better economic performance and the fast growth of the modern sector observed there. 30. Poverty and unemployment: High Table 12: Poverty and unemployment rates by unemployment rate has long been one of the province most important policy issues in Sri Lanka. There is no doubt unemployment of main Poverty Headcount Unemployment income earners constitute a serious challenge to Ratio (%) rate (%) household incomes. Unemployment is however Western 11 8.9 not necessarily correlated with high incidence of 8.4 poverty in Sri Lanka for two reasons. Firstly, North Central 21 evidence suggests that many college graduates Central 25 8.9 and highly educated young people are North Western 27 7.8 unemployed in Sri Lanka mainly because they Southern 28 10.6 can afford to wait for better job opportunities. In Sabaragamuwa 35 9.8 its analysis unemployment in Sri Lanka, Rama Uva 37 6.0 (1999) concludes that a large proportion of the Total 23 8.8 unemployed are young, relatively educated Source HIES 2002 WB HIES 2002 DCS individuals who live with their parents and Notes:"HIES 2002 DCS" refers to "Poverty Statistics/ benefit from family support to continue their job Indicators for Sri Lanka" by DCS (2004). search. Such unemployed individuals tend to seek a relatively good job, either in the public sector or private sector activities protected heavily by product and labor market regulations. Secondly, however low the wages are, the extremely poor must work simply because they cannot survive without income. In fact, as Table 12 shows, unemployment figures for provinces do not show any pattern of association with their ranking in terms of poverty incidence in Sri Lanka. For example, the Uva province, which has the highest poverty incidence, records the lowest unemployment rate.13 The unemployment figures presented here however constitute a very partial picture. To understand better the links with poverty, it will be more useful to examine the relationships between wages, underemployment and poverty, which will require additional work since statistics on underemployment are not readily available. 31. Poverty and the agricultural sector: As discussed earlier, the agricultural sector has been stagnant during the 1990s. A question is whether the stagnation in the agricultural sector can explain the geographical differences in poverty incidence. In other words, does a province with a high share of population working in the agricultural sector have high poverty incidence? 32. In addressing this question, Table 13 (next page) shows a decidedly mixed picture: on the one hand, Western province records the lowest share of population working in the agricultural sector, while Uva, which has the highest poverty incidence, records the highest share of population working in the agricultural sector. On the other hand, Sabaragamuwa has the second lowest share of population working in the agricultural sector, but its poverty incidence is among the highest in Sri Lanka. 13Sri Lanka Poverty Assessment (World Bank, 2002) indicates a similar conclusion ­ poverty is not closely correlated with unemployment. 11 33. Table 13 shows another notable fact: with the exception of Western Province and Sabaragamuwa to a lesser extent, agriculture is the primary source of livelihood for a majority of households in all provinces. This reiterates the point made in the previous section ­ while the share of agriculture in employment has fallen over the years, agriculture continues to be quite important in terms employment and Table 13: Poverty and the agricultural sector by province contribution to Share of Per capita monthly household income household incomes in Poverty population Household heads rural areas. This is Head- Household heads count working in the working only in even more true when working in the Ratio (%) agricultural agricultural sector the non- one excludes Western sector agriculture sector Province, which is an Western 11 18 3646 4632 outlier in this respect North Central 21 70 2589 3578 because of the rapid Central 25 55 2592 3642 growth of the modern North Western 27 51 3463 3124 sector that has Southern 28 55 2750 3113 occurred there. Table Sabaragamuwa 35 48 2236 2484 13 however presents a Uva 37 79 2526 3921 partial picture, and a better understanding of Total 23 45 2876 3905 the contribution of Source: The World Bank staff calculates the figures using HIES 2002 agriculture can be Notes: "Share of population working in the agricultural sector" includes obtained by measuring individuals whose primary occupation does not belong to the agricultural sector. its share in household incomes, which requires additional work. 34. Thus there appears to be no strong correlation between the share of population working in the agricultural sector and poverty incidence. This however may be a result of pooling occupations that are quite different in terms of income opportunities ­ like wage workers, marginal and large landholders ­ in the same category. Therefore, it is useful to see if some clearer associations emerge when one actually disaggregates agriculture sector households by different occupations. Table 14: Poverty and the share of employees in the agricultural sector Poverty Per capita monthly household income Head- Share of In the agriculture sector Household head count agricultural wage Household head Household head working working only in Ratio employees in total working as wage only as own account the non- (%) employment (%) employees worker/employer agricultural sector Western 11 4 2448 3871 4632 North Central 21 17 2120 2728 3578 Central 25 28 2035 2907 3642 North Western 27 10 2264 3700 3124 Southern 28 20 1934 3054 3113 Sabaragamuwa 35 16 1456 2521 2484 Uva 37 29 2027 2758 3921 Aggregate 23 15 2015 3166 3905 Source: World Bank staff calculations using HIES (2002) 35. In Table 14, we examine correlations between poverty incidence and the share of wage employees in the agricultural sector. The category of wage employees include those who do not own any land, as well as marginal farmers who need to work for wages because their own farms 12 do not yield sufficient incomes. As shown in the table, not all types of households (by occupation of household head) working in the agricultural sector have low incomes. In fact in some provinces, households whose heads are working as an own account worker/employer in the agricultural sector are even better off than those whose heads are working in the non-agricultural sector. Those who appear to be the most poorly off on the average are households whose heads are working as wage employees in the agricultural sector. This implies that provinces with a higher share of agricultural employees in total employment tend to be poorer ­ Uva has a share of 29 percent employed as agricultural wage workers, compared to 4 percent for Western Province. This also partly explains why the Central and the Southern provinces record relatively high poverty incidences­ although the share of the agricultural sector in employment is relatively low in these provinces, the share of agricultural wage employees is high in comparison to other provinces. 36. There are a few caveats that qualify these results. Firstly, due to limited information on income source, it will take more work to decompose the share of income for each sector. Doing so for a number of survey periods will be particularly useful for quantifying the extent of stagnation of the agricultural sector, including households that are primarily agricultural wage workers. Secondly, evidence from other data sources (see World Bank 2003) suggest that poverty incidence is also high among non­agricultural wage workers in rural areas. Given that the non-agricultural sector also provides income diversification opportunities for households engaged in agriculture, a study of rural poverty must also incorporate analysis of the non- agricultural sector. 37. In summary, comparisons among provinces in this section provide us with some evidence that poverty is concentrated in geographically isolated areas (in terms of distance to markets and cities), the estate sector, and among households with agricultural wage employees. At the same time, there is no clear evidence that inequality between provinces is explained by differences in unemployment rates, educational attainments and access to infrastructure such as electricity, telephone line, and banks. 38. However, it is worth noting that there exist a high degree of heterogeneity within a province, which severely limits our ability to understand the true relationship between poverty and its important correlates without further geographical disaggregation. This is attempted below, using maps indicating poverty incidence and other geographical information at the level of DS division.14 Comparison among Division Secretary's (DS) divisions: 39. The role of poverty maps in analysis: As shown earlier, Sri Lanka's overall poverty rate is reasonably low for a developing country ­ around 22 percent at the national level. However, there are good reasons to believe that much higher poverty rates can be found in specific areas in Sri Lanka, even within districts that on the aggregate show a relatively low incidence of poverty. Thus there is a long-standing demand for an understanding of poverty and inequality at finer levels of spatial disaggregation than what is available as direct estimates using HIES data (district level). Reasonably accurate estimates of poverty, at the DS Division or lower administrative level, can greatly facilitate monitoring and evaluation of the existing poverty alleviation programs and geographic targeting of future government interventions. This is possible through an exercise in Poverty Mapping ­ a technique developed in Elbers et al (2003) and since implemented in many countries around the world ­ whose objective is to provide statistically reliable estimates of 14Sri Lanka has a four tiers of administrative units: province, district, divisional secretary's division (DS division), and Grama Niladhari division (GN division). In total, there are 9 provinces, 25 districts, 324 DS divisions, and around 14,000 GN divisions. 13 consumption-based welfare indicators. The DCS of Sri Lanka has initiated this exercise ­ the first-ever attempted in South Asia ­ with technical assistance from the World Bank. 40. The poverty mapping method takes advantage of strengths of both the HIES ­ that includes consumption aggregates but lacks enough sample size to estimate poverty at the geographical unit below district ­ and the Population CENSUS ­ that has enough sample size but lacks consumption aggregates. Using this method, members of the DCS and World Bank staff produced a preliminary map (using 5 percent of Census data) of poverty headcount ratios at the DS division level. The results of this exercise, presented below (along with other GIS-based maps of characteristics like infrastructure and rainfall/drought), suggest the important role that poverty maps can play in the analysis of spatial inequality and its correlates. These results will also motivate the ongoing exercise to produce further disaggregated poverty maps (for instance, at GN division level), using full Census information when that becomes available in the coming months. Such "finer" poverty maps, when combined with other GIS maps constructed at a similar level of disaggregation, can yield even more useful insights into the linkages between poverty and its potential determinants. 41. Poverty incidence at the DS division level: Figure 7 ­ a map of poverty headcount ratios at the Division Secretary Division level ­ illustrates some interesting geographical characteristics of poverty incidence. First, as expected, poverty headcount ratios are substantially lower in Colombo district and its neighboring areas. Second, areas with high rates of poverty are much more prevalent in areas in deep south (Southern, Uva and Sabaragamuwa provinces) than in areas more to the center and north of the country (North-West and North Central provinces).15 Having said that, the poverty map is particularly useful in highlighting that pockets of extreme poverty exist in almost all parts of Sri Lanka, including in districts with relatively low aggregate poverty rates. For example, some DS divisions in the southern part of Western Province (Kalutara district) suffer from severe deprivation; and similar pockets of extreme poverty exist in North- west and North-Central provinces (e.g. in parts of Puttalam, Anuradhapura and Kurunegala districts). Thirdly, there is a wide variation in poverty incidence in Central province, while poverty headcount ratios are high almost uniformly in the Sabaragamuwa province and, especially, Uva province. 42. Furthermore, high headcount ratios do not necessarily indicate large population of poor in a DS division, since the number of poor people in an area depends on its total population as well as the poverty headcount ratio. Figure 8 illustrates this fact quite clearly: even though the headcount ratio in Colombo district is only 6 percent, the number of poor people in the district is high, especially in Colombo city areas, due to the large population; and the coastal areas from southern Gampaha to the western part of Hambantota record high numbers of poor people despite their relatively low headcount ratios. On the other hand, many of the DS divisions in Monaragala district record the highest headcount ratios in the nation, but lower numbers of poor people due to the low density of population. This illustrates the danger of relying exclusively on the poverty headcount index in designing poverty alleviation programs that intend to achieve the maximum impact. In Sri Lanka's case, targeting all anti-poverty programs to the poor districts in deep south, for instance, will run the risk of missing large numbers of the poor in districts that are better-off on the average, including the capital city of Colombo. 43. Accessibility and poverty: As discussed earlier, geographical isolation measured by the distance to nearest market/city seems to be highly correlated with poverty incidence. To examine this relationship in detail, Figure 9 shows an accessibility index for each DS division. The accessibility index is calculated for every point as the sum of the population totals of surrounding 15Note that the darkest areas of the map denote projected poverty headcount rates of 36 percent and above, compared to the country's average of 22 percent. 14 cities and towns, inversely weighted by the road network travel time to each town. The map shows the mean of the access values for all points that fall into a given DS unit. Figure 9 clearly shows that areas surrounding the Colombo district in Western Province are well connected to cities/markets ­ the southeastern coastal areas surrounding Colombo and the areas between Colombo city and Kandy city ­ while most of Uva province is geographically isolated. In general, it appears to be the case that the further one goes away from the area surrounding Colombo, lower is the accessibility index. 44. It is clear from comparing between Figure 7 and Figure 9 that there are clear indications of a negative correlation between the poverty headcount ratio and the accessibility index. For example, the coastal areas surrounding the Colombo district record a high accessibility index as well as a low poverty headcount ratio, while many DS divisions in the Monaragala district are very poor and geographically isolated. A simple regression between these two indices verifies the observation above ­ there is a significant negative correlation between these two indices (Figure 10).16 45. Droughts and poverty incidence: The agricultural sector remains one of the major sources of livelihood in all provinces except for Western Province, with one group, namely agricultural wage employees, being particularly vulnerable. Therefore any natural disaster such as flooding or droughts can have serious consequences on their livelihoods, resulting in a sharp increase in poverty incidence. 46. Figure 11 shows rainfall anomalies in 2001, which are defined as a percentage of deviation from 30 years average annual rainfall.17 It indicates some areas were severely affected by droughts in 2001, especially most of Hambantota district and southern part of Matara district. Drought do not necessarily increase poverty incidence ­ the impact also depends on other factors such as availability of proper irrigation system, kinds of crops cultivated, and the level of diversity in occupations, that determine the extent of vulnerability of the population to rainfall anomalies. For these reasons, it is difficult to hypothesize about the links between rainfall anomalies and poverty incidence ­ especially in the absence of information about the other factors mentioned above, and panel data that allows for the measurement of impact over time. 47. Nevertheless, we can find some rough correlation between poverty incidence and drought- affected areas by comparing the poverty map (Figure 7) and the drought map (Figure 11) visually. For example, Hambantota district and southern parts of Kalutara district were affected by severe drought, and at the same time, record high poverty incidence in terms of poverty headcount ratio. While these visual links are indicative of the specific areas of the country that are likely to be vulnerable to such events, more careful analysis needs to be done to measure the impact of such vulnerability on poverty. 48. To summarize, the poverty maps at the DS division level reveal (i) DS divisions with severe level of deprivation are more prevalent in the southernmost areas of the country; but at the same time pockets of high poverty exist in even relatively better off districts; (ii) large numbers of poor people are found not only in Central province and the southern part of the Badulla district, but also in Western Province including Colombo city area due to the high density of population there. The comparison between accessibility to towns and markets and poverty headcount ratios shows quite clearly that poverty in Sri Lanka is closely related to geographical isolation, which in turn is consistent with the pattern of higher poverty and fewer economic opportunities found in rural 16The R2 for a regression of poverty rate of DS divisions on the accessibility index is 0.21, which is very high considering that this is a regression with a single variable to explain DS level variations in poverty rates. 17Maps depicting elevation and 30 years average rainfall are presented in appendix 2. According to these, rainfall is concentrated in the south-east of the country, while high mountains cover the south central part of Sri Lanka. Note that there does not seem to be an obvious visual association between poverty incidence and elevation or rainfall. 15 areas, especially in remote districts/provinces. One pattern seems quite clear ­ that accessibility to markets declines as one moves further away from the economic growth center that is Colombo. Finally, the impact of drought on poverty is observed for certain areas of the country, but this issue needs further analysis and clarification ­ in terms of the extent of vulnerability and its impact on poverty. Conclusion 49. Poverty trends in Sri Lanka over the last decade reveal an uneven pattern of poverty reduction, with rapid improvements in Colombo and its surrounding districts co-existing with stagnation in more remote areas of the country. Widening inequality, between rural and urban areas, and across provinces and districts, has also resulted in overall reduction in poverty incidence being less than what would have been the case with more equitable distribution of the benefits of economic growth, as evidence in Section II of this paper has clearly indicated. 50. Available evidence seems to indicate strongly that low productivity and growth in agriculture is one of the important reasons why rural incomes have lagged far behind urban incomes. Poverty incidence is especially strongly associated with employment as agricultural wage labor, which accounts for a significant share of employment in the poorer provinces. Poverty is also associated strongly with access to markets and connectivity with urban centers, and the poorest areas of the country also tend to be the most under-served in terms of such infrastructure. 51. In concluding, it is important to note that that while this paper provides some clues about what characteristics are associated with the wide differences in poverty incidence in Sri Lanka, much more detailed analysis is necessary to understand the processes that determine how these factors act as constraints to income opportunities. While employment in agriculture, particularly as wage labor, is associated with lower incomes, a key question is what explains persistent low productivity and incomes (and holds down agricultural wages) in the agricultural sector. Evidence also suggests that lack of access to markets acts as an important obstacle to income opportunities in remote rural areas, but the critical question that remains is what would be the most optimal way to improve links with urban centers and markets in lagging regions and provinces. Although access to primary education does not appear to have any correlation with poverty, it is quite likely that attainment of tertiary education ­ which is far from universal ­ is linked with better opportunities in the labor market, as well as greater diversification across occupations. While unemployment may not have direct links with poverty incidence, an important question is to what extent does a rigid labor market (especially the regulations that impose the rigidity) hamper the growth prospects of the country, and affects the potential expansion of the modern sector to areas beyond Western Province where it is currently concentrated. 52. Future analyses for the Poverty Assessment, using different data sources, will examine in greater detail some of the above questions. That said, even the preliminary evidence presented here on poverty and its correlates, at a high level of geographical disaggregation, can be useful for policymakers ­ for instance in identifying poor areas and informing targeted interventions to address specific needs of those areas. 16 Reference Central Bank of Sri Lanka. Annual Report, various years. Datt, G. and M. Ravallion (1992). "Growth and redistribution components of changes in poverty measures: A decomposition with applications to Brazil and India in the 1980s." Journal of Development Economics 38: 275-95. Deaton, A. (1997) The analysis of household surveys: A microeconomic approach to development policy. The World Bank. Washington, DC. Department of Census and Statistics, Sri Lanka (2004). Poverty Statistics/Indicators for Sri Lanka Department of Census and Statistics, Sri Lanka (2004). Bulletin on Official Poverty Line for Sri Lanka Elbers, C., J. O. Lanjouw, and P. Lanjouw (2003). "Micro-level estimation of poverty and inequality". Econometrica, 71(1): 355-364. Rama, Martin. The Sri Lankan unemployment problem revisited, Policy Research Working Paper No. 2227. The World Bank. Ravallion, M. and S. Chen (2003). "Measuring Pro-Poor Growth", Economics Letters 78: 93-99 World Bank. Sri Lanka (2002). Poverty Assessment. Report No. 22535-CE World Bank. Sri Lanka (2003). Promoting agricultural and rural non-farm sector growth, Volume 1: Main Report. Report No. 25387-CE. World Bank. Sri Lanka: Development Policy Review (2004). Report No. 29396-LK World Bank and Asian Development Bank (2004). Sri Lanka: Improving the Rural and Urban Investment Climate World Bank (2005). "A Poverty Map for Sri Lanka--Findings and Lessons". Policy Note (draft). South Asia Region. 17 Figures and Appendix Figure 7: Poverty Headcount Classification based on natural breaks Note: this is a preliminary poverty map, based on 5 percent Census data 18 Figure 8: Number of poor persons Each dot is randomly placed within a DS unit and represents 500 poor persons Note: this is based on a preliminary poverty map, based on 5 percent Census data 19 Figure 9: Accessibility Potential The accessibility index is calculated for every point as the sum of the population totals of surrounding cities and towns, inversely weighted by the road network travel time to each town. This map shows the mean of the access values for all points that fall into a given DS unit. The index is a measure of potential market integration reflecting the quality and density of local transportation infrastructure. 185 cities and towns were included in this analysis. 20 Figure 10:Poverty rates and access to urban centers in DS units 60 50 ) R2 = 0.21 %(etaryt 40 30 veroP20 10 0 4.5 6.0 7.5 9.0 Mean accessibility index Note: The accessibility index (also known as "population potential") is calculated for every point as the sum of the population totals of surrounding cities and towns, inversely weighted by the road network travel time to each town. The index is a measure of potential market integration reflecting the quality and density of local transportation infrastructure. 185 cities and towns were included in this analysis. No poverty data were available for Northern and Eastern regions. 21 Figure 11: Rainfall anomalies in 2001 Annual rainfall in 2001 minus avg. annual rainfall over 30 years - red areas are drier in 2001 - blue areas are wetter in 2001 computed using only stations that have data for 30 year period and 2001 Data source: Department of Meteorology 22 Appendix 1: A map of administrative units in Sri Lanka 23 Appendix 2: Elevation and 30 year average rainfall in Sri Lanka Data Source: Department of Meteorology 24 Appendix 3: Similarities between per capita consumption expenditures and per capita income in poverty analysis To analyze poverty and inequality, per capita consumption expenditure has been used in text. Consumption expenditure is preferred to income since per capita income is not comparable between before and after 1995-96 due to different treatments used for computing income from self-employment, and consumption is less affected by measurement errors than income (see Deaton, 1994). However, the following tables and figures assures that similar trends in inequality since 1995-96 are observed irrespective of the choice between per capita consumption expenditure (pcexp) and income (pcinc). Table A-1: Comparison between per capita consumption and income Per capita income Per Capita consumption Quintile 95-96 2002 95-96 2002 1 703 766 991 1068 2 1215 1381 1445 1596 3 1698 1984 1881 2168 4 2472 2952 2578 3117 5 5966 7809 5274 7325 Total 2411 2978 2434 3055 GINI 0.43 0.46 0.32 0.40 Source: The numbers are computed by the World Bank staff using HIES 95-95 and 2002 Notes: Quintiles are computed based on per capita consumption expenditure Figure A-1: Comparison between per capita income (pcinc) and consumption (pcexp) in Growth Incidence Curve between 1995-96 and 2002 National (95-96 to 2002) National (95-96 to 2002) 10 12 9 11 2002 8 2002 10 to 7 to 9 6 8 95-96 95-96 5 7 pcinc 4 pcexp 6 in 3 in 5 2 4 growth 1 growth 3 0 2 annual annual 1 0 1 10 20 30 40 50 60 70 80 90 100 1 10 20 30 40 50 60 70 80 90 100 % of population ranked by pcinc % of population ranked by pcexp 25 Urban (95-96 to 2002) Urban (95-96 to 2002) 10 12 9 11 2002 8 2002 10 to to 9 7 8 95-96 6 95-96 7 pcinc 5 pcexp 6 in 4 in 5 4 3 growth growth 3 2 2 annual 1 annual 1 0 0 1 10 20 30 40 50 60 70 80 90 100 1 10 20 30 40 50 60 70 80 90 100 % of population ranked by pcinc % of population ranked by pcexp Rural (95-96 to 2002) Rural (95-96 to 2002) 10 12 9 11 2002 8 2002 10 to 7 to 9 6 8 95-96 95-96 5 7 pcinc 4 pcexp 6 in in 5 3 4 2 growth growth 3 1 2 annual 0 annual 1 0 1 10 20 30 40 50 60 70 80 90 100 1 10 20 30 40 50 60 70 80 90 100 % of population ranked by pcinc % of population ranked by pcexp 26