Poverty alleviation in Mexico

Policy, Research, and External Affairs WORKING PAPERS Country Operations Country Department II Latin America and the Caribbean Regional Office The World Bank May 1991 WPS 679 Poverty Alleviation in Mexico Santiago Levy T he main determinants of poverty in Mexico are macroeconomic uncertainity, an urban bias in social and infrastructure spending, atnd institutional arrangements and government policies in rural areas that discriminate against the poor. Benefits to the poor should be administered under a single program that simulta- neously delivers food (through coupons rather than price subsi- dies), preventive health services, and information on hygiene, birth control, and food handling. The lolicy, Research, and Extermal Affairs Complex distributes PRE Working Papers to disseminae thc hfndings of work in progress and to cicourage the exchange of ideas among Batik staff and all othcrs interested in dcvclopment issucs. Thesc papers carry thc names of thc authors, rcelect only their views, and should be used and cited accordingly. The findings, interpretations, and conclusions are the authors' own. 'hey should not be attnbuted to the World Bank, its Board of Directors, itL management, or any of iLs mcmber countries.  Plc,Research, and External Affairs Country Operations WPS 679 This paper is a product of the Country Operations I Division, Country Department 11, Latin America and the Caribbean Regional Ofrice. Copies are available free from the World Bank, 1818 H Street NW, Washington DC 20433. Please contact Margarel Stroude, room 18-155, extension 38831 (94 pages). Among the liindings is this ambitious analysis of increasing investment in rural roads, irrigation, poverty in Mexico: extension serviecs, and the like). Mexico's moderately poor lack some goods * Eliminating urban bias in social and infra- and services that everyone should enjoy, given structure spending Mexico's wealtlh. The extremely poor have so few resources as to be at risk of undemutrition * Bringing private costs of production in and illness. urban areas in line with social costs. At most, 1 peIrcent of ihe population is Policies to alleviate poverny must allow for extremely poor (probably an overestimate), and the fact that the extremely poor are less able to extreme poveity is mostly a rural problem. The bear risk, have higher fertility rates, have higher extremely poor have larger households, more price and income elasticities of demand for children, and the higlhest dependency ratios. food, and may experience more household inequality. The moderately poor, on the other The tlhree main determinants of poverty are hand, can migrate, can benefit from educational urban hias, macroeconomic uncertainty, and opportunities, and can participat; more fully in institutional arrangements and government the labor market. policies in rural arcas that discriminate against the poor. Urban bias in social and infrastructure There is a strong case for direct targeting of spending reduces the rural poor's ability to benefits onlt to the to the extremely poor. Such increase their human capital. Macroeconomic benefits should be administered under a single uncertainty and stop-go cycles depress the program that simultaneously delivers food permanent demand foi unskilled labor and the (through coupons rather than price subsidies), steady stream of social spending. Institutional preventive health services, and education about arrangements and resource allocation policics to hygiene, birth control, and food preparation and increase agricultural output deliver substantial conservation. Food pricing policies should be rents to high-income agricultural producers divorced from poverty considerations. A while depressing retums to land and the demand poverty program for the extremely poor should for unskilled rural labor, the two main assets of direct its efforts at reducing fertility, morbidity, the rural poor. undernutrition, and infant mortality. Development policies to help the poor Intertemporal, incentive, and administrative should focus on: considerations all argue that the govemment can best help the moderately poor indirectly. This * Furthering the process of institutional can be done through policies that increase the reforni of the incentive structure in rural areas. permanent demand for unskilled labor, returns to land, and the poor's access to education and * Changing the way resources are channeled social infrastructure. to rural areals (eliminating price subsidies and I The P'RE Working Paper Scrics disseminaics the findings of work under way in thc Bank's Plolicy, Research, and External AffairsCornplx. An ohjccLivc oflhc scries is to get thcsc findings out quickly, cven if prcsentations arc Icss than filly polished. The findings, interprctations, and conclusions in ilicse papers do not necessarily represent official Rank policy. Produced by the PRE Dissemination Ccntcr  CONTENTS I. Introduction ........................................ p.3 II. The Setting ........................................ p.4 III. Poverty: Concepts and Aieasurement III.1 Concepts ..................................... p.6 III.2 Measurement ..................................... p.10 III.3 Operational Measures of Poverty ......................... p.15 IV. Quantification of Poverty IV.1 Data ....... p.20 IV.2 Socicec'Jnomic Characteristics of Households .............. p.25 IV.3 Estimates of Moderate and Extreme-Poverty ................ p.27 V. Determinants of Poverty V.1 Rural and Agricultural Development ........................ p.32 V.2 Urban Bias ................................................ p.41 V.3 Macroeconomic Policy ...................................... p.43 VI. Policies for Poverty Alleviation VI.1 Needs, Behavior, and Policy .............................. p.45 VI.2 Determinants of Intervention in Poverty Alleviation ...... p.50 VI.3 Objectives in Poverty Alleviation ........................ p.53 VI.4 Policies for the Extremely-Poor .......................... p.55 VI.5 Development Policies for the Poor ........................ p.64 VII. Government Programs for Poverty VII.1 Description of Current Programs ......................... p.72 VII.2 Preliminary Assessment .................................. p.76 VIII. Concluding Remarks VIII.1 Summary of Results ..................................... p.83 VIII.2 Issues for Further Research ............................ p.86 References .........................................p. 89 * This paper was prepared for the World Bank, Latin Am.rican and Caribbean Operations Department. I want to thank Sweder Van Wijnbergen for inviting me to work on this topic, and Hans Binswanger, Ken Chomitz, Santiago Friedmann, Ravi Kanbur, Nora Lustig, Lyn Squire and Paul Streeten for useful conversations. Fidel Jaramillo provided excellent research assistantship.  -3- I. Introduction. Tnis paper is concerned with the problem of poverty in Micico. Its four objectives are to: (i) present evidence, (ii) analyze economic determinants, (iii) discuss policy options, and (iv) assess existing poverty programs. This is an ambitious agenda. In fact, one could argue strongly against attempts to cover so many issues at once. The objectives mentioned present difficult theoretical and empirical challenges, and require detailed and systematic work. On the other hand, something can be gained from a self- contained paper that presents an overview of the problem: it can help identify the key issues, point out areas where immediate action is likely to be beneficial, locate possible errors in current policies, and indicate gaps in our knowledge, so that future research can focus vhere its pay-off is high. This is what I attempt here. As the reader will notice, many hypotheses are introduced, but not tested; most topics are discussed, but none are thoroughly dealt with. The paper is therefore sui generis: neither a polished academic piece nor a policy document with concrete recommendations for action. Hopefully this approach will be useful. I divide the paper into seven sections. Section II giv-es a very brief discussion of recent economic events, as these set the stage for poverty programs in the 1990's. Section III discusses the concept and measurement of poverty. Section IV presents evidence of the extent of poverty in Mexico. It aims at answering the question: who, where and how poor are the poor? Section V asks the question: why are the poor poor? I review issues of rural development, urban bias and macroeconomic policy. Section 'I turns to policy centering on two issues: what should government's objectives with regards to poverty be, and what are the appropriate instruments to use. To answer these questions I review the 'stylized facts' about the behavior of the poor, as well as information and incentive issues that bear on the design of poverty alleviation programs. I then suggest policies for alleviating extreme and moderate poverty. Section VII assesses, in the light of previous findings, current government programs to alleviate poverty and offers some suggestions for improvement. A summary of results and unanswered questions is presented in Section VIII.  -4- II. b,,; Settin. After a period of rapid growth during the oil boom of 1978-1981, the Mexican economy entered a prolonged crisis, triggered by a negative terms of trade shock and increases in world interest rates. The crisis proved deep given tne large fiscal deficit, the over valuation of the exchange rate, and the external debt. Since 1983 the bulk of policy makers' efforts have centered in the areas of short run stabilization and adjustment to the change in the direction of external capital flows. As the adjustment process was une--'.ay, it became clear that short run macroeconomic policies, per se, were insufficient to achieve sustained growth. In consequence, concomitant with the adjustments in macroeconomic policy, further structural changes have been progressively introduced. A program of trade liberalization in the manufacturing sector was initiated in 1985 and accelerated in 1988. Subsidies granted through the pricing policies of public enterprises have been reduced. Regulations on direct foreign investment have been relaxed. Regulations on internal trade have been liberalized, allowing freer entry and exit (e.g., transportation). Private investment has been authorized in areas previously reserved for the government (e.g., petrochemicals). Tax laws have been modified to reduce evasion a:ad increase efficiency. A program of privatization of public enterprises is currently underway; commercial banks nationalized in 1982 are being re-privatized. 1 Finally, a renegotiation of the private external debt has been accomplished These impressive changes are a manifestation of a change in the government's role in the economy. Far from a simple retrenchment caused by a temporary scarcity of funds, the Mexican government is embarked on a radical redefinition of its role in the economy, its responsibilities, and the nature of its interventions. Its role as producer is diminished (with a few exceptions like oil and electricity). Its role as regulator is changing: at the macroeconomic level to set credible and sustainaule policies; at the micro level to promote the efficient operation of markets. At the same time, the government maintains its commitment to improve the welfare of the poor. lVan Wijnbergen (1990, p. 32) argues that the 1989 debt re-negotiation package "..seems sufficient to establish a basis for sustained growth in Mexico."  -5- Three characteristics of the economic environment in the 1990's are important in thinkir- about poverty. One, servicing the foreign debt will continue to be an important constraint on policy, despite the renegotiation of the private component of the debt. This, together with the overriding need to keep the fiscal deficit under control, implies that stringent resource constraints will continue (barring a large and sustained positive terms of trade shock). And while the government's role as producer and investor is reduced, there are large gaps in infrastructure that need attention, particularly so given the cutbacks imposed during the 1980's. An attack on poverty requires resources. Yet, given this environment, poverty programs must only reach the target population, and do so in a cost effective way. Two, structural reforms have enhanced the role of market forces in resource allocation. In this context distortions in relative prices prove counterproductive as the economy produces the wrong bundle of goods, or does so with the wrong techniques: it is essential that prices, including the product wage, reflect opportunity costs. Poverty programs must recognize this. To the extent possible, these programs must avoid, or minimize, subsidies and price controls. Three, the structural reforms implemented so far are lopsided. The environment for industry has been substantially more liberalized than tha f agriculture, both with regards to external and internal regulations. Trade reform has concentrated mainly in manufactures; this sector now operates under a mostly tariff trade regime2 with relatively few restrictions on entry and exit. Its legal and institutional environment is well defined; property rights are identified. The same is not true of agriculture. The external trade regime is more controlled; there are significant restrictions on entry and exit; property rights are at times uncertain. The reform gap is, at present, one of Mexico's key problems. If not attended soon it can deepen disparities in income levels. Reform of agriculture is now essential to redress the balance. Poverty programs must not only aim at providing minimum welfare levels for the poor. They must also be part of a larger strategy that incorporates lagging regions into the rest of the economy. 2At the time of writing most sectors of manufacturing had no prior permits, licenses or QR's. Significant sectors where restrictions remain are automobiles, pharmaceuticals and computers. Production weighted tariff rates average 12.6%, with a standard deviation of 4.3X (World Bank, 1989c, Vol. II, p. 79).  -6- III. Poverty: concePts and Neasurement. IXI.1 CI eRts . A recent study of poverty in Mexico argues that approximately 60S of the population could be classified as poor (Hernandez Laos, 1989a, p. 29). Of this total, between 20 to 25X were estimated to live in extreme poverty, with the remaining 35 to 401 being poor, but not extremely-poor. Given an estimated total population of 81 million in 1990, this study implies that 20.2 million Mexicans were livi.lg in extreme poverty, with an additional 28.4 million living in poverty. A separate study estimated the number of poor people at 21.6 million, but made no distinction between poverty and extreme poverty (World Bank, 1989a, p. i). Another study states that in 1982 211 of all Mexican households were desperately poor (World Bank, 1989b, p.1). And yet another World Bank study states chat 25 million rsople in Mexico are poor, with seven million at the destitute level (World Bank, 1990b, p. 5). Data problems aside, disparities in estimates arise from different definitions of what poverty is, and of different methods of estimating it. It is very important to determine just how many are poor and how many are extremely-poor. Overestimates can make the task of eradicating poverty look almost impossible; they may also imply wastes, as resources that could be used for the extremely-poor are spread across larger groups. UnderistSmates leave some desperate people without help. Equally important, confusing poverty with other phenomena like income inequality leaus to wrong policies. To provide a framework for measurement, and to properly identify the target populations for what could potentially be different policies, one question is unavoidable: what is poverty? Tc answer this question, I begin by observing that poverty and income inequality are two distinct problems. Figure 1 illustrates the point: the vertical axis measures yi, the income of the ith individual3, while the horizontal axis measures the total population of the country (n), ordered by increasing income. In turn, z denutes the 'poverty line', or the income level below which a person is considered to be poor; finally, f(y) denotes the 3The important distinction between individuals and huuseholds is discussed in section IV.1. At this point no harm is done by thinking of either households or individuals as the unit of reference.  Figure 1 Income z I I Y) .. I I .OR ~~~~~~~~II f 2 (yf1(y)II I I I I I I I I .. ........ 2 ,~~~~ I ,, 1 2 n - total q q * Qopulati  -7- distribution of income. Assume that ff1(y)ey - ff2(y)dy. By some measures of inequality (e.g. the Gini coefficient) f2ty) is a more equal distribution of income than f1(y). Yet, by some measures of poverty, e.g. the head-count ratio4, poverty is greater under f2(y). Policies that reduce income inequality may, but need not, reduce poverty. In this paper I focus on poverty. Next, I make a distinction between moderate and extreme-poverty. A working approximation to this distinction is to say that extreme-poverty is at. absolute condition, while moderate-poverty is a relative condition. The extremely-poor are those who cannot secure enough nutrition to function adequately. People that are undernourished are more vulnerable to disease, are at risk of developing anthropometric deficiencies. are at times letharqfL and, in general, are less able to lead a healthy life with sufficient energy to satisfactorily perform tasks in the labor market and/or participate in educational activities. Extreme-poverty in this sense is invariant to time and, within limits, space5. The moderately-poor, on the other hand, cannot avail themselves of what, at the given stage of the country's development, are considered basic needs. However, their situation is fundamentally different to the extent that their health and nutritional status allows them to actively participate in the labor market; to take advantage of educational opportunities; to have mobility; to bear more risk. Their poverty is relative in the sense that they lack some goods and services which, given the national vealth, everybody should enjoy. Lack of primary education can be seen in Mexico in the 1990's as a condition of poverty. This was perhaps not the case 100 years ago. The same is true with lack of access to electricity. The line of moderate-poverty, as opposed to that of extreme-poverty, has a larger subjective component; this is unavoidable. The line of moderate-poverty 4The head-count ratio, H, is defined as the proportion of the total population living in poverty. In figure 1 H - q/n; clearly, H2 > H . 5Note that by this definition extreme-poverty is, in principle, an individual-specific condition, as the nutritional needs of individuals of the same age and sex performing similar tasks may differ. In addition, for a given individual nutritional needs are defined as a bald, not a point; see section III.2 below.  attempts to answer the question: when do people cease to be poor? Clearly, cultural and political issues are inexorably linked here6. From the point of view of policy distinguishing between moderate and extreme-poverty is very important. As argued more fully in 3ection VI, the moderately-poor can best be helped .Jy policies that widen the set of opportunities open to them. But the extremely-poor must first improve their health and nutritional status to be able to fully profit from such policies. The concept of extreme poverty thus identifies a set of individuals who need direct help to be able to fully benefit from general policies designeu to reduce po-rerty; the extremely-poor constitute the target population for special programs. The concept of extreme poverty, in addition, helps to identify what benefits need to be delivered, where, and in what priority (cf. section VI.4). Unfortunately, the distinction between moderate and extreme-poverty cannot be made with great precision. Some ambiguity is vnavoidable since it is impossible to draw a sharp line to separate those who, as a result of their better nutritional status, can 'function adequately', from those who cannot. Section IV.2 shows that demographic, exr diture, and other characteristics of households follow a continuum when households are ranked by per capita household income. Households with similar incomes have similar characteristics, but differences in these characteristics become significant as differences in income get larger. Hence, although low and high income households are distinct, households that are slightly above the line of extreme-poverty (denoted by z) and households that are slightly below are not. Yet, for operational purposes policy makers need a cut-off line. The challenge is to set this line at the point; where it minimizes the probability .f leaving out of directly tar,.jted programs truly destitute people (whicii is achieved by setting a high), while at the same time recognizing that resource constraints and incentive considerations imply that not all individuals can, or should, recieve directly targeted benefits (which is achieved by setting E, low). A line of extreme poverty based on nutritional status is useful because 6Sen (1984a) and Streeten (1989a) provide a useful discussion of this issue. In particular, Sen argues that a distinction needs to be made between the space of 'capabilities' and the space of commodities. roverty can be absolute in the former (what is needed to perform well), while relative in the latter (as the list of commodities required to perform well varies through time and location).  -9- nutritional status is closely associated with mauy of the characteristics of households that require them to receive direct benefits before they can fully participate in the development process. I end this sub-section with two remarks. The first concerns the temporal dimension of poverty. There is a life-cycle and some people are poor while young, but accumulate through life so thet poverty reduces with age. Unfortunately, there is no longitudinal data to trace poverty across time. I assume here, however, that the life-cycle component of poverty, particularly for the extremely-poor, is less important than the permanint componeut. Most of the extremely-poor are born and stay poor throughout their lives7 Second, the discussion has centered on moderate al... extreme-poverty as a general manifestation of lack of resources. This is sometimes labelled primary poverty to distinguish it from secondary poverty. The latter is a condition derived from inefficiency in the use of resources: people that are poor because they misallocate their income in the wrong diet, suifer from alcoholism or, perhaps more significant, have some form of physical (e.g. very old age, crippled) or psychological (e.g. mental retardation) problem that interferes with their productive life. Without negating the importance of this phenomenon, it is most likely the case that its quantitative relevance for Mexico is significantly lower, and that the policies (and institutions) required to alleviate it are also different. In this paper I focus on primary poverty: the set of individuals that, for reasons explored below, are able but have welfare levels below those deemed acceptable. With the right set of policies most of these individuals can increase their productive potential and, therefore, Mexico's national income. Political and ethical considerations aside, reducing primary poverty should not be seen as a burden on the government's budget. Rather, it should be seen as a socially profitable investment. This investment, as any other, will have a gestation period and, to produce results, must be pursued systematically. The challenge is t- wind policies that, given the setting of the Mexican economy in the 1990's, will make the investment p-ofitable. 7Schultz (1981) finds that for Colombia the life-cycle component of poverty is less significant than the permanent component.  -10- I L 2 Mea. Is extreme-poverty mostly a rural phenomenon? Political considerations aside, which regions should be targeted for poverty alleviationi programs? Policy makers need to know the number of people livinz in moderate and extreme-poverty. But they also need to know how poor are the poor, where they are located, what the regional composition of national poverty is, and how much of total poverty is accounted for by the moderately-poor and the extremely-poor. One often finds large numbers of indicators quoted in povert- discussions (life expectancy, literacy rates, child mortality rates, access t_ piped water, number of hospital beds per region, etc.). This information is useful as evidence of poverty, but is less useful for policy. To the extent that these indicators are highly correlated with each other, any one would suffice. To the extent that they are not, policy makers .: -d to know which should guide resource allocation. In addition, many of these indicacors confuse inputs with outputs: we do not really care about the number of hospital beds per region but about health status; longer life expectancy is a result of good nutrition and healthy ives. |tsultiple indicators of poverty have been used in previous studies of poverty in Mexico. In particular the influential study by the Coordinacion General del Plan Nacional de Zonas Deprimidas y Grupos Marginados (henceforth Coplqmzr), computes an 'indice de ma6Slnalizacionl from a list of 19 diffe-ent socioeconomic indicators. For four rea'ons this is a doubtful procedure. One, the list of indicaters is to some extent arbitrary, and mixes issues of lack of infrastructure with evidence of extreme-poverty. Two, the weights on the 19 indicators calculated at the state level are not equal to the regional weights, which in turn are not equal to the municipal weights (see Coplamar, 1985c, pp. 30-56). Three, the sign of the weights varies as the geographical level of aggregation changes8. Fourth, the index of marginalizaUion provides no information about the depth or distribution of poverty, and cannot be used 9 to rank regions to allocate resources in a poverty alleviation program 8The weights for the marginalization index are obtained using the method ef principal components, i.e., the weights are the elements of the eigenvector associated with the dominant root of the square matrix formed from the product of the region/indicator matrix times its transpose; there is no guarantee that this eigenvector is non-negative (and indeed it is not). 9To the extent that Coplamar's indice de marginalidad is used by current government programs for poverty alleviation, there is a risk of mis-targeting the population; I return to this point in section VII.2.  Measuring poverty consists of two tasks. First, a poverty line must be determined. Second, the poverty level of individuals has to be aggregated. This sub section deals with the first issue, sub-section III.3 tackles the second. Data and numerical results are presented in section IV. Two methods can be employed to set the poverty line. In the first one a single indicator is used (e.g. nutritional intake)10. In the second one a list of commodities considered essential is made. Following the previous discussion, I argue that access to adequate sources of nutrition should provide the benchmark for setting the line of extreme-poverty. Hence, I define the extremely-poor as those individuals who are unable to purchase enough nutrients which, given age and sex, allow to maintain health and performance. The required level of nutrient intake is set at 2250 calories per day for an adult. Three comments are relevant in relation to this definition. First, caloric intake is taken as the reference point for nutritional status. The bulk of the evidence shows that protei.n and caloric intake are highly correlated: people that fulfill their caloric needs will most likely also satisfy their protein requirements1l. Second, the definition makes no 10The minimum wage ia sometimes used in Mexico to set the poverty line (e.g. World Bank (1989a)), Lustig (1984)): poor households are those earning less thani -he minimum wage. This poverty line is at times also used by the government to discriminate among households for some of its food subsidy programs (see section VII.2, below). This procedure is probably inappropriate. The real wage is subject to transitory deviations arising from macroeconomic shocks, and may also respond to political considerations. For example, if the real wage declines then, ceteris paribus, the number of poor people falls. Conversely, if for political considerations the minimum wage is increased, poverty increases. 11"Malnutrition is not primarily a problem of an imbalance between calories and proteins. Most surveys have found that if energy intake is adequate, protein needs are also satisfied, and if not, protein is burned up for energy requirements" (Streeten, 1989b, p. &). Also: "There is today relative consensus that the indicator 'intake of calories' is more representative of the whole nutritional oroblem than the quantities of proteins or other nutrients" (Garcia et. al., 1986, p. 33; my own translation, S.L.). This is not to say that there will be no deficiencies of some vitamins and other key nutrients like iron, iodine, and calcium. But these deficiencies may arise from cultural factors that determine the type of diet, and not from lack of resources; people with higher incomes can show these deficiencies too. As mentioned below, a distinction between undernutrition and malnutrition is required.  -12- reference to the composition of the diet; extreme-poverty is measured with respect to a diet based on the preferences of individuals as well as the prices ruling in the areal2. Third, the target level of 2250 calories a day per adult equivalent is taken from WHO/FAO food adequacy standards appropriately modified to the climatic conditions of developing countries13. Satisfaction of this calorie intake avoids undernutrition and permits individuals to, in principle, stay healthy and participate in an active life (including satisfactory performance of tasks in the labor market). Conversely, risks of severe undernutrition with permanent effects (particularly for children under five), above-normal vulnerability to disease, and anthropometric deficiencies appear when calorie intakes are, for a sustained period of time, below this level (Lipton (1983a)). The use of nutritional status as a benchmark for measuring extreme- poverty is not without problems. This is because the same nutritional intake is taken as reference point for all the (adult equivalent) population. This procedure is problematic because even after making corrections for climate and work ronditions, it fails to account for intra and inter-individual variations 14 in nutritional requirements . Some researchers, moreover, claim that the 12This contrasts with 'minimum cost' diets found using linear programming techniques that also satisfy a nutritional objective. This technique was used by Coplarar to measure undernutrition (Coplamar, 1985a, pp. 101-23). But note that the calculated diet may not be desired by households, since it is derived independently of their preferences. Significant econometric evidence shows that issues like palatability, status, and odor matter in food selection, even at very low income levels (see the discussion in section VI.1, below). The approach also contrasts with exogenoulsy given diets either chosen by nutritionists or government agencies that specify target levels of individual foods. Rationality of the extremely-poor is assumed, given their information. 13See the extensive discussion in Lipton (1983a). Many other studies for LDC's also use this reference point (e.g. Greer and Thorbecke (1986a) for Kenya). However, the World Bank (1989b, p. 32) uses a reference point of 2,120 for Mexico; a similar number is used by Lustig (1984) and Cepal (1990)). On the other hand, the National Institute of Nutrition sets the standard at 2,600. 14Srinivasan (1981, p. 17) argues that: "Many of the widely used procedures for assessing nutritional status, by classifying all individuals in a population as malnourished who have intakes below a single average norm for the population as a whole, thereby ignoring intra and inter individual variance in intakes and requirements, will misclassify individuals to varying degrees. This misclassification bias need not cancel out for the population as a whole and there is danger of overestimating the proportion of truly malnourished" (emphasis added, S.L.).  -13- autoregulatory homeostatic nature of the body allows for substantial variability in energy intakel5, and that such variability creates a band of up to 20X around the reference nutrient requirement where intakes can vary. It is only when intake is below the lower bound of the band for a sustained period of time (e.g. below 1,800 calories per day per adult) that undernutrition appears as a permanent condition, with its associated risks. These arguments imply that longitudinal data is required to identify the extremely-poor and, in addition, that corrections for individual variances in nutritional requirements should be made. Unfortunately, the data to carry out these computations in Mexico is at present unavailable. As a result, if 'access to adequate sources of nutrition' is used to measure the line of extreme-poverty, the use of averages is unavoidable and, as with any other procedure, some errors of classification will be made. Estimates of undernutrition in Mexico vary widely, from 20 to 50X of the population (cf. Lustig, 1984, pp. 443-47). Most of these estimates are made comparing the monetary costs of a 'desired' reference intake of nutrients (given, in most cases, by an exogenously determined diet as in the Coplamar study mentioned before) with actual food expenditures by households16. These estimates are: one, subject to the Srinivasan-Sukhatme type of criticisms 15"There is considerable evidence to show that a healthy, active individual engaged in fixed tasks and maintaining near constant body weight enjoys wide flexibility in intake" (Sukhatme, 1988, pp. 374-75). On the other hand, Behrman and Deolalikar (1988, pp. 654-55) argue that this evidence is based on small sample sizes that are not representative. (But Edmundson and Sukhatme (1990) present more evidence to reinforce Sukhatme's thesis.) 16The Coplamar estimates imply that undernutrition in Mexico is almost non-existent for the lowest decile of the income distribution: urban (rural) households consume 92.65 and 95.71X (117.68 and 98.89X) of the recommended daily intake of 2082 calories and 63 grams of protein, respectively (see Coplamar, 1985a, tables A-3.14 and A-3.15). On the other hand, the World Bank (1989b, p. 32, table V.1) presents data to show average intake of calories in rural households in 1979 in the following states as: Chiapas, 1609; Oaxaca 1483; Guerrero 1638; Coahuila and Nuevo Leon 1684; Hidalgo 1703; Veracruz 1746; Yucatan 1755. Unfortunately, it is not clear what the reference point is: if it is average intake at a point in time, or average intake for a sustained period of time. In the latter case the situation is very serious, but need not be in the former.  -14- mentioned above17. Two, do not allow for differences in preferences across households. Given these difficulties, a strong case can be made for measuring undernutrition using anthropometric indicators like height for age and weight for height. Unfortunately, the available evidence of undernutrition is not systematic, although there are signs that it does exist'8. It is difficult to put together these conflicting pieces of evidence and to ascertain the significance of undernutrition in Mexico. But two points can be made: first, a sharper distinction between under and malnutrition is required (cf. Schiff and Valdes (1990)). Undernutrition reflects lack of resources to buy adequate amounts of food; malnutrition reflects improper choice of food given incomplete information on the part of the consumer and other problems of the environment. As section VI argues, the policy implications differ. Second, systematic data on nutritional status in Mexico that differentiates between under and malnutrition is needed. Until such evidence is gathered there is a risk that resources for poverty alleviation be targeted to the wrong population, or wrong policies applied to the right population. Pending such evidence, however, second-best methods to determine the poverty line for the extremely-poor and the moderately-poor have to be used. Before I discuss these methods, I turn to describe the construction of operational poverty indices. 7A related problem is generated by the presence of intra-household inequality (of which more in section VI). Even if sufficient food is available, it may be unequally distributed within the household. Sen puts the matter aptly: "The requirement-intake comparisons also suffer from the serious problem of getting accurate information on the food intake of each individual member of the family. Obviously, food-purchase data are not adequate for this. Family members need to be observed eating, and -more than that- the food partaken would have to be weighed in the process of its journey from the plate to the mouth... So the actual intake figures may well be no more reliable than the alleged 'requirement' figures, and the blind shall lead the blind. All of this is, in fact, quite the wrong way of going about the problem. If nutrition is what we are concerned with, then nutrition is what we must observe. We have to look not at food intakes, but at signs of undernourishment" (1984b, pp. 382-3). 18For example, a World Bank report based on 1988 data (World Bank, 1990a, p. 1, table 1) shows that 14X of all children show below normal weight for age (which can be thought of as a measure of cumulative nutrition), while 15X of all babies show low birth weight. In addition, one-third of women of reproductive age are underweight.  -15- 111.3 Operational Measures of Poverty. Denote by A the monetary line of extreme-poverty: the minimum income required for a household of given age and sex composition, and the environment in which it lives, to purchase sufficient food to avoid undernutritionl9. Denote by z the monetary line of moderate-poverty; this line exceeds the line of extreme-poverty by the cost of the necessities beyond those included in A that society (policy makers? economists?) deems are required so that people are not considered poor. Given a poverty line (z or z), it is necessary to aggregate the level of poverty of individuals. An index of poverty that can serve as a summary statistic about the level of poverty is useful because: one, it can provide quantitative answers to the questions posed at the beginning of sub-section III.2. Two, it can help determine how poverty changes overtime; a poverty program needs to monitor how much progress is being made, and tracking an index of poverty is a simple and practical way of systematically doing such evaluation. Three, it can help to rank regions in the allocation of resources for poverty (as discussed in section VI.4). Of course, a single index may fail to be a sufficient statistic of all dimensions of poverty; unavoidably, relevant information gets lost in the aggregation process. Yet indices, properly used, give useful insights. To provide quantitative assessments of the extent of poverty, and to provide a mechanism by which progress in poverty alleviation can be monitored, this sub-section discusses some properties of poverty indices20. I begin by noting two desirable axioms that an index of poverty should satisfy (Sen (1976)): Monotonicity: given other things, a reduction in the income of a poor household must increase the poverty index. Transfer: given other things, a pure transfer of income from a poor household to any other household that is richer must increase the poverty index. These two axioms are discussed with the help of figure 2, where three distributions are plotted. Note that in all cases the number of people living 19Note that z is made contingent on the environment. This is so because even if relative prices are the same across regions, the same income level may translate into different nutritional status. As I discuss in section VI.4, the relationship between income and nutrition is mediated by other factors. 20There is a large theoretical literature on the measurement of poverty and the construction of poverty indices. A seminal paper is Sen (1976); further contributions are by Takayama (1979), Foster, Greer and Thorbecke (1984), and Atkinson (1987).  Figure 2 ly.l(y) z ... . .IPanel (a) _~~~~ ~~~~ -_ ---------- f (y)~~~~~~~~~~f1y tl n I X Panel (b) z 0 A /!- -I qi  -16- in poverty is the same, so that the three dLstrLbutions have the same head- count ratlo (H - q/n). Yet, it is intuitive that the poverty situations descrlbed by these dlstributions are different. Contrast distributions fl(y) and f2(y) in panel (a): clearly, fl(y) is better in the sense that the poor are, all of them, less poor than in f2(y). Yet, the head-count ratio would not indicate this. Horeover, if the income of any poor household (or a111) increased, but still remained below z, the head-count ratio would remain invariant. Evidently, however, poverty would be less. Panel (b) illustrates a different phenomenon. The poverty level of all the people below z is not the same. Those close to q are almost non-poor, while those close to the origin are the poorest of all. Yet, the head-count ratio provides no information about this. Statements like "x number of Mexicans are poor" or "y X of Mexicans are living in poverty" are therefore only partly useful. The problem with the head-count ratio is that it says nothing about the severity (or depth) of poverty nor about the distributlon of poverty. Put differently, the head-count ratio satisfies neither the monotonicity axiom (severity of poverty) nor the transfer axiom (distributicn of poverty). Foster, Greer and Thorbecke (1984), FGT, have developed a class of poverty indices that incorporate these concerns. Define the poverty gap for the ith individual, gi, as: (1) gi - max [(z - YL), 01 The FGT poverty index, denoted by P(a,z), is: q a (2) P(a,z) - 1/n E (gi/z) for a 2 0. i-l where q is the number of individuals for which gi > 0, i.e., the number of people below the poverty line. The parameter a is interpreted as a measure of societies' aversion to poverty; as it increases greater weight is attached to the poverty gap of the poorest individuals. FGT show that for a > 0 P(.) satisfies the monotonicity axiom, while for a > 1 it satisfies both axioms21. 21In addition, for a > 2 P(.) satisfies a further axlom known as the transfer sensitivity axiom: If a transfer t > 0 of income takes place from a poor individual with income y to a poor individual with income (YL + d), d > 0, then the magnitude of the increase in poverty must be smaller for larger Yi. In other words, this axiom gives more weight to transfers at the lower end of the distribution than at the higher end.  -17- The FGT index has a number of properties that make it useful for policy 22 purposes . In particular, P(.) is additively decomposable, with population shares as weights. Hence, the national poverty index can be decomposed into a series of regional poverty indices which measure the contribution that poverty in each region makes to the national total. In addition, for special values of a various well-known measures of poverty are obtained. Note first from (2) that for a - 0 we have: (3) P(O,z) - q/n - H e (0,1] ,i.e., the head-count ratio. Similarly, for a - 1 we have: q (4) P(l,z) - (1/n.z). E gi, i-1 Now, the total income required to eliminate poverty is given by Egi, or area A in panel (b) of figure 2. This allows us to define the income-gap ratio, I, as: q (5) I - Z gi/q.z ; I e [0,1] i-1 or A/(A + B). It follows that: (6) P(l,z) - H.I - P(O,z).I i.e., P(l,z) is the income-gap ratio normalized by the head-count ratio. As opposed to P(O,z), P(l,z) is sensitive to the severity of poverty: it increases when more people become poor (H goes up), and when on average people become poorer (I goes up). However, neither is sensitive to the distribution of poverty, i.e., they do not satisfy the transfer axiom. However, the index: q 2 (7) P(2,Z) - (1/n).E (gi/z)- i-1 22Atkinson (1987) shows that the FGT class of poverty indices belong to a larger class of poverty indices that satisfy a restricted form of a second- degree stochastic dominance condition. This is a useful condition in making comparisons overtime as to whether poverty has decreased or not when the distribution of poverty changes.  -18- satisfies both axioms. As a result, I use it as the basic index of poverty in the remainder of the paper. Since P(.) is an index, it provides, by itself, little information. However, as mentioned above, P(.) is additively decomposable. Thus, let the total population of the country, n, be divided into m regions, with nj (j - 1,2,....m) individuals in each region. For concreteness, think of m as the number of states in Mexico, so that nj is the population of each state. FGT show that P(2,z) can be re-written as: m (8) P(2,z) - E (nj/n).Pj(2,z), where: 1-1 (9) Pj(2,z) - (1/2)- (gl/z i-l with qj denoting the number of poor individuals in the jth state, and gij the poverty gap of the ith individual in the jth state. Alternatively, let: m (10) P(2,z) -Z Qj where Qj - (nj/n).Pj (2,z) J=1 so that TJ Q/P(2,z) is interpreted as the (X) contribution of the ith state to national poverty. Of course, this decomposition can be repeated a second time: the poverty index for any state, Pj(2,z), can be written as the weighted sum of the poverty indices of the individual municipios (counties) within the state. Since the index is additively decomposable at this second stage, the procedure yields a decomposition of total national poverty into the components accounted for by each state and, within each state, by each municipio. Thus a geographical poverty profile is constructed; this provides key information to identify target regions for poverty programs (see section VI.4 below).  -19- Figure 3 develops an obvious extension of the FGT poverty index to separate the moderately-poor from the extremely-poor. Note that I label 6 the income difference between the two poverty lines; in addition, the poverty gaps for each group are now denoted as: (11) gi - max [(X - Yi), 01 (12) gi - max [(z - Yi), 01 - max [(zy + 6 - Yi), 0] I label P(a,7) the index of extreme-poverty and P(a,z) the index of moderate-poverty. These two indices can then be decomposed across states and municipios along the lines indicated above. It is clear that the increase in the poverty index when raising the poverty line from z to z depends on two factors: first, on the size of 6. Second, on the shape of the distribution function between a and q, which determines how many people are added to the ranks of the poor when raising the poverty line by 6. As figure 3 makes clear, if the distribution function f(y) is relatively (steep) flat, for a small change in 6 the number of people 'n poverty increases substantially (minimally). Of course, the change in the poverty index also depends on a, that captures how the poverty gap of the extremely-poor is weighted vis-a-vis the poverty gap of the moderately-poor.  Figure 3 Income z z~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ I q n  -20- IV. Ouantification of PoErty_P. ILA IV.1.A. The Income-EXDenditure Survey of,1984. The most recently available income-expenditure survey (IES) for Mexico, carried out by the Instituto Nacional de Estadistica, Geografia e Informatica (INEGI), is for 1984 (SPP (1984))23. I make five observations on the data. First, about 5,000 households (the unit of observation) during each trimester of 1984 were surveyed. However, since some questionnaires were improperly filled out, the actual size of the sample for the whole year is 18,958. Second, the IES does not directly distinguish between urban and rural areas. Rather, households are characterized by the population density of the area in which they live, and grouped into two categories: 'high' and 'low' density areas. The former are municipios with at least one of the following characteristics (see INEGI, 1984, p. 30): (i) at least one locality with more than 15,000 inhabitants, (ii) a total of more than 100,000 inhabitants, (iii) be the capital of the state, or (iv) be part of any of the twelve largest metropolitan areas of the country. In this paper I use 'low' ('high') density areas as equivalent to rural (urban) areas24. Third, the sample was designed such that results are representative of the urban and rural regions only at the national level; unfortunately, the sample is not representative at the state level. Table 1 summarizes the number of households surveyed in each state and region; note that there are eleven states where rural households were not surveyed 23Although a similar survey was carried out in 1989. Unfortunately, the results of this latter survey are not yet available. 240ther authors divide the population into urban and rural using the occupation of the household head: rural households are those where the household head is a 'Jornalero rural o peon de campo' (roughly, landless agricultural worker). I find this procedure inadequate because it misses an important category of rural inhabitants, the ejidatarios, who are sometimes classified as 'patron' (employer with one to five employees) or as 'trabajador por cuenta propia' (self-employed worker); see INEGI (1984, p. 144). On the other hand, if self-employed workers or employers with one to five employees are included in the definition of rural, one then adds incorrectly all the urban self-employed as well as small scale urban employers. Thus, the low/high density approximation seems to be better since it focuses directly on a geographical concept.  Table 1 Sample Characteristics State State Name Number of households sampled: Number Urbin Rural 01 Aguascalientes 163 0 02 Baja California Norte 302 0 03 Baja California Sur 152 0 04 Campeche 271 0 05 Couhuila 418 0 06 Colima 157 355 07 Chiapas 278 553 08 Chihuahua 563 355 09 Distrito Federal 1407 0 10 Durango 145 0 11 Guanajuato 447 398 12 Guerrero 313 0 13 Hidalgo 141 180 14 Jaliso 1016 181 15 Mexico 989 502 16 Michoacan 470 180 17 Norelos 226 208 18 Nayarit 141 0 19 Nuevo Leon 961 168 20 Oaxaca 160 186 21 Puebla 410 385 22 Queretaro 286 195 23 Quintana Roo 153 184 24 San Luis Potosi 165 210 25 Sinaloa 315 326 26 Sonora 432 345 27 Tabasco 152 0 28 Tamaulipas 430 545 29 Tlaxcala 157 189 30 Veracruz 653 786 31 Yucatan 159 0 32 Zacatecas 284 221 Total 12306 6652 Total 18958  -21- Fourth, the IES makes explicit allowance for own-consumption. For each of the 14 expenditure categories considered (food, transportation, clothing, housing, medicines, etc.) data was gathered to evaluate the monetary value of consumption from household own production, and from non-monetary payments and gifts. The sum total of monetary expenditures plus the monetary equivalent of own-consumption and gifts was considered as total expenditure. Unfortunately, I had no information to divide the total monetary value of own-consumption into its components. Thus I am able to rank households on the basis of total expenditures, but not on the basis of total food expenditures25. Fifth, as is generally the case with most IES, income seems to be underreported26. This, together with consumption smoothing considerations, propelled me to use total expenditures rather than reported income as the relevant variable (including, as just mentioned, the monetary value of own- consumption). Thus, while for expositional convenience in the paper I refer to the variable y1 as total income, the reader should have in mind that total expenditures is the proxy variable used. Two adjustments were made to the data. The first concerns inflation. 1984 was a period of substantial inflation in Mexico (49.5X according to the CPI), implying that average nominal values of income and expenditure for the beginning of the year were lower than for the end of the year. This generates the need for a correction. Fortunately, the IES included information on the date on which a household was surveyed, allowing me to apply appropriate deflators to express all monetary flows in prices of January of 198427. 25Differently put, if yi and yf is total expenditures and total food expenditures (including own-consumption) of the ith housnhold, respectively, I can obtain f(y) but not f(yf). While clearly f(y) ; f(yy), the difference between the two increases with income given Less than unitary food expenditure elasticities. 26In particular, total expenditure exceeds total income for the lowest 16 out of the 20 household groups (ordered by per capita household income; see below). 27The IES divided the year into 36 intervals of 10 days, and recorded the interval in which a household was surveyed. I divided the Banco the Mexico national monthly consumer price index for 1984 into three components (assuming a linear trend within each ten day period), constructed a price series for the year with 36 observations, and made the base for mid-January 1984 - 1.00. All nominal variables were then converted into prices of mid-January 1984 using the respective date and corresponding price index.  -22- The second adjustment concerns the unit of observation. Data was collected for the household, but information was also available on household size (i.e., number of members in each household). Since household size is not the same across income levels, a measurement based on household income is an inaccurate reflection of individual poverty (cf. Anand, 1983, pp. 63-7). Moreover, since lower income households are larger (see table 3 below), 28 estimates based on household incomes underestimate poverty . To correct for this phenomenon, I rank households on the basis of per capita household income, obtained by dividing the income level of each household by household size, and measure poverty at the level of the individual. One final remark. Although the IES has very detailed coverage of expenditures, with special emphasis on food, it provides no information on asset ownership or ethnic characteristics29. Thus, while the information is very valuable, it is insufficient to make direct connections between income and expenditure patterns, on the one hand, and asset ownership and productive activity, on the other. In particular, it is important to know whether agricultural workers' income levels are correlated with type of crop (cereals vs. other; within cereals corn vs. wheat, etc.), or with type of land tenure (ejido vs. private land). This information is essential to test hypotheses on the relationship between poverty, crop pattern and land te-ure status. IV._.B. The Lines of Moderate and Extreme-Poverty. My departure point for constructing x and z is the Coplamar (1983) study on basic needs, where the annual cost of a basket of necessities for an average family of 4.9 members, made up of 2.7 adults (older than 15 years of age), 1.66 children (between 3 and 14 years of age) and 0.47 babies is calculated. Coplamar followed a three part procedure to construct this 28Consider households A and B with, say, household incomes 2 and 4 (pesos) and household size 1 and 3, respectively. Ranked by household income household A is poorer, but ranked by per capita household income household B is poorer. As table 3 below shows, there are substantial differences in household size across income levels in Mexico, so that reference to the 'average family' may for some purposes be quite misleading. 29A potentially important issue of poverty in Mexico concerns the fact that some poor families belong to indigenous communities; for some of them language and other dimensions may be barriers to income growth.  -23- basket, labelled the 'Canasta Normativa de Satisfactores Esenciales', CNSE. First, it used the expenditure patterns of households in the seventh income decile of the 1977 IES. Second, it added a few goods deemed essential considering "..the rights that the national laws grant to the population, their expectations and the objective necessities that society imposes .." (Coplamar, 1983, p. 133; my translation and emphasis, S.L.). Third, the costs of food and housing were obtained from separate studies also elaborated by Coplamar (1985a, 1985b). In particular, the food component consisted of two parts: (i) the cost of a basket of food that satisfied exogenously given nutritional requirements, and (ii) the cost of additional food items also consumed by households in the seventh decile. The nutritional food basket was labelled the 'Canasta Normativa Alimentaria', CNA, and is composed of 34 food items that satisfy a minimum of 2082 calories and 35.1 grams of protein per day for an adult. In fact, Coplamar constructed fifteen different baskets that satisfied the minimum requirements of calories and proteins, but varied in the number of food items included and the origins of the nutrients (animal vs. vegetal)30. The chosen CNA was not the least cost diet, exceeding the minimum by 36X (Coplamar, 1985a, pp. 102-12). The Coplamar study did not distinguish between moderate and extreme- poverty, although other studies (e.g. Hernandez Laos. 1989a) have taken a subset of the CNSE to construct a Canasta Sub--inima, CSN, to set a line of extreme-poverty (thus interpreting the CNSE as the line of moderate-poverty). on the other hand, the Coplamar study distinguished between an urban and a rural basket, but found insignificant cost differences so that only one basket was used for the whole population (op. cit., p. 146). In this paper I follow a mixed procedure to construct E and z. Consider first the line of extreme-poverty. In principle this line is given by the cost of .the nutritional basket, and the extremely poor are those whose food expenditures are below this cost: these are the individuals who lack 'access to adequate sources of nutrition'. (Differently put, given their preferences and the information at their disposal, the extremely-poor maximize welfare allocating their income across different goods; if the endogenously determined food demanded is less then the nutritional minimum, the individual is 30The linear program reached different optimal points because the number of variables (food items) and the number of contraints (mix of foods) was changed in different solutions.  -24- classified as extremely-poor.) But as mentioned in the previous section, because own-consumption cannot be separated into its components, I can only rank households by total expenditures, Yi, and not by total food expenditures, yfLI This requires that an indirect procedure be followed. Because yi 2 yf1, a comparison of the monetary costs of the nutritional basket with yi would underestimate poverty. As a result, it is necessary to scale-up the cost of the nutritional basket, and I do so by 25131,32. The resulting line of extreme-poverty, y, can then be compared against the distribution of total expenditures f(y) to identify the extremely poor. On the other hand, I simply take the monetary cost of the CNSE as z33. (From my subjective point of view this is somewhat too high. But because I do not propose any specific policy measures based on this line, I do not pursue the point further.) Table 2 summarizes these calculations. One final remark. It is clear that the procedure used in this and other papers to set both z and z involve some arbitrariness. Yet, only if there is general consensus on these poverty lines can measurements be widely accepted and used to monitor progress in poverty alleviation. Further discussion on the values for , and z by policy makers and others concerned with poverty alleviation would be called for. 31This procedure can be rationalized assuming that there is an irreducible minimum of expenditures that must be allocated to non-food items. Streeten (1989b) and Lipton (1988a) present evidence to show that this minimum is around 201, implying a 'scaling factor' of 1.25. It should be clear that this procedure, while plausible, is somewhat arbitrary, since expenditure shares on diffe:.ent goods are endogenous. One can conceive of situations where households do not purchase sufficient food to satisfy a nutritional requirement, but still allocate less then 801 of total expenditures to food. To avoid this arbitrariness, further work needs to separate the components of own-consumption to obtain f(y ), and include only the monetary cost of the nutritional basket in y. 32Ny line of extreme-poverty differs from the one used in other studies. For example, Hernandez Laos also defines households in extreme poverty "..as those households that have such a small income that, even if it was all allocated to food, would not allow 'hem to satisfy their nutritional needs" (1989a, p. 2; my translation, S.L.). But his Canasta Sub-Minima is composed as follows: 551 food, 351 housing, 8.5X health and 1.51 education. 33Since the CNSE was expressed in prices of March of 1982, I use the respective components of the national monthly consumer price index to express the cost of the basket in prices of January of 1984; see table 2.  Table 2 Lines of Moderate and Extreme-Poverty Monetary Cost List of Necessities Moderate Poverty Extreme-Poverty 1. Food 1.1 nutritional basket 41,863 41,863 (Canasta Normativa Alimentaria) 1.2 other food consumed at home 14,073 1.3 food consumed outside home 6,680 1.4 eating & preparation utensils 6,398 2. Housing 2.1 maintenance & depreciation 12,237 2.2 financial amortization 28,342 2.3 water & electricity 4,498 2.4 real estate taxes 1,800 2.5 furniture, blankets and similar 4,576 3. Health 3.1 Medicines 527 3.2 House and personal cleaning 9,108 4. Education 4.1 tuition 791 4.2 school materials 1,287 5. Culture and Entertainment 5.1 books 6,403 5.2 movies, vacation and similar 24,516 5.3 radio, T.V., and similar 3,679 6. Transport and Communication 6.1 transport 10,107 6.2 mail and phone 224 (continued)  (table 2 continued) 7. Clothbna 7.1 clothing 26,020 7.2 shoes 6,270 7.3 belts, bags and similar 355 8. Per6onal Needs 8.1 shaving materials, deodorants and similar 4,840 8.2 items for the house 128 8.3 legal and other services 411 Total in prices of March 1982 215,133 41,863 (25Z expansion factor for extreme-poverty) - 52,328 Total in prices of January 1984 635,512 151,753 Total per trimester (*.25) 158,878 37,938 Total per capita (*1/4.9) - Poverty lines 32.424 7.742 Source: For the line of moderate poverty Coplamar (1983); for the line of extreme-poverty own construction.  -25- IV.2 Socioeconomic Characteristics of Housgholds. I turn to briefly discuss households' demographic (table 3), expenditure (table 4), income (table 5) and occupational characteristics (table 6). At this point I simply describe some 'stylized facts'; section VI links these facts to behavior and policies for the poor. To accomplish this, households are classified by location and income level. In particular, 20 groups of households are constructed in intervals of 5X of the total sample (so that each interval contains 948 households) ordered by per capita household income and, within each group, households are divided into urban and rural. I make five observations on table 3. One, there are sharp differences in household size across income levels, but not along rural-urban lines. Poorer households are significantly larger than the average34. Two, poorer households have both a larger absolute number of children (individuals under 12 years of age) and a larger proportion of children in the household. Three, poorer households have a smaller share of income earners. Differently put, the dependency ratio -the number of people who do not work over total household size- is highest for poorer households; differences in this ratio between the poorest and the richest households are dramatic. Four, average education of the household head increases steadily with income, and within each household group is always higher for urban households35. Five, there appears to be no systematic relationship between the share of households headed by females and either income or location. Turn to allocation of expenditures and sources of income. Table 4 reveals that: one, even for the poorest households the share of monetary 34The sample contained a total of 96,380 individuals, giving an average household size of 5.08. 35The IES reports education for the household head as a discrete variable between 0 and 10 with the following values: 0, no schooling; 1, finished first year of primary education; 2, between two and five years of primary education; 3, finished primary education; 4, incomplete secondary education; 5, finished secondary education; 6, unfinished high-school or vocational education; 7, finished high-school or vocational education; 8, unfinished university education; 9, completed university education; 10, master's or doctoral education. The figures reported in table 3 are obtained by taking simple averages of this value for household heads in each household group.  Table 3 DemoaraDhic Characteristics of Households Proportion Average Average Average Share Share Household Proportion of Female Education Houeehold Number of of Group* of Headed of Size of Children Earners Households Households Household Children Head 1 Urban 0.210 0.160 1.345 7.59 3.49 0.459 0.211 Rural 0.789 0.065 1.081 7.11 3.25 0.457 0.208 2 Urban 0.345 0.100 1.756 7.52 3.26 0.433 0.229 Rural 0.654 0.077 1.229 6.65 2.88 0.432 0.234 3 Urban 0.381 0.104 1.613 7.00 3.04 0.434 0.229 Rural 0.618 0.090 1.549 6.25 2.73 0.437 0.239 4 Urban 0.430 0.125 1.853 6.87 2.71 0.395 0.242 Rural 0.569 0.085 1.418 5.89 2.38 0.405 0.261 5 Urban 0.485 0.152 2.026 6.37 2.42 0.380 0.261 Rural 0.514 0.092 1.539 5.84 2.20 0.376 0.279 6 Urban 0.564 0.142 2.190 6.40 2.46 0.384 0.267 Rural 0.435 0.118 1.479 5.59 2.06 0.368 0.281 7 Urban 0.604 0.165 2.209 5.87 2.06 0.352 0.282 Rural 0.395 0.141 1.776 5.22 1.89 0.362 0.295 8 Urban 0.659 0.147 2.481 5.98 1.99 0.333 0.284 Rural 0.340 0.148 1.783 5.07 1.61 0.318 0.308 9 Urban 0.666 0.120 2.625 5.51 1.84 0.334 0.301 Rural 0.333 0.123 1.974 5.03 1.69 0.336 0.288 10 Urban 0.683 0.120 2.683 5.58 1.83 0.328 0.302 Rural 0.316 0.143 2.000 4.61 1.47 0.319 0.314 11 Urban 0.706 0.140 2.817 5.21 1.62 0.310 0.310 Rural 0.293 0.143 2.287 4.41 1.33 0.302 0.331 (continued)  (table 39 continued) 12 Urban 0.699 0.144 3.126 5.03 1.52 0.303 0.324 Rural 0.300 0.136 2.375 4.49 1.38 0.308 0.314 13 Urban 0.746 0.156 3.124 4.79 1.34 0.279 0.340 Rural 0.253 0.212 2.133 3.86 1.06 0.276 0.360 14 Urban 0.787 0.152 3.293 4.71 1.23 0.262 0.352 Rural 0.212 0.189 2.413 3.99 0.88 0.220 0.369 15 Urban 0.792 0.167 3.584 4.46 1.13 0.253 0.366 Rural 0.207 0.223 2.568 3.60 0.86 0.239 0.405 16 Urban 0.810 0.179 3.865 4.32 1.09 0.251 0.397 Rural 0.189 0.161 2.733 3.42 0.72 0.210 0.440 17 Urban 0.848 0.199 4.107 3.86 0.92 0.238 0.417 Rural 0.151 0.215 2.784 3.41 0.83 0.243 0.426 18 Urban 0.828 0.201 4.592 3.55 0.81 0.227 0.460 Rural 0.172 0.159 3.073 3.14 0.76 0.243 0.444 19 Urban 0.858 0.218 4.926 3.29 0.71 0.217 0.482 Rural 0.141 0.201 3.425 2.71 0.44 0.162 0.534 20 Urban 0.872 0.225 6.006 2.68 0.53 0.199 0.571 Rural 0.128 0.173 4.735 2.47 O.i5 0.224 0.498 Households are grouped in twenty intervals with 5Z of the sample each, and are ranked by increasing household per capita income.  Table 4 Allocation of Monetary Expenditures* Household Food & Clothing Housing Transport & Education Other Group Beverages & Shoes Comuunication I Urban 0.61404 0.05759 0.09506 0.04298 0.03446 0.15584 Rural 0.59666 0.07961 0.07092 0.03731 0.02526 0.19021 2 Urban 0.60266 0.07403 0.08724 0.05445 0.03428 0.14732 Rural 0.61258 0.08463 0.06126 0.04384 0.03044 0.16723 3 Urban 0.59726 0.07278 0.09225 0.06436 0.03812 0.13521 Rural 0.60959 0.08064 0.06420 0.05734 0.03043 0.15776 4 Urban 0.60074 0.06208 0.08977 0.06828 0.04503 0.13408 Rural 0.61080 0.07867 0.06060 0.05957 0.03262 0.15771 5 Urban 0.59481 0.07332 0.08231 0.06776 0.05096 0.13082 Rural 0.58500 0.09581 0.05482 0.06535 0.03735 0.16164 6 Urban 0.58582 0.08108 0.08098 0.07270 0.04767 0.13172 Rural 0.59812 0.09331 0.05295 0.06690 0.03774 0.15096 7 Urban 0.58817 0.06964 0.08489 0.07239 0.04736 0.13752 Rural 0.58669 0.09861 0.05675 0.06787 0.03651 0.15354 8 Urban 0.56718 0.07443 0.08308 0.08314 0.05526 0.13689 Rural 0.57297 0.10126 0.05506 0.07198 0.04119 0.15752 9 Urban 0.55899 0.08052 0.08410 0.08107 0.05622 0.13907 Rural 0.54828 0.09428 0.05130 0.09895 0.04017 0.16699 10 Urban 0.56468 0.08409 0.08123 0.07907 0.05403 0.13688 Rural 0.55209 0.10872 0.05090 0.08173 0.04041 0.16612 11 Urban 0.54973 0.08913 0.07413 0.09044 0.05882 0.13771 Rural 0.54146 0.10080 0.06053 0.08932 0.04015 0.16771 (continued)  (table 4, continued) 12 Urban 0.53467 0.08617 0.07905 0.09?17 0.05871 0.14420 Rural 0.51974 0.09365 0.05317 0.10534 0.04541 0.18267 13 Urban 0.51844 0.08655 0.08225 0.10438 0.06200 0.14635 Rural 0.50851 0.09424 0.04766 0.11714 0.04507 0.18734 14 Urban 0.50327 0.08745 0.07579 0.12288 0.06608 0.14450 Rural 0.50566 0.09567 0.05574 0.11322 0.06765 0.16205 15 Urban 0.47749 0.09143 0.08027 0.12754 0.07006 0.15319 Rural 0.48534 0.10312 0.05339 0.12411 0.05173 0.18228 16 Urban 0.46976 0.09627 0.07827 0.12567 0.06853 0.16148 Rural 0.46577 0.09507 0.04932 0.16346 0-04449 0.18187 17 Urban 0.44086 0.08867 0.08093 0.14422 0.07586 0.16944 Rural 0.45166 0.10298 0.04727 0.12743 0.07317 0.19746 18 Urban 0.39871 0.09516 0.08113 0.15021 0.08896 0.18580 Rural 0.38918 0.10652 0.04011 0.17169 0.05148 0.24100 19 Urban 0.35585 0.09145 0.07498 0.18310 0.09644 0.19815 Rural 0.35983 0.09311 0.04136 0.16189 0.07408 0.26970 20 Urban 0.26344 0.07463 0.06881 0.23970 0.10353 0.24986 Rural 0.22459 0.06447 0.02755 0.36034 0.04499 0.27804 * Rows add up to lOOS except for rounding errors.  -26- expenditures devoted to food and beverages is around 60236. Two, even at the lowest income levels a significant portion of expenditures (between 15 and 20%) is allocated to items other than food, shelter and clothing. Three, as expected, the food share declines as income increases. Table 5 shows that: one, in each household group mean household income is always lower for rural households. Two, for all groups a significant share of income derives from 'imputed' sources although, as noted, its nature is probably quite different for each group. Three, transfers -which in the IES include migrant remittances- are a relatively less important source of income for the poorest households. Four, income from own business is, in each household group, relatively more important for rural than urban households. Conversely, in each household group wage income is relatively more important for urban than rural households. Since the majority of the poorest households are rural (table 7 below), it becomes clear that wage income is not the most important source of earnings for the poorest groups. Finally, table 6 shows that: one, the largest number of the poorest head of households work as self employed in the rural areas (almost 40X), with the next category being agricultural worker (21X)37. Although there is no direct information, self employed workers in the rural areas are most probably small scale agricultural producers (but note that other activities like handicrafts are also included here); hence, even owners of some land are among the very poor. Two, if we take the lowest three household groups as constituting the extremely-poor (see section IV.3 below), and if we take self employment and employers with one to five employees in the rural areas to be mostly small scale agriculture38, and add to this agricultural workers, then 631 of all 36I note here that to construct this table only monetary expenditures are considered since, as noted, information on the components of own-consumption was not available. As can be seen from table 5 below, all household groups show a significant share for imputed income (the income equivalent of own consumption). For the poor it may be non-marketed food grown by them; for the rich it may be the use of a company car. Thus, the poorest groups may consume more food than what they purchase. The importance of this resides on the potential impact of food subsidies: to the extent that not all food consumed by the poor is purchased, the effectiveness of food subsidies is diminished; see the discussion in section VI.4. 37For each household group the rows of urban and rural add up to 100%. 38Recall that the IES did not report a separate category for ejidatarios. I assume here that employers with one to five employees in the rural areas are either ejidatarios or are also involved in some agricultural activity.  Table 5 Income Sources of Households* Household Imputed" Wages Own Property Coopera- Transfers Other Mean** Group Business Income tives Income 1 Urban 0.167 0.532 0.252 0.003 0.000 0.044 0.000 41,531 Rural 0.240 0.345 0.364 0.001 0.001 0.047 0.030 36,630 2 Urban 0.146 0.566 0.214 0.006 0.006 0.059 0.001 58,133 Rural 0.231 0.374 0.339 0.004 0.002 0.049 0.000 48,859 3 Urban 0.133 0.588 0.234 0.004 0.004 0.035 0.000 63,473 Rural 0.227 0.357 0.355 0.006 0.001 0.051 0.001 54,723 4 Urban 0.149 0.542 0.249 0.004 0.002 0.052 0.000 72,306 Rural 0.213 0.38i 0.326 0.011 0.001 0.066 0.000 58,217 5 Urban 0.142 0.581 0.214 0.010 0.002 0.048 0.000 79,417 Rural 0.215 0.346 0.342 0.007 0.004 0.083 0.002 69,464 6 Urban 0.144 0.614 0.185 0.007 0.009 0.037 0.001 86,078 Rural 0.207 0.405 0.311 0.008 0.003 0.064 0.0^^ 70,891 7 Urban 0.154 0.562 0.217 0.008 0.005 0.052 0.000 90,186 Rural 0.190 0.373 0.342 0.007 0.001 0.085 0.000 76,109 8 Urban 0.161 0.572 0.212 0.005 0.000 0.046 0.002 104,876 Rural 0.174 0.356 0.388 0.005 0.002 0.073 0.000 91,531 9 Urban 0.157 0.601 0.174 0.014 0.003 0.042 0.005 104,896 Rural 0.184 0.329 0.404 0.014 0.016 0.047 0.005 90,563 10 Urban 0.161 0.585 0.186 0.009 0.000 0.055 0.003 114,270 Rural 0.182 0.385 0.318 0.027 0.004 0.081 0.000 90,279 11 Urban 0.167 0.563 0.194 0.010 0.005 0.057 0.002 120,046 Rural 0.207 0.413 0.308 0.010 0.001 0.058 0.001 97,343 (continued)  (table 5, continued) 12 Urban 0.168 0.589 0.169 0.014 0.002 0.055 0.002 127,215 Rural 0.196 0.367 0.343 0.014 0.006 0.071 0.000 106,920 13 Urban 0.192 0.576 0.166 0.015 0.000 0.048 0.001 131,448 Rural 0.209 0.292 0.396 0.009 0.001 0.090 0.001 110,084 14 Urban 0.195 0.536 0.193 0.012 0.002 0.058 0.002 149,820 Rural 0.185 0.382 0.368 0.023 0.000 0.040 0.001 122,989 15 Urban 0.191 0.562 0.175 0.018 0.006 0.044 0.003 160,323 Rural 0.186 0.410 0.281 0.009 0.018 0.095 0.000 124,833 16 Urban 0.199 0.563 0.151 0.025 0.001 0.055 0.004 178,959 Rural 0.180 0.369 0.351 0.029 0.000 0.064 0.006 136,712 17 U:-u 0.212 0.539 0.152 0.022 0.000 0.069 0.004 186,883 Rur&: 0.192 0.350 0.315 0.038 0.016 0.086 0.001 164,343 18 Urban 0.204 0.512 0.182 0.032 0.002 0.063 0.004 211,614 Rural 0.219 0.342 0.304 0.030 0.001 0.092 0.010 189,852 19 Urban 0.217 0.535 0.141 0.029 0.001 0.065 0.012 249,924 Rural 0.242 0.378 0.249 0.038 0.000 0.089 0.004 207,328 20 Urban 0.238 0.443 0.159 0.078 0.001 0.064 0.016 385,978 Rural 0.173 0.256 0.394 0.055 0.001 0.084 0.037 330,926 * Within each household group income shares for urban and rural add up to IOOZ except for rounding errors. ** Imputed income is the monetary equivalent of own-consumption and gift.s see text. For a trimester, measured in pesos of January of 1984.  Table 6 Ocgaoational Characteristics of Rousehold Heada* un- non- Employer with non- member Household employ agri- agri- I to 5 > 6 self- remun- of Group ed cultural cultural employees employed erated coope- worker worker -worker rative 1 Urban 0.032 0.063 0.043 0.001 0.000 0.069 0.001 0.000 Rural 0.075 0.072 0.215 0.028 0.004 0.393 0.001 0.000 2 Urban 0.048 0.135 0.043 0.006 0.000 0.104 0.003 0.005 Rural 0.063 0.087 0.165 0.029 0.007 0.298 0.000 0.002 3 Urban 0.040 0.160 0.064 0.003 0.001 u.110 0.001 0.001 Rural 0.062 0.094 0.137 0.026 0.008 0.290 0.000 0.000 4 Urban 0.059 0.182 0.046 0.009 0.000 0.131 0.001 0.001 Rural 0.069 0.104 0.120 0.022 0.002 0.248 0.002 0.001 5 Urban 0.083 0.222 0.050 0.008 0.001 0.118 0.001 0.000 Rural 0.065 0.095 0.112 0.028 0.007 0.203 0.001 0.002 6 Urban 0.066 0.286 0.057 0.005 0.001 0.145 0.000 0.003 Rural 0.055 0.091 0.088 0.021 0.004 0.172 0.003 0.001 7 Urban 0.096 0.304 0.034 0.014 0.000 0.152 0.001 0.003 Rural 0.047 0.082 0.063 0.025 0.000 0.173 0.003 0.001 8 Urban 0.104 0.342 0.033 0.023 0.003 0.151 0.002 0.000 Rural 0.042 0.085 0.048 0.019 0.006 0.139 0.000 0.000 9 Urban 0.097 0.375 0.025 0.022 0.001 0.141 0.002 0.002 Rural 0.042 0.073 0.049 0.013 0.000 0.149 0.001 0.004 10 Urban 0.120 0.375 0.030 0.015 0.001 0.140 0.001 0.000 Rural 0.049 0.088 0.054 0.017 0.001 0.099 0.003 0.003 (continued)  (table 6, continued) 11 Urban 0.095 0.406 0.021 0.027 0.001 0.152 0.001 0.003 Rural 0.040 0.091 0.040 0.016 0.003 0.100 0.003 0.000 12 Urban 0.097 0.425 0.017 0.020 0.004 0.132 0.002 0.002 Rural 0.052 0.086 0.031 0.009 0.002 0.115 0.001 0.002 13 Urban 0.114 0.441 0.014 0.021 0.003 0.148 0.004 0.000 Rural 0.042 0.054 0.029 0.022 0.003 0.099 0.002 0.000 14 Urban 0.117 0.443 0.021 0.023 0.003 0.174 0.005 0.001 Rural 0.037 0.064 0.016 0.021 0.003 0.068 0.001 0.000 15 Urban 0.138 0.436 0.012 0.029 0.005 0.161 0.004 0.004 Rural 0.035 0.067 0.013 0.010 0.002 0.075 0.002 0.001 16 Urban 0.149 0.488 0.010 0.033 0.002 0.122 0.003 0.000 Rural 0.036 0.050 0.016 0.016 0.002 0.069 0.001 0.000 17 Urban 0.169 0.508 0.007 0.041 0.002 0.116 0.003 0.000 Rural 0.040 0.046 0.005 0.011 0.004 0.041 0.000 0.003 18 Urban 0.158 0.466 0.008 0.046 0.005 0.138 0.002 0.003 Rural 0.036 0.055 0.010 0.018 0.003 0.047 0.001 0.001 19 Urban 0.178 0.517 0.009 0.037 0.004 0.107 0.005 0.001 Rural 0.031 0.049 0.009 0.011 0.002 0.037 0.000 0.000 20 Urban 0.160 0.563 0.006 0.045 0.021 0.072 0.002 0.001 Rural 0.017 0.049 0.005 0.017 0.007 0.029 0.001 0.001 * Share of total household heads in each occupational category; within each household group shares add up to 1OOZ except for rounding errors.  -27- extremely-poor household heads are principally engaged in agriculture39. Three, while the urban poor constitute the minority of the extremely-poor (31X of the three lowest household groups), their main occupation is either as nont- agricultural workers or self-employed (28%). IV.3 Estimates of Moderate and Extreme-Poverty. Poverty indices Pj(a,z) for a - 0,1,2, z - z, z and j - rural, urban are presented in tables 7 and 840. Consider first panel (a) of table 7. When a - 0 I find, first, that 19.5X of the sample population could be classified as extremely-poor. Second, that 37% of the rural population is below the line of extreme-poverty, while only 9.9X of the urban population falls in this category. Given the respective share of each population in the sample total, t:is implies that the rural areas account for almost 671 of national extreme- poverty. Based on the head count ratio, extreme-poverty is mostly a rural problem. When correction is made for the depth of poverty, a - 1, the proportion of extreme-poverty accounted for by the rural population increases to 72.8%. Finally, when account is made of the distribution of poverty, a - 2, the proportion of extreme-poverty accounted for by rural groups increases to 76.6X. The fact that the share of rural extremely-poor increases with higher order indices illustrates that not only is extreme-poverty mostly a rural phenomenon, but the poorest of the extremely-poor are almost all in the rural areas41. (The significance of this is highlighted recalling that in Mexico more than two thirds of the population can be classified as urban.) It also illustrates the importance of appropriate weighting and the misleading 39Table 6 refers to the main occupation of the household head during the month in which the household was surveyed. To the extent that household heads may migrate between urban and rural locations with the agricultural cycle these results may be misleading. In addition, micro studies for poor land owning households show a diversification of income sources, with some earnings coming from off-farm labor (cf. Roberts (1982)), section V.1 below). Also note that household heads that report to be unemployed were so with reference to that month only, so that these figures cannot be used to properly assess unemployment statuis. 40The distinction between P[a,a(C)1 and P[a,&(M)] found in table 7 is explained below; at this point the discussion focuses on P[a,(C)]. 41In fact, one could argue that table 7 underestimates rural poverty to the extent that some publicly provided services, whose effect on poverty is not captured by the values of P (.), are relatively more available in urban areas; see footnote 44 below.  17 t Table 7 Indices of-Extreme-Poverty Panel (a): CNA recommended by Coplamar, a(C). a - 0 I a - 2 Pi Ti Pi Ti Pi Ti Rural 0.3719 0.6689 0.1232 0.7280 0.0572 0.7658 Urban 0.0995 0.3311 0.0248 0.2720 0.0094 0.2342 National 0.1951 1.0000 0.0594 1.0000 0.0262 1.0000 Panel (b): Minimum Cost CNA, z(M). a - 0 a -2 p Ti Pi Ti Pi T Rural 0.2113 0.7357 0.0617 0.7875 0.0266 0.8124 Urban 0.0410 0.2643 0.0090 0.2125 0.0033 0.1876 National 0.1008 1.0000 0.0275 1.0000 0.0115 1.0000 Table 8 Indices of goderate-Poverty. a 0 a - 1 a 2 pi Ti pi Ti pi Ti Rural 0.9667 0.4178 0.6351 0.4860 0.4662 0.5307 Urban 0.7281 0.5822 0.3631 0.5140 0.2195 0.4693 National 0.8119 1.0000 0.4586 1.0000 0.3037 1.0000  -28- picture that can be obtained from the simple head-count ratio: at least in the case of Mexico there are sharp differences In the level of poverty among the extremely-poor. It is important to note that for four reasons the above measures overestimate extreme-poverty. One, the estimate of X took as a basis the CNA recommended by Coplamar. But as already remarked, this CNA, denoted CNA(C), is not the least cost diet to give 2,082 calories and 35.1 grams of protein per day. If instead the 'true' minimum cost CNA, denoted CNA(M), is used as the basis for computing E I obtain, P[O,I(M)] - 0.1008, with 73.5% of this poverty total accounted for by the rural population; see panel (b) of table 7. More generally, given a value for a, as E falls the share of extreme-poverty accounted for by tbe rural population increases. This suggests that: first, P[a,y(C)] and P[a,g(M)] should be interpreted as upper and lower bounds on the indices of extreme-poverty, respectively, with the 'quality' (in terms of food variety and perhaps palatability) of the diet that satisfies the same nutritional requirement decreasing as we move from CNA(C) to CNA(M)42. Second, that while there is some uncertainty determining the exact proportion of the population that is extremely-poor, the proposition that extreme-poverty is fundamentally a rural phenomenon is very robust. Two, the above measures of individual poverty are based on per capita household income, and while as seen above poorer households are larger, they also have a greater proportion of children, requiring a conversion into adult equivalent income levels. Three, the existence of economies of scale in consumption reduces income requirements for larger households (cf. Behrman and Wolfe (1984)). Four, finally, only sample data was used for the estimates of P(a,y). But since the total value of consumption expenditures implicit in the IES is lowar than the corresponding figure of the national accounts, this could bias the results upwards (assuming the national accounts are closer to the 'true' expenditure totals). No correction was introduced for this factor since I had no information to distribute the discrepancy across income 42The CNA(C) costs 6,193.00 pesos per capita per trimester, so that i(C) equals 7,742 pesos; the CNA(M) costs only 4,554.10, so that z(M) equals 5,692 pesos. On the other hand, if we assume that for the lowest deciles of the population all own-consumption consists of food, the implied share of expenditures allocated to food would be 70X. Multiplying CNA(M) by 1/0.70 gives a line of extreme-poverty of 6,505.85, which is still 19X below z(C).  -29- levels43. Aside from this potential for overestimation, there is an additional source of bias in the measurements that arises from differences in relative prices across regions, since in these circumstances the same monetary income may translate into different consumption levels44. It is left for further research to determine how significant these omissions are. I emphasize that a value for P[O,z(C)] of 0.19 does not imply that in 1984 19X of the population was undernourished. As discussed in section III.2, measurements of undernutrition based on requirement-intake comparisons fail to correct for intra and inter-individual variability in nutrient requirements. In addition, the diet implicit in the monetary value for z(C) need not coincide with a freely chosen diet: even if an individual is given a monetary income of y(C) there is no guarantee that he will in fact spend 80X of it on food and that, in addition, the food chosen will have a composition equal to CNA(C). Moreover, note that an equally nutritious diet but with a different composition, CNA(M), gives substantially lower estimates of extreme-poverty. Of course, although table 7 provides no direct evidence, the presumption is t.hat the individuals below a(C) are in fact those that have the highest probability of being undernourished, are more vulnerable to disease, and suffer from anthropometric deficiencies, with this probability increasing as incomes fall further below X(C). Until additional evidence of undernutrition is available, these are the individuals who policy makers must assume are most in need. Additionally, it is clear that these individuals do behave differently, particularly with regards to key variables like fertility, household size and dependency ratios. 430ne could argue that since the IES paid particular attention to food expenditures and own-consumption, it is likely that most of the under- reporting occured at higher income levels. Since P(a,y) only requires information on the distribution of income up to y, it follows that the potential for upward bias arising from this factor is not too significant, however. "An example of this is the provision of water for the inhabitants of Mexlco City and those of surrounding poor neighborhoods (like Ciudad Netzahualcoyotl): the former get it free, while the latter must pay for its delivery. Solis (1984) presents evidence to show that, on average, prices in Mexico are higher in remote rural regions compared to urban. Greer and Thorbecke (1986a,b) have developed a simple methodology that allows to compute the regional monetary cost, y (J-1,2,...m), of a freely chosen diet (i.e., given individual preferences) that has the same caloric content in a context where there is regional price variation. In this case the monetary poverty line varies across regions, although the underlying reference to the level of nutritional poverty is the same.  Table 8 presents the values for P(a, z). I note, first, that if the CNSE is accepted as the appropriate reference point, 81.2X of the population (and 72.8X of all households) would be classified as moderately-poor. While, as argued in section III.1, moderate-poverty is a subjective concept, such a large number calls into question the components of the CNSE (and of other studies that have also used the CNSE as a reference point)45. Second, when a - 0 I find that the urban areas now account for the largest share (58%) of national poverty. When a - 2 these proportions are reversed, with the rural areas accounting for 53% of moderate-poverty. As with extreme-poverty, when account is taken of the depth and distribution of moderate-poverty the rural areas come to the forefront. To sum up, based on the 1984 IES: (i) at most 19% of the population was below the line of extreme-poverty, although it is probably the case that this is an over-estimate, (ii) the extremely-poor are mostly located in rural areas, (iii) the poorest of the extremely-poor are also found mostly in rural areas, (iv) the extremely-poor have very large families, have the largest share of children, the highest dependency ratio and the lowest educational levels, (v) not even the extremely-poor allocate more than 60X of total monetary expenditures to food, (vi) most of the extremely-poor are in agricultural activities, (vii) the urban extremely-poor are relatively better off than the rural, but have similar demographic, expenditure and educational characteristics. 45The complete basket is found in Coplamar, 1983, pp. 134-45; a careful look shows that an important part of the basket is made up of items like refrigerator, T.V., automatic laundry and dry cleaning, vacations and personal entertainment, etc.. Recall that this basket was formed on the basis of the expenditure patterns of the seventh decile; not surprisingly, the value for P(O,z) calculated on the basis of household incomes is, as mentioned in the text, 0.728.  -31- V.pDeterminants of Poverty. This section discusses the determinants of poverty. The central hypothesis is that lagging rural and agricultural development lies at the root of Mexican poverty. Independently of the geographical distribution of the extremely-poor, this hypothesis is important in a behavioral sense: urban poverty is not only quantitatively less important, but is to a large extent a reflection of rural poverty, as migration is a key mechanism through which the rural poor attempt to reduce their income differences vis-a-vis the rest of the population. The poverty profiler of section IV showed that the extremely- poor, aside from being located mostly in the rural areas and having the lowest levels of education, derive most of their earnings from self employment and wage labor, presumably in agriculture and related activities. An obvious implication is that to study the determinants of poverty is to study the determinants of the returns to unskilled labor and land -the main assets owned by the poor46. Given preferences, technology and the size and distribution of endowments, the returns to land and to unskilled labor depend on: (i) government policies broadly defined to include pricing, intra and inter- sectoral resource allocation, and (ii) the institutional environment in which agents make their decisions. Given its importance, most of this section is devoted to intra-rural policies, including a brief discussion of the institutional framework (sub-section V.1). However, I also briefly mention inter-sectoral (sub-section V.2) and macroeconomic policies (sub-section V.3). Two caveats: one, discussing all these issues here involves a substantial risk of over-extending and thinning my arguments. I emphasize that my aim is only to raise some key points and discuss their bearing on poverty; there is 46Actually, some of the land exploited by the extremely-poor under the ejido tenure form has important restrictions attached to it, so it should be thought of as a very peculiar asset; see below.  -32- no claim to a complete or systematic analysis47. Two, I focus attention in those factors that have characterized the Mexican economy over the last decades, since the current structure of poverty is the cumulative result of past policies. As I point out below, progress has been made in reforming rural regulations, in changing the pattern of subsidies, and in reducing macroeconomic uncertainty. But these changes have only occurred recently, so that their effects on poverty will only be felt in the future. V.1 Rural and Agricultural Develo2ment. There is a certain paradox in the argument that the root cause of poverty in Mexico lies in the rural areas. The paradox lies in the fact that for a substantial period of time Mexico's agriculture was a success story. Indeed, using the language of Timmer (1988), during the period of 'the agricultural transformation' the agricultural sector in Mexico helped to: ".. (i) increase the supply of food for domestic consumption, (ii) release labor for industrial employment, (iii) enlarge the size of the market for industrial output, (iv) increase the supply of domestic savings, and (v) earn foreign exchange" (op. cit., p. 290). As argued by Yates (1981, p. 7-8), between 1940 and 1965 agricultural output in Mexico increased at an average annual rate of 5.71; and while the population was growing substantially, output per capita increased over 2X annually during this time. Various factors account for this magnificent performance, but two need to be singled out: one, there was an increase in the extensive margin, with harvested area growing at about 31 annually from 1940 to 1960. Two, technical change generated an increase along the intensive margin with yields growing at about 2X annually. Much of the growth in irrigated land occurred in the orth 47Major research challenges lie in these areas. At the theoretical level it is necessary to develop models that capture the incentive and efficiency effects of the ejido/private land dichotomy. At the empirical level more evidence is needed on the determinants of crop choice, tenancy arrangements and the demand for rural labor. Unfortunately, the data for such analysis is particularly scarce in Mexico. In addition, because some tenancy arrangements are illegal (although apparently common), the quality of the official data is suspect. ("However, given the clandestine renting out of ejido land to private farmers, the census data probably fail to reflect the real distribution of inputs and outputs between the two sectors; this places in doubt any conclusions about relative productivity that are derived from the census data." Heath, 1990, p. v.)  -33- and Northwest, where for geographical reasons public sector investments had higher rates of return; fewer irrigation projects were undertaken in the relatively denser South and Central parts of the country. Rain-fed agriculture, on the other hand, had a much wider growth. Irrigation, the development of high-yielding varieties, and increased use of fertilizers also allowed Northern agriculture to diversify crop choice. Yates, op. cit., estimates that about 1% of agricultural growth was accounted for by specialization according to comparative advantage (i.e., a switch towards higher value crops). As of the mid-1960's this performance deteriorated, however. Between 1967 and 1980 agricultural output grew at an annual average rate of 2.6Z, less than the growth rate of population (around 3.5% at that time); since then agricultural growth has slowed down even more: between 1982 and 1987 output grew 1.6% annually on average. This slowdown is a complex phenomenon but four causes can be singled out. First, the extensive margin was exhausted; harvested area peaked in 1966 (Yates, op. cit., p. 117). Second, public sector investment in irrigation projects diminished, which partly reduced the potential for growth along the intensive margin: only by switching towards higher value crops could agricultural growth be enhanced. Third, the terms of trade between agriculture and industry turned increasingly against agriculture: "..the price/cost relationship began to deteriorate just at a time when it was becoming more expensive per acre to bring additional land into cultivation" (Yates, op. cit., p. 65). Fourth, private investment in * agriculture fell. The supply contraction was therefore not caused by weather variations, but by an inelastic supply of land, lower public and private investment, and deteriorated terms of trade. Aside from price and public investment policies, institutional factors also play a key role. In particular: (i) the land tenure system divides agriculture in two separate forms of tenure, private and ejido agriculture, and (ii) a complex system of regulations applies to the use of land, labor and credit in both types of agriculture. In private agriculture the are limitations to the size of land holdings, so that entry is limited; restrictions apply also to the uses of land48. In ejido agriculture land 48For example, private farmers may own up to a maximum of 100 hectares of irrigated land, but up to 150 if they grow cotton, and up to 300 if the land is used for coffee, sugarcane, grapes and other fruit trees. They may own more land if it is not irrigated, or if it is used for cattle-grazing. But land used for cattle-grazing cannot be used for crops.  -34- cannot be sold or mortgaged; restrictions apply to the type of labor contracts that can be implemented, and to the sources and uses of credit49. A full analysis of these institutions _s beyond the scope of this paper but some remarks are necessary since they are key determinants - earnings in the rural areas. I focus attention not only on the ejido-private agriculture dichotomy (which is certainly very important), but also on other regulationis that bear on the rural sector. It is this complex of regulations and land tenure institutions that hurts the rural poor. Unfortunately, it is very difficult to discern which individual regulations are more binding than others. The difficulty in identifying the relative importance of each reguiation stems from the lack of empirical studies, and from the fact that they have been in place simultaneously, and across very different regions. In addition, some regulations (particularly on sharecropping and renting out of ejido land) are often by-passed; this probably varies from region to region and on the 'political climate' that determines when and how carefully the law is enforced50. Thus it is probably the case that at times the regulations on credit to the ejido are the major deterrent to agricultural growth. But at other times restrictions on sharecropping and rental might play a key rVle limiting rural incomes, while yet in other cases it is the absence of investment which is the limiting factor (which in turn might be depressed due to uncertainty or to deteriorated terms of trade). I now turn to four particular issues that merit special attention. One, restrictions on the use of (private and ejido) land imply the absence of a mechanism to equate the marginal productivity of land across various uses. In private agriculture, land (particularly in large farms) can at times not be switched across different crops. In ejido agriculture land cannot be rented or exploited through sharecropping and other tenancy 49See Yates, op. cit., chps. 7 and 8 for a good description of the legal framework, of the incentive problems it creates, and of the amazing array of restrictions and regulations. Since then some modifications have been made to the agrarian laws, although it appears that the main impediments and contradictions remain (cf. Heath, 1990; Velez, 1990). 50"Although, in spite of the law, renting is currently widespread, it is reasonable to argue that, in the absence of any ban on leasing, the incidence of this practice would be even greater: there are pressumably a number of ejidatarios (particularly those on bad terms with the ejido leadership) who are deterred from renting by the prospect that their parcels may be confiscated" (Heath, 1990, p. 11).  -35- arrangements. This need not imply that ejido agriculture is less efficient than private agriculture (in plots of similar size, with similar access to water, etc.). In fact, the empirical evidence on this score is ambiguous: "It is almost impossible to draw hard and fast conclusions about the relative productivity of private farms and ejidos in the recent period" (Heath, 1990, p. 3). Yet, when the restrictions are imposed they imply that the returns to land are below the maximum attainable. Moreover, the rural poor own land under both tenure systems, and poverty is associated with both51. In addition, they are also affected by regulations on large private farms through its effects on the demand for labor. Two, regulations in agriculture have an effect on the market for credit, particularly in the ejidos. Since ejido land cannot be mortgaged, most credit for the ejidos comes from public sources52; in addition, individual ejidatarios are not allowed to contract credit on their own, but must do so as a joint operation of the community (although this changed very recently). Joint provision of credit creates three problems. First, there is no 53 mechanism to insure that credit is allocated to efficient producers Second, the joint nature of credit creates a free rider problem resulting in large default rates. In effect, credit becomes a subsidy to production and consumption (Yates, op. cit., p. 209). The associated subsidies to public agricultural credit institutions are probably regressive, since there is no mechanism to insure that the subsidized credit goes to the poorest 51As mentioned in section IV, the IES does not allow one to connect earnings with asset ownership. Nevertheless, household studies report low incomes for small scale private agricultural producers and ejidatarios, so it seems safe to assume that some of both are among the poor; see Finkler (1978) and Roberts (1982). 52Commercial banks can lend to ejidatarios, but do so accompanied by guarantee schemes provided by second tier institutions. 53A related problem is that credit is given in kind which appears to create problems with timely delivery, an issue of particular importance in agriculture. In addition, "..partly in order to achieve economies of scale in procurement, (Banrural) operates with a standarized input package that is insensitive to regional variations in input requirements and prices" (Heath, op. cit., p. 21).  -36- ejidatarios54. And while agricultural credit is not the first best in3trument to subsidize consumption, this subsidy, by construction, cannot reach landless rural inhabitants, who are probably among the poorest of the rural poor. Third, credit to the ejidos is made contingent on crop choice, with emphasis on corn, beans and basic cereals. There is almost no credit for livestock, nor for other crops that might be more labor intensive or have higher value per acre55. This limits the returns to both ejido land and labor. Three, the rural labor market suffers from at least five distortions. First, regulations on private agriculture (particularly large farms) can at times depress the demand for labor, since land used for catcle-grazing may not be used for crops (which are probably more labor-intensive), even if this is feasible and profitable. Second, certain practices that ejido owners might engage to reduce risk and diversify their earnings sources are prohibited; in particular, sharecropping is not allowed, nor is the renting of land to 54"Empirical data suggests that not even the institutional credit channeled through Banrural is reaching a significant portion of the poorest population." (World Bank, 1989a, Vol I, p. 56). Heath, op. cit., also notes a "..tendency to use credit for political rather than economic purposes" (p. 5). In addition, he notes that "State lending policy reflects a confusion between the objective of poverty alleviation and the objective of enhancing agricultural productivity" (p. 40). 55Crop diversification is one of the key mechanisms to increase the returns to land. Yet changes in cropping patterns have been very uneven. In the North and North-West there seems to be a "..remarkable degree of responsiveness to market incentives" (Yates, op. cit., p. 53), but in the Altiplano and Center the dependence on corn and beans has been increasing, although these are among the crops with the lowest value in terms of output per acre. Of course, crop choice depends not only on access to credit, but also on risk considerations, access to water, and timely access to fertilizers and transport; see below, section VI.5.  -37- private producers56. Third, crop choice limitations in ejido land (derived from the crop-contingent nature of public credit) reduce the demand for labor if ejidatarios wanted to switch to higher value and more labor intensive crops. Fourth, labor mobility of ejidatarios is reduced given the risk of losing their right to exploit a given parcel of land57. Fifth, the law "..prohibits ejidatarios from using hired hands as a