Unformatted text preview: de Janvry and Sadoulet Chapter 4 Poverty and vulnerability analysis
September 7, 2009 Take home messages for chapter 4 1. Measuring poverty is essential to identify the poor, understand who they are, design and target anti-poverty interventions, and monitor and evaluate the performance of these interventions. 2. Any statement about poverty is relative to the choice of a poverty line. Key is to be explicit and consistent about the choice of a poverty line. 3. The Pα indicators are the most commonly used poverty measures, specializing in the headcount ratio, the poverty gap, and severity of poverty indices. 4. Vulnerability to shocks is an important source of new poor, potentially leading to irreversibilities and poverty traps. Reducing transitory poverty requires safety-net policies that differ from the policies used to combat chronic and permanent poverty. 5. Interventions to reduce poverty include pro-poor growth with emphasis on access to assets, improved opportunities and productivity in use of the assets, and presence of safety nets and social assistance programs for the poor. Reducing poverty is likely to be the most widely shared development objective. Hunger with emaciated children shown stunned or dying on our television screens is simply repulsive. Few can remain indifferent. Vulnerability to shocks – a tsunami, a drought, a flood, a pandemic, a civil war – similarly arouses broad compassion of the Rawlsian “it could be me” type (Rawls, 1971). It is also a major source of new poor. Reducing vulnerability and the associated irreversibilities is thus a major instrument for poverty reduction as well. But reducing poverty may have more than intrinsic value if maintaining a share of the population is inefficient, imposing a social cost shared by all. As with all aspects of development, we need to address poverty from both positive and normative analyses: How do we explain poverty? What can be done to reduce poverty? In order to answer these two questions, we need to progress in a logical sequence of nine steps. First, we need to agree on a monetary indicator of well-being (y) that can be used to define poverty. Second, we need to agree on a threshold level (z) for this indicator, called a poverty line, below which the poor will be found. Third, we can characterize the households with y < z, obtaining a description of the poor: Who are they? What do they do? Where are they located? Fourth, we can develop a number of poverty indicators that will help us measure poverty: How extensive? How deep? Fifth, we can look at special aspects of poverty such as its intra-household incidence by gender and age and its intergenerational transmission. Sixth, we can characterize what we mean by vulnerability, distinguishing between those who are chronic poor and those who are transitory poor, calling on different types of policy instruments to deal with each type of poverty. Seventh, we can attempt to explain poverty: Where does it come from? How is it being reproduced over time? Knowledge of causality will allow us to switch from positive to normative analysis. Eight, what is the effectiveness of alternative strategies to reduce poverty? Ninth, in implementing anti-poverty programs that require transfers, how can we target beneficiaries? And finally, how effective is economic growth in reducing poverty? Are there types of growth that are more pro-poor than others? As we shall see, there is considerable expertise in characterizing, measuring, and analyzing poverty. By contrast, we are poorly equipped in explaining poverty, particularly with a pretense at 1 9/7/10 de Janvry and Sadoulet establishing causalities. We are also quite weak at understanding what works and does not work in reducing poverty. While central to development economics, understanding and acting on poverty remain major challenges that limit our ability to act decisively on this major scourge of humanity, one of today’s greatest challenge for all of us. The World Bank and country governments engage in periodic poverty assessments, in particular to help countries design Poverty Reduction Strategy Papers, so-called PRSP (World Bank, 2009). Conducting such assessments requires a number of well defined skills and techniques. This chapter proposes to give the reader the knowledge and instruments needed to conduct such assessments. I. Characterize welfare: choice of an indicator of well-being (y) 1.1. What indicator (y) to use to measure well-being? Income vs consumption There are two monetary indicators we could use to measure an individual’s level of well-being at a particular point of time: income per capita or consumption expenditure per capita. While the most intuitively obvious may be income, it is in fact better to use consumption. Advantages over income are that it is closer to welfare as households engage in consumption smoothing across years. Consumption is also smoothed across the life cycle, exceeding income during youth, inferior to income during working age, and exceeding income again during old age (Figure 1). Individuals will attempt to stabilize consumption when income fluctuates. Temporarily low and high incomes would thus give an erroneous characterization of well-being. Consumption is also relatively easier to measure as there are many sources of income quite difficult to capture, such as the return to assets, transfers across households, and remittances from abroad. There are however particular difficulties to using consumption as a welfare indicator that we must be aware of. They include the following: 1) Consumption expenditures are generally not available for individuals but only for households. This is because the household is the unit of decision-making for purchases and the production of so-called z-goods (home produced goods for consumption, such as the elaboration of dishes out of purchased food ingredients). It is very difficult to observe how consumption is distributed across individual household members. 2) There are large errors in measuring consumption. Think only for yourself of how you would answer questions about how much you spent on food during the last week, on services such as utilities or rents during the last month, and on durable goods during the last year. Households are sometimes asked to record their daily expenditures in diaries (this is done by China’s National Bureau of Statistics), but this becomes quickly tedious to them, and attrition rates are not random: people with high opportunity costs tend to drop out of the sample first, biasing the representativeness of the data. 3) Consumption expenditures vary with tastes. It is consequently difficult to use them to make inter-personal comparisons of well-being, something we need to do in measuring poverty. 4) How to measure the expenditure on durable goods? In principle we would measure the service provided by the durable good over the period of time, namely its depreciation at the current price (plus the opportunity cost of the money tied up in the value of the asset).
2 9/7/10 de Janvry and Sadoulet 5) Finally, households are differentially able to smooth consumption, in particular by wealth level that determine savings and access to loans. This creates measurement biases across households by ability to smooth, with more low and high measurements among those less able to smooth. Figure 1. Income or consumption to measure well-being 1.2. Need adjust y for a number of factors In order to use the information on consumption expenditures as an indicator of wellbeing, we need to make the following adjustments: 1) Changes in prices for comparisons over time: use a deflator to calculate real expenditures, such as the Consumer Price Index. 2) Spatial price differences for country/regional comparisons: use a PPP-adjusted exchange rate across countries; and CPI-adjusted expenditures across regions or for urban/rural differences in cost of living. 3) Many commodities consumed are home-produced. This includes food and zgoods. They need to be valued in monetary terms to be added to purchased commodities, but at what price? In principle it should be at their opportunity cost: the purchase price for commodities that would have been bought had they not been home produced; the sale price for commodities that would have been sold had they not been consumed. In practice, we tend to use the average price for local transactions of that commodity. 4) Need account for the imputed value of public goods and services received (e.g., free or subsidized health care, school lunches, public education). Failing to include these in y can lead to mistakes in assessing changes in welfare. For example, in Ghana poverty fell according to the World Bank Poverty Assessment during the period of structural adjustment; while it rose according to local NGOs. This NGO assessment was due to a decline in access to public goods that had not been measured by the expenditure indicator used in the World Bank assessment. 1.3. Household vs. individual well-being We said that consumption expenditure is measured at the household level. However, a given expenditure for a household of 10 is not the same as for a household of 2. Wellbeing is individual, requiring to measure per capita consumption. How do we do this? 1) Per capita is measured using adult equivalence scales, to take into account differences in demographic composition for comparison across households. This consists in giving consumption weights to household members according to their gender and age that correspond to their relative consumption levels. For example, the OECD gives a weight of 1 to adult males, 0.7 to adult females, and lower weights to children according 3 9/7/10 de Janvry and Sadoulet to their ages with an average of 0.5. The gender and age-adjusted family size in adult equivalence scale is then: n ! = " wk nk , where nk are the number of members in category k
k and wk the consumption weight of demographic category k. The OECD scale to calculate the number of adult equivalents (AE) in a household uses the following formula: Number of AE = 1 + 0.7 (Number of adults – 1) + 0.5 Number of children. A household with two adults and two children would thus have an AE number equal to 2.7. 2) Per capita consumption should also allow for the existence of economies of scale in consumption for household-level public goods (housing, durable goods, heating, electricity) versus private goods (food, child education). In this case, a public good is one that is non-excludable and non-rival for household members. This done as follows: Say that total household consumption is: y = C f + ph Ch , where: C f = consumption expenditure on private goods f such as food, Ch = consumption expenditure on household-public goods h such as housing, ph = price of housing (with food price as numéraire, p f = 1 ). Per capita consumption is then: y pc =
! + ph Ch , where: n!" n* = number of adult equivalent household members, ! = degree of “privateness” of the good: ! = 0 for a pure public good such as food, ! = 1 for a pure private good such as food. Hence, per capita expenditure in the household is y pc = goods, and y pc =
Cf n! C f + ph C h n! if the are only private + ph Ch if Ch is a pure public good. This adjustment can obviously make a big difference to the measure of per capita well-being. 1.4. Specific types of poverty: hunger Poverty takes on specific forms, such as hunger. In this case, y would be measured as daily per capita caloric intake, and z as the daily threshold caloric consumption for that category of individual. Other forms of nutritional deficits can be measured similarly, for instance for proteins or specific types of micronutrients. 1.5. Data availability Measuring poverty is data intensive, particularly if we want to have measurements over time and across many geographical settings. Significant progress has been made in recent years in making publicly available large data bases that can be used to measure poverty. They allow to characterize poverty and test many hypotheses about correlates or determinants of poverty, and about impact on poverty of specific programs and policy reforms. Some of the most important data sources are the following: 1) Living Standards Measurement Surveys (LSMS): they can be cross-sectional, repeated surveys, or panel data. See: http://econ.worldbank.org/WBSITE/EXTERNAL/EXTDEC/EXTRESEARCH/EXTLSM S/0,,contentMDK:21478196~menuPK:3359066~pagePK:64168445~piPK:64168309~the SitePK:3358997,00.html 4 9/7/10 de Janvry and Sadoulet 2) Household income and expenditure surveys: NSS data for India, ENIG data for Mexico, PNAJ data for Brazil, etc. 3) Population census (5-10% release of individual records): they sometimes have data on income (Brazil). More often they give correlates of poverty such as quality of housing and ownership of durable goods (Mexico). 4) Demographic and Health Surveys (DHS): they give data on housing, ownership of assets and durable goods. See: http://www.measuredhs.com/aboutsurveys/dhs/start.cfm II. Separate the poor from the non-poor: Choice of a poverty line (z) Deciding who is poor and non-poor based on knowledge of y for individuals in a particular population requires choosing a threshold level z of y that can be called a “poverty line”. This is quite difficult because there is no agreement on what would be a poverty threshold: Is it absolute? Is it relative to a level of development? Is it relative to the welfare of other households? Clearly z for a poor US citizen cannot be the same as for a poor Zambian. For the first, z should likely allow for a car, while for the second it would allow for a pair of shoes. The important messages here are consequently the following: • Any statement about poverty is relative to the choice of a poverty line. There is consequently no possible absolute statement about poverty. A statement about poverty must explicit relative to which poverty line it is made. • Because there is no agreement on the choice of one poverty line, best is to use several clearly defined alternatives poverty lines. • Important is to maintain consistency in the definition of a poverty line when comparisons are made across time, geography, or subsets of a population. The most commonly used definitions of poverty line are the following four: 2.1. Nutrition-based poverty line In Figure 2, the calorie expenditure function relates calories purchased to a level of food expenditure. There are decreasing returns because consumers spend on increasingly expensive sources of calories as expenditures rises: beans and corn at the lower end, and poultry and meat at the upper end. Figure 2 also shows the Engel curve: it is the level of food expenditure as income (or total expenditure) rises. At very low levels of income, most expenditure is on food. As income rises, the share of expenditure going to food also has decreasing returns. Define: • Extreme poverty line or indigence: zabs = monetary cost of the recommended minimum food caloric intake, such as 2000 ca/day/adult. • Normal poverty line: z = expenditure level necessary to consume the recommended minimum food caloric intake of 2000 ca/day/adult, along with the non-food expenditures normally associated with that level of food expenditures. In poor countries, z will be two times zabs if the food budget share of the poor is 50%. In the United States, z is three times zabs , but under discussion is to raise it to 7 times as the food budget share of the poor is only about 15%. 5 9/7/10 de Janvry and Sadoulet Food expenditures Budget constraint Calorie expenditure function Non-food expenditures Engel curve zabs Food expenditures Calories 2000ca/day/adult zabs z
Income to consume 2000ca y Figure 2. Nutrition-based poverty line 2.2. Poverty line for international comparisons This is the poverty line used by the World Bank in the World Development Indicators. It uses either a $1 or a $2/day measured in PPP dollars. Recently (see chapter 2), these poverty lines were raised to $1.25 and $2.5, respectively, basically because the old $1/day measure is now lower than what any single country uses as poverty line. 2.3. Relative poverty line Poverty is not only an absolute, but also a relative concept. A relative poverty line is commonly used in Europe to make poverty statements. In this case, z is a fixed share k of the mean expenditure y of the population: z = k y . For example, Atkinson (1995) uses k = 0.5 for poverty and k = 0.33 for extreme poverty. Poverty will then be a measure of inequality: it remains constant if all incomes go up by the same percentage, when it would fall with the two previous poverty lines. 2.4. Subjective poverty line Finally, we can say that poverty is not only an absolute or a relative standard, but importantly a perception. In this case, we can ask households “what expenditure do you consider to be absolutely minimal, in that you could not make ends meet with any less?”, and compare it to their actual expenditure level. Those below the stated minimum would be considered in poverty (Figure 3). Like all subjective statements (for instance in contingency valuation where people are asked about a willingness to pay for an object or a service), this measure suffers from a “framing effect”: it very much depends on how the question is asked. While logically appealing, such measures must consequently be used with caution because we have little control over framing. 6 9/7/10 de Janvry and Sadoulet Figure 3. Subjective poverty line III. Describe poverty: Poverty profile and correlates of poverty 3.1. Poverty profile We have a household survey such as an LSMS that gives us per the capita expenditure level yi, i = 1, ..., n, for each of the n members of a population. We can rank these n individuals by increasing level of expenditure, from the poorest to the richest. We also have a poverty line z. If we compare yi to z, we find that q individuals have yi < z, i.e., are in poverty. These data allow us to draw in Figure 4 a very useful representation of poverty in the population: the poverty profile. On the horizontal axis is the population from 1 to n ranked by expenditure level. On the vertical axis are their corresponding yi. The horizontal line is the poverty line at z. This poverty profile can also be represented as a poverty incidence curve by flipping the axes, and counting the percentage of the population on the vertical axis as opposed to the number of people in the population on the horizontal axis of the poverty profile. The poverty line is now vertical at z. The curve represents the cumulative distribution of the population by expenditure level. The poverty incidence curve crosses the poverty line at the headcount ratio q/100: the percentage of the population in poverty.
Cumulative percentage of the population y 100 H(y) Poverty gap z ! yi z yi i q n Households ranked by income level z y Poverty profile Poverty incidence curve Figure 4. Poverty profile and poverty incidence curve 7 9/7/10 de Janvry and Sadoulet There are two useful observations to make in looking at the poverty profile. The first is that the profile tends to be quite flat precisely where it crosses the poverty line because we have a lot of people with similar income levels around z. This implies that small changes in the level of the poverty line will have large implications on the level of poverty q. This stresses again the importance (and sensitivity) of the choice of a poverty line on any particular poverty statement. The second is that there is a lot more information about poverty into a poverty profile than simply the number of people q below z: we can also read the poverty gap z – yi for each poor person. This is an important piece of information: it tells us how much expenditure the individual is short of to be out of poverty. The poverty gap is the amount that would have to be transferred to each of the poor (i = 1, ..., q) to eliminate poverty if we knew exactly their actual yi. The sum of these transfers over the poor is the minimum welfare budget that we would need to have to implement cash transfers that eliminate poverty under perfect information about current levels of y. 3.2. Correlates of poverty: Who are the poor? Where do they live? What do they do? How do they live? With people classified below and above the poverty line, we can also look at the average characteristics of poor versus non-poor individuals or households. These are very useful descriptive statistics to get a diagnostic of poverty and to help target anti-poverty interventions based on these characteristics. The variables that we use are those found in the LSMS or household surveys that we used to draw the poverty profile. They typically give us information on the following aspects of the population: 1) Demographic characteristics: age, gender, ethnicity, family size, dependency ratio (number of children and elders per working-age adult). 2) Asset position: land, livestock, education, social capital. 3) Activities: sector of economic activity, choice of crops, type of employment. 4) Location: rural-urban, region, neighborhood effects (what other households in the same social network do). 5) Access to public services: health facilities, school, social protection programs. 6) Access to market: distance, access to financial services. These are very useful. Looking for instance at Guatemala, we immediately see that the poor are disproportionately rural, indigenous, with large families, low levels of education, and little access to public services. It is interesting, however, to note that most of the poor are not in remote areas: the poor vote with their feet, seeking better options where economic opportunities are present. Most of the poor are in fact next door to the nonpoor. They are poor because they lack access to assets, and in addition may suffer from discrimination in using these assets productively, for instance due to ethnic or gender differences. For Ecuador, we see the following contrasts between the living conditions of the poor and non-poor (Table 1). Most notable is that the poor have larger families, much lower education, work disproportionately in the informal sector, in agriculture, and are self-employed. Their incomes are on average one fourth that of the non-poor. And they have much less access to public goods such as water, electricity, and sanitation. Clearly, the correlates of poverty are quite multidimensional 8 9/7/10 de Janvry and Sadoulet Ecuador, 2000 Poor Non-poor Household characteristics Age of household head 46.4 49.4*** Number of members 5.5*** 3.9 Education of househod head Primary or no education (%) 63.1*** 38.1 Secondary (%) 32.2 36.6* Higher education (%) 4.6 25.3*** Sector of employment Informal sector (%) 80.4*** 55.8 Formal private sector (%) 16.4 31*** Public sector (%) 3.1 13.2*** Sector of activity Primary (agriculture) 56.9*** 26.4 Inustrial 15.3 15.2 Services 27.8 58.4*** Occupational category Self-employed 67.1*** 42.9 Worker 22.5 26.2 Employer 10.4 30.9*** Annual income 137.3 509.4*** Access to public services Water 43.5 70.7*** Electricity 54 85.1*** Sewage 26.3 60.9*** Significantly larger at the *** 1% confidence level, ** 5% level, * 10* level. Table 1. Characteristics of the poor and non-poor in Ecuador If we have two surveys over time, we can analyze the change in the correlates of poverty. This will tell us how the poor are changing, for instance if they are becoming more urbanized or more feminized. IV. Measure poverty: choice of a poverty indicator What we now like to do is to summarize the information contained in the poverty profile in a few scalars that are easy to measure and communicate. A profile is nice to look at, but hard to describe in words. This is why we need indicators. There are two commonly used categories of poverty indicators: the P! class, and the average exit time from poverty for a population. 4.1. Members of the P! class: Incidence, depth, and severity of poverty Foster, Geer, and Thorbecke (1984) proposed a general class of poverty indices,
i the P! class, defined as, P! = ) % ( . It specializes into three indicators according n i =1 $ z ' to the value given to α: 1 q #z "y & ! 1) If α = 0, P0 = . This is the headcount ratio or the incidence of poverty. It measures the percentage of poor in the population. It is the simplest and most commonly used poverty indicator. 2) If α = 1, P1 = where:
9 9/7/10 q n "(z ! y )
i =1 i q nz . This is the poverty gap index or the depth of poverty, de Janvry and Sadoulet
q 1 " ( z ! y i ) is the total expenditure deficit of the poor, or poverty gap. It measures
the cost of eliminating poverty with perfect targeting (i.e., if we knew exactly the expenditure level yi of every poor person), nz is the cost of eliminating poverty without targeting (i.e., if we had no information on the expenditure level of anyone in the population). Hence, P1 is the ratio of the targeted to the untargeted budget needed to eliminate poverty. Alternatively, 100*(1 ! P1 ) is the percentage saving in the poverty budget due to ability to target the poor.
i% 3) If α = 2, P2 = ( " $ ' . This is the severity of poverty index. It weights the n 1# z & poverty gap as a square, giving greater weight to expenditure deficits further away from the poverty line. In that sense, P2 is sensitive to the distribution of expenditure among the poor. A population with a greater share of extreme poor among the poor will have a higher P2, even if it has the same P0 and P1 than another population. As an exercise, we can consider the following policy question: How should a country target a given welfare budget B in order to minimize P0, P1, and P2? For P0, it should start spending on the least poor first until B is exhausted. For P2, it should start spending on the poorest first until B is exhausted. But there is no rule for P1 as a dollar of poverty gap is the same across all poor. Note that the first MDG stresses reducing P0. It can consequently create a policy bias in poverty reduction programs, giving priority to the least poor among the poor, the opposite than what would have happened had the MDG stressed reducing P2. 1 q z!y 2 4.2. Average exit time from poverty Another useful poverty indicator with easy interpretation is the number of years t that it would take to eliminate poverty at a given growth rate in per capita expenditure (Morduch, 1998). Consider the following question: What is the number of years ti needed by a poor woman i with expenditure yi to reach the poverty line z if her expenditure grows at the annual rate g? In discrete time, it would be given by the solution to t yi (1 + g ) = z . Hence, taking logarithms, the exit time out of poverty is:
i ti = ln z ! ln yi . g If, for instance, the poor woman’s current expenditure level is yi = 500 and z = 1,000, and her income grows at the rate of 2% per year, it will take her 35 years to escape poverty. For the whole population of poor, the average exit time is consequently:
Tg = 1 N !t
i =1 N i . This is a simple indicator that, like P2, is sensitive to redistribution among the poor. As opposed to P2, it is measured in meaningful units, namely years. It however requires making a statement about a growth rate g that we may want to avoid. 4.3. Robustness of a poverty profile over a range of z: Poverty comparisons without a poverty line A major inconvenient of poverty statements is that they are conditional on the choice of a poverty line. There may be situations, however, where we can compare 10 9/7/10 de Janvry and Sadoulet poverty across two situations A and B without concern for a poverty line. This would be the case if the poverty profile of A lies above (or the poverty incidence curve lies below) that of B at all levels of y. In this case, we say that there is first-order dominance. If this is the case, irrespective of the poverty line z, P0A > P0B , P1A > P1B , and P2A > P2B . We say that poverty comparisons are “robust” to the choice of a poverty line. We may also have lower and upper limits for the poverty line. In this case, all that is needed to make a poverty statement irrespective of the specific choice of a poverty line is that the poverty profiles do not cross over the range of poverty lines to make a poverty ranking robust to the choice of poverty line. 4.4. Comparing population sub-groups: The relative risk of being poor When comparing poverty across population sub-groups, a useful statement is the relative risk of being poor in each sub-group. If P0u is the urban poverty rate and P0r the rural poverty rate, the relative risk of poverty for the rural vs. the urban population is: (P0u –P0r)/P0u. For Madagascar, where P0u = 0.47 and P0r = 0.77, this would be (0.47 0.77)/0.47 = 0.63. We can then say that a rural inhabitant is 63% more likely to be poor than an urban dweller. 4.5. Decomposition of Pα by population sub-groups We often would like to know how much do different population sub-groups contribute to total poverty measured by a P (or a T) indicator. Let j = 1, ..., k be k exclusive population sub-groups each with a poverty index Pj! . A convenient feature of these indicators is that we can write: k nj is the population share of group j. P! = " m j Pj! , where m j = n j =1 Dividing by Pα and multiplying by 100 to have percentages, we can then calculate the percentage contribution of group j to the total poverty index as 100*
m j Pj! . P! Using the LSMS for 2000 and 2006, Table 2 characterizes changes in poverty in Guatemala for the whole country and across population sub-groups by area, ethnicity, and gender of the household head (World Bank, 2008). Poverty rates are higher among the rural, indigenous, non-capital city, and male-headed households. The poverty rate declined overall and in all groups. However, as the population became increasingly urbanized, the contribution to poverty of non-indigenous and female-headed households increased. We can thus say that poverty in Guatemala is becoming more urbanized, more non-indigenous, and more feminized. 11 9/7/10 de Janvry and Sadoulet Total Guatemala By area Urban 27.1 30 Rural 74.5 70.5 By ethnicity Non-indigenous 41.4 36.2 Indigenous 76.2 75.7 By gender of household head Male 57.6 53.4 Female 47.9 40.8 Source: The World Bank, Guatemala Poverty Assessment, 2008 Headcount ratio 2000 2006 56.2 51.0 Population share 2000 2006 100 100 38.6 61.4 57.4 42.6 85.3 14.7 48.1 51.9 62.4 37.6 81.2 18.8 Contribution to poverty 2000 2006 100 100 18.6 81.4 42.3 57.8 87.4 12.5 28.3 71.7 44.3 55.8 85.0 15.0 Table 2. Poverty patterns in Guatemala, 2000 and 2006 Consider in Table 3 poverty in Buenos Aires before (1980) and after (1989) the debt crisis (Morley, 1995). The headcount ratio increased from 6% to 22%. Decomposing the population in sub-groups be educational levels shows that large increases in the headcount ratio occurred among the middle education levels. Indeed, looking at how the percentage distribution of poverty changed, we see that the contribution of the illiterates declined, while that of the grade school and high school sub-groups increased. For the high school sub-group, this was due to both an increasing population share (as education continued to progress) and a rising headcount ratio. The debt crisis has been widely described a creating a class of “new poor” among the educated, people who lost their jobs in formal sectors of employment, both private industry and government, which previously had low poverty rates.
Headcount ratio 1980 1989 0.34 0.51 0.05 0.27 0.03 0.13 0.01 0.04 0.06 0.22 Population share 1980 1989 7 5 61 57 23 27 9 11 100 100 Contribution to poverty 1980 1989 40 12 51 70 12 16 2 2 100 100 Illiterate Grade school High school University Total Source: S. Morley Table 3. Impact of the debt crisis by educational levels, Buenos Aires V. Dynamics of poverty and vulnerability: Transient and chronic poverty 5.1. Types of poverty Poverty is a dynamic condition and we observe that there is high mobility in and out of poverty: some people fall into poverty, others escape poverty. For this reason, the overall poverty rate may be misleading: we could have ! P0 = 0 over a certain period, and yet many may have entered and left poverty during the period. For that, it is useful to distinguish between three categories of poor in terms of the dynamics of poverty: 1) Transitory poor/Temporarily poor: people who are on average above z, but sometimes in poverty. 2) Chronic poor: People who are on average below the poverty line, but sometimes out of poverty. 3) Persistent poor: People who are always in poverty. Adding the never poor, households can be placed into four categories in Figure 5. In this figure, we see the poverty line z and the average expenditure yi for each category. 12 9/7/10 de Janvry and Sadoulet Expenditure is subject to shocks, with the result that there is considerable mobility in and out of poverty. yi yi
z yi yi
Never poor Transitory poor Chronic poor Figure 5. Never-poor and three categories of poor Persistent poor An indicator of vulnerability to poverty would be based on the ex-ante probability of falling into poverty. We can say that a household is vulnerable if the probability of being poor the next period is greater than an arbitrary threshold, say 50% An example is given by Jalan and Ravallion (2000) for South-West rural China, Over the period 1985-90, the percentage of households that fell into the four categories were: Never poor: 41% Transient poor: 36% Chronic poor: 18% Persistent poor: 5% The headcount ratio over the period would thus have been 21%. However, as many as 59% of the households were at a time in poverty during the period. 5.2. Types of shocks Shocks that affect a household’s expenditure level can consist in shocks to its assets (illness or accident affecting capacity to work, loss of animals, land expropriation, loss of reputation), shocks to the context where it uses its assets (drought, fall in producer prices, unemployment, civil war), shocks to the context where it transforms its income into consumption (rising food prices), and shock to transfers (loss of access to a safety net, economic crisis at the destination of migration affecting remittances). These shocks can be idiosyncratic (illness) or covariate (drought, recession, policy changes, political cycles). Idiosyncratic shocks can be insured locally (e.g., through mutual insurance) but not covariate shocks. Exposure to a shock can create irreversibilities. This is the case when a short run shock transforms a transient poor into a chronic or permanent poor. This happens when the shock leads to asset decapitalization, e.g., the sale of productive assets, taking children out of school, exposing infants to malnutrition, becoming homeless and losing the capacity to work. Irreversibilities can thus create poverty traps (if households cannot 13 9/7/10 de Janvry and Sadoulet escape poverty) or require very long periods to accumulate sufficient assets to escape poverty (Carter and Barrett, 2006). To understand the concept of a poverty trap, we can introduce the concept of an asset (or income) threshold for reversibility (Figure 6). Below this threshold, a household becomes captive to poverty, sometimes referred to as entering a “vicious circle of poverty”.
$ z y
Reversibility threshold Poverty trap Figure 6. Shock and poverty trap 5.3. Policy instruments to combat persistent and chronic poverty: cargo nets and safety nets Reducing persistent and chronic poverty requires raising the average level of expenditure. This requires improving the household’s asset position, the quality of the context where it uses its assets and transforms income into consumption, and/or increasing permanent transfers. This has been referred to as “cargo net” policies that can lift average income (Carter and Barrett, 2006). A key component of cargo net policies is overall economic growth that improves the context where the assets are used, in particular by creating new and better employment and investment opportunities. Reducing transient poverty can be done by raising the average level of expenditure, but also by reducing exposure to risk, managing risk, and improving access to risk coping instruments. This includes accumulating savings (often under the form of liquid assets such as animals) that can be dis-saved when needed, quick access to credit, and access to insurance schemes. This has been referred to as “safety nets” policies that can place a floor on expenditures so households do not fall below the poverty line. The safety net can be on income and expenditures, but also importantly on assets to avoid decapitalization and preserve future income earning capacity. A low cost one-time intervention in response to a shock can have long term benefits with huge cost-saving in terms of future poverty. Here are some examples: Employment guarantee scheme (India) Emergency health coverage (Indonesia) Emergency scholarship program (South Korea) Livestock mortality insurance scheme (Tibet) Rapid inclusion in a conditional cash transfer program (Chile Solidario program) What is important to remember here is that different causes of poverty call on different policy instruments. While a great deal of attention has been given to combating chronic poverty, not enough attention has been given to protecting the vulnerable non- 14 9/7/10 de Janvry and Sadoulet poor to fall into poverty. Low cost, well targeted interventions can have huge long term benefits if they can help the transitory poor from becoming permanent poor. 5.4. Dynamics of poverty: entry and exit With poverty a dynamic state of nature, we can categorize in Table 4 the households in a particular population into four categories in terms of the probability of entry into poverty and the probability of exit from poverty. From a policy standpoint, the critical categories of individuals are those with a high probability of entry into poverty and a low probability of exit from poverty. We can then look at the characteristics of the individuals that belong to each of these four categories to establish how to target the most critical category. This has been done for example for Poland in 1993-96 in the context of the transition from a centrally planned to a market economy. The key variables characterizing individuals are characteristics (age of household head, education, type of household, gender, marital status), location (rural, large town), sector of employment (public, private), socio-economic status (employees, farmers, self-employed, welfare recipients, pensioners), financial assets (savings account), access to transfers (private, public), and access to social benefits (family allowances, unemployment benefits). What we see is that the critically vulnerable category of individuals with high probability of entry into poverty and low probability of exit are more likely to be those with low education, many children, disabled, employed in the public sector, and recipient of welfare transfers.
Poland, 1993-96 Probability of exit from poverty High Low High poverty mobility High poverty Single persistence Married w/out children Low education Has savings account Married with many High Participate to transfer children network Disabled Employed in public sector Welfare recipient Low poverty Low poverty mobility persistence Widowed University d egree Divorced Low Single Indebted Employees Farmers Self- employed Pensioners Table 4. Entry and exit from poverty Probability of entry into poverty 5.5. Other important issues in characterizing poverty 1) Economic mobility: upward and downward mobility. Transition matrices where population is categorized by expenditure quintiles in an initial and a terminal period allow to track economic mobility. If the period of observation is long enough, this allows to measure long term changes in poverty status. Indicators of mobility are for example: 15 9/7/10 de Janvry and Sadoulet The percentage of households that remain in the same expenditure category. The percentage of households that move up or down by one or more quintiles. 2) Inter-generational transmission of poverty: the inheritance of poverty. Children born from poor parents are much more likely to be poor themselves as they will likely receive low education and health attention. This is a key aspect of inequity. Conditional cash transfer programs such as Oportunidades in Mexico, Bolsa Familia in Brazil, and many others across the world pay poor mothers to send their children to school. This is very much motivated by the objective of breaking the intergenerational inheritance of poverty. 3) Intra-household poverty: the role of gender in poverty. Consumption is unequally distributed within a household. Men and earning members of the household are typically given priority, for example in terms of who eats first. As a consequence, women and girls tend to consume less than their share when food is scarce. Measuring intra-household poverty shows that household-level measures of poverty under-estimate the true extent of individual-level poverty, perhaps by as much as 25% (Haddad and Kanbur, 1989). VI. The geography of poverty: constructing poverty maps 6.1. Constructing a poverty map Poverty maps are useful in helping visualize the geographical location of poverty. They can in turn be overlaid with other maps such as agro-ecological quality, distance to major agglomerations, and fiscal expenditures on welfare programs to observe the correspondence of these variables with poverty. The most commonly used methodology used to construct a poverty map consists in combining household survey data with household-level population census data (Elbers, Lanjouw, and Lanjouw, 2003). The approach follows three steps: Step 1: Use household survey data (e.g., LSMS) to explain the per capita expenditure of an individual i with characteristics Xi. The Xi are individual-level correlates of poverty that are available both in the household survey (called intensive data) and in the census data (called extensive data). This includes individual characteristics (such as age, gender, education, professional activities), household characteristics (such as family size, dependency ratio), and variables that characterize the context where the household is located (such as rural/urban, population density, and employment structure). Step 2: Use the population census information on every individual in the country to predict this individual’s expenditure level using the function estimated with the household survey. Average the expenditure predictions over individuals in “small areas” (e.g., census tract, municipality) to reduce the variance. This gives per capita expenditure predictions for population groups (e.g., of no less than 5,000 individuals to have reasonable accuracy). Calculate the desired poverty indicators such the members of the Pα class. Step 3: Map the poverty indicators for each small area using Geographical Information System techniques. 6.2. Using a policy map for poverty analysis: Poverty scope versus poverty density 16 9/7/10 de Janvry and Sadoulet A poverty map for Viet Nam in Figure 7 shows that headcount ratios (left panel) tend to increase with distance from markets. Clearly, poverty is less common in market areas, and poverty headcount ratios appear to rise with distance from markets. Policies targeted at reducing poverty headcounts, also called poverty scope, could be quite expensive, including extensive commitments to transport, communication, health, and educational infrastructure. Now contrast the concept of poverty density, which measures the actual numbers of poor people (Figure 7, right panel), to that of poverty scope (left panel). Although poverty is very common in remote provinces, nationally it is clear that the majority of the poor live in reasonable proximity to urban areas. This evidence would suggest a very different strategy for poverty reduction, one which builds market access incrementally from urban areas, rather than laying out extensive (and expensive) new corridors to remote areas. Figure 7. Mapping the incidence and the density of poverty in Viet Nam VII. Reducing poverty: The growth-poverty relation 17 9/7/10 de Janvry and Sadoulet How effective is growth to reduce poverty? This is measured by the elasticity E of income of the poor with respect to aggregate income, or of income of the poor relative to income of the rich. If E is larger than one, growth is “pro-poor” in the UNDP sense of benefiting the poor more than the average population or the rich. 7.1. “Growth is good”: E = 1 Dollar and Kray (2001) found that E tends to be equal to one. This means that the percentage changes in income of the poor (y) and of the rich (Y) are equal, namely:
" !y % " !Y % =$ ' . $' # y & Poor # Y & Rich They use this result to claim that “growth is good” for the poor, with the policy implication that accelerating growth is good for poverty reduction. But, if y = 10 and Y = 100, and " !y % , then !y = 1 and !Y = 10 . Of the aggregate " !Y %
=$ = 10% ' $ y' # & Poor # Y & Rich income gain of 11 points, 9% goes to the poor (1 point) and 91% to the rich (10 points). Is growth really “good” for the poor? Perhaps if growth is very high (as in China), but clearly absolute income disparities are rising, fueling discontent. Rising disparities in the context of high growth is the story of booming middle-income countries. Political acceptance of rising disparities has been thin both in India (where promise of rural poverty reduction was key to the electoral wins of the Congress party) and in China (where mass protect movements have induced reduced taxation and increased subsidies for farm households). Where growth is low, pro-poor growth with E > 1 becomes all the more important to achieve poverty reduction. 7.2. E depends on the quality of growth Can we do better than E = 1? It depends on the “quality” of growth. Squire and Walton in the WDR 1990 (World Bank, 1990) suggested that poverty reduction requires laborintensive growth complemented by targeted transfers and safety nets. Datt and Ravallion (1998) find that in India growth of agriculture is the most poverty reducing among economic sectors. This is not surprising given the fact that most of the poor are rural and most of them depend on agriculture for their livelihoods. Sadoulet and Ligon (2007) find that GDP growth originating in agriculture is 2 to 3 times more effective in raising the income of the households with the 50% lowest expenditure levels compared to GDP growth originating in the rest of the economy. This is shown in Figure 8 where the five poorest expenditure deciles in the population achieve expenditure gains that are larger when GDP growth originates in agriculture than in non-agriculture. 18 9/7/10 de Janvry and Sadoulet Source: Ligon and Sadoulet (2007) Figure 8. GDP growth originating in agriculture is more effective to raise the expenditure of the poor than growth originating in non-agriculture Growth can thus be made to be more pro-poor. Focusing public investment on sectors with larger E is an instrument. This elasticity can be raised by increasing the asset position of the poor, improving the context where they use their assets, and providing them with safety nets so they can take risks. Poverty reduction also requires targeted interventions since not all poor are able to benefit from growth through employment and investment opportunities. Social assistance should thus be an integral component of propoor growth. References Atkinson, Anthony. 1995. Incomes and the Welfare State: Essays on Britain and Europe. Cambridge University Press. Carter, Michael, and Christopher Barrett. 2006. “The Economics of Poverty Traps and Persistent Poverty: An Asset-Based Approach,” Journal of Development Studies 42(2): 178-199. Datt, Gaurav, and Martin Ravallion. 1998. “Farm Productivity and Rural Poverty in India.” Journal of Development Studies 34(4): 62-85. Dollar, David, and Aart Kraay. 2001. “Growth is Good for the Poor”. World Bank Policy Research Working Paper No. 2587. Elbers, Chris, Jean Lanjouw, and Peter Lanjouw. 2003. “Micro-Level Estimation of Poverty and Inequality.” Econometrica 71(1): 355-364. Foster, James, Joel Greer, and Erik Thorbecke. 1984. “A Class of Decomposable Poverty M easures.” Econometrica 52: 761-766. Haddad, Lawrence, and Ravi Kanbur. 1989. “How Serious is the Neglect of IntraHousehold Inequality? (What difference does it make to the measurement and decomposition of inequality and poverty?).” Warwick University, Development Economics Research Centre. Jalan, Jyotsna, and Martin Ravallion. 2000. “Determinants of transient and chronic poverty: Evidence from rural China.” Journal of Development Studies 36(6): 82-99. Morduch, Jonathan. 1998. “Growth, Poverty, and Average Exit Time.” Economics Letters 58: 385-390. 19 9/7/10 de Janvry and Sadoulet Morley, Samuel. 1995. Poverty and Inequality in Latin America: The Impact of Adjustment and Recovery in the 1980s. Baltimore: The Johns Hopkins University Press. Rawls, John. 1971. A Theory of Justice. Cambridge, Mass.: Harvard University Press. Ligon, Ethan, and Elisabeth Sadoulet. 2007. “Estimating the Effects of Aggregate Agricultural Growth on the Distribution of Expenditures.” Background paper for the World Development Report 2008. Washington D.C.: The World Bank. World Bank. 1990. Poverty. World Development Report 1990. The World Bank. World Bank. 2009. Guatemala Poverty Assessment. Washington DC: The World Bank. World Bank. 2009. PovertyNet. http://web.worldbank.org/WBSITE/EXTERNAL/TOPICS/EXTPOVERTY/EXTPA/ 0,,contentMDK:20153855~menuPK:435040~pagePK:148956~piPK:216618~theSite PK:430367,00.html 20 9/7/10 ...
View Full Document
- Fall '10
- Poverty, poverty line, de Janvry, Sadoulet