Poverty and Inequality Changes in Turkey ( )

State Planning Organization of the Republic of Turkey and World Bank Welfare and Social Policy Analytical Work Program Working Paper Number 1: Poverty and Inequality Changes in Turkey (2003-2006) Meltem Aran Oxford University & The World Bank Sırma Demir State Planning Organization, Republic of Turkey Özlem Sarıca Turkish Statistical Institute Hakan Yazıcı State Planning Organization, Republic of Turkey Ankara, March 2010

State Planning Organization of the Republic of Turkey and World Bank Welfare and Social Policy Analytical Work Program Working Paper Number 1: Poverty and Inequality Changes in Turkey (2003-2006) Meltem Aran Oxford University & The World Bank Sýrma Demir State Planning Organization, Republic of Turkey Özlem Sarýca Turkish Statistical Institute Hakan Yazýcý State Planning Organization, Republic of Turkey Ankara, March 2010 State Planning Organization World Bank

Copyright @ 2010 The International Bank for Reconstruction and Development The World Bank 1818 H Street, NW Washington, DC 20433, USA All rights reserved The World Bank enjoys copyright under protocol 2 of the Universal Copyright Convention. This material may nonetheless be copied for research, educational or scholarly purposes only in the member countries of The World Bank. Material in this report is subject to revision.

Poverty and Inequality Changes in Turkey (2003-2006) iii Poverty and Inequality Changes in Turkey (2003-2006) Table of Contents Abstract... v 1. Introduction... 1 2. Literature Review... 1 3. Data and Methodology... 2 3.1 Data... 2 3.2 Methodology... 2 (i) Construction of the consumption (or welfare) aggregate... 2 (ii) Construction of the food and non-food poverty lines... 3 4. Analysis of Changes in Poverty and Inequality in Turkey... 4 4.1 Findings on Overall Changes in Poverty (2003-2006)... 4 4.2 Stochastic Dominance Tests... 6 4.3 Changes in Inequality Measures (2003-2006)... 8 5. Poverty Profiles and Relative Risk of Poverty Over Time...10 5.1 Poverty by Status and Sector of Employment...10 5.2 Poverty by Educational Attainment...11 5.3 Poverty by Age Groups... 11 5.4 Poverty by Household Composition and Gender of Household Head...11 5.5 Changes in the Relative Risk of Poverty in Turkey...12 6. Correlates of Poverty: A Multivariate Analysis...15 6.1 Household Composition Variables...15 6.2 Educational Attainment... 17 6.3 Sector of Employment for Household Head...17 7. Conclusions... 17 Annex-Tables... 19 References... 26 Figures Figure 1: Growth Incidence Curves... 5 Figure 2: Probability Density Functions for Consumption in Turkey (2003-2006)... 7 Figure 3: First Order Stochastic Dominance Test: Poverty Incidence Curves... 7 Figure 4: The Gini Coefficient and Lorenz Curves for Turkey (2003-2006)... 9 Figure 5: Poverty by Age Groups and Gender (2006)...12

Poverty and Inequality Changes in Turkey (2003-2006) v Abstract Poverty in Turkey has declined significantly between 2003 and 2006, as a result of rapid poverty reduction in urban areas. In the same time period, the reduction in poverty in rural areas has been slow or non-existent. As a result, the relative risk of poverty has increased in this time period for those employed in the agricultural sector, living in rural areas and in large households. Inequality in urban areas has decreased as a result of higher growth in the consumption levels of the urban poor compared to richer deciles, while no significant changes to inequality measures have been noted in rural areas. In fact, the consumption levels of the poorest groups in rural Turkey have declined between 2003 and 2006. Child poverty has also been persistent in this time period, with the relative risk of poverty for children (ages 0-19) increasing over time. This paper constitutes part of a collaborative analytical work program between the World Bank and the State Planning Organization. The findings of this paper have been previously presented at the Welfare and Social Policy Conference organized by these institutions in Ankara on October 22, 2008. The findings and statements in this research paper are the responsibility of the authors and do not reflect the official views of their respective institutions. The authors would like to thank Jesko Hentschel, Francisco Ferreira, Erwin H. R. Tiongson and Edmundo Murrugarra for their valuable comments during the conference and in the process of writing this paper.

Poverty and Inequality Changes in Turkey (2003-2006) 1 1. Introduction 1. This paper illustrates the changes in poverty and consumption inequality in Turkey in recent years. The official calculations for poverty in Turkey indicate that there has been a significant decline in poverty levels from 23% in 2003 to 17% in 2006 (using the complete poverty lines for food and non-food spending). By looking at changes in inequality and consumption patterns, the paper also aims to analyze who has benefited most from the recent rapid changes in poverty reduction in Turkey. In many ways, this diagnostic paper updates the results of the Turkey Joint Poverty Assessment report which was carried out using data from 2001, in terms of the profiles of the poor and also provides more detailed analysis on relative changes in welfare. 2. The paper is laid out as follows: A brief literature review is provided in Section 2. The data sources and methodology for constructing the poverty lines and consumption aggregate are discussed in Section 3. In Section 4, we consider the overall changes in poverty and inequality levels in Turkey and provide analysis on the changes in the consumption distribution over time. The distributional changes in consumption levels by urban and rural categories are analyzed in detail in this section. This section also provides sensitivity analyses for poverty measures with changes to the poverty line as well as stochastic dominance tests of the first and second order in order to establish the robustness of poverty measures. Section 5 goes into detailed analysis on the poverty profiles in Turkey by various characteristics of household heads and household members, such as the sector of employment, age groups, educational attainment and demographic composition of the household. Section 6 considers some important correlates of poverty in Turkey using multivariate regression analysis and a simulation that predicts changes in probability of being poor given certain life events and changes to individual and household endowments. Section 7 concludes with main findings. 2. Literature Review 3. In the last decade, several studies have been undertaken in Turkey that tried to explain and quantify the levels of poverty in the country. Most of the studies have approached poverty measurement with consumption and income measures (Dumanli (1996), Erdogan (1996), Dumanli and Bulutay (2000), Alici (2000), Oguzlar (2006)) while some studies follow a human development approach focusing on the social settings and natural endowments (See Akder (1999 and 2000)). Studies dating before 2000 have mainly used income aggregates whereas after 2000, consumption expenditure is heavily employed as a welfare indicator in poverty measurement. 4. In the 2000s, following the 2001 economic crises, recent academic work has focused on the changing forms of poverty in Turkey (Buðra and Keyder, 2003, 2005; Keyder and Ustundag, 2006). Besides, some studies have touched on the determinants of poverty both at rural and urban levels (Atahan 2006, Pamuk 2002). For instance, Buðra and Keyder (2003 and 2006) focus on the changing nature of immigration and how the new immigrants in urban areas are unable to link up with social networks as the ones that came before were able to do. They name this phenomenon the end of gecekondu-nization whereby new arrivals in urban towns are no longer able to locate themselves in newly created slum neighborhoods and have to find rental apartments in isolation. This reduces their ability to access informal safety nets in the cities. 5. Adaman and Keyder (2005) focus on the poor and marginalized people (the disabled, elderly, street vendor children, Roma people and internally displaced people) who are among the most vulnerable and socially excluded. The study s focus is on the poor, but also considers the marginalized (disabled, elderly, street vendor children, Roma people and internally displaced people) who are the most vulnerable and socially excluded. The results drawn in this study are based on many interviews and meetings with other socially excluded groups in slums of the selected cities. High incidence of social exclusion is also associated with unemployment or employment in the informal sector, where the marginalized groups have no social security or health insurance. This study also suggests that informal social networks are loosening and the traditional mechanisms of cooperation and insurance against risks, such as family and other social networks are becoming less relevant for the marginalized poor in urban areas. 6. Studies on poverty in rural Turkey have examined the dimensions, changes in characteristics of poverty

2 Poverty and Inequality Changes in Turkey (2003-2006) in rural areas. Yalman (2006) has noted that the highest risk of poverty is mostly seen in households in which heads are employed in casual or seasonal jobs. He finds that the average age of the populations in the selected southeastern regions is lower than the national average and unemployment among them is rampant. He also finds that women and children are the most vulnerable ones that suffer deeply from poverty. Yalman (2006) has put forwarded a striking feature of the impoverished families in the southeast. Most of the households were absent of the male figure as a breadwinner. Besides some households were lacking a husband or father, some were in a situation where the husband/father was incapable due to poor health, labor conditions. 7. The World Bank s Turkey Rural Finance Study (RFS), based on rural household survey 2004 and on financial data compiled in 2005, demonstrates that the rural financial markets are ill-performing and a very low-proportion of rural households are using the financial services. The study finds that over 70 percent of the rural sector is credit constrained. The dynamics lying behind this poor performance indicate high interest rates, significant price and output variability in the agricultural sector, poor credit supply given uncertainty about potential government interventions and policies, and high transactions (RFS 2006:5). 3. Data and Methodology 3.1 Data 8. The analysis in the paper uses the Household Budget Survey datasets from years 2003-2006 collected by TUIK. The household budget survey is collected annually in Turkey in order to provide estimations of consumption levels in various categories at the household level. The poverty analysis carried out in this paper takes into account poverty and inequality measures that depend on consumption poverty and the HBS survey is selected for the analysis in this paper as it is the only data source that provides accurate estimations of consumption by spending categories. The survey size in 2003 was large enough to provide regional estimations at the NUTS1 level. In the following years, the survey size was reduced to 8,640 households and provided estimations at the national as well as urban and rural levels in Turkey. The survey is collected over the course of 12 months in each year whereby 720 households are visited each month. The sampling strategy uses a 2- stage cluster sampling method whereby the first stage blocks are selected from the sample frame of the 2000 census and at the second stage households are selected randomly within each block. 3.2 Methodology 9. This paper builds on poverty calculation methodology officially used by TUIK and builds the poverty profiles and determinants on the officially calculated consumption and poverty line levels. The paper makes use of the ADEPT poverty analysis software program developed by the World Bank. 1 10. The variables used in the analysis were defined in a standard way across time (2003-2006) in order to have comparable results. 2 More specifically, the variables used in the analysis were the following: (i) at the household level: per adult equivalent consumption aggregate (adjusted by regional price variations, economies of scale in the household), education level of household head, employment status of household head, sector of household head and urban/rural variable; (ii) at the individual level: gender, age, education level, employment status, sector of employment and (iii) at the national level the food and non-food poverty line combined. 11. The methodology used in this paper for constructing the poverty lines and the consumption aggregate to compare to this level is the same methodology that was implemented in 2002 for the Joint Poverty Assessment report between the World Bank and TUIK. This methodology follows closely along the lines of Deaton & Zaidi 2002. Since then TUIK has annually published official poverty statistics using this methodology. (i) Construction of the consumption (or welfare) aggregate: 12. The total monthly spending is calculated from the household expenditures module in the dataset. The 1 This software package allows for the standardized analysis of poverty data across countries. The program has been developed by Martin Ravallion, Michael Lokshin, Zurab Sajaia in the Development Research Group of the World Bank. For more information on the ADEPT software please visit http://econ.worldbank.org/programs/poverty/adept. 2 The food basket used for constructing the poverty line and the methodology for deflating the consumption aggregate were different in 2002. Therefore the analysis has only taken the years 2003 onward into consideration.

Poverty and Inequality Changes in Turkey (2003-2006) 3 spending on durable goods (as defined by COICOP standard codes) is taken out of this total spending level. Adjustments for Cost-of-Living Differences: The cost of basic needs may vary between areas and over time. In order to have a nationally comparable consumption aggregate over time, the spending data needs to be adjusted for regional and over time price differences. HBS data is collected every year over a 12 month period and an adjustment needs to be made for data collected within a year in order to reflect inflation over time. 3 Adjustments for Household Composition: Households of different size and composition have different needs and two main adjustments need to be made: (i) to reflect the age of the household members (number of children versus adults), and (ii) to reflect household size and economies of scale at the household level. The number of adult equivalents in the household is determined by the formula: AE i = (A + ac) q where A is the number of adults and C is the number of children. The parameter a is the cost of a child relative to that of an adult and lies somewhere between 0 and 1. The other parameter q, which also lies between 0 and 1, controls the extent of economies of scale. In the case of this analysis these parameters were taken at a= 0.9 and q= 0.6. Children are defined as individuals ages 14 or below. In the analysis in Turkey, a modal household of 2 adults and 2 children was taken as the point of departure therefore the formula was adjusted in the following way, where A 0 =2 and C 0 =2: AE ADJ i = A 0 + C 0 (A 0 + ac 0 ) q AEi The consumption aggregate at the household level, adjusted above for price differences over time and location, was then divided by the adjusted adult equivalence measure in order to get the adult equivalence and economies of scale adjusted consumption aggregate at the individual level. This individual level consumption aggregate is then compared to the national poverty line for the determination of poverty status. (ii) Construction of the food and non-food poverty lines 13. The national (or complete) poverty line used in this analysis has a food and non-food components. The food component was determined according to the same basket that was used in the previous Turkey poverty report (JPAR 2005). The food basket was determined as follows: Households were ranked by monthly food expenditures and broken into 10 deciles. The 3rd and 4 th deciles according to food expenditures were taken as the reference group for the poverty food basket and 80 food items they consumed most were taken as the reference items that formed the minimum 2100 calorie requirement per adult. The cost of the food basket was taken as the food poverty line in Turkey. The basket that was determined in 2003 was used in the analysis for consequent years. The cost of the basket was revised according to the average prices per item coming from consequent years of HBS data. The non-food component of the basket was determined by looking at the non-food share of expenditures in household spending just above the food poverty line. This Engel s coefficient was allowed to vary over time and the food component was between 58% and 61% between 2003 and 2005. 4 14. The national (complete) poverty line was calculated by adding the food and non-food components calculated above. The poverty line was calculated on a daily basis and then multiplied by 30 days before being compared 3 In 2002 (initial analysis for JPAR) an index was built reflecting price differences for 7 geographical regions, 2 urban /rural dummies and 12 months resulting in an index that took on 168 different values (7x2x12=168). In the consequent years starting in 2003, an index was built using 12 NUTS1 regions, 2 urban/rural dummies and 12 months. This has resulted in an index that takes on 288 values (12x2x12=288). In order to make sure results were fully comparable over the years, this paper has focused only on the micro data from 2003 onward. Note that in 2003 the food basket in the poverty line was also changed, therefore dictating that the over the years comparable analysis start in 2003. The monthly adjustments were made using the CPI food price index base year (1994 for 2003-2004 and base year 2003 for 2005-2006 analysis.) 4 An alternative method would keep the Engel s coefficient constant over time. The paper has followed the methodology followed by TUIK for those years and hence has allowed the Engel s coefficient to vary in the poverty line. For future analysis the coefficient will be kept constant.

4 Poverty and Inequality Changes in Turkey (2003-2006) to the consumption aggregate which was also adjusted for 30 day spending. Both the 30 day poverty line and the 30 day adult equivalent consumption aggregate were deflated up from June prices of each year to June 2006 overall TUFE prices in order to make all years of poverty lines and consumption aggregates comparable over the years in real terms. 4. Analysis of Changes in Poverty and Inequality in Turkey 4.1 Findings on Overall Changes in Poverty (2003-2006) 15. Turkey has experienced a rapid decline in consumption poverty between 2003 and 2006. The poverty headcount, which provides the share of the population that has consumption below the national poverty line, fell from 28.1% of the population in 2003 to 18.3 % in 2006. This represents a 9.8 percentage point reduction in poverty in this 3 year period. Poverty has also declined in terms of the poverty gap and poverty severity measures. 5 The poverty gap (P1), often considered as representing the depth of poverty, is the mean distance separating the poor population from the poverty line (where the average is taken over the entire population, with the gap set to zero for incomes above the poverty line), and has come down from 7.3 to 5.4 from 2003 to 2006. Poverty severity (P2) measure which squares the distance of each poor individual to the poverty line and gives more emphasis to the welfare of the poorest in the distribution has also declined from 2.8 in 2003 to 1.8 in 2006. 6 These measures suggest that poverty has declined in Turkey in this time period, for those households who were right below the poverty line although it is difficult to say what happens to the poor in lower percentiles of the distribution. The poverty numbers for the period 2003-2006 are summarized in Table 1. 7 16. By all measures of poverty, the reduction in poverty was sharper in urban areas and quite weak in rural areas. Urban areas experienced a 12.9 percentage points reduction in the poverty headcount, from 22.3% in 2003 to 9.4% in 2006. The poverty gap and poverty severity indices also came down more strongly in urban areas than in rural areas. Rural poverty was reduced at a slower rate: with the poverty headcount coming down only by 4.1 percentage points from 37.1% in 2003 to 33.1% in 2006. The poverty gap and poverty severity measures in rural areas also remained high with the poverty severity measure declining only from 4.1 to 3.9 in rural areas, while this measure declined from 1.9 to 0.6 for urban areas in the same time period. 17. Sensitivity analysis for different values of the poverty line shows that the poverty level is sensitive to the poverty line in Turkey. Table A-1 shows that: a 5% increase in the poverty line increases the Table 1: Overall Changes in Poverty Note: Changes shown between 2003 and 2006. 5 The poverty gap (P1) is a measure of the poverty deficit of the entire population in which the notion of poverty deficit captures the resources that would be needed to lift all the poor out of poverty through perfectly targeted cash transfers. 6 Squared Poverty Gap (P2) is described as a measure of the severity of poverty. When using the squared poverty gap, the poverty gap is weighted by itself, so as to give more weight to the very poor. In other words, the squared poverty gap takes into account the inequality among the poor. 7 Please note that while these numbers follow closely the official poverty figures of Turkey published annually by TUIK, there is a slight difference between these numbers and the official figures. This is due to a definitional issue. The numbers in Table 1 consider all people with adult equivalent consumption aggregate below the national poverty line as poor. The TUIK definition also looks at the income levels of these households and redefines those people in households in the top quintile in terms of per capita income, as non-poor. This may be a legitimate redefinition that takes out richer households who have consumed less in a given month perhaps as a result of being away from the house for an extended part of the month. In this paper, in order to carry out further analysis on the decomposition of poverty changes, however, we stay with the original definition.

Poverty and Inequality Changes in Turkey (2003-2006) 5 headcount rate by 2 percentage points in 2006 (a change of around 12%). A 10% increase in the poverty line increases the poverty headcount by 4.5 percentage points (a change of 25%). This suggests that the nearpoor in Turkey are highly vulnerable to changes in price level that may impact the poverty line. The recent increase in the food prices may therefore significantly hurt he near-poor and push them below the poverty line assuming no substitution effects in their food consumption bundle (especially in urban areas where there are no net benefits to higher food prices). The sensitivity analysis also shows that vulnerability to falling below the poverty line may be reduced just as a result of the changes in the consumption distribution in recent years. Whereas in 2003, a 20% change in the poverty line implied a 12 percentage point increase in poverty (about 8.4 million more poor people) in 2006 it implies a 9.5 percentage point increase in poverty (about 6.8 million more poor people). 8 real consumption levels. The mean growth rate of real consumption in rural areas is only around 1%, with the poorest quintile of the population in rural areas having experienced negative real growth in Figure 1: Growth Incidence Curves 18. Urban poor have benefited more from growth and redistribution of consumption in the time period analyzed. Following Ravallion and Chen (2002), the growth incidence curves in Figure 1 show the changes in the consumption aggregate between the two selected years for the percentiles of the consump-tion aggregate. Three panels on the figure show the growth-incidence curves for the country as a whole and separately for urban and rural samples. The steep negative slope of the growth incidence curve in urban areas for Figure 1 shows that as a percentage of their initial consumption level, the urban poor has seen a higher percentage increase in their consumption between 2003 and 2006. In urban areas, the percentage change in the welfare indicator for the poorest quintile is between 7-11% in real terms while for the richest quintile it varies between 0-3 %. In urban areas, the mean growth rate (expressed by the orange line going across Figure 1) is around 9% in real terms (total from 2003 to 2006). 9 19. In contrast, the poorest percentiles in rural areas have experienced negative growth in their 8 Note: The calculation assumes that the population of Turkey was around 70 million people in 2003 and 72 million people in 2006. We can revise this calculation to get more accurate numbers once we have the finalized population figures from TUIK. 9 In this analysis all prices have been adjusted to the June 2006 prices therefore the percentage change in consumption reflects the real change net of inflation.

6 Poverty and Inequality Changes in Turkey (2003-2006) consumption in the time period analyzed. Similarly, the richest 10% of the rural population usually identified with larger landholdings, have also experienced a decline in real consumption levels. Only the middle-classes in rural areas have benefited slightly from growth in this time period. The changes in rural areas that affect the richest consumption decile, may be due to the shift in agricultural policies away from input and price based subsidies for agricultural products, that tended to favor large landowners in the past. The negative real growth in the consumption of the poorest portions of rural society, is a serious concern, the reasons for which should be considered in further detail in the rest of the report. 20. The decomposition of the changes in poverty levels allows us to differentiate between the growth and redistribution components of poverty reduction. Lower poverty could result either from a general increase in the income of all households (without change in the income distribution) or from a decrease in inequality (redistribution from the rich to the poor without change in mean income or consumption). Following Datt and Ravallion (1992) and Ravallion (1994), Table A-2 shows the poverty rates in two years, actual change in poverty rate (percentage points), changes in poverty due to the growth in the mean, changes in poverty due to changes in the distribution, and the interaction components. 21. Growth has played a dominant role in overall poverty reduction, with the redistribution 10 component of poverty reduction also being strong, particularly in urban areas. The drivers of poverty reduction in urban and rural areas in Turkey for the time period analyzed can be summarized as follows: Overall, growth has played a dominant role in poverty reduction, with 7.03 percentage points in poverty reduction in this time period are attributable to growth. Redistribution accounts for 3.82 percentage points in overall poverty reduction, while the interaction term increases poverty by about 1 percentage point. This is not surprising to see in that this time period represents a period of high growth around 8.5% annual increase in GDP. 11 In rural areas, growth has been weaker and the 4.05 percentage point reduction in rural poverty is attributable mostly to a redistribution effect. This is consistent with the fact that the annual increase in GDP for this time period in agriculture has been low at 3.5%. In urban areas, growth and redistribution both account for poverty reduction with 7.9 percentage point reduction in urban poverty being attributable to growth and 5.9 percentage point reduction being attributable to redistribution of consumption. 22. The probability density function for consumption provides the distribution of consumption levels across the population (See Figure 2). The distribution of consumption visibly shifts to the right in the time period analyzed, implying higher levels of real consumption, especially in urban areas. In 2006, the shape of the distribution curve is noticeably wider than the curve in 2003 particularly for urban areas, creating a stronger middle section in the consumption distribution chart. The mean level of consumption in rural areas is closer to that in urban areas in 2003, while rural areas fall behind and the gap between mean levels of expenditure widens between urban and rural areas by 2006. 4.2 Stochastic Dominance Tests 23. Stochastic dominance tests have been applied for the consumption aggregate between 2003 and 2006 in order to establish the robustness of ordinal poverty rankings for the two years. First order stochastic dominance tests are presented in Figure 3 in the form of poverty incidence curves for 2003 and 2006. Figure 3 depicts the distribution functions for 2003 and 2006 (which is in essence the integral of the area under the probability density functions provided in Figure 2). First-order stochastic dominance involves comparing the cumulative distribution functions for the indicator of well-being (in this case adult equivalent consumption) for different survey years. The x-axis in these charts presented below is the consumption aggregate figures and the y-axis is the percentage of the population that has access to a particular level of per capita consumption (for each consumption aggregate level in the x-axis). Whenever a growth incidence curve is entirely to the right of the y-axis (in positive territory), there has been positive 10 Note that redistribution is used here as a term that defines changes in the Lorenz Curve and inequality rather than any specific programs that target the poor and redistribute wealth. The redistribution component can be positive even in the absence of such programs, as long as the consumption of the poorer percentiles increases faster than the consumption of the richer portions of the population. 11 Source: Turkish Statistical Institute (TUIK) Gross domestic product in constant prices (By kind of economic activity at basic prices at 1998, 1998-2008) taken from www.tuik.gov.tr

Poverty and Inequality Changes in Turkey (2003-2006) 7 Figure 2: Probability Density Functions for Consumption in Turkey (2003-2006) Figure 3: First Order Stochastic Dominance Test: Poverty Incidence Curves

8 Poverty and Inequality Changes in Turkey (2003-2006) growth everywhere along the distribution. 12 This implies first order stochastic dominance, which in turn implies that poverty comparisons are robust with respect to the choice of poverty line. 24. In urban areas the poverty incidence curve in 2006 first-order stochastically dominates the curve for 2003. In other words, for all percentiles in the distribution there is a higher level of per capita consumption in 2006 than in 2003 13 since the lines do not cross each other. This means that no matter where the poverty line was drawn, the FGT poverty measured (P0, P1 and P2) would be lower for 2006 than in 2003. The comparison between the poverty measures we obtain, do not depend on where the poverty line is drawn in these years. In rural areas, on the other hand, the poverty incidence curves cross each other, indicating that first order stochastic dominance condition does not hold for rural areas. Depending on where the poverty line was drawn, we could get an increase or decrease in the poverty headcount measure in rural areas. 25. Second order stochastic dominance tests involve analyzing deficit curves or integrals of the consumption distribution functions. This analysis similarly helps determine whether poverty has improved or worsened over time for all poverty measures P1, P2 and higher. Figure A-1 provides the poverty deficit curve for Turkey in 2003 and 2006 for urban and rural areas. In the graph, we can see that for urban areas the 2006 second order stochastically dominates 2003 (an already established fact in the poverty incidence curves as poverty incidence curves speak to all poverty measures P0 and higher). For rural areas, it is difficult to say if there is second order stochastic dominance since the poverty deficit curves lie on top of each other for these two years. 26. Considering the poverty incidence and poverty severity figures, therefore, it is possible to say that changes in poverty in Turkey between 2003 and 2006 are only stochastically dominant for urban areas for the poverty headcount, poverty gap and poverty severity measures (P0 measure and higher). The reduction in poverty on these poverty indicators does not depend on the choice of where the poverty line is drawn for each year. In contrast, for rural areas there is no evidence of stochastic dominance for any of the poverty measures P0, P1 or P2. The reduction in poverty between 2003 and 2006 for rural areas depends on where the poverty line is drawn and we could have seen an increase in the poverty levels had the poverty line been drawn elsewhere on the distribution. This observation is also consistent with the way that growth incidence curves in Figure 1 for rural areas cross the x axis, suggesting that some people are in fact worse off in rural areas in 2006 than they were in 2003. Drawing the poverty line in the middle of the first decile in that figure would have resulted in an increase rather than decrease in rural poverty. 4.3 Changes in Inequality Measures (2003-2006) 27. Inequality in terms of consumption has also come down in Turkey in urban areas, while it has stayed more stable in rural areas in the period 2003-2006. The level of overall inequality in consumption as measured by the Gini coefficient has been reduced from 34.34 in 2003 to 30.95 in 2006. In urban areas, the Gini coefficient has come down from 33.41 to 28.45 while in rural areas there has been only a slight decrease in the Gini coefficient from 31.59 to 31.19. 28. The reduction in urban inequality is depicted clearly in the Lorenz curves presented in Figure 4. The Lorenz curve, which is a standard tool in inequality research, plots the proportion of income earned by various portions of the population, when members of the population are ordered by their consumption levels (Gastwirth, 1972). The 45 degree line on the Lorenz curve represents the line of total equality and the further away is the curve from the diagonal, the higher is inequality. In 2006, the distribution of the consumption aggregate is closer to the diagonal line than in 2003 in urban areas. The inward shift in the Lorenz curve for urban areas reveals that the bottom percentiles of the population has access to more of the total consumption in 2006 than they did in 2003. 12 The anonymity axiom is maintained here such that some individuals in the distribution may have been worse off but overall each percentile is better off when compared to the previous time period. 13 Note that this does not mean that all individuals are better off in 2006, only that each percentile has a higher level of consumption, although the percentiles do not have to include the same individuals.

Poverty and Inequality Changes in Turkey (2003-2006) 9 Figure 4: The Gini Coefficient and Lorenz Curves for Turkey (2003-2006) 29. The changes in inequality levels by urban and rural areas are also confirmed by comparisons of consumption levels by percentile across time. 14 Figure 4 shows the relative comparison of per capita adult equivalent consumption levels by percentiles in detail, while providing the Gini coefficients by urban rural areas over this time period. In urban areas, while the per capita adult equivalent adjusted consumption level for the 90 th percentile (in the richest decile) was 4.24 times that of the 10 th percentile, the ratio has come down to 3.56 in 2006. In rural areas, there is an increase in the ratio of consumption levels between the poorest and richest deciles as well as the consumption level of the 10 th percentile and 25 th percentile suggesting that the poorest part of the population in rural areas has been most disadvantaged in this time period (Table A-3a for details). 30. The gap between urban and rural areas in terms of consumption levels has declined for the top decile in both areas, while it has widened for the poorest decile and the median consumption group. Table A- 3b shows the ratios of selected expenditures percentiles by urban and rural areas and compares their consumption levels. As of 2006, the median level of consumption in urban areas is 1.5 times the median consumption level in rural areas (after regional price and adult equivalence adjustments). This level is up from 1.23 in 2003. In fact for the 5 different percentile ranges in urban and rural areas, the relative consumption of urban households, when compared to rural households has increased. This is especially significant for the comparison of the bottom 10 th percentile, where the consumption ratio has increased from 1.24 in 2003 to 1.64 in 2006. 31. The decomposition of inequality measures shows that between group inequality has increased in Turkey (between urban and rural areas) but within group inequality still makes up the majority of inequality in the country. GE class inequality measures and their decomposition by urban/rural areas in Turkey 14 Source: Blackorby et al. (1981), Shorrocks (1984).

10 Poverty and Inequality Changes in Turkey (2003-2006) are provided in Table A-4. In Turkey in this time period, the inequality that results from across group differences has increased from 4.1% of total inequality in 2003 to 10.5% of total inequality in 2006. However, as of 2006 within urban and within rural inequality still account for the rest (89.5%) of inequality in Turkey. 5. Poverty Profiles and Relative Risk of Poverty Over Time 32. This section of the paper focuses on the profiles of the poor in Turkey and considers the changes in poverty profiles over time. The profiles of poverty considered in this section of the paper include poverty levels and the distribution of the poor across status and sector of employment, levels of educational attainment, age groups and the demographic composition of household. The tables for each characteristic at the individual and household level are provided at the annex of the report (in Table A-5 - Table A-9) in terms of (i) the percentage of the group who are poor (P0 headcount) in the given population, (ii) the share of total poor who are in the given group (the distribution of the poor) and the (iii) the share of the total population in the given group (the distribution of the population). 5.1 Poverty by Status and Sector of Employment 33. The poverty headcount rate is highest in Turkey among people employed in the agricultural sector. While only 12.3% of the population is employed in this sector, 27% of the poor are employed in this sector. 35.3% of those in the agricultural sector have consumption levels below the poverty line in 2006, compared to 18.3% for the whole population. Table A-5a provides the poverty rates by sectors of employment and work status as well as the distribution of the poor in those categories. Between 2003 and 2006, the percentage of the people working in the agricultural sector who are poor has declined from 40.3% to 35.3% following the general decline in poverty. However, the decline in poverty in the agricultural sector was slower than other sectors and therefore the likelihood of the poor in 2006 to be working in the agricultural sector is higher than it was in 2003. 34. In fact as of 2006 the poverty headcount rate in the agricultural sector is even higher than the poverty rate among the unemployed and inactive people in the country. Among those people whose household heads were unemployed in 2003, 43% were in poverty while the headcount rate for those in the agriculture sector was close to 45% (See Table A-5b). In the years that followed, the likelihood of being poor among those employed in agriculture did not decline much while for those people who were unemployed or inactive, the probability of being in poverty was lower. In 2006, while the headcount rate for people with household heads in the agricultural sector came down only to 41% (from 43% in 2003), among the unemployed and inactive this level was down to 36% (from 43%) and 14% (from 23%) respectively. 35. Following the agriculture sector, the construction and mining sectors have the second and third highest poverty headcount rates. In 2006, the headcount rate among people where the household head was employed in the construction sector was 28.5% and in the mining sector it was 20.5%. 12.1% of the poor live in households where the household head works in the construction sector, while only 7.8% of the population lives in such households. The mining sector employs a very small percentage of the population (with only 0.6% of the population living in households where the household head is employed in the mining sector) though the headcount rate among those employed in this sector is high. 36. Between 2003 and 2006, the largest reduction in poverty has occurred in the commerce/tourism, manufacturing and construction sectors, all of which have experienced output growth of around 10-14% 15 annually compared to 8.5% annual growth in overall GDP. In terms of the distribution of the overall population across these sectors, there has been little change between 2003 and 2006: about 45% of the population lives in households where the household head is employed in one of these three sectors. However, the share of the poor employed in these sectors has declined rapidly in the same period: whereas in 2003, 33% of the poor were employed in these sectors, by 2006 this percentage has declined to 25%. Not surp- 15 Source: TUIK Growth rates of gross domestic product in current prices - By kind of economic activity at basic prices, 1999-2008 (www.tuik.gov.tr/veribilgi.do?tb_id=55&ust_id=16 )

Poverty and Inequality Changes in Turkey (2003-2006) 11 risingly, in the same time period, the headcount rate in these sectors has also come down more rapidly than for other sectors in the economy. (See also relative risk of poverty calculations for these sectors in Table 2). 5.2 Poverty by Educational Attainment 37. Educational attainment is a strong correlate of poverty in Turkey. 84% of the poor in Turkey are either illiterate or have not completed basic education. Poverty headcount rate is highest among the illiterate population and 35% of those who are illiterate and 19% who have not completed basic education are under the poverty line. The poverty headcount rate among the illiterate population is twice the level for the general population of Turkey. The poverty headcount rate declines with higher levels of educational attainment: 6% of those who graduated from senior secondary school and 1% of those who completed higher education experience poverty as of 2006. Table A-6a provides poverty levels and distribution of the poor for different categories of educational attainment. The education level of the household head is also closely correlated with poverty levels for household members. In 2006, the poverty headcount rate was 48.2 % among those living in households where the household head was illiterate, and 21.5% in households where the household head has not completed basic education. Table A-6b provides poverty levels and the distribution of the poor by the levels of educational attainment of household head. 38. In recent years, the largest reduction in poverty (headcount) has occurred for those who have not completed basic education, or have only completed basic education. Those who are illiterate or whose household heads are illiterate have benefited less from the rapid poverty reduction in the country between 2003 and 2006. In the overall distribution of the poor, those who are illiterate have come to represent a higher percentage in 2006 than in 2003: whereas in 2003 the 16.9% of the poor were illiterate, this level has gone up to 22.2% in 2006 while the percentage share of those who have some level of schooling has come down. 5.3 Poverty by Age Groups 39. The age profiles of the poor in Turkey reveal the dramatically high rate of child poverty in the country. As of 2006, one-in-four children under the age of 14 and one-in-five children between the ages of 15-19 lived under the national poverty line. Turkey has a young population and children under 14 make up 29% of the total population in the country. But children make up a disproportionate percentage of the poor: making up 40% of the poor population in the country. While the poverty headcount among children under the age of 14 has come down along with national poverty levels from 2003 to 2006, the distribution of the poor who are children has increased in the same time period. 40. Among the elderly, the poverty headcount is only slightly higher than the national average. Close to 20% of the elderly above the age of 65 are below the poverty line in terms of their consumption. Of the poor in Turkey, the elderly over the age of 65 make up 6.4%. Similar to children, the share of the elderly in the poor population has also increased from 2003 and 2006. Figure 5 provides an age pyramid for Turkey showing the levels of poverty by age group and gender. The analysis is carried out for 2006. Table A-7 provides a similar set of information on poverty headcount rates by age group as well as the distribution of the poor across the ages. 5.4 Poverty by Household Composition and Gender of Household Head 41. In households where the dependency ratio is high, with a large number of children and less number of adults who are able to support the household, the poverty rate is also high. In fact in Turkey, among households where there are three or more children between the ages of 0-6, the poverty headcount rate is around 60%. Table A-8 summarizes poverty headcount rates and the distribution of the poor by household composition type. According to this table, a substantial share of poor people lives in large households: as of 2006, 44% of the poor lived in households where there were 7 or more people (up from 35% of the poor in 2003). While the share of the total population living is such large households has slightly declined in this time period, the share of the poor living in such households has increased. 42. In Turkey, the percentage of the population living in households where there is a female household head is only around 8% as of 2006. The share of the poor living in female headed households is also low at

12 Poverty and Inequality Changes in Turkey (2003-2006) around 9.3% of the poor population. Table A-9 summarizes poverty by the gender of the household head. The percentage of female headed households has increased from 2003-2006 both in the overall population and among the poor. The poverty headcount rate is higher in Turkey among female headed households, with 22.3% of people living in female headed households, being below the poverty line in 2006. Poverty among female headed households has also declined less rapidly than male headed households in the time period analyzed. Figure 5: Poverty by Age Groups and Gender (2006) 5.5 Changes in the Relative Risk of Poverty in Turkey 43. While poverty has been reduced in Turkey in the time period analyzed, there are certain categories in the population whose relative risk of poverty has increased. 44. The calculation of the relative risk of poverty helps one see the changes in the probability of the poor being allocated to a certain group over time. It is calculated by dividing the percentage of poor in a certain category (the headcount ratio), by the ratio of the poor in all the other categories. The formulation for the relative risk of poverty in this calculation is as follows: Probability of the poor in Category A Relative Risk of Poverty = Probability of the being poor being in all other Categories The results of the above calculation are presented in Table 2 for different sub-categories of the population. 45. The relative risk of being poor in Turkey is highest in households mapped to the agricultural sector, among those who are illiterate, among children and in large households. Between 2003 and 2006, the relative risk of poverty has also increased for these most vulnerable groups.

Poverty and Inequality Changes in Turkey (2003-2006) 13 By sector: The relative risk of poverty in the agricultural sector has increased from 1.54 to 2.63 between 2003 and 2006, while it has declined most significantly in households where the household head is in commerce, manufacturing and construction sectors. By education level: The risk of poverty has increased among the least educated: The relative risk of poverty has increased from 1.57 to 2.06 among the illiterate population while for the other groups the risk of poverty has been reduced. For instance in 2003, relative risk of poverty among those who had some schooling but no primary school diploma was higher than those who are illiterate but their relative risk of poverty has declined in recent years coming down from 1.65 in 2003 to 1.55 in 2006. By age group: The relative risk of poverty for children has increased in Turkey in these years for children (in the 0-19 year old group). For the 0-5 age group, the relative risk of poverty has increased from 1.40 to 1.44 and for the 6-14 year old group it has increased even further from 1.40 to 1.59. This means that while child poverty rates have declined with overall poverty in Turkey from 2003 to 2006, children s relative well-being and poverty rates have worsened when compared to the rest of the population. This is an important result that points toward the persistence of child poverty in large households over time. The elderly in Turkey have also experienced an increase in their relative risk of poverty from 0.90 in 2003 to 1.09 in 2006. By household composition: Households where there are more than 7 people or more than 3 children between the ages of 0-6, are more likely to be poor in Turkey. A person living in a large household with more than 7 people was 2.16 times more likely to be poor than the rest of the population in 2003 and this level has increased to 3.55 times in 2006. Similarly, in households where there are many children (at least 3 children under the age of 6) the relative risk of poverty is high in 2003 and has also increased by 2006 to 3.72. 46. This analysis shows that despite reductions in poverty, people in large agricultural households, people who have the lowest levels of education and children (between ages 0-19) face higher risks in 2006 than in 2003. These groups represent a hardening core of poverty in Turkey: given their location and characteristics, these groups may be more difficult to reach with general policies that promote overall growth in the economy and therefore it may be necessary to have more targeted programs in place in order to reach them.