What's new
		New Volume The promise and perils of participatory policy making by Lucio Baccaro, Konstantinos Papadakis Discussion Paper 190 Executive compensation: Trends and policy issues by Franz Christian Ebert Raymond Torres Konstantinos Papadakis Discussion Paper 192 Labour, Globalization and Inequality: Are Trade Unions Still Redistributive? by Lucio Baccaro Discussion Paper 193 Impact of changing work patterns on income inequality, by Uma Rani Subscribe to our mailing list and we will keep you informed about our new publications

2.1 Bivariate characteristics and multivariate models of poverty

Enquiries into "characteristics of poverty" abjure counterfactual models of causal sequences. Policy analysts therefore cannot slice into the correlations and conclude (for example) that poverty may be effectively attacked by first reducing hunger or vice versa. The characteristics approach may, however, be a first step to causal modelling. For example, Klitgaard [1991, chs.10-12] rightly confronts economists for ignoring links between poverty and ethnicity (and possibly discrimination); yet in fact there is a three-stage sequence of enquiry. First, we need to establish the bivariate relationship: for example, in Malaysia [Anand 1984], ethnic origin is associated with much less poverty than is intra-Chinese or intra-Malay inequality, but in Côte d'Ivoire [Glewwe 1990; Kakwani 1993]. Voltaic nationality does seem strongly correlated with poverty risk. Second, we need to embed these relationships into a multivariate model: is ethnicity in Côte d'Ivoire associated with poverty independently, or as a proxy for certain sorts of low-wage casual work? Third, we need to order the model causally. The last two steps can be combined, e.g. by testing a simultaneous-equation system. However, sorting out the bivariate relationships between poverty and its characteristics, and comparing bivariate and multivariate results, are essential prerequisites to formal modelling of causal sequences involving poverty.

Moreover, even if no causal statements can be made simply by establishing bivariate attributes of poverty groups, these attributes may provide useful indirect indicators of which persons are likely to be poor. This, in turn, can permit successful "indicator targeting" of programmes to help poor people. If such indicators, e.g. shortness, slum residence -- are not readily altered or simulated, they avoid the incentive and information problems associated with targeting on persons claiming to have low income or wealth [Besley and Kanbur 1993; Glewwe 1992]. At the cost of a few high-profile errors, the overall incidence of both Type A and Type B errors -- alongside targeting costs, and stigma -- may all be reduced.

If there is a large and significant zero-order correlation between poverty and another characteristic, we usually investigate higher-order partials and/or perform multiple regressions. However, conclusions about poverty can be misleading. Suppose that, on large samples in Country X, rural areas show a significantly higher P_2. Noting that a larger proportion of rural than of urban people are elderly, in big families, or female -- also (let us assume) characteristics linked to high P₂-- we then regress P₂ on these characteristics as well as on a rural/urban dummy. If the beta on this dummy becomes non-significant, we should not be misled into concluding that rurality, as such, cannot be a cause of poverty. Features of rural (urban) life, created by nature or by policy, may harm (improve) the prospects of women, the old, or those in big families to reduce their poverty.

Sorting out such causal sequences is the business of both anthropology and econometrics. Only a handful of countries has big enough household samples for robust multivariate work, let alone for causal modelling and testing. To lay the basis for such work, we need bivariatey results. Some are presented below as "characteristics of the poor". Pending multivariate modelling and testing, analysts can explore causal alternatives, but policymakers should not depend too heavily on the answers.

2.2 Demographic characteristics

2.2.1 Malthus

His challenge is often analysed (and, nowadays, rejected) at the macro-level, in real space, and relevant to GNP per person (via a population-resource "race"). Malthus himself saw demographic change -- more interestingly -- as a micro-economic problem. Except in the very long run, he normally interprets the problems in pricey space. And he sees it is a problem, not for "growth" or the limits to it, but for the prospects of improving the welfare of the poor.

We thus examine the "characteristics of the poor" as the centre of gravity of a rectangle -- population, labour, food, assets -- created by Malthus's challenge and the responses to it. Population growth, for Malthus, increased labour supply relative to demand, and (because of diminishing returns to labour) increased food demand relative to output. These increases tended, respectively, to push down the money wage-rate (and/or employment-per-person) and to raise food prices. These tendencies damaged poor people's nutrition. Therefore, "The object of those who really wish to better the condition of the lower classes of society must be to raise the relative proportion between the price of labour and the price of provisions" [Malthus 1803/1960: 499]. Partly for ethical reasons and partly because of incentive effects, Malthus rules out artificial contraception, and most forms of income and asset redistribution (he advocated the very gradual phasing out of poor relief). We do not share these constraints. However, to a greater or lesser extent, many poor states "act as if" they do so. This should link poverty to those households with most members, especially children, and least land or other assets. This leads, in the demographic arena, to the question: to what extent is poverty associated with large or fast-growing households, high fertility, and (in some sense) consequential high mortality?

There are two other relevant sets of demographic characteristics of poverty groups: structural and cyclical. Does the structure of the poor population place it disproportionately, or increasingly, in female-headed households, or among women, children, or the old? Does the life-cycle move people, in large numbers, in or out of high risk of poverty?

2.2.2 Household size, poverty, assets, status

Typically, larger household size is strongly associated with much greater risk of poverty [Birdsall 1979, Meesook 1979, Musgrove 1980, Visaria 1980, Lipton 1983a, Lanjouw and Ravallion 1993]. For example, in the poorest quintile of households in rural Thailand in 1975-6, 42.2 per cent had 9 or more members, as against 12.2 per cent in the richest quintile [Meesook 1979: 65]. In urban Colombia in the 1970s, 78 per cent of households in the poorest decile contained 8 or more persons, as against 12 per cent overall [Birdsall 1979]. ^{(Endnote 6)} There is now growing evidence that even the alleged exceptions (in West Africa and elsewhere) to the rule that "big households tend to be poor households" in fact conform to that rule [e.g. Glewwe 1990; Kakwani 1993].

There is a "family size paradox" [Krishnaji 1984], or "demographic paradox of poverty" [Lipton 1983a]. (1) Small households are less likely to be poor than large households. (2) Households with little status or few assets (especially land) are less likely to be poor than others. Yet (3) higher-status -- higher-caste, non immigrant, "better" job-holding, male-headed, etc. -- and/or more abundantly asset-holding households tend to be larger than others [ibid.; Krishnaji 1980a, 1984, 1989; Shapiro 1990]. For example, in 1961-2, rural households in India operating no, or below 0.2 ha., of farmland averaged 2.7 persons; 4-5 ha., 6.5 persons; and above 20 ha., 8.7 persons [Krishnaji 1987: 893]. Since all these effects are probabilistic, there is of course only an oddity, not a contradiction, among them.

Historically, no oddity or paradox existed, because (1) above was false: poor households in most countries tended to be smaller than others. Children from non-poor households married younger; and, if poor households became large, adolescents were often put in service with non-poor households, "transferring" large size to them [Hajnal 1982]. However, in today's LDCs, adolescents from non-poor households are likelier to delay marriage through education, and subsequently to have lower marital fertility, than those in poor households. These in turn, have become less likely to be put out to full-time servant membership in non-poor households [Lipton 1983a]. Hence today's demographic paradox of poverty: poor households tend to be bigger; households with land, status or assets tend to be less poor; yet landed, etc., households tend to be bigger, not smaller.

It is land or assets per person or per adult-equivalent (AE), not per household, that one would expect to be associated with low risk of poverty. In most cases in the above sources, however, even farmland-per-person increased with household size, though less sharply than land-per-household.

Exceptionally, Shapiro [1990] shows that in the land-abundant conditions of Southern Zaire an extra household worker was associated with a small (0.055 ha.) but significant fall in land farmed per worker, and attributes this to increasing marginal costs of supervising even family labour in extensive agriculture, as portrayed by Binswanger and McIntire [1987]. However, this sequence fails in more intensive agricultures. There, how can poverty be strongly related to larger family and household size, and to low status and asset-holding (even per person), when the latter characteristics are strongly related to smaller family and household size? The three tendencies imply that, as a rule, each asset-holding or status-receiving group, e.g. the landless or the high-caste, must show a very strong positive linkage of large family size to poverty. The above evidence supports this, but does not dispose of the oddity.

2.2.3 Mortality

Evidence cited in Lipton and Ravallion [1994] and Lipton [1983a] shows that inter-group mortality differences are (a) heavily concentrated in the first five years of life, (b) increasing quite steeply with poverty, but also with asset deprivation, (c) sometimes apparently discontinuous, or at least strongly non-linear, around a very low level of nutritional status, income, and asset-holding. On their own -- i.e. without allowing for actions by poor couples to (over)compensate for high child mortality, viz early marriage and high marital fertility -- these facts render large size of poor families more surprising, but of assetless families less so.

Cross-regional and cross-cultural mortality differences, however, are also big, even with similar severities of poverty. Much of this relates to disease and food environments. However, an unknown part of the difference is due to working requirements and circumstances, in particular those of pregnant women. Joint research between medical and social scientists might find promising routes to "uncouple" poverty from overwork, poor child care and high death risk; and hence to help poor people to escape from poverty.

2.2.4 Fertility

Higher mortality on its own would render large size among poor families more surprising. However, it is more than offset by their higher marital fertility. That is partly to replace, and insure against risks of, infant and child deaths; and partly because lack of access to affordable education, plus the need for income from child labour, often render it difficult for poor couples to "substitute (child) quality for quantity" [Becker and Lewis 1974]. This helps to explain the "paradox of family size". It is further illuminated by the fact that -- although income per person is negatively associated to fertility over most of the range -- operated (but not owned) landholding per person is positively associated to fertility [Mueller and Short 1983, for Botswana; Stoeckel and Chowdhury 1980; Schutjer and Stokes 1982, for Thailand and Egypt; Mitra 1978: 209-210, for India]. This may be because family, rather than hired, labour per hectare economizes transactions-costs in own-account farming; or because [Chayanov 1966; Nakamura 1986] extra hands and mouths in a family raise the marginal disutility to it of incomeless leisure, relative to that of drudgery divided over the total family.

ILO/IILS/WEP has done much important population-related research [e.g. Rodgers et al. 1992]. Probably, its comparative advantage lies in exploring the links between changed -- especially, improved -- prospects for wages and employment, and subsequent fertility declines. There is reason to believe that the favourable links are strong, but delayed and perhaps non-monotonic. Can policy speed up or otherwise support the linkages?

2.2.5 Age of marriage

Poorer households tend to have more births per marriage partly because (contrary to historical experience [Hajnal 1982]), they marry earlier than other households in today's developing countries. However, the effects on household size are ambiguous. Couples start to have children sooner, increasing it; but children marry sooner, reducing it [Lipton 1983a: 24-27]. High-prestige occupations, joint or extended family structures, education, and income-per-person are independent correlates of later marriage of offspring [ibid.; Singh and Richard 1989; Reddy 1991]; it is liklier to pay them to stay at home, working part-time for their parents, to the extent that they have expectations of inheritance.

2.2.6 Nuclear or extended?

Most of the above discussion this assumes nuclear, not extended, families. For the great majority of the world's poor, this is valid. Nuclear households are proving to be the norm in more and more cases -- developing as well as developed -- where the debate switches from literary to empirical. Nuclear families are especially dominant among the poor, where parents have few assets to "will-shake" over -- or to benefit from the complementary labour of -- married children who seek to leave home [Krishnaji 1980, 1984; Lipton 1983a: 27-32, and sources therein]. Hence complexity, i.e. jointness or extendedness, of households cannot explain poor people's larger household size, because complexity is linked to high (landed, chiefly, Brahman, etc.) status [ibid.: 30-1], perhaps as large landholders diversify income sources [Krishnaji 1980].

Labour participation, employment, and wage-rate, as they might differ between nuclear and extended families, are little researched. This is an interesting topic, but probably not a priority for policy research.

2.2.7 Is poverty feminizing?

Our conclusion, then, is that poorer couples' greater (need for) replacement and other fertility exceeds their greater (exposure to) mortality and hence raises their family size. More physical assets for the poor reduce it only if they substantially cut poverty. Education, of course, cuts both poverty and fertility (in normal circumstances: these are caveats). Since high fertility among the poor also causes high mortality via sib crowding -- and impedes women, in particular, from advances out of poverty through productive work -- the importance of appropriate intervention in this sequence is obvious.

This issue apart, until the Latin American evidence presented to this symposium by Buvinic [1994] was collected, the evidence was very strong that women are noty in general over-represented in poorer households; nor among heads of households that are likelier to be poor [Visaria 1977 and 1985, on Asia, Drèze 1990, on Karnataka, Maharashra, and Gujarat in India; H. Standing 1985, on Calcutta; Haddad 1991 and Lloyd and Brandon 1991, on Ghana; and earlier sources cited in Lipton 1983b: 48-53]. Further -- except for some subsets of girls under five in North India [Levinson 1974; Bardhan 1982; Das Gupta 1987] and Bangladesh [Chen 1981; Muhuvi and Preston 1991] -- females are not normally exposed to excess poverty-induced nutritional risk within households, as Harriss [1991] shows for South Asia. As for Africa, among 12 African countries, only in Nigeria do pre-school boys have better anthropometric status than girls; in eight of nine country samples with significant gender differences in infant or child nutrition it is the girls who do better; and adult women generally show a better body mass index than men [Svedberg 1990].

Nevertheless, women are especially severe victims of poverty, in three respects. First, they work for longer hours (household plus "economic") than men to achieve the same level of living. For example, in Peru, the excess female burden was worst for single-earner households, where female heads had to work 39 per cent more "market" hours than male heads; but even in multiple-earner households, market plus domestic work occupied female heads for 76 hours per month more than male heads [Rosenhouse 1989]. Such burdens are heaviest for the poorest. The relative demands of child-rearing can be assessed by observing that the ratio of under-fourteens -- and of under-fives -- to adult women, doubles between the best-off and the poorest household quintiles in most samples [Visaria 1977, 1980, on Asia; Lipton 1983a: 43-4]. ^{(Endnote 7)} As women participate more in market work under pressure of poverty, their domestic labour contribution is not substantially reassigned to men [K. Bardhan 1985; G. Standing 1985].

Second -- in part because women's culturally assigned, large share of domestic commitments prevents them from seizing new and profitable work opportunities as readily as men [Haddad 1991, for Ghana] -- women's chances of independent escape from poverty are much worse than men's. Many LDC job markets appear to be largely segregated -- into "progressive", poverty-escaping, and usually male; and "static", poverty-confirming and usually female [Standing 1985; Anker and Hein 1985; Guhan and Bharathan 1984, on silk-weaving in South India; von Braun, Puetz and Webb 1989 on irrigated rice-farming in the Gambia]. Even more important than the domestic burden, in explaining this poverty trap, may be cultural discrimination against females in both education and -- given education -- in job assignments. (On Taiwan, see Greenhalgh [1985]; on Ghana, see Haddad [1991]; on Bangladesh, see Safilios-Rothschild [1991]). In rural India in 1981, mens probability of being literate exceeded womens by a larger proportion among the far poorer scheduled castes (22 per cent-6 per cent) and scheduled tribes (28 per cent-8 per cent) than among the population as a whole (40 per cent-18 per cent) [Bennett 1991].

Third, in many cultures, widows face barriers against employment or remarriage, leading to especially high risks of poverty [Drèze 1990].

Though the severity of income-based poverty is usually no more among women than among men, male-dominated societies make it harder for women (widows being an extreme case) than for men to escape from poverty [Alam and Martin 1984; Schiegel 1976]. These two facts imply that poverty is more likely to be chronic for women, and transitory for men. The "feminization of poverty" happens, not so much via higher incidence, but in the sense that turnover is lower among poor women than among poor men, so that expected lifetime disutility is higher.

The policy relevance for research in the UN system, above all ILO and UNESCO, is strong. What labour market conditions, and what material and cultural constraints, are associated with the trapping of poor women into either low participation rates -- urbanization reduces the rates in the poorest quintile by 20-35 per cent [Lipton 1983b] -- or into higher unemployment (see section 2.4), or, above all, into types of work offering few prospects for self-advancement? How can the organization of work, homestead micro-horticulture, and other occupations compatible with both skilling and domestic work, be arranged so that women can better escape poverty, yet employers are not faced with disincentives to labour-intensive activities?

2.2.8 "Juvenization" of poverty

It is a huge, and in Africa at least a proportionately worsening, problem. Yet the concentration seldom arises because their nutritional requirements are more neglected than that for adults in a given household [Schofield 1979]. It arises because child/adult ratios are much larger in poor households. This is both because higher infant and child mortality (leading to even higher replacement fertility) is caused by undernutrition, and because higher child/adult ratios cause severer income-based poverty. In the early 1970s, the poorest quintile of households contained 25 per cent of children in rural India, and about 30 per cent in Colombia, Malaysia and Brazil [Birdsall 1980: 39]. In five Andean countries, the probability that an urban household would be in poverty in the late 1970s was more elastic to child/adult ratios than to the proportion of adults working [Musgrove 1980]. In rural India in 1972-3 child/adult ratios were above 0.45 among the poor, and around 0.35 among the non-poor [Lipton 1983a: 71, 102-103].

Heavy female burdens from the "double day" and child poverty often go hand in hand; hence, where children most need intensive parenting to grow up healthy [Zeitlin et al.y 1987], they are least likely to get it. Typically, in poor Indian households in 1972-3, each adult was associated with 1.8 children; in non-poor households, with 1.1 [ibid.: 104-105]. In one Indian village, widow-headed households with no adult male derived 56 per cent of earnings from child labour [Drèze 1990]! In a Bombay slum in the mid-1980s, a large majority of working children came from widow-headed households [Bharat 1988].

The linked questions of the "double day" and child labour are (a) important causes and effects (not just correlates) of poverty, (b) areas where research has been distorted by rhetoric and "political correctness". There is serious empirical and analytical work to be done. Much of it relates to the interfaces between labour, health, education, and survival chances.

2.2.9 Greying of poverty

In LDCs, the old have until recently comprised a much smaller proportion of the poor than of the non-poor [Lipton 1983a], but this may be changing. Overall the over-65s comprised 3.8 per cent of South Asians in 1980, but are projected at 4.8 per cent in 2000 and 8.2 per cent in 2025; in other developing regions the expansion is faster, except in Africa where even by 2025 the proportion is projected at only 3.9 per cent [Deaton and Paxson 1991: 2]. The old in the Ivory Coast and Thailand have below-average income -- but only because they are more rurally concentrated [ibid.]; and this does not prove that proportions in poverty are higher for the old. In Nigeria and India in the 1970s, the evidence was against this [Gaiha and Kazmi 1982: 56; Hill 1982: 187-188].

However, more poor people survive into old age nowadays. Also, because of the income-dispersing effects of assetless widowhood (creating poverty) and spousal inheritance (creating wealth), inequality among the old may well be greater than among those of prime age. If so, similar average income per person in these two groups -- as in the Deaton-Paxson evidence -- would mean higher, perhaps much higher, poverty among the elderly. With the impending "greying of poverty" in most LDCs -- though not in Africa -- these issues merit further research. Their interaction with labour markets depends heavily on the extent to which (a) the poor can and do continue work into old age, (b) family, group, or state social-security mechanisms exist through which the working population support the elderly poor. European experience suggests that such mechanisms should be financed by general progressive taxes (including profits taxes), not by charges to employers based on -- and hence discouraging -- large payroll size. Analogous effects in family and community old-age protection are largely unresearched.

2.3 Labour and poverty

Population growth is supposed to worsen poverty through two routes in the Malthusian macro-model. First, it drives food prices up (Section. 2.4). Second, it raises labour supply, driving down wage-rates and/or employment. The counter-arguments, in the spirit of "new institutional economics", are that both fertility [Schultz 1981, Simon 1983] and technical progress [Boserup 1965, Hayami and Ruttan 1985, Simon 1986] are adjusted, by rational individuals, to these harmful effects -- thereby preventing them.

Only recently has it become possible to assess what happens to poor people as workers in LDCs. Tautologically, income per person is the product of (2.3.1) the proportion of people who are of working age, (2.3.2) the proportion of their time spent in workforce participation, (2.3.3) the proportion of their participating time during which they are self-employed or employed, and (2.3.4) their income per unit of time worked. Poverty depends on the average levels of these four variables, on their distribution between poor and non-poor, and (except for the widest poverty incidence measure) on their distribution among the poor.

2.3.1 Proportions of working age

As we saw in Section 2.2.8, the age structure of poor households is unfavourable to workforce participation. In three large rural and urban state samples in India, the proportion of under-14s in 1972-73 decreased sharply as household expenditure per person fell, while the proportion of over-sixties rose only gently. For example, in urban Maharashtra the proportions in poorest and richest household quintiles respectively were 49.6 per cent and 20.5 per cent for under-fifteens, and 5.2 per cent and 6.4 per cent for over-sixties. Thus the ratio of dependants to prime-age workers, the dependency ratio, among the ultra-poor was 54.8 per cent/45.2 or 1.21 for the poorest quintile -- well over three times the ratio, 26.9/73.1 or 0.37, for the best-off quintile [Visaria 1977; Lipton 1983: 43, 101-105]. Similar relationships prevail in numerous Asian data sets, rural and urban [ibid.; Visaria, 1980: Table 4]. Of course, the age at which economic work starts and ends is a behavioural variable, cultural and individual; and poverty compels some of the young and the old to work for reward, even when the norm is education or rest.

Yet a high ratio of young and old people to prime-age workers is a great drag on poor people's participation. This drag increases, at least relatively, with early development and urbanization. The rich-poor gap in the dependency ratio is proportionately greater in cities than in villages, and in more than in less developed countries and regions [ibid., Table 4 and p. 65; Lipton 1983: 45]. These demographic dynamics of urbanization, and their impact on poverty, appear to be an important field for research, perhaps jointly by UNFPA with IILS/ILO/WEP.

2.3.2 Age- and sex-specific participation rates (ASPRs)

Three things drive the poor to seek higher ASPRs than others. Poverty itself increases the marginal rate of substitution of income for leisure [Robbins 1930]. Second, as poverty deepens, so does the proportion of income derived from labour. Third, dramatically high dependency ratios increase the marginal utility of income-per-worker [Chayanov 1966]; thus the Malthus effect, that high fertility raises labour supply (and thus depresses the wage-rate), operates long before the children reach working age.

In all income-groups, 90-95 per cent of prime-age men are in the workforce; it is, therefore, only for women, children and the old that ASPRs can vary greatly in response to poverty. Women's ASPRs increase, but only modestly, as household poverty deepens, but with limited benefits for the poor. (a) The effect fades out among the ultra-poor; the poorest 5-15 per cent of households show female ASPRs usually no more than the moderately poor, probably in part due to bad health and nutrition. (b) At a given income per person, the household's female ASPRs decline with rising ratios of under-fives to women and older children [Lipton 1983: 16-17; Dasgupta 1977: 153]. (c) High female ASPRs may harm pregnant women (and unborn children) in extreme situations. (d) Women's participation and employment are in worse-rewarded tasks -- and subject to more fluctuation -- than men's (Section 2.10). (e) Above all, female urban ASPRs -- given the poverty level -- are quite dramatically lower in urban than in rural areas; so their tendency to be higher among the poor is less useful to poor women as they urbanize [Lipton 1983: 23-25]. ^{(Endnote 8)}

In urban Gujarat, India, only 1 in 4 women in the poorest decile participated in market work in 1972-73 [Visaria 1981: 13]. By 1983, among rural Indian women (over 15), 32 per cent were workforce participants, as against 35 per cent in the lowest four deciles; the urban proportions were 18 per cent and 23 per cent respectively [Hanson and Lieberman 1989]. ^{(Endnote 9)} This seems puzzling: urban women are usually better qualified educationally than rural women, and less often pregnant or lactating; and their children are likelier to be at school. Explanations may include greater underenumeration of women's urban work [ibid.], lower propensity of urban women to be household heads, factory scale-economies (and zoning laws) that militate against "economic" work at home [Lipton 1983: 24], and greater urban risks of violence against women at work and travelling.

Urban proportions of populations (including poor populations) are rising. So are urban female/male ratios (once very low), especially for poorer adults [Lipton 1983a: 51]. Therefore, the reasons -- and, if appropriate, cures -- for urban women's low ASPRs should be high on the poverty research agenda.

Child labour is demonstrably understated by large offical surveys, and much more prevalent in poorer households [Lipton 1983: 17-18]. More evidence is needed on causes and cures. Weiner [1991] develops a powerful case for enforcing India's paper protections for children. However, legislation cannot (should not?) prevent child labour, if it offers poor parents the only safe route to survival, unless alternative routes are offered -- and financed.

2.3.3 Employment and unemployment

For Arthur Lewis [1954], mass rural unemployment prevailed in LDCs, creating the chance for "economic development with unlimited supplies of labour". For Hansen [1965], Myrdal [1968] and others, unemployment in areas of owner-farming was meaningless, empirically small, or a "bourgeois luxury" irrelevant to poverty. Since about 1980, with the emergence (a) of hired work as the main economic activity of the urban and rural poor, (b) of the economics of search costs, information and risk, and (c) of major new evidence from household surveys, a more measured view of unemployment's role in poverty has emerged. It is less important than low incomes at work, but, selectively by time and place, important still; and increasing.

Unemployment as a usual status over a long period, in countries without social security, truly is a bourgeois luxury [see e.g. Udall and Sinclair 1982]. However, the time-rate of unemployment (TRU) -- i.e. the proportion of days or half-days in a reference period, typically the week before survey, spent workless and seeking work -- is substantially higher among poor workers and sharply so among the poorest, especially in towns, among casual workers and women, and for places, groups, and periods when most people are assetless and landless, unable to fall back on asset-based self-employment [Sundaram and Tendulkar 1988]. In view of the extremely strong impact of farm and firm size upon labour/land ratios [Lipton 1993a] the linkage of "unemployment as a cause of poverty" to the case for land reform -- and for urban and rural micro-enterprise -- is blatant. Unemployment itself is concentrated among the assetless and in areas, age-groups, etc. that are likely to over-represent the poor.

Relevant data for many countries are reviewed in [Lipton 1983: 42-54]; important new sources are Minhas and Visaria [1991] and Krishnamurthy [1988: 302, 306, 309-312]. He analyses a huge and careful household sample for India in 1977-78. Only 3 per cent of the rural workforce were unemployed as a "main activity" during the previous month, and the incidence was very slightly higher for the non-poor. However, time-rates of male unemployment "for rural areas...steadily decline from about 15 per cent for very poor households to about 8 per cent ... just below the poverty line, and continue to decline" above it. In towns the respective rates are about 20 per cent and 11 per cent -- corresponding to the greater urban prevalence of hired (including casual) labour. Female TRUs were 1.3 (rural) to 1.5 (urban) times male, but among the urban poorest female TRUs were well below male rates (probably reflecting lower female urban ASPRs among "discouraged workers"). In rural areas, the mainly self-employed suffered a TRU of only 3 per cent, as against 15 per cent for those deriving income mainly from hired work; the respective urban figures were 6 per cent and 13 per cent. Much higher TRUs were suffered by workers in "unskilled" households, i.e. earning most income from rural or urban construction (15-19 per cent), transport (10-12 per cent), or mining (11-12 per cent), or urban agriculture (10-11 per cent). There are striking regional differences; the TRU was over 25 per cent in urban and rural Kerala, 14-15 per cent in Tamil Nadu, but 4-7 per cent in the Northern States. More research is needed -- not only in India -- into why "time unemployed", clearly a correlate of poverty, varies so greatly among places without obviously different relative factor prices or labour information régimes.

A huge research agenda, in which largely neglected issues of workforce participation should be considered alongside much more widely discussed unemployment data, awaits those concerned to improve the information base on labour and poverty. Liaison with information from agricultural production functions, and on the relations between size of firm, self-employment, and labour demand and supply, is part of this agenda. Of course the primary data base is much more in need of strengthening, prior to such research, in Africa than Asia. The JASPA work in Africa [JASPA 1990] appears to confirm the strong impression from Asia, Latin America, and Europe that unemployment is a growing problem, and a growing explanation of poverty; but African TRUs are hardly ever estimated, and in several African countries (including South Africa) the distinction between unemployment and informal-sector activity is not made clear in the data available.

2.3.4 Wage-rates

We have seen that workforce participation rates, especially those reflecting supply of unskilled casual hired labour, increase with poverty; and that time rates of unemployment, reflecting slackness of labour demand, also tend to do so. The prospect for wage-rates of poor people's (unskilled) labour is therefore bleak. This is especially the case in a Malthusian world of growing overall labour supply, unless offset (1) by rises in labour's marginal value-product due to intersectoral shifts or to sector-specific (and not too labour-saving) technical progress, and/or (2) if falling relative prices of food staples offset falling relative wage-rates.

In a world where growing proportions of rural people depend on wage incomes, the imperfect linkage of poverty to farm wage-rates in many countries is interesting, and suggests growing alternatives -- urban, non-farm, or state-mediated -- for the rural poor. Thus rural poverty fell in Indonesia, Egypt, Kenya, and (to a smaller extent) India around 1950-75; yet real farm wage-rates showed no clear uptrend [Lipton 1983: 86-87]. In India, "the marked increase in [agricultural growth after Independence] to 2.4 per cent p.a., over its long-term trend rate of just under 1 per cent p.a., [had by 1978 not accelerated] rural real wages [beyond] the long-term trend rate of less than 0.5 per cent p.a." [Lal 1988: 283], probably due mainly to rising rural labour supply. In 1976-88 agricultural wage-rates appear to have risen by 2.5 per cent in India, 1.1 per cent in Sri Lanka and 11.5 per cent in Pakistan -- though by 0 per cent in Bangladesh and 1.1 per cent p.a. in the Philippines [Gaiha and Spinedi 1992]. The real wage rises were probably due mainly to world food price trends, not to labour-using technical change; elasticity of employment to cereals yield has fallen sharply since the early 1980s [Lipton with Longhurst 1989: 84-85]. A model that reportedly "tracked agricultural wages closely" suggests normal (negative) and substantial effects of Asian agricultural labour supply on farm wage-rates, except in Pakistan, where a strong positive effect was found [Gaiha and Spinedi 1992: 468].

It looks as if it is not rising real farm unskilled wage-rates -- but rather skilling, sectoral shifts, increased cereals yields even on handkerchief-sized farms, remittances, or (seldom) rising employment -- that accounts for falling rural poverty in most of Asia. Skilling, and associated human capital formation, raise productivity "even" in basic farm tasks [Jamison and Lau 1982], and help people to escape from low real wage-rates in unskilled agriculture by shifting or diversifying sector, or place, of work. In Malaysia, Thailand and Korea, it was arguably this skilling process that eventually led to the "Lewis-Fei-Ranis" outcome: absolute decline in farm labour supply. Post-primary education also helps this process by inducing lower fertility.

The emphasis of the World Bank [1990] on technology and incentives via domestic institutions and policies that absorb labour is justified. It appears obvious that minimum wage laws, restrictions against redundancy, etc., harm the poorest by discouraging employers from using labour. But ILO and other research suggests a more complex reality. In some circumstances, the poorest can be net gainers, even if a minimum wage law for those in work does (inevitably) reduce employment. It depends on, for example, whether those in (or at risk of) unemployment tend to be secondary workers, in households with a securely employed or self-employed primary worker -- and, of course, on the elasticities involved. However, very careful enquiry is needed, into the employment and wage structures of poor (and potentially poor) households from an adequate sample survey, before adopting policies that set a minimum wage significantly above the level justified as a market signal. Experience from Zimbabwe to Kerala does suggest a serious impact on the job prospects of the poor.

Perhaps even more important, what happens to poor people's wage-rates (and welfare) if world-scale technical progress, and the relative factor and product prices with which it interacts causally, tend to displace labour (and the output-mixes that are made labour-intensively? Technology and incentive outcomes are ultimately determined globally for all but the biggest LDCs; and the global trends (as Western unemployment suggests) are extruding labour. The bad news is that unskilled farm wage-rates therefore seldom rise much (at least so long as farm labour supply grows). The good news is that, as shown, poverty can nevertheless fall sharply.

2.3.5 Structure of work

Total work done can be classified by type of contract, worker, work, or employer. Related to type of contract are: casual/long-term, factory/outwork, piecerate/timerate, informal/formal, and employee/self-employed/family. Related to characteristics of worker are: migrancy, nationality, age, gender, and educational level. Related to type of work are: location (e.g. urban-rural), economic activity (e.g. by SITC classification), skill level, and part-time/full-time. Related to type of employer are size of unit and (again) location and nationality.

Some of these characteristics show systematic links to poverty incidence or severity. In particular, casual work is strongly linked to high poverty incidence, apparently in large part because casual workers face a relatively high unemployment risk [Visaria 1980; Lipton 1983 and citations therein]. Clear inverse-U-shaped age-wagerate relationships, and positive education-wagerate relationships, have also been established, matching poverty risks. So has the fact that apparently strong gender-wagerate relationships break down when task and day-length are held constant [ibid.]. More will be said about the crucial economic-activity/poverty linkage below, but first a word of caution is needed.

Unlike analyses of participation and unemployment, even as correlates of poverty, the analyses of "structure of work" do not, even implicitly, distinguish between supply and demand factors. Nor do these analyses usually separate the poverty linkages of cross-sectional differences among individuals (or households) in a variable such as education, from those of its increase over time in an economy. Taken together, these two omissions create a serious danger of drawing wrong policy conclusions from a "fallacy of composition". For example, from the almost universal finding that individuals with more education ^{(Endnote 10)} enjoy lower poverty risk, it does not follow that educating more people in a nation will lower poverty. The first, partial-equilibrium, question is: will a minority of poor persons, if it becomes able to supply a higher level of education, increase expected earnings, income from self-employment, or ability to achieve welfare from a given income? Suppose the answer is yes. Then the second, general-equilibrium question is: does demand for educated services grow commensurately with supply? If not, the extra efficiency-units of labour-supply due to a more educated workforce may -- even if enriching the persons whose level of education rises -- crowd out from the labour market, and hence impoverish, other workers. This is especially likely to affect poorer, less-skilled workers, in a process of "qualification escalation". Conversely, suppose that demand for educated work rises faster than supply. This will normally accompany (1) rising GNP but also (2) a rising share, in GNP, for educated and usually non-poor workers; if (2) is faster than (1) then the incidence of poverty -- and probably higher-order indicators such as P₂ -- must worsen.

Obviously, these remarks are not made in order to denigrate education as a remedy for poverty. Both cross-section and time-series studies reveal its power, at least given rising demand for the commodities produced by the educated (Section 2.7). The point is to warn against over-simple policy readings of the growing, and in some respects exciting, evidence on work-structure and poverty. To give another important example: it is becoming clear that in towns informal-sector employment is not strongly associated with poverty; and that in rural areas non-farm activity (RNFS), partly because less strongly associated with casual labour than is farm activity, is strongly associated with reduced poverty risk. In most countries, such findings do greatly help in locating "markers" of what sorts of areas, groups, etc. are likelier to be severely affected by poverty. But the findings do not, as a rule, allow us to conclude that -- for example -- a growing share of urban labour supply to the informal sector will not increase urban poverty, or that State acts of stimulus should be transferred from agriculture to other rural activities in order to reduce rural poverty. First, there are interaction effects -- e.g. RNFS growth appears to be fastest in areas where farm growth has been fastest [Hazell and Ramasamy 1992; Dev 1991]; and in urban areas informal-sector growth may depend on formal-sector growth. Second, we do not know if it is extra supply of informal (relative to formal) urban labour -- or of non-farm (relative to farm) rural labour -- that is associated with lower poverty, or extra demand.

The usual assumption behind many "new" policies is that supply management is required. Yet it is at least probable that growing demand (and not at all supply) for urban informal or rural non-farm activity is growth-linked, and in turn responsible for poverty reduction. Absent growing demand, attempts to reduce poverty by encouraging the supply of urban informal-sector [de Soto 1989] or RNFS activity may be like pushing on a piece of string. Moreover, both RNFS and the urban informal sector are (a) heterogeneous and (b) infrastructure-dependent. (a) In Egypt, Indonesia and India, it is rural commerce, transport and construction -- not crafts and manufactures -- that have proved dynamic, labour-absorptive, and "linked" to farm growth [Dev, 1991; Unni, 1991]. (b) Across Indian Districts, RNFS growth -- much more than farm growth -- has been highly responsive to the density of local rural bank branches [Binswanger and Khandker 1992].

No data set, by relating work structure (e.g. by activity, status, outsidership, formality, etc.), to the incidence and severity of poverty, so far enables us to disentangle the supply-demand and cause-effect conundrums. In presenting a few possible implications of recent findings, we suggest a tentative interpretation only, in the hope of stimulating proper modelling and testing.

There is quite widespread evidence that rural households which earn a high share of income from the non-farm sector have relatively low risk of poverty. But that statement skates over the (very rudimentary) causal evidence. Three things seem to be clear. First, agricultural labour is especially poverty-prone -- either as such [on India see Dev et al. 1991] or because it is especially likely to be casual labour and therefore liable to unemployment [Visaria 1980]. In this case, those engaged mainly in the RNFS are helped to escape poverty because RNFS work is likelier than agricultural work to be self-employed (entrepreneurial), or else long-term employed (craft, apprentice, family) -- in some cases because of the greater permanence of RNFS enterprises (for Mauritania see [Coulombe and McKay 1991]). Second, diversity of income sources, and hence multiple family bases (locations), are associated with both membership of the RNFS and reduced risk of poverty due to downward income fluctuation, as in Zimbabwe [Jackson and Collier 1988]; this is especially likely to link the RNFS to the escape from poverty where the agriculturally self-employed are worse off than labourers, as in Mauritania [Coulombe and McKay 1991], and/or where, as in Burkina Faso, "larger" landholding signifies concentration on agriculture, lack of access to RNFS, and hence greater poverty risk [Delgado et al. 1991]. Third, much RNFS activity (52 per cent in a study of 288 households in Zimbabwe [Helmsing 1991]) is highly seasonal, thereby offsetting seasonal poverty for rural people with impaired capacity to save, borrow or store. ^{(Endnote 11)} The linkage of RNFS activity and poverty escape is therefore associated with either overall farm poverty, as in parts of Africa, or seasonal and annual fluctuations associated with casual (and principally agricultural) employee status.

Apart from the general finding that RNFS activity at household level reduces poverty risk -- via diversification, outsidership, reduced risk of casual work and unemployment, or simple avoidance of low-returns farming -- there is some evidence on poverty correlates in respect of the structure of work within agriculture:

-- Involvement in cash cropping is in numerous studies associated with reduced risk and severity of poverty. Often, this reduction clearly occurs after cash-crop involvement, and is not found in control groups not similarly involved [Maxwell and Fernando 1989; von Braun and Kennedy 1986]. However, the impact of higher real income, based on cash-cropping, upon poor people's nutrition is small and slow, and sometimes absent or not significant [ibid.]. This may reflect the fact that work on cash crops increases the energy requirement, or [Kumar 1977] reduces mother's capacity for child care, at least seasonally. An early compilation of village studies suggests that the incidence of undernutrition -- and perhaps of poverty as a cause of it? -- is significantly (at 5 per cent) less in villages with a mixture of cash crops and self-consumed food crops, than in villages with a great preponderance of either [Schofield 1979].

-- Reliance on landless labour, as opposed to farm self-employment, is a major poverty correlate in Bangladesh [Ravallion 1989]. A very small sample of elderly people in the same country [Cain 1991] suggests that wage work is much more closely related to landlessness among the elderly than among younger landless people, who more often engage in trade. In India as a whole, poverty incidence is much higher among landless workers than among small farmers [Dev et al.: 2.14]. However, this does not apply in semi-arid areas of India for farmers with below 3 ha. [Lipton 1985], nor in similarly dry African countries such as Mauritania [Coulombe and McKay 1991: 3.4, 3.12] and Burkina Faso [Delgado et al. 1991], where farming often proxies lack of access to poverty-reducing opportunities for rural trade or artisanship.

-- Greater reliance on (i.e. income or employment share from) common property resources is associated with higher poverty incidence in rural Tanzania [Collier, Radhwan and Wangwe 1986], as in rural India [Jodha 1986]. This is despite the substantially smaller inequality in CPRs than in private resources, but probably due in part to the major reduction in CPRs [ibid.]. Communal land tenure is associated with lower poverty risk in Mauritania [Coulombe and McKay 1991: 3.21], probably because both are associated with cattle ownership.

-- Formal-sector work (indifferently whether public-sector or private-sector) is associated with lower poverty risk, but almost certainly only because such work proxies educated labour [Coulombe and McKay 1991]. In Tanzania, education militates against low wage-rates even for unskilled farm labour [Collier et al. 1986]; this carries over to effects on total farm household income, and poverty risk, in the Indian Punjab [Chaudhri 1979] and more generally [Jamison and Lau 1982].

We have much to learn about the impact of work structures on poverty risk. For example, it is clear that almost everywhere the proportion of poor people dependent mainly on rural or urban employment is rising, relative to the proportion dependent mainly on income from farming. Yet a little own-farm activity may provide a "reservation wage" that increases immunity from poverty both directly and via improved bargaining capacity with the employer.

ILO will presumably want to concentrate on answering questions that could be relevant to policy and where the poverty impact of policy improvements is, or can be made, unambiguous. On the last issue, the recently contentious question of child labour [Weiner 1991] illustrates the problems. If the horrendous abuses (millions of children denied schooling; probably hundreds of thousands with jobs impairing life expectancy, health, or eyesight) are remedied by a legal crackdown alone, family (including child) nutrition may well suffer. Yet some countries did successfully stamp out child labour --Sri Lanka, China? One needs to ask how a damaging short-run poverty impact was avoided in such cases. More generally, if a particular labour structure -- such as a small RNFS -- is associated with poverty, we need to know why, and in particular whether societies with that structure differ mainly in respect of labour supply or of labour demand.

2.4 Food, nutrition and poverty

Part 1 reviewed the case for using an expected food energy adequacy' level of expenditure income or (per adult equivalent) as an ultra-poverty cut-off. The overall relationships between food, work, energy, income and welfare are shown in Diagram 1. Much recent research has been concerned with thresholds, turning-points and non-linearities in these relationships. If such thresholds, etc., tend to occur, it is at very low levels of (for example) body mass index, energy intake, or income; above such levels, affecting some 15-25 per cent of non-famine populations in low-income countries, adaptations -- e.g. in the speed with which a task is performed -- are possible (though often costly) in the event of energy stress. In brief, the ultra-poor need "food first"; the moderately poor can more quickly get less poor if they obtain assets and opportunities [for summaries of evidence see Lipton 1983b; Payne and Lipton 1994].

All this relates to labour research in three main ways. First, those at risk of extreme poverty adapt in many ways -- some more costly, or more likely to reduce prospects of escaping poverty, than others; these alternative modes of adaptation may suggest forms of work structure, options, or organization likely to help or harm the poor. Second, relatedly, the income-elasticity of demand for calories (CIE) has in several recent studies proved "surprisingly" low [Behrman and Deolalikar 1988; Bouis and Haddad 1992]; one of many plausible explanations relates to the fact that leisure will normally be substituted for income as income rises, thereby reducing energy requirements, but with an offsetting "substitution effect" of work for leisure if the wage-rate is rising. Third, the timing and intra-household allocation of energy stress among the poor may have major, researchable implications, differing among types of labour, and of policy to affect the balance among those types; the food-related issues are treated here in sec. 2.4.3; issues of timing of work and labour income are deferred to sec. 2.10.

2.4.1. Adaptation and labour research

In an important, challenging contribution to the economics and moral philosophy of poverty, Dasgupta [1993: 474] makes the surprising suggestion that -- because unable to adapt to undernutrition without severe damage by other means -- the very poor (landless) are driven to price themselves out of the (rural unskilled) labour market, where the landed deficit farmer "can undercut" them. This is supposed to happen because the landless: "must" earn sufficient to meet their total energy requirements; have no land of their own to help in this task; and are too weak to do so from agricultural labour (and to meet its extra energy costs) unless they receive a quite high piece rate. ^{(Endnote 12)} At a lower rate -- which it pays a deficit (part-time) farmer to accept, because he does have land of his own to help feed him -- the very poor (on Dasgupta's account) find that farm labour is a less efficient way to use energy than is CPR activity, scavenging and begging. These latter activities, however, "earn" too little energy to enable the very poor to escape poverty. Rationed out of farm labour by the only means of adaptation available to them -- viz. by rejection of (farm) work with relatively low work-to-food conversion efficiency -- the very poor are "rationed into" residual activity that earns so little as to lock them into their poverty.

The Dasgupta argument (1) is one of a big set of positive-feedback, vicious-circle explanations of the "too poor and hungry to work hard" type, (2) thus belongs to a set of explanations with much to commend it, and of high relevance for labour research, but (3) in my judgement, is not a very convincing example of these explanations, because it is hard to reconcile with some widely observed facts, and in particular with the nature and scope for adaptation [Dubos 1965; Payne and Lipton 1994]. ^{(Endnote 13)} Dasgupta (chs. 14-15) may well be right to cast grave doubt on the argument [Sukhatme and Margen 1982] that the undernourished can significantly respond to energy stress by biologically raising either metabolic efficiency (e.g. lowering BMR) or working efficiency. However, these intrapersonal, short-term intertemporal and biological adaptations to energy stress are much less important as "weapons of the weak" than three others [Payne and Lipton 1994]. These are (a) short- and medium-term behavioural adaptations (to physiology and other sources of energy balance), especially at work; (b) long-term intertemporal (including intergenerational) biological adaptations of physiology to the nature and timing of energy stress; (c) cross-sectional (i.e. interpersonal), static (atemporal) specializations -- by households, in response to the fact that they comprise different sets of persons, of given physiology -- that involve selecting many things, including types of work, along the lines of least comparative disadvantage. A crucial area for food/labour research is to identify sets of policies, behaviours, and labour-market environments that, without disastrously disrupting the above coping strategies, create options for people, now locked into them, to find poverty-escaping alternatives.

(a) Lean, short, or hungry people do a given job in different ways from heavy, tall, or well-fed people. The former group, unless "severely undernourished" according to conventional anthropometry, need not be less energy-efficient (i.e. need not use up more food calories per unit of work done), and certainly need not be less task-efficient (i.e. need not use up more food calories per piece of output produced). Jobs done slowly may be done better, per calorie of food used up, than jobs done energetically.

(b) Families, if poor and hungry for generations, select against genes (and children) that carry high dietary energy requirements. Mild to moderate stunting in youth is in effect selected for -- so as to avoid the much graver risk of wasting in adulthood. Stunting, even if mild, is indeed "no more healthy than scar tissue is healthy" [Martorell 1985], but until the risk of burning (undernourishment) is abolished, scar tissue (shortness, leanness, and low BMI) may be the least bad option. Its consequence is that:

(c) The "hereditarily" landless/poor/small will specialize in work where they have least comparative disadvantage. Need this trap them into poverty? No, because fortunately some such types of work actually often show absolute advantage for the small. Dasgupta rightly emphasizes that big people have high VO₂max -- prolonged maximum oxygen processing capacity. This indeed provides big people with absolute productivity advantage over small people in work mostly involving heavy lifting, such as lumbering or cane-cutting. However, small people have as high, or even higher, VO₂ max per kg of body weight than large people. Also, being small, they need to spend lower proportions of working effort in moving their bodies around, alongside the hoe or the ploughshare. This provides small people with absolute, not just comparative, advantage over large people, in terms of calories of food required per piece of work performed, if that "piece" requires mainly movement of the body, not heavy lifting, pulling or pushing. Most agricultural tasks in fact fall into the former category. This is consistent with the nearly universal observation that the very poor and landless spend a larger share of work time in hired farm labour than do the moderately poor or non-poor small farmers, and receive a smaller income per hour, though not usually per piece. Dasgupta's argument that the poor and undernourished are driven biologically to be on offer to the farm employer only at an efficiency-wage that -- in competition with the deficit farmer -- prices them out of unskilled labour markets (because CPR, etc., work is more energy-efficient for the poor) runs against these facts -- and against the steady decline of CPRs and of poor people's income and work from them [Jodha 1986].

I hypothesize that the conflict between theory and fact arises because Dasguptas outcome, with the very poor rationed out of farm labour and into CPR work or begging, is in fact a knife-edge, or at best a middle régime. In "bad" cases, people who cannot afford the calories for farmwork are liable to starve. In "good" cases, such people engage in adaptation -- short-term behavioural, and long-term biological (though probably rather little short-term biological adaptation à la Sukhatme-Margen) and hence we see a set of chosen specializations by very poor (small) people in unskilled labour markets. The test of my hypothesis is the forms taken by such specializations; if it is correct, its labour policy implication is the need selectively to train, empower, or "enasset" precisely workers in those specializations. If technology is flexible enough -- and if derived demand for trained, etc., labour is reasonably price-elastic! -- this is quite a promising way forward. Skill and education are complementary with higher wage-rates even in "unskilled" farm labour [Jamison and Lau 1982]. Neither the acquisition nor the exercise of such skills usually requires a high or rising VO₂ max, though exceptions do exist.

2.4.2. Income-elasticity of demand for energy

Recently, a major challenge to the view that undernutrition is due mainly to poverty has arisen from Behrman's work, and more generally from a number of observations that, properly measured, energy intake at the mean rises by only 1 to 3 per cent in poor populations when income (or expenditure) rises by 10 per cent [Bouis and Haddad 1992; Behrman and Deolalikar 1988]. This finding -- together with the fact that expenditure-elasticity of food outlay among the very poor is increasingly found to disobey Engel's Law, i.e. to be not significantly below unity [Bhanoji Rao 1981; Lipton 1983b; Poleman and Edirisinghe 1983; Hassan and Babu 1991] -- strongly suggests that the poor as a whole do not perceive extra calories (as opposed to pleasanter, more varied diets) as an overriding unmet need. However, more carefully considered, these findings redirect our attention to three critical distinctions: between (a) the undernourished ultra-poor and the "under-opportunitied" moderately poor; between (b) the effects on calorie requirements of extra income due to extra work, to higher wage-rates, or to non-labour sources; and between (c) use by poor people of extra resources to increase calorie intake and use to increase calorie adequacy relative to requirements. All these distinctions, in the light of Behrman's work, are of crucial importance for labour research.

A large majority of studies, investigating elasticities of (i) health and work performance with respect to energy intakes, (ii) calorie intakes with respect to income or expenditure (CIEs), find sharp increases -- perhaps simple thresholds [Lipton 1983b], but probably more complex non-linearities [Pelletier 1991] -- around low energy intake levels, usually below those typical of "poverty line" levels of income per consumer unit. CIEs estimated at the mean, therefore, considerably understate CIEs for, say, the lower quartile. This is true even if the "mean" is for total populations around, or even somewhat below, the poverty line. Furthermore, even if the lower quartile's CIE is also as low as 0.25 or thereabouts, the response of health and work performance to even a modest increase in energy intake -- and hence in income, even with a fairly low CIE -- could be substantial, either because of thresholds in (say) immune response functioning around some level of energy adequacy, or because energy intake is closely clustered around the apparent requirements level, as appears to be the case for the rural poor in an excellent Indonesian data set [Ravallion 1990b].

The above helps to explain the importance of income and expenditure to energy and health/work-performance, even with a low observed CIE at the mean -- i.e. a low beta in (log-linear) regressions of properly-measured energy intake on properly-instrumented ^{(Endnote 14)} total income (or expenditure), together with other explanatory variables. However, there is also reason for concern about the low r ² in such regressions (typically, "poverty" plus other variables "explain" only 5-15 per cent of interpersonal variation in energy intakes), and about the often low or marginal t-statistics.

I conjecture [1989] that this is largely explained by the aggregation of such equations across three types of source of extra income-and-expenditure. These three sources modify the effect of extra income (and total expenditure) on the choice of amount of energy, i.e. calories, purchased, because the sources have quite different likely impact on energy requirements. Income can increase (or be more for some people than for others):

-- because more effort is put into work at the same rate of reward, so that energy requirements rise;

-- because the wage-rate rises, so that energy requirements may rise or fall, according to whether substitution-effect (of income for leisure) outweighs or is outweighed by income-effect [Robbins 1930]; or

-- because non-labour income, e.g. remittance income, rises, normally with the result that the duration and/or intensity of work effort are reduced, so that energy requirements fall -- in which case some of the extra welfare is taken by replacing labour-income by leisure.

These three cases respectively amplify, complicate, and diminish the direct effect to extra income, via extra demand, upon extra energy intake. Since measured CIEs are estimated in equations that seldom include dummies (or other methods) to separate the cases, we may conjecture that low r ^2' s and t-statistics are likely. Field research is needed to quantify (or refute) this conjecture.

Such research will help to show what sort of work situations do -- and do not -- help which poor people to escape undernutrition. Other relevant variables involved may include intra-household energy allocations (and hence perhaps worker/dependent ratios); manner of working (piece, duration, etc.); non-energy nutritional content of foods (iron, vitamin A or zinc may "potentiate" the capacity of energy intake to enhance health or work performance); and timing or frequency of energy intake. However, before testing for the impact of many complicating variables, the first step is to estimate CIEs -- in various poor urban and rural situations -- allowing for changes, or differences, in rates of wages (or self-employment earnings), in worked time, and in non-wage income. This is, in my view, a high priority for research by, or supported by, ILO and/or FAO.

2.4.3. Timing, location and allocation of energy stress

A major recent trend in applied development research has been the emphasis on food security -- in the extreme case, from famine -- and on the reduction of vulnerability, achieved mainly by stabilizing and increasing "food entitlements" of groups at risk [Sen 1981], as the main way to attack poverty. Accompanying this has been emphasis on localized, and intra-household, variation in the adequacy of food to meet energy requirements. Most of this literature, though not Sen himself, tends to run together two separate issues:

-- To what extent is undernutrition ^{(Endnote 15)} caused by -- or (not the same thing) cost--effectively treated by improving -- levels of nearby food availability, food production, energy requirements of work or illness, or "simply" real income (usually the main determinant of "entitlements" to available food) for groups at risk? Note that membership of such groups varies over the life-cycle; and that the "best" way to raise real income (and hence market-based food entitlements) of the poor may be to improve incentives ^{(Endnote 16)} or technologies for food production, which is usually more unskilled-labour-intensive (and with a higher employment multiplier) than are most alternative rural or urban production activities.

-- To what extent are undernutrition and poverty caused by fluctuations, local differences, or intrahousehold misallocations, either in access to work and income (Section 2.10) or in food availability or need (see this section)? Note that, even if a village, household, or person has adequate average command over food to meet requirements -- i.e. even if nutritional inadequacies happen only when there are fluctuations, highly localized differences, or intra-household misallocations of foods, tasks, or both -- it does not follow that the most cost-effective remedy for inadequate nutrition is to reduce intra-household misallocations, localized variations, and/or fluctuations. Raising the typical poor household's average income may be a much cheaper, or even the only feasible, remedy -- whether for the household itself or for policymakers.

This said, timing, spacing and intra-household allocation -- both of food and of the sources of energy needs, notably work and illness -- are associated with a lot of energy stress. "Only" 3-7 per cent of under-fives in LDCs are severely undernourished, and the proportion has been falling [Bengoa and Donosa 1974; Kelley and Fillmore 1983; Garcia and Mason 1992]; yet some 15-25 per cent of populations in low-income countries have incomes carrying serious risk of severe undernourishment [Lipton 1983b]. Who gets "caught" varies over seasons [Chambers, Longhurst and Pacey (eds.) 1981], localized harvest failures, life-cycles, and with price-induced fluctuations in entitlements [Sen 1981]; across space; and according to allocation rules and decisions.

On timing, Edmundson and Sukhatme [1990] conclude that the main problem of low energy intakes is that normal reserves may not suffice to cope with extra stress. In rural India, the very poor at any time -- i.e. those whose energy adequacy is likeliest to be threatened -- are less likely to be in chronic poverty (throughout the period of the panel), and much more likely to be in transient poverty, than are the moderately poor [Gaiha 1989]. Bhattacharya et al. [1991], for six villages in West Bengal, confirm that the poorest suffer much more fluctuating -- but not less -- daily intake of staple food than the moderately-poor. In Bangladesh, where girls aged 2-5 are clearly likelier to die than boys, it is in times of acute need that this differential is clearest [D'Souza and Chen 1980]. Sen [1981] points out that the largest death-rate gaps between labourers and farmers usually appear during, or just after, periods of acute failures in entitlements to food.

The fact that energy stresses, or its consequences, fluctuate does not, of course, prove that it is more cost-effective to treat (or to help people treat) energy-balance declines, rather than low average adequacy. The lag structures need more research: in Côte d'Ivoire higher food prices (i.e. lower real incomes) do more damage -- especially for the poorest households -- to shorter-run indicators of nutrition status such as children's weight-for-height and adult BMI, but amount and quality of localized, precisely defined health infrastructure does more to affect long-run child nutrition indicators such as height-for-age [Thomas et al. 1992: 5].

Compared to the mass of work on timing of energy stress (see the reports in [Sahn (ed.) 1989; Payne and Lipton 1994: ch. 2]) -- and although casual empiricism suggests locational differences in stress are almost as important for the poor -- very little work exists on such differences. Yet there is:

-- a known geography of diseases (e.g. malaria) and of deficiencies of micronutrients (e.g. iodine);

-- some evidence that diseases interact, to product bad health outcomes, with energy stress [see Stephenson et al. 1986 on schistosomiasis];

-- reason to believe that micronutrient deficiencies, e.g. of zinc, iron or vitamin A, do so also.

On food allocation by households, Harriss [1986] surveys evidence demonstrating bias against girls aged 0-5 for Bangladesh, parts of Northern India, but not elsewhere in Asia; Svedberg [1990] demonstrates the absence of nutritional gender bias in Africa. Girls' inferior access to health care, rather than to food itself, is often the main cause of their worse nutritional status and performance where it exists [Sabir and Ebrahim 1984, for dysentery treatment in Bangladesh]. This illustrates both that nutritional status depends on illness as well as food intake and work, and (as per [Drèze and Sen (eds.) 1991]) that poverty can often be uncoupled from undernutrition by appropriate public action.

How food security is allocated depends on who allocates, not just on who earns. What can be called the liquidity theory of marginal income -- that men's becomes alcohol, but women's becomes breastmilk -- is naïve about fungibility, income pooling, and household decision structures. Female-headed households do seem to show a significantly higher calorie-income elasticity than the typical male-headed household [Garcia 1991, Table 5, for the Philippines] and the effect is stronger in poor households, in both Kenya and Malawi [Kennedy and Peters 1992]. However, this may well be because female-headed households are smaller, especially among the poor [Greer and Thorbecke 1986: 86, for Kenya].

Year-to-year variations in a family's access to income and other claims, and hence to food adequacy -- being dependent on weather and prices as well as on life-cycle events -- are not individually (or collectively) predictable, but may be directly or indirectly insurable. Seasonal variations are individually (and collectively) predictable, but not insurable. A large number of studies [Sahn (ed.) 1989; Chambers et al. (eds.) 1981] show that poor people experience substantial variations in seasonal and annual food adequacy. This is partly because variations in labour requirements for energy often aggravate (rather than offsetting):

-- variations in food intake, and in its determinants such as employment income, food prices, and home stores; and

-- variations in non-labour requirements for energy, e.g. to fight infections.

These "perverse" correlations do not indicate failure either of individual adaptation or of market functioning. In Bangladesh, individual adaptation to seasonal energy stress is -- as it must be -- so finely tuned that births, and therefore conceptions, are so timed as to reduce risk of overlap of the most critical times for energy adequacy (second half of wet season) and for child immunity (during the switch from passive to active immunity, i.e. age 6-12 months) [Schofield 1974].

There is growing evidence that seasonal fluctuations in body weight are much too small to "capture" the required responses to varying energy stress [Ferro--Luzzi 1992]. It probably has to be discretionary activity, work methods, or (to a small extent only) BMR that is "adapted". Over the longer span, plasticity of response - the ability of an organism to adjust growth-rates to stress, and then catch up -- declines rapidly in the early years of life [Eveleth and Tanner 1976].

Policymakers are often rightly advised to seek labour-intensive options, so as to pull up demand (and thus wage-rates, in some cases, as well as employment) for the labour of the poor. A familiar caveat is the need to avoid doing so just when seasonal peaks of employment already create islands of labour scarcity, even in oceans of labour surplus. Less widely recognized is the need to check that the timing and allocation of extra labour requirements does not damage poor people at periods of energy stress. For example, Kumar [1977] has shown pronounced seasonal variation -- from positive to negative -- in the effect, on the nutritional status of children, of extra work and income for their mothers.

2.5 Land ownership and poverty

The modest aim of most of the research reported here is to track the bivariate links between poverty -- its incidence, severity, location, timing -- and other variables, ideally at individual or household level, but otherwise through cross-sections of places, or time-series of a particular area. Where causal inferences seem possible, they are suggested. This partial-equilibrium approach is at best risky and doubtful, but it is all we have; general-equilibrium models are interesting, but have many problems and anyway exclude many key "political economy" variables. However, in the case of land, it is especially dangerous to equate modest, partial-equilibrium statements such as "Landless X is much likelier to be poor than landed Y" to interesting, general-equilibrium statements like "Areas of nations are likelier to contain many poor people -- other things being the same -- if they have unequal land rights and much landlessness".

Land distribution affects the demand for consumer loans, and hence the market for savings and investment. Land distribution also affects the transaction-costs of labour use by farmers, and hence not only the demand and supply of labour, but also the direction of indeed technical progress. Finally and less quantifiably, land distribution affects the gains and losses to rural élites from most sorts of State action. So the direct effect of land-rights on poverty cannot be even approximately reduced to the "crude question" of whether the landless are poorer than the landed -- especially since a growing majority of the rural poor, and almost all the urban poor, derive more income from work than from operating farms, and are net buyers of farm products. The land-poverty relationship -- and the argument for (or against) land reform as a remedy for poverty -- must be evaluated increasingly by its effects on those who remain poor workers and net food buyers, rather than on potential land-reform beneficiaries.

Nevertheless, the "crude question" is important in itself. In 1983, of Indian rural households living mainly from farm employment income, 45 per cent were poor -- as against 24 per cent of households living mainly from farming. In 1987-8, of the 35 per cent of Indian households mainly engaged in agriculture, but owning or renting in no (or below 1/10 ha.) of land, some 43 per cent were poor; of those cultivating 0.1-0.4 ha., some 26 per cent were poor; and the proportions fall steadily to 10 per cent for 8 ha. and above [Dev et al. 1991: 58, 88]. Landholding size is a quite good correlate of reduced poverty risk in Bangladesh, though poverty reduction by land-contingent targeting has limited scope [Ravallion and Sen 1992]. Smaller holdings tend to comprise better land [Bhalla 1988], so that this cross-section relationship would be even stronger if land were measured in efficiency-units. The landlessness-poverty link is so strong, in part, because those earning income mostly as farm employees do not, in many cases, have more access to non-farm income if they are landless than if they also operate a small farm (see Shankar [1993] for a study of three villages].

The landlessness-poverty link is generally confirmed by single-village studies [Lanjouw and Stern 1991]. However, in arid and semi-arid areas, it breaks down. In Rajasthan and Maharashtra in 1977-8 even 3 hectares of owned land -- normally cropland -- did not confer a lower incidence of poverty than did landlessness. In Northern Nigeria and Burkina Faso, landlessness may be a proxy for ability to work (at better returns) in the non-farm sector, and is not well correlated with poverty [Delgado et al. 1991].

The rural poor are (a) increasingly earning income from farmwork rather than from land operation, (b) increasingly net food buyers, (c) much more often than was once believed in transient poverty, especially as drought reduces demand for hired farm labour. The impact or landlessness and smallholding on poverty, therefore, needs to be judged by its effects on the poor through these three channels.

Easiest is (a). Smallness and equality of operated landholdings almost always raise per-hectare labour input by a larger proportion that they raised the ratio of family to hired labour [Booth and Sundrum 1984]. Thus even the demand for hired labour per hectare is greater on smaller farms, or after a redistributive land reform. Moreover, supply of hired labour from farmers themselves is less if land operation is more equal -- because there are fewer farm families with farms so small that they have a lot of "spare" working time to hire out. For both reasons, the landless poor gain farm employment, if farms are more equally distributed among the landed.

As for (b), smaller farmers generally produce a higher ratio of food to non-food, but also retain a larger proportion of product for home consumption. If the former effect is stronger -- and I hypothesize that it is -- then the poor gain as net food buyers, if land is more equal among the landed.

Least is known about (c). A priori reasoning suggests a mixed outcome, for the landless poor, from equal/smaller-scale farming. It is likely to be somewhat lower-risk farming. However, in bad years, small farmers are likelier than big farmers to find it feasible and profitable to displace a substantial proportion if their hired workforce with family labour. Yet, even in a bad year, the total workforce per hectare will be larger with small-farm systems.

All this assumes no change in "relations of production" between landowner, operator (manager), and worker. Tenancy almost always comprises a net transfer of land from big to small farms [Singh 1988] thus reducing poverty both among the latter and -- via (a) above -- among landless farmworkers. Little is known about employment-per-hectare with various types of tenancy contract, holding land size constant. Systematic changes in labour contracts have been associated with changing landholding conditions during the "green revolution" in Indonesia [Hart 1986] and the Philippines [Hayami, Quisumbing and Adriano 1991], but the net impact on poverty is not clear.

In brief, the history of East Asia, Latin America, and even West Bengal does tell us that more equal smallholdings benefit the poor. But this involves much more than the direct gain of land, even together with benefits from the inverse size-yield relationship, evidence for which is strong [Binswanger, Deininger and Feder 1993; Lipton 1993]. General-equilibrium effects, not just from flows of incomes and inputs but from (partly interlinked) non-land factor markets and from redistribution of power, also affect the small-farm poor. And it is on the poorest of all -- the landless net food buyer -- that analysis of the effects of land distribution should increasingly concentrate.

2.6 Other tangible assets

Even more than that other negative, the RNFS, "tangible non-land assets" (TNLAs) are very heterogeneous. Can anything useful be said abut their impact on the poor? We should bear in mind two distinctions: between TNLAs (or rights to them) that multiply income from work (e.g. land farmed by a worker-owner) and those that add to it (e.g. interest-bearing bonds) [Lipton 1985]; and between TNLAs (or rights to them) that, actually or potentially, tend to be owned or controlled by the poor and those (such as infrastructure) that may benefit the poor though seldom owned by them.

As regards TNLAs (including rights to them) tending to be owned or controlled by the poor, there are three issues. What types of TNLAs are "desirable" by and for the poor? What types of TNLAs are in fact likelier to be owned or controlled by the poor? What can research tell us about the likely fate of schemes to get TNLAs to the poor -- and, in particular, are there implications for new labour research?

Presumably, the poor -- as compared with other people -- would select, and benefit from, TNLAs that

-- multiply, rather than merely adding to, unskilled work income, because workforce participation increases with poverty down to the ultra-poverty threshold (Section 2.3.2);

-- stabilize income flows, since the poor are most exposed to variability in income and nutrition, and most harmed by a given downward fluctuation (Section 2.4.3, 2.10);

-- reduce risk (not the same thing), since the poor are more risk-averse, and less able to bear or diffuse a given risk, than the non-poor;

-- are divisible into small units, preferably with diseconomies of scale at large holding sizes;

-- are labour-intensive in maintenance, enhancement and use;

-- tend to raise the returns to other assets, including land, over which the poor have claims.

There is astonishingly little evidence on the distribution of claims on total assets, let alone on specific TNLAs. Obviously, overall, the poor have a much smaller share of such claims than of income; of income, than of expenditure; of expenditure, than of food expenditure; and of food expenditure, than of dietary energy intake or adequacy. One of the very few studies, in 1971-72 in India, showed that total tangible assets -- financial, land, other physical assets -- were 294 times larger per household in the richest decile than in the poorest, i.e. at least 700 times larger per person [Pathak et al. 1977: 507]. Yet calories-per-person can hardly vary by more than, say, 3 to 1.

Contrary to widespread belief, the poor are not likelier, relative to the rich, to own cattle than to own land in India [Lalwani 1991], and are much less likely to own cattle than to have significant (claims on common) land in Botswana [Watanabe and Mueller 1982: Table 3].

Paths:

1-5 Central path
6-9 Other income-related paths
10-12 Work-enhancer paths
13-20 Welfare paths

Smallstock, however, are more "divisible" blocks of value (and therefore embody less risk) than cattle, and are much likelier to be owned by the poorer rural groups. Yet India's main programme for subsidized asset distribution to the rural poor (IRDP) has concentrated heavily on livestock. Seabright [1991] shows that landless households were usually unable to manage the two cows allocated profitably, because of the cost of acquisition even after subsidy; yet, although this was too large a cash commitment for the poor, per-animal labour costs were inflated by the small number of cattle per household. Owned oxen may well be more useful, as sources of income, to the landed poor than to the landless poor, both because of lower feeding costs and because ox ownership (absent a perfect draught-hire market) may be required to enable a mini-farmer to cultivate on own account, rather than renting out.

With growing pessimism about land reform, poor people's acquisition of RNFS assets, usually supported by credit, has become a major thrust of rural anti-poverty policy in South Asia. Attempts to replicate Bangladesh's Grameen Bank are under way in several African and Latin American countries. There is a need for prior review of the existing ownership structure, labour-intensity, and potential manageability by the poor of the particular assets, which schemes such as IRDP and Grameen are to encourage the poor to acquire.

Just as ownership of RNFS assets, notably livestock, has been widely identified as the road to far rural poverty reduction, so ownership of one form of TNLA -- housing -- has been identified as an urban priority for helping the poor. Indeed, urban anti-poverty policy has until very recently been very substantially focused on site-and-service and slum upgrading. A better house is a relatively low-risk asset for the poor, but only because it does very little to enhance their capacity to earn income from (inevitably somewhat risky) production. Until recently, indeed, most housing-focused programmes did little to build in physical support facilities for informal urban production.

There remains the question of whether TNLA helps the poor most cost-effectively if they own or control it (i.e. if it is heavily concentrated on items suitable for the poor to manage) or if it is present in other forms, even if owned by the non-poor or the public sector. Work in progress at IFPRI indicates that, in Bangladesh, Zambia and elsewhere, the incidence of rural poverty is strongly and inversely correlated with both the per-person value of total infrastructure, and the presence of key specific items of infrastructure. There is a chicken-and-egg problem about such findings -- maybe widespread enrichment justifies the provision of infrastructure, rather than being induced by it -- but they are suggestive. Harriss [1993], analysing the causes of poverty reduction across three surveys (1958-66, 1982, 1991) in a village in West Bengal, gives first place (above even land reform) to extra employment for the landless due to groundwater infrastructures. Given wage behaviour in India, irrigation infrastructure appears to be a much more "pro-poor" (and pro-growth) recipient of public-sector subsidies than either fertilizers or even food [Ratha and Sharma 1992]. A simple model of the rice and labour markets in Bangladesh produces a similarly encouraging result for infrastructures of "irrigation-induced technical change" [Ahmed and Sampath 1992].

This is not meant to denigrate TNLA asset ownership or control as a cure for poverty. (Indeed micro-irrigation capital -- land or pedal pumps -- has proved in Bangladesh to be much likelier to remain owned and controlled by the very poor, selling water to larger farmers, than is land [Howes 1982; Shankar 1992].) However, the "cure" has been advocated -- and supported with large amounts of scarce credit -- with astonishingly little analysis of which TNLA assets the urban and rural poor now own and can manage productively. Nor has there been enough concern for secondary employment effects of "TNLA reform": effects known to be highly favourable for land redistribution.

2.7 Education and skills

There is a mass of evidence about the impact of a person's education on private earnings, employment (and thus income), and productivity. There is also much evidence of the response of a person's output to education. Further, East Asian countries achieving rapid and relatively egalitarian growth are well known for high and widespread initial literacy before, and for persistent outlay on education during, growth accelerations.

Surprisingly little analysis exists, however, of the household-level or national-level causal sequences that are supposed to lead from education outlays, via educational "assets", to reduced poverty risk (as opposed to higher expected income for the educated). It is familiar that even literacy declines as poverty increases, and that the higher the indicator of education the faster is such a decline. One's first thought is that the rich buy a fair amount of education for their children, so that publicly provided or heavily subsidized education is likely to benefit mainly the non-poor. However, this assumes that public-sector spending on education in developing countries is initially mainly on primary education [see, to the contrary, Schultz 1988: 606-7]; and that such education provides clear gains to the poor.

There is little doubt that farm productivity [Jamison and Lau 1982], and even the incomes of farm labourers [Chaudhri 1979], are increased by education. However, many of the gains, especially from female education, may be indirect. It has been shown to increase expected age of marriage, to reduce marital fertility, and reduce infant and child mortality [Rosenzweig and Schultz 1982, for Colombia; Behrman and Deolalikar 1988; Bourne and Walker 1991]. Completed primary education seems to be required to produce some of these effects [World Bank 1984]. Kerala's much better performance on indicators of "social development", despite very low average incomes, seems to be explained, in a careful cross-section of Indian States, by higher public expenditure on social sectors, mainly education [Raut 1993].

There is much evidence that it is opportunity-cost relative to perceived benefits, not direct cost, that often constrains poor parents from having their children educated. For example, Botswana is one of the few countries where boys are likelier than girls to drop out of primary school, because it is the boys who are expected to mind the cattle. However, educational planning can itself reduce those opportunity-costs, e.g. by timing school into the agricultural slack seasons. And the laws could be implemented -- e.g. on child labour in India [Weiner 1991] -- if the State so decided. Too little is known about the trade-off between short-run losses and long-run gains, to poor parents and their children alike; and labour-market pressures interact strongly with those losses and gains.

2.8 Health

There are big differences in life expectancy, infant mortality, and risk of illness between rich and poor countries in terms of average income per person [Behrman and Deolalikar 1988: 633-4]. "Outliers" can often be explained in terms of either income distribution or the degree of public-sector health and nutrition-linked activity [Sen 1980]; indeed, recent research by Ravallion (pers. comm.) shows that it is poverty incidence, not average GNP, that explains the bulk of many of these differences. Further, there are some clear indicators within countries that groups exposed to poverty suffer higher death-rates than other groups: Mitra [1978] cites striking micro-evidence for landless labourers vis-à-vis farmers in India, and Lipton and de Kadt [1988: 51-3] summarize evidence that death and disease are linked to rurality, low access to land, and shortage of nonfarm assets in several countries.

Yet we are, frustratingly, unable to identify -- let alone to interrupt -- causal chains from poverty to ill-health, except via undernutrition. Partly, this is because demand for health-inducing services is variable, by choice, among people, even poor people, with different trade-offs between (say) longevity and enjoyment while alive. Partly, it is because supply of such services varies within countries in ways very imperfectly linked to poverty: urban-rural differences in age- and sex-specific mortality rates are substantial [Lipton and Ravallion 1993] and strongly related to access to services [Lipton 1977]. Further, there exist "equal-opportunity diseases" from which income, and all it can buy, provide no protection. Finally, a study from Colombia [Rosenzweig and Schultz 1982: 58] indicates that urban (though not rural) "public health institutions are substitutes" for the health knowledge and child-core associated with education (and non-poverty).

An excellent study [Behrman and Deolalikar 1988] teases out the links connecting supply and demand for health to the main outcomes. Yet they find that variation in incomes, prices, and household characteristics such as mother's schooling are seldom associated with more than 10 per cent of interpersonal variation in health status [ibid.: 660]. In practice, poverty is linked as cause or effect to numerous variables -- large households with sib crowding, residence in unhealthy or remote places, highly seasonal work, periods of energy stress -- known to be themselves causes of illness or risk of death. Correctly specified simultaneous models may understate the direct and indirect health damage from poverty, just as crude bivariate correlations overstate them. Infection and disability are also causes of poverty via lost labour-time [for data see Lipton 1983: 11-15], and the official data measure the extent to which sufferers report a condition in the hope of cure, at least as much as the incidence of that condition. Where, as in Kerala, primary health care is relatively good, the proportion of casual working time "lost" to reported sickness goes up to 10-15 per cent, from the otherwise typical Indian figures of 4-6 per cent [ibid. : 13].

The emphasis in health policy on (a) pesticides, water quality, and other environmental hazards [Lipton and de Kadt 1988: 33-40]; (b) nutrition; and (c) AIDS have left almost neglected some areas of central interest for labour research in the area of poverty. (d) Until recently, almost no attention has been paid to the health (and financing) effects of the ageing of populations -- even among the very poor -- in Asia and Latin America. The upper boundary of "working age", especially in casual and informal work, is bound to rise. What can be done to mitigate the health effects on the poor? (e) A big and neglected problem in "poor people's health" comprises industrial, and even more agricultural, injuries, especially during wet weather and in peak seasons; added to this is the revival of TB and malaria. What innovations in the workplace (from scorpion boots via DEET to simple retiming of operations) could reduce the damage?

Lanjouw and Stern [1991] provide one of several careful micro-studies to show that poverty risks -- and risk of decline into poverty -- are strongly associated with illness. In their Indian village, households with land and a fit adult male were at much smaller risk of poverty than were other households. Of course the causation is mutual. But labour-related policy, and research against poverty, need to attend more to health issues.

2.9 Variation in space: rural-urban and other issues

It is useful to structure the analysis of poverty in LDCs into "rural" and "urban", to the extent that we can answer "Yes" to four questions.

First, do LDCs usually comprise urban and rural places that are clearly distinct, and in which most people reside and work? On the whole, the answer seems to be Yes [Lipton and Ravallion 1994]. Second, do levels of poverty differ systematically and substantially between rural and urban places? The answer is clearly Yes. For 13 developing countries with reliable data for the 1980s, rural/urban poverty incidence ratios range from 1.3 to 2.5 in nine Asian and Latin American cases, plus Indonesia 3.7, Ghana 4.2, Ivory Coast 4.6 and Kenya 6.0. These data understate rural-urban poverty differences [ibid.: 40-41], mainly because state-provided goods are more unequally distributed between city and village than are the goods normally included in assessing rural and urban poverty. ^{(Endnote 17)} They also ignore the much more serious rural impacts of fluctuations in poverty. Third, do rural and urban poverty differ in structure or type, relatedly to a specifically rural and/or urban physical or socio-economic environment? Clearly, this is so. The rural poor are more dependent on agriculture -- central to rurality -- than the rural non-poor in South and East Asia and in West Africa [Quibria and Srinivasan 1991: 49; Hill 1982; Delgado et al. 1991]. Finally, is poverty importantly related to interactions, equilibria, or imbalances between rural and urban (i) people or groups, e.g. consumers or producers; (ii) impacts of decisions by governments; or (iii) land, water, or other physical factors? This is still controversial, but the balance of evidence seems to me strongly favourable, especially in Africa [Lipton 1977; and see papers in Harriss and Moore (eds.) 1983 and Varshney (ed.) 1993].

Within rural areas, the poor are heavily concentrated in remote and unreliably watered places. In 1972-3, of the 56 rural regions -- each treated as (more or less) homogeneous in India's National Sample Survey -- all nine with poverty incidence above 60 per cent averaged below 1540 mm/year of rainfall and had below 17 per cent of farmland irrigated. Not one of the 6 regions with poverty incidence below 15 per cent (and only 4 of the 15 below 30 per cent) had rainfall below 1700 mm and irrigation below 19 per cent of farmland [Lipton 1992: 3]. By the early 1980s, across 58 rural regions, "a strong correlation between agricultural output per worker and daily wage rate is found" [Quibria and Srinivasan 1991: 72], and this also applies to all-India time-series [Gaiha and Spinedi 1992]. Economists should expect capital and labour mobility, and therefore should find all this surprising. Presumably the transactions costs of migration for the poor, and of capital mobility towards low-wage areas, have proved prohibitive. Certainly these findings underscore the evidence that (location in areas of) transient, specifically drought-induced, poverty is a major component of poverty overall.

Within towns, there is much quantitative geographical work (not discussed here) on the location of poverty, transport costs, etc. So far as I know, little has been done on the distribution of poverty incidence or severity among towns of different size or types. In Mauritania in 1989-90 only 10 per cent of households in "main economic centres" were poor, as against 40 per cent in "other urban areas", 49 per cent in "rural centres" and 61-65 per cent in "other rural" [Coulombe and McKay 1991].

Nor is a lot known about urban poverty by occupational structures. This may be a serious omission. For example, census and survey data for developing countries show that 8-15 per cent of urban populations, even in large cities, live mainly off income from (urban) agriculture. Is this -- as casual empiricism suggests -- strongly associated with poverty, female workers or gardeners, and casual labour?

2.10 Fluctuations, variations and compensations

We saw in Section 2.4.3 that much poverty is due to fluctuations in income and hence command over food. Because declines are from a barely adequate average, they push the victims into poverty. In labour markets, the poor are normally characterized by greater short-run, seasonal and year-to-year variability -- in real wage-rates, time-rates of unemployment, and age- and sex-specific participation rates -- than the non-poor. Especially for the poor, such fluctuations are not mutually compensating. In slack times (and places), reduced participation (downward shift of labour supply curve) tends to raise the real wage-rate, but this is outweighed by reduced employment (downward shift of labour demand curve). Thus employment, participation and the real wage-rate for the poory tend to rise together and to fall together [Lipton 1983].

Similarly variations in location or treatment (e.g. between city and village; progressive and "backward" regions; women and men; or persons of different caste or ethnic group) tend to saddle the poorer area or group with low wage-rates, employment prospects, and participation rates, i.e. to worsen the spacial distribution of income among the poor. Once again, most economists would predict that such spreads would "normally" be reduced by factor-market adjustment. Their persistence is due partly to informational issues; only to a minor extent to segregation or wage-discrimination against groups that over-represent poor people; and substantially to access discrimination, in which outsiders, low-caste persons, and women are associated with low-income tasks. In short, the summary of evidence in Lipton [1983: 23-38, 54-60, 69-84] still applies, though an up-to-date treatment would place more emphasis on search costs and other informational issues [Stiglitz 1988], on adjustment of labour disequilibria through non-labour markets [Binswanger and Rosenzweig 1981], and on general equilibrium [de Janvry and Sadoulet 1987].

The surveys, revealing the importance of fluctuations and transient poverty, seldom tell us why income fluctuates downwards among the poor or near-poor. There are three broad causes: life-cycle variation; decline in labour income, due to bad harvests (so smaller farm employers neither need nor can afford to hire workers), falling relative prices of labour-intensive products, or recession and adjustment; and random events, especially illness. Quite remarkably little research has been done on the sorts of rural work that are more -- or less -- vulnerable to these three sources of downward income fluctuation. What sorts of urban, farm, and RNFS activity are more readily "kept up" by (for example) a near-poor household comprising a pregnant mother, a young husband, and two small children? What if such a family is in a moderately drought-affected village, or a town affected by retrenchment? There is huge scope, here and elsewhere, for the sort of research in which ILO/WEP/IILS has shown absolute (not just comparative) advantage in the past two decades. Such research has much to contribute to our understanding of the causes, correlates and cures of poverty.

Endnote 6:
"Poor" means "having household income or consumption, per adult-equivalent, normally associated with inadequate nutrient (or energy) intake". No account is taken of possible economies or diseconomies of scale in consumption. For the USA [Lazear and Michael 1980], substantial economies of scale have been demonstrated even for the lowest income quintile. However, this is very unlikely to apply to low-income countries, because 70-80 per cent of poor people's income (consumption) is used for food purchases.

Endnote 7:
The increase, with deepening poverty, vanishes among the very poorest 5-10 per cent of households in most samples [Lipton 1983a: 43-45].

Endnote 8:
At the symposium, Buvinic pointed out that Latin American rural-urban migrations -- unlike those of most other developing countries -- featured a high female/male ratio. This point (e), therefore, may not apply in Latin America; more research is needed.

Endnote 9:
Only to a very small extent can this be explained by the fact that the poorest urban deciles have somewhat higher real income and consumption per person than the corresponding rural deciles.

Endnote 10:
Secondary-school completers (and above) in cross-section studies show a persistently higher rate of "usual status" or "long-term" unemployment than others [Dev et al. 1991 for India; Vandemoortele 1991 for Africa]; but this proxies, not poverty, but ability to live off family income while searching for secure or well-paid work. In cross-sections of workers below secondary completion level, time-rates of unemployment (and wage-rates and poverty) decrease with increased education.

Endnote 11:
This widespread observation (e.g. Hopper [1955]) that rural people increase RNFS work and income in agriculturally slack times, is quite consistent with the evidence against "distress diversification" (e.g. Dev [1991]; Unni [1991]), i.e. that localized fast RNFS growth follows fast, not slow, growth of local farm output.

Endnote 12:
The argument is complicated by Dasgupta's assumption that, for the big farm-employer to maximize profit, he has to pay an efficiency wagerate -- maximizing his net return from the worker, allowing for the impact of the wage-payment on her nutrition and hence work -- to each work-seeker (landless and deficit-farm landed alike).

Endnote 13:
Dasgupta (pers. comm.) stresses that the one-period sketch of his model, reported above, does not do justice to the full, two-period formulation. On day one, initially "equal" landless workers present themselves for day-labour. At the efficiency-wage, some are rationed out of work. On day two and subsequently, these increasingly undernourished workers become progressively less prepared to accept (harder) day-labour in preference to (less well-rewarded) residual work on begging, CPRs, etc.

Endnote 14:
See Behrman [1991], who proves that lower (and more accurate) estimates of CIEs are normally obtained by (a) proper measurement, involving, among other things, allowing for meals eaten at work (largely by the poor) or served to workers (largely by the non-poor), (b) proper instrumenting of total expenditures -- necessary because food expenditure is so large a part of them that their residuals and its own (and therefore, to some extent, those of energy intake) usually show strong positive correlation.

Endnote 15:
Despite a modest amount of recent protein revisionism, it remains the consensus that the overwhelming majority of sufferers from PEM (protein-energy malnutrition) would get sufficient protein, if only their current food intake were boosted by raising energy levels. Micronutrient deficiencies are of widespread importance, but the most cost-effective and sustainable treatment is seldom (as it often is for PEM) to increase poor peoples incomes.

Endnote 16:
Most research, however, indicates that dearer food harms the poor on balance in the short-to-medium term; even among the rural poor (even in most of Africa), a substantial majority comprises net food buyers.

Endnote 17:
At this symposium, Dr. Jamal drew attention to his work with Dr. Weeks, illustrating big falls in urban formal-sector wage-rates -- and hence, it was suggested, in the urban-rural disparity -- in Africa in the 1980s. However: (a) there is no evidence that urban-rural disparity in mortality or in levels of living has fallen; (b) apart from trends in the informal sector (and in rural incomes), public-sector outlays on rural and agricultural objectives have fallen, offsetting the undoubted decline in officially-sponsored distortions of prices and exchange rates against rural people [Lipton 1990: 75]; (c) continuing (slow) rural-urban migration in Africa, despite negative or slow growth in real income per person, casts serious doubts on claims that urban bias has been substantially corrected overall.