Jump to ContentJump to Main Navigation
African Economic DevelopmentEvidence, Theory, Policy$

Christopher Cramer, John Sender, and Arkebe Oqubay

Print publication date: 2020

Print ISBN-13: 9780198832331

Published to Oxford Scholarship Online: July 2020

DOI: 10.1093/oso/9780198832331.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (oxford.universitypressscholarship.com). (c) Copyright Oxford University Press, 2021. All Rights Reserved. An individual user may print out a PDF of a single chapter of a monograph in OSO for personal use. date: 15 June 2021

Working Out the Solution to Rural Poverty

Working Out the Solution to Rural Poverty

(p.188) 8 Working Out the Solution to Rural Poverty
African Economic Development

Christopher Cramer

John Sender

Arkebe Oqubay

Oxford University Press

Abstract and Keywords

Ideas about poverty and poverty reduction policy are clouded by misleading measures and unreliable evidence. National statistical organizations (NSOs) are under-resourced and the collection and dissemination of data are compromised by political pressures. Allegedly pro-poor policies have had an inegalitarian impact in rural Africa. Conventional views expecting that support spread across smallholder farmers will reduce poverty are based on skewed evidence and ideology. Large farms have a bigger effect on poverty reduction through labour markets. ‘Gold standard’ poverty measures based on consumption surveys are unreliable and misleading. Composite indices are even less useful. There are better ways to assess deprivation. The poorest typically live in small households with few men in them. Women and children in these households suffer the risks of teenage pregnancy; they risk undernutrition because of a monotonous and undiversified diet; they can acquire hardly any basic consumer wage goods; they depend on access to wage employment opportunities.

Keywords:   poverty, national statistical organizations (NSOs), poverty reduction, smallholder farmers, composite indices, labour markets, undernutrition, teenage pregnancy, wage goods

8.1 Introduction

What people think they know about poverty is clouded by misleading measures, unreliable data, and ideology masquerading as common sense. Powerful common-sense ideas—that poor people are rural, that they live in large, female-headed households, and most of them live on small farms—appear to be confirmed in official surveys generating the raw data for a string of poverty measures and indexes. However, most of these indexes are confused; the surveys have serious shortcomings; and the ‘rhetorical commonplaces’ of poverty in Africa are wrong. It is no wonder too little poverty reduction has really been achieved in recent decades.

This chapter explains why we argue in this way. We begin by discussing issues of measurement and definition. We then try to blow off the common-sense chaff from the more useful grain of what we term stylized facts: facts that are broadly supported by carefully collected evidence and provide policy-relevant ideas about the characteristics of extreme poverty. We show that some, but not all, of these stylized facts are uncontroversial and similar to the conventional wisdoms used to identify the poor in Africa. The poorest people live in female-dominated households (not ‘female-headed’ households) and lack regular access to support from an adult male; the adult females and children in these households probably have not received an adequate education; they are severely threatened by the risks of teenage pregnancy; they also rely on a monotonous and undiversified diet; and the poorest people have been able to acquire barely any basic consumer wage goods. More controversially, we also show that these households are relatively small; and we stress the stylized fact that the poorest people in Africa depend on low-paid wage labour (not self-employment on their own farms) for their survival.

Policymakers and researchers must question the data and arguments underpinning proposals to reduce poverty in Africa. Part of the problem is that national statistical organizations (NSOs) are under-resourced and often under political pressure. There is nothing uniquely African about this: it is a common issue elsewhere and historically. We give some recent examples of difficulties experienced by African NSOs, focusing on problematic census and agricultural data.

(p.189) This chapter provides an overview of evidence about the impact of policy interventions on the poorest people in Africa (Section 8.3). We then try to explain the skewed impact of many very fashionable, but obviously failing, poverty reduction policies. We show how bad data and hegemonic ideas about how to reduce poverty in Africa are intimately connected. Our conclusion is that commonly advocated policies to reduce poverty and increase agricultural output are likely to continue to have disappointing results. The recommendations from fashionable randomized control trials (RCT) are also very unlikely to make more than very modest inroads on poverty.1 What these RCTs actually show is often stretched into far larger claims than the method justifies.2 At their best, RCTs can be useful in answering limited types of question; but they are not designed to generate understanding of distributional dynamics or the political economy of production, accumulation, and labour, let alone to generate effective policies to make structural differences to the conditions of development and poverty reduction.3 It has also been shown that some of the flagship RCTs used to promote a much wider use of the method have not conformed to the standards of research protocol that they set out; and that their claims to statistical significance are often unfounded.4 The policy implications that we draw from the stylized facts on rural poverty are elaborated through other chapters in this book (especially Chapters 4, 5, 7, 9, and 10). The single most important mechanism in poverty reduction (globally and in Africa) is, as it has been throughout the history of capitalism, labour market development—the growth of demand for labour in higher-productivity activities and the hard-won achievement of higher wages and some legal protections for employees. We argue that a greater difference can be made—including in the relatively short term—by prioritizing policies and investments designed to promote a rapid rate of growth of labour demand in particular types of activity (in rural and urban areas) than by pouring resources into safety nets that invariably fail the poorest or by ‘nudging’ the behaviour of poor people themselves through small interventions based on very sketchy empirical work.

Although a small number of countries have failed to make any progress at all in reducing monetary poverty rates since 2002—mainly small countries classified by the World Bank as ‘Fragile States’5—boasts that poverty is falling in Africa are easily made. Some widely published data do support a less gloomy, if not a rosy, interpretation of trends in the prevalence of poverty (Figure 8.1). Although the (p.190) prevalence of extreme poverty in sub-Saharan Africa appears to have fallen a great deal between 1993 and 2015—from 59 per cent to 41 per cent—we will alert policymakers to serious problems that limit the usefulness of this particular measure. We emphasize, in Section 8.5, the difficulties (and the costs) of identifying poor households by estimating whether or not their consumption per capita falls below the cut-off of an international poverty line.

Working Out the Solution to Rural Poverty

Figure 8.1 Percentage of people living below the ‘extreme poverty’ line in sub-Saharan Africa, 1993–2015

Note: Extreme poverty de­fined by the World Bank using 2011 PPP and $1.9/day poverty line.

Source: Author calculation using PovCalnet, World Bank (2019).

Boasts about poverty reduction will be contested by those who mistrust governments or are critical of their policies and wish to change them. Claims about trend reductions in poverty are also questioned by rigorous applied economists, such as Tony Atkinson, who insists that ‘any estimate—of level or of change—is surrounded by a margin of error. This is often lost from sight in public pronouncements, and it is important to convey to policymakers and other users that they are operating with numbers about which there is considerable uncertainty’.6 The broad aim of this chapter is to encourage policymakers to probe very carefully the arguments and available data underpinning fashionable proposals to reduce poverty in Africa.

(p.191) 8.2 Who Wants to Know? Struggles over Data

African political leaders such as President Kagame may be prepared to dispute statistics measuring poverty, but few are prepared, as Stalin was, to execute the nation’s most senior statisticians if data in a population census fail to support implausible claims about output growth.7 More generally, as Angus Deaton put it:

The state decides what it needs to see and how to see it. That politics infuses every part of this is a testament to the importance of the numbers; lives depend on what they show.8

The NSO in Malawi was subject to intense pressure after 2005, because it had become politically imperative to exaggerate the success of the Minister of Agriculture’s flagship farmer-support initiative, the Fertilizer Subsidy Programme. This political pressure was difficult to ignore because the Minister of Agriculture was also Malawi’s President, who was facing an impending election. This political pressure, intensified by fiscal dependence on donors sceptical about the benefits of subsidies, perhaps explains delays in the publication of the Agricultural and Livestock Census (2006/7), which inconveniently showed that maize output in Malawi was about 60 per cent below previous estimates provided by agricultural extension officers. The embarrassing census also showed that the Ministry of Agriculture had been allocating expenditures on the basis of hugely inaccurate administrative estimates of the total number of farm households and their regional distribution. The result was a skewed distribution of subsidized inputs to ghost farmers and to the Northern Region, which was ‘the swing voting region’ critical to the electoral success of President Mutharika in 2009.9

A better-known African example of the political sensitivity of the census involves Nigeria, where there has not been a population census since 2006. Estimates of the size of the current population range between 140 and 200 million.10 Political parties cultivating support in northern and southern Nigeria are continuing to obstruct and contest the production of more definitive statistics, because new and disaggregated estimates of the relative size of regional populations would have major resource (re)allocation implications. More recently, in Tanzania a political leader was arrested in November 2017 for publishing an analysis of anomalies in the statistics provided by the Bank of Tanzania and suggesting that the government had been manipulating (and exaggerating) gross (p.192) domestic product (GDP) growth rates.11 And in 2019 (an election year) the Head of Mozambique’s National Statistical Institute was forced by the President to resign after defending national census figures for the Frelimo-voting province of Gaza against what appeared to be inflated numbers in the electoral register.12

Africa’s political leaders usually allocate limited and fluctuating resources to NSOs, leaving the funding and much of the design of statistical investigations to external donors.13 Most NSOs in Africa are said to be ‘constrained by budget instability and a lack of autonomy that leave them vulnerable to political and interest group pressures’. It is also reported that: ‘ad hoc donor-funded projects generate significant revenue for statistics offices and individual NSO staff. Increasing take home pay by chasing donor-funded per diems via workshop attendance, training, and survey fieldwork is the order of the day. As a result, NSOs lack incentives to improve national statistical capacity … ’14 Labrousse concludes:

Dominated by external actors (private, public, supranational), African ‘statistical sovereignty’ is profoundly incomplete.15

The World Bank has published an index of national statistical capacity for sub-Saharan African countries. Among the reasons why this index is unreliable is that it excludes important indicators, for example, data about labour force surveys.16 Another index aims to measure the degree to which the websites of African NSOs follow best international practice in terms of coverage, accessibility, and quality of publicly accessible data.17 Nigeria’s score on this index might be regarded as surprisingly high in view of the absence of up-to-date population data; and Rwanda’s relatively good scores on both these indices may raise eyebrows in the light of recent controversies over the politicization of poverty data.18

The meaning of all the scores on these indices, including the high score and apparent superiority of Statistics South Africa—so fiercely defended against academic critics by its recently retired leader19—needs to be questioned and examined alongside other evidence. In five major South African surveys, interviewer cheating and other sources of error were widespread.20 The most politically embarrassing statistics, that is, the trends in South African inequality and real (p.193) wages, are difficult to analyse because of serious data quality issues and because the officially published series has been interrupted by breaks in the measurement process.21 The demographic impact of HIV AIDS in South Africa has also been politically sensitive; different sources of data provide very different estimates of the pattern and trends in new-born deaths, partly because ‘there is uncertainty regarding the degree of completeness of reporting for young children’.22 Earlier, President Mbeki’s obsessive AIDS denialism encouraged the Department of Health to refuse medical researchers access to survey data while this department published misleading statistics on trends in the epidemic between 2006 and 2008.23

8.3 Ignoring Pro-Poor Policy Failures

There is accumulating evidence that allegedly pro-poor policies have an inegalitarian impact in rural Africa. Aid officials and the political elite have publicly championed a variety of anti-poverty policies and social welfare expenditures; but most of the poorest 20 per cent of the population in sub-Saharan Africa (as defined by the World Bank) failed to benefit from any of these anti-poverty programmes between 1998 and 2012.24 An analysis for the period after 2012 examines African Development Bank and World Bank project aid to 17 African countries and also ‘reveals that aid does not favor the poorest. Rather, aid disproportionately flows to regions that hold more of a country’s richest people.’25

Data on education spending in Africa provide a particularly clear illustration of the degree to which large volumes of donor and state expenditure have been channelled to benefit the rich: the children of the rich, defined in terms of a Demographic and Health Survey (DHS) Wealth Index, are far more likely to gain access to higher levels of education; only about one in ten young people in sub-Saharan Africa are able to enter higher education, but hardly any of this small group of young people come from the poorest decile of households. Young people in poor households are particularly disadvantaged in some countries: in Ghana, for example, the poorest households only receive $16 for every $100 spent on rich households, while in Malawi poorest can expect to receive less than $10 for every $100 spent on the richest decile.26

Those countries having the most extreme pro-rich spending patterns usually show a consistent pattern of pro-rich public expenditure allocation at both secondary and higher education levels. By far the most extreme pro-rich bias in public expenditure on secondary schooling is found in Ethiopia (Figure 8.2). (p.194) Current public spending patterns on education in Africa generally allocate large amounts to higher education at the expense of the primary and secondary sectors; these spending allocations are clearly much more regressive (pro-rich) in some African countries than in others.

Working Out the Solution to Rural Poverty

Figure 8.2 Public spending on the richest decile relative to the poorest decile, by level of education, selected sub-Saharan African countries

Note: The darker shaded bars in Table (b) indicate countries where the richest:poorest ratio exceeds 1,000.

Source: Adapted from Ilie and Rose (2017), based on UNESCO Institute of Statistics (UIS) and DHS data.

By 2018, most African countries had established policies ‘to protect and promote the poor and the vulnerable’; but, on average, these ‘social safety net’ programmes only cover a tiny percentage of the population. For example, less than 3 per cent of the population participate in public works programmes and most safety nets provide hardly any coverage of the huge internally displaced populations in Africa. And a significant proportion of Africa’s social safety net expenditures benefit the richest quintile of households: Malawi’s Social Action Fund and Ghana’s Livelihood Empowerment Against Poverty programme are both good examples of benefit leakage to the richest households. A World Bank evaluation concludes that most of the poor in Africa have not benefitted from social safety net expenditure.27

Ethiopia’s Public Safety Net Programme (PSNP) is the largest in Africa outside South Africa and its impact has been evaluated several times, leading to some (p.195) rather positive conclusions.28 Despite the size of Ethiopia’s safety net programme, the bottom 10 per cent of rural households actually suffered declining consumption between 2005 and 2011—at an annual average rate of close to 2 per cent in real terms robust to the choice of deflator, while inequality as measured by a Theil Index, a measure sensitive to the share of the poorest, increased between 2000 and 2011.29

The beneficiaries of PSNP interventions to reduce malnutrition have been concentrated in selected districts (woredas), with predictably skewed nutritional results:

Ethiopia’s malnutrition rate could likely be substantially reduced by shifting some of the programs from the woredas with a high concentration of major programs into those with high malnutrition rates but no major programs.30

The PSNP ‘excludes—by design—at least 52 per cent of vulnerable Ethiopian households’.31 Econometric evidence suggests that this flagship anti-poverty intervention has had no effect on household dietary diversity or children’s height-for-age in participating households.32 Perhaps this is because the people benefitting from Ethiopia’s PSNP (about one tenth of Ethiopia’s population) were not selected by applying simple anthropometric rules or quantitative criteria.33 People have been excluded (graduated) from the PSNP for not supporting the political elite and also because ‘the criteria were subjective and no household data existed to support decision making’.34

This type of outcome is predictable whenever local officials beholden to politically appointed leaders are required to distribute scarce resources. The political imperatives underlying resource allocation in rural Ethiopia have been described by René Lefort and have also been illustrated by ethnographic research in seven different kebeles, which concludes that the ‘social protection programme is being implemented in a way to eliminate opposition and … entrench power of the existing elite’.35 Similarly, in Uganda, donor-funded technical reports on ‘how to select’ the beneficiaries of anti-poverty programmes have been brushed aside to pursue ‘vote-buying clientelism’, while in Tanzania the rollout of the Productive Social Safety Net and of earlier donor-funded schemes aimed at poverty reduction has been manipulated for electoral purposes and has served to strengthen the ruling party (the Chama Cha Mapinduzi; CCM).36

(p.196) It has been argued that in rural Zimbabwe the government intentionally adopted policies to achieve ‘Extreme scarcity and dependence on the state for agricultural inputs and food aid … [to] create an enforced loyalty that supports regime perpetuation.’37 The extreme bias in the pattern of distribution of food aid to the rural poor by the rural cadres of Zimbabwe African National Union—Patriotic Front (ZANU–PF) has been ‘discretionary, clientelistic and exclusionary’.38 More recently, Emmerson Mnangagwa spearheaded the Command Agriculture Programme to allocate resources to securocrats and other politically important beneficiaries in rural areas.39 Both the President and the Minister of Agriculture are said to ‘have demonstrated a willingness to (ab)use food for political gain in Zimbabwe’.40

If policymakers and foreign advisors ignore the prevalence of violent struggles to maintain political power and to monopolize access to state resources in rural Africa, their proposals for interventions to reduce poverty are likely to be irrelevant. Resources ostensibly allocated to anti-poverty programmes, smallholder development, and food distribution are routinely diverted—to subsidize accumulation by larger farmers and/or conspicuous consumption by cronies.41 The policy remedies proposed by the Food and Agriculture Organization (FAO), for instance, are unlikely to benefit the poor, because they so studiously avoid any discussion of African political realities: instead, FAO resorts to platitudes about the need for ‘strong political commitment’ as a precondition for ending extreme rural poverty.42 Proclamations from Rome, together with the conventional surveys of rural household poverty (criticized in Section 8.5) and the increasingly complex multidimensional indicators of poverty promoted by other donor agencies, can readily be manipulated or ignored by administrative elites within Africa.

8.4 Conventional Strategies for Rural Poverty Reduction

Policymakers are repeatedly advised to focus support on smallholder farmers not only because of their allegedly superior yields per hectare but also because of a belief that, when African smallholder output increases, poverty will fall.43 We discuss the data on the yield of small farmers in Chapter 9; here we focus on poverty reduction mechanisms.

John Mellor, the founder and first Director of International Food Policy Research Institute (IFPRI) and a Chief Economist for United States Agency for International Development (USAID), has argued, since 1961, that increases in (p.197) smallholder output ‘generate rapid, equitable, geographically dispersed growth owing to agriculture’s substantial labor intensive linkages with the non-farm economy’. The rural non-farm sector is a key source of income for the landless or semi-landless poor and opportunities in this sector are positively linked to the growth of small farmer income, because ‘smallholders spend substantial portions of increased income … on employment-intensive, nontradable goods and services produced by the rural non-farm sector’. In contrast, according to this argument, much weaker rural consumption linkages are generated when large farmers’ output and income grows; these farmers have ‘urban-orientated consumption patterns’.44

Anyone who has spent time in rural Africa will agree that many poor people do depend on activities other than farming, although the poorest usually work on farms as casual agricultural wage workers rather than in the non-farm economy. But trading and village centres as well as small rural towns often provide poor rural women and men with opportunities for part-time and unskilled non-farm employment. These include many opportunities for low-paid wage employment: in construction and transport, food processing and packaging, carpentry, alcohol preparation, hairdressing, sexual services, guard work, and, of course, as domestic servants in private households. In some rural areas the main impetus for the growth of these local opportunities for wage employment may, as Mellor and many others suggest, be an increase in the agricultural revenues earned by scattered small famers. But elsewhere very different causal mechanisms have been much more important, with dramatically better outcomes for the poor.

For example, when public sector expenditure constructs a district hospital, a health clinic, a teacher training college, or a new irrigation structure, then salaried staff cadres, together with the more numerous wage workers employed as cleaners and security guards in these institutions, are likely to provide the breakthrough to a new level of local demand for wage-labour-intensive goods and services. The consequence may be increased competition for unskilled rural labour between, for example, the construction and the agricultural sector and a tightening of the labour market, leading to a rise in seasonal agricultural wages and a reduction in rural poverty. This link between public sector investment in rural areas, tightening labour markets, and rising agricultural wage rates is well documented for rural India.45 In other rural areas, the impetus may arise when large-scale agribusinesses invest and create a new group of formal on-farm unskilled wage employees, who then stimulate a surprising array of non-farm employment. This is precisely what happened in Ziway, a town in Ethiopia that has grown very (p.198) dramatically since 2001 in the shadow of a few very large-scale floricultural and wine-producing firms.46 Figure 8.3 illustrates some of this growth: part of the (p.199) massive expansion in the area covered by greenhouses by 2018 can be seen in the bottom right-hand corner; and the degree to which the original (blacked-out) area of the town has been dwarfed by new residential areas constructed to the west and to the north is also obvious. Google’s coloured satellite pictures make it clearer that the density of buildings in the original area covered by Ziway in 2001 has been transformed, with every available scrap of land converted to provide rental accommodation for migrant wage workers.

Working Out the Solution to Rural Poverty

Figure 8.3 Mapping agribusiness and workers’ accommodation in Ziway, 2001–18

There are other examples of this type of poverty-reducing dynamic. In northern Senegal, where five large agribusiness estates have been established since 2003, there was rapid growth in the incomes of agribusiness wage employees.47 In Tanzania, several emerging urban centres (EUCs) became ‘hotspots for rural migration’, often providing a range of services for large-scale agribusiness managers and their workers. When the population of salaried employees increased in the 1990s, access to schooling, health, and financial services improved—and non-crop-related wage employment opportunities for unskilled migrants from nearby and distant rural areas expanded.48

Strategies designed to increase the output of all small farm households will result, according to conventional theory, in new wage employment opportunities for people who are much poorer, because the beneficiaries will not only need to hire additional seasonal agricultural labour inputs at harvest time, but will also devote their incremental income to additional consumption of rurally produced labour-intensive goods and services. But we are not convinced about the strength of this indirect mechanism and its likely impact on poverty reduction. Even if some farmers do achieve a marginal increase in their consumption of purchased goods and services, only a few will have the capacity to use more hired labour inputs. The well-connected individuals farming on a relatively large scale will account for much of any growth in consumption stemming from state and non-governmental organization (NGO) interventions.49 These individuals are also likely to purchase a high proportion of all hired labour days: the largest size tercile of smallholder farmers in Tanzania, for example, are much more likely to use hired labour than the smallest size tercile, and farmers who have acquired larger extents of irrigated vegetables use more hired labour than farmers with access to less irrigated land.50 Where employers are oligopsonists in local labour markets, then the bargaining power of and the daily wage rate paid to their agricultural wage workers may be very low.51

If, instead, workers could find employment on much larger-scale farms, they would reduce the risk of failing to find daily employment outside peak seasons (p.200) and, as our own fieldwork over the past 30 years in Ethiopia, Mozambique, South Africa, and Uganda has shown, they would typically benefit from better wages and working conditions than those provided by smaller farmers.52 Policy initiatives focused on improving the real annual wage of rural wage labourers—rather than on increasing the profits of ‘small’ producers—are rarely discussed or advocated in Africa by agricultural economists or donors. The need to monitor the real wages of agricultural wage workers as an indicator of poverty trends in Africa is hardly ever acknowledged.

There is no a priori reason to suppose that the non-farm linkages generated by the expenditures of rural wage workers, especially if they are paid a better wage, will be weaker and less poverty reducing than the linkages generated by the mini-farmers who struggle to survive by cultivating a few plants. On the contrary, we have good evidence that even the very poorest casual agricultural wage labourers in Africa aim to purchase a few locally produced consumer goods—such as a new dress, a table, bed, or cupboard.53

The assumption that poverty-reducing linkages can only stem from small farm development is also unconvincing because it has relied so heavily on strange and slippery definitions of the ‘small’. Available data on the size distribution of farms in Africa is inconsistent and unreliable, but in many countries the median operated farm area per household is probably below one hectare.54 So it is surprising that Mellor and his colleagues regard some farmers cultivating as much as 75 hectares as small. In Africa, they choose to define the top end of the distribution of all family farms (i.e. the ‘commercial’ and ‘middle’ farmers—with reliable access to water) as ‘small’. In Ethiopia, for example, these authors adopt an idiosyncratic ‘small’ category that includes farmers who cultivate farms as large as 33 hectares.55 Such quirky definitions can be useful when boasting that flagship policies supporting smallholders are pro-poor.56 But it is disingenuous and extremely unhelpful for policy officials.

8.5 The Fog of Poverty Reduction: How Not to Focus on the Poorest

Angus Deaton argues that the ‘gold standard’—the World Bank’s central monetary measure of poverty—is ‘inherently unreliable’.57 He points out that it continues to be difficult to explain the large discrepancies between measures of consumption derived from household surveys and from national income statistics. (p.201) He also argues conventional methods of poverty estimation are ‘clouded in a fog of technicalities’.58 We will try to dispel this fog in our discussion of the most widely used conventional method.

The ‘traditional basis’, ‘the core indicator’ or the gold standard for identifying the poor, is the detailed information on consumption per household member collected in household surveys, especially the Living Standards Measurement Surveys (LSMS). On the basis of this costly, rather unreliable, and often dated information, everyone who lives in a household failing to achieve some minimum (cut-off) level of consumption per day is defined as ‘poor’. If an individual’s daily consumption expenditure is below the cut-off, defined by an international poverty line (set by the World Bank at US$1.90 in 2011 purchasing power parity (PPP) dollars), they suffer from ‘extreme poverty’ (see Figure 8.1).

It is not considered a good idea to use official exchange rates to compare the standard of living between countries, especially between low-income and OECD countries. At the official exchange rate, one dollar (for example) will usually be able to purchase far more goods and services in Dar es Salaam—cups of tea and haircuts—than it could in New York. A more ‘correct’ exchange rate would make one dollar worth the same, that is, have the same purchasing power in the low-income and the OECD country. The PPP rate is an estimate of the ‘correct’ exchange rate in this sense and this estimate requires the collection of millions of prices of comparable items all around the world. There is probably a 25 per cent margin of error on either side of PPP exchange rates.59

The mix of methods and countries used to select the US$1.90 poverty line—originally a ‘dollar a day’—appears to have been strongly influenced by the Bank’s own public relations requirements and a desire to achieve consistency with previous estimates, rather than by examining the lived experience of the poor. An earlier version of the current international poverty line was slightly below $1.90, but the World Bank decided to round this number up.60 Angus Deaton labels the level of daily expenditure that has been selected as the poverty line cut-off ‘a miserable pittance’, while Jean Drèze points out that (even above the cut-off) the amount a ‘non-poor’ person could afford to spend per month on health, for example, would barely cover the cost of purchasing a single aspirin; their total monthly expenditures could not possibly ‘meet the requirements of dignified living’.61 It has been suggested that the apparently arbitrary choice of such a low cut-off line was influenced by a desire to applaud the success of governments and international agencies in achieving an impressive reduction in poverty.62

(p.202) In sub-Saharan Africa the consumption level in 2015 of a huge percentage of the population (41 per cent) fell below this cut-off line and all these people are defined by the World Bank as living in extreme poverty. In 11 sub-Saharan African countries ‘more than half of the population live in extreme poverty’.63 No less than 86 per cent of the rural population of South Sudan is, according to the latest estimate (for 2016), living in extreme poverty.64 A large proportion of African households are consuming at levels only very slightly above or below the level of the international poverty line.65 One implication is that a recalculation of the PPP rate by economists in Pennsylvania, or a proposal in Washington for a very small increase/decrease in the poverty cut-off, will immediately shift millions of Africans into or out of poverty. For example, a tiny increase in the international PPP poverty line (from $1.90 to $2.00) would add 100 million people to the extreme poverty headcount.66

The World Bank argues that ‘In 2015, more than half of the global poor resided in Sub-Saharan Africa’ and forecasts that by 2030 about 87 per cent of the global poor will live in sub-Saharan Africa.67 But the proportion of global poverty that is accounted for by sub-Saharan Africa (as opposed to South Asia) is determined by the Bank’s questionable decision to set the international poverty cut-off at a relatively low level. Whether those living just above this line really are now ‘non-poor’ depends on whether or not we accept that someone can live decently when their level of consumption is so low. Policymakers should also be aware that trends in poverty reduction can easily be exaggerated when household survey questionnaires or methods change between surveys and when the available consumer price indices do not accurately reflect the changing consumption basket of very poor households. For example, poverty reduction has been exaggerated in Uganda for both of these reasons; and it has been exaggerated in Rwanda by manipulating the Consumer Price Index.68

Any optimistic assessment of trends in poverty also depends on whether or not we accept claims that household surveys reliably capture the consumption of individuals at the tail end of the distribution, for example, the meagre consumption of migrant construction workers, squatters, and agricultural wage labourers. Quite apart from the problem of these excluded individuals at the tail end, even officially included ‘poor’ individuals have not enjoyed the consumption gains reflected in the aggregate trend for sub-Saharan Africa as a whole: growth rates of consumption for the ‘poor’ were negative in Uganda, Zambia, Ghana, Niger, and South Africa between about 2010 and 2015 and, over roughly the same period, were barely positive in Ethiopia and Rwanda.69 Consequently, it is not (p.203) surprising that the share of national income currently received by the poor is minuscule. The latest available disaggregated data on the income share of the poorest (defined as the lowest 20 per cent in Figure 8.4) show large differences between African countries, but also that the income share of the poorest (p.204) 20 per cent is always disproportionately small; it is below—in some countries way below—10 per cent.

Working Out the Solution to Rural Poverty

Figure 8.4 Income inequality in sub-Saharan Africa (shares of the lowest and highest 20% of population)

Source: World Bank PovCalnet (2019).

8.6 Alternatives to the ‘Gold Standard’?

We have discussed some of the problems of using conventional consumption per capita and cut-off lines to analyse poverty. Many development economists are aware of these (and other) difficulties with LSMS data in Africa. Nonetheless, when economists advise policymakers on how to select the ‘best’ method of identifying the poor the only selection criterion continues to be accuracy in predicting household per capita expenditure.70 It is not necessary for policymakers to accept this definition of the ‘best’ indicator: they can turn to well-established alternative indicators of poverty or socio-economic status, such as wealth or asset indices based on DHS data.

Asset indices have been used rather successfully in Africa to identify rural households containing less educated people with more health problems, higher fertility, and greater risk of child stunting than other households.71 They can also offer other important advantages to policymakers: they are usually constructed from a few dichotomous indicators of asset ownership and housing characteristics (e.g. does the house have access to a radio, improved sanitation, or have an earth/dung floor?). Answers to these questions do not require respondents to perform unrealistic feats of recall, and can be collected more cheaply, more quickly, and probably with less measurement error than the huge number of questions about past and current expenditures in the LSMS. (The costs of LSMS in Uganda and Tanzania amounted to about US$400 per surveyed household.72)

The accuracy of responses to questions about ownership of inanimate items of furniture, a bed for example, can be confirmed visually by enumerators. But responses to questions about other key assets in rural Africa—the number of chickens and animals owned—are always much more unreliable.73 And there are other important problems faced by all attempts to construct these asset indices: if the aim is to use the asset index as a proxy for wealth, researchers must either use depreciated values or make arbitrary choices about how to weight different assets; but the relevant time series of prices to estimate depreciation is unlikely to be available in poor rural areas. Besides, it will always be difficult to account for differential quality: it may be obvious to all enumerators that an Apple iPhone is not the same thing as the most common mobile phone in Africa (a Tecno T201—costing about US$15), but less obvious that the quality of one cooking utensil is very much lower than another (home-made) item, or that the (p.205) quality of access to sanitation varies dramatically, for example, between those with access to piped water 24/7 and those with only intermittent access.74 Also, huge seasonal price variations have been recorded for rural assets—in Tanzania, for example, the price of a bicycle can more than double in the dry relative to the wet season.75

The theoretical and practical problems of how to justify the selection of the specific items to be included in an asset index—and of how to weight these items—are often side-stepped in published research on rural poverty. Most studies are content to follow conventions and procedures established by earlier social scientists. Researchers admit that there is little underlying theory to support the choice of variables to include when using statistical procedures—such as the principal components analysis (PCA) used by the DHS.76

Besides, unweighted indices of socio-economic status have often been found to perform just about as well in identifying low socio-economic status rural households as the indices constructed using PCA to estimate weights.77 Since many policymakers may not be inclined to unpick the meaning of the weights used or to query the arguments for including one asset rather than another, PCA (and other statistical techniques such as multiple correspondence analysis) would seem merely to lend a spurious aura of scientific, quantitative precision to conventional policy prescriptions.78

A bewildering smörgåsbord of implicitly or explicitly weighted indices and recondite definitions of poverty now confront policymakers in Africa. Apart from asset/wealth indices and the ‘gold standard’ monetary measures, a mixed assortment of new non-monetary indices of poverty has been added to the buffet table, including indices that make brave attempts to quantify levels of ‘empowered decision-making’ such as the Relative Autonomy Index (RAI), providing a direct measure of women’s motivational autonomy. Another, though less fashionable, non-monetary index facetiously claims to track the quality of governance in public institutions—the Gross Toilet Index.79 Some poverty experts mix and match, using these novel indices in combination with the increasingly fashionable Multidimensional Poverty Index (or other ‘mash-up indices’) to report, for example, the percentage of women ‘deprived in at least three dimensions’, or the proportion of children deprived in three to six dimensions.80

(p.206) One noteworthy feature of the Multidimensional Poverty Indices is a failure to include any indicators of the fundamentally important determinants of the standard of living of rural children and adults, that is, their wages and working conditions or, in the case of children, their exposure to teenage pregnancy.81 The World Bank has made an effort to explain its failure to include any employment indicators in its 2018 Multidimensional Poverty Index, but inconsistent and unfounded arguments citing relevance and data quality smack of ideology:

Employment is not part of the multidimensional poverty measure presented here for two reasons. First, many of the frequently used indicators of employment in high-income countries, such as … wage employment, are not as relevant in low-income countries, which have very different labor market structures … Second, whatever relevant indicators of employment exist, these indicators are not available or not sufficiently harmonized in the different surveys considered here.82

Since 2009, the World Bank has promoted an index of development that is sensitive to inequality of opportunity, that is, an index that can take account of the fact while average access to education or maternal health services for women in Africa has been improving, some (rural) women have not been able to benefit very much from expanded coverage. But this Human Opportunity Index (HOI) is very highly and significantly correlated with the United Nations Development Programme’s (UNDP’s) Human Development Index (HDI); and the Bank’s Multidimensional Poverty Indices produce a similar global poverty profile to the monetary ‘gold standard’ measures based on the $1.90 poverty cut-off.83 Our conclusion is that many of the growing number of newly minted poverty indices are redundant—they may appear to satisfy NGOs’ demand for non-monetary, more ‘human’ development measures, but they do not get us closer to identifying the realities of the economic lives of the poorest people and they do not help in sharpening the focus of policymakers on these poor people.

8.7 The Most Vulnerable People: ‘Common Sense’ and Uncontroversial Stylized Facts

Nicholas Kaldor used stylized facts to criticize theories and explanations published by neoclassical economists; he regarded them as guilty of producing theories based on assumptions that were not even approximately true and, therefore, could not justifiably be used for purposes of explanation or policy analysis.84 Something (p.207) similar affects much economic analysis of poverty and of contemporary African economic development. Most economists agree on several conventional wisdom propositions about African poverty. In this section we begin by listing and questioning the policy relevance of three of these popular, conventional propositions about poverty. We then introduce a selection of stylized facts about poverty—uncontroversial facts that are much less likely to mislead policymakers. Finally, we introduce a particular stylized fact that has too often been ignored or dismissed as irrelevant for understanding broad trends in poverty or making policy but that we argue is at the very core of the problem.

Conventional Wisdom I: Poverty is Rural

In most African countries most of the people who are poor live in rural rather than urban areas. It makes little difference how poverty is defined—with a money metric, an asset index, an anthropometric cut-off, or a mix of multidimensional indicators: rurality dominates the poverty profile. The percentage of the poor who live in rural areas in the 2010s is about three times greater than the percentage of the poor who live in urban areas.85 About two thirds of the total population of sub-Saharan Africa live in rural areas and in some countries, for example, Burundi, Uganda, Malawi, Niger, South Sudan, and Ethiopia, more than 80 per cent do so.

But this does not help policymakers to design very precise or effective interventions.86 A geographically defined strategy claiming to benefit ‘rural’ households will, for political economy reasons already discussed in Section 8.2, probably exclude the poorest households. Besides, a growing number of rural households in Africa’s rapidly differentiating countryside are by no means poor; they operate medium and large-scale capitalist enterprises.87

The policy relevance of a simple rural–urban distinction is also questionable for another reason: the boundaries between the rural and the urban are shifting very rapidly in Africa and are often contested.88 Many non-poor households are falsely classified as ‘rural’ when they live just outside the dated and arbitrarily drawn administrative boundaries of a city.89 On closer examination, they are primarily engaged in urban occupations and well integrated into urban economic activities, including, in some cases, real estate speculation. If relatively prosperous urban people can make false claims to be living in ‘rural’ households with customary (p.208) land rights, then the poorer rural households may again be at the back of any queue for allocating valuable resources.

The rural–urban gap in consumption per capita is, like the headcount rate of poverty, much larger in some countries than others. For example, Integrated Surveys of Agriculture (LSMS-ISA) suggest that the level of consumption per capita in urban Tanzania is 2.2 times higher than in rural Tanzania, while in Uganda urban consumption per capita is 3.3 times rural consumption per capita.90 Non-monetary poverty indicators can also provide evidence of inter-country differences in the urban–rural welfare gap: for example, while the under-5 mortality rate (U5MR) is very high in rural Sierra Leone, the urban mortality rate is not much lower; but in Nigeria the urban rate is much lower than the rural rate; in Kenya, the rural–urban gap has already disappeared and the urban U5MR is now actually higher than the rural rate.91 While these variations are important, average data for a group of 23 African countries can be used to illustrate the general disadvantages of rurality (Figure 8.5). This figure highlights the huge disadvantages faced by rural women, especially in their ability to achieve adequate education and antenatal care.

Working Out the Solution to Rural Poverty

Figure 8.5 Urban and rural inequality: mean comparison for selected variables (for a group of 23 sub-Saharan African countries)

Source: Stifel et al. (2018).

(p.209) Geographical inequality is only one form of inequality that can be used to investigate disparities in standards of living and to design policy, but it is conventional common sense to focus on geography and on the mapping of administrative units, rather than political economy and class analysis. The widely accepted view is that geographical mapping of poverty indicators can make a major contribution to the design of poverty reduction strategies.92 But there is good evidence that the pattern of U5MR, as well as other indicators of risks to health, cannot be explained in terms of rural or urban residence. Instead, much if not all of the urban health advantage can be explained by socio-economic factors such as household wealth and maternal education.93 For example, urbanization does not appear to have a direct and positive association with child nutrition outcomes and diets. When socio-economic controls are introduced into linear probability regressions, the results suggest that the rural–urban child nutrition gap does not remain significant; instead of geography, ‘the key nutritional disadvantage of rural populations stems chiefly from social and economic poverty’.94

Influential populists have claimed for decades that the poor will benefit if interventions are rebalanced to favour rural rather than urban areas, but policymakers face a much more difficult task—to reduce unacceptable differences in the standard of living within rural and urban locations.95 The broad claim that most of the poor in Africa are young is also not very useful for (or surprising to) policymakers; it is also true by definition because children and young people account for such a very high proportion (about 60 per cent) of Africa’s total population.96 The scatter-gun targeting of the poor as ‘youth’ has fuelled spurts in expenditure on tertiary education and training that, while claiming to be pro-poor, have mainly benefitted the children of richer parents and very few of the poorest young people. It is also driven by a supply-side fantasy that more education will lead neatly to more growth and employment, another nugget of conventional wisdom contradicted by the evidence.

Conventional Wisdom II: The Poorest People Live in Large Households

There is a long tradition emphasizing the negative social consequences of excessive fecundity among poor women; and these views have often fuelled regressive policy interventions and moral panics.97 One of the founders of the London School of Economics believed that children living in large households in Britain (p.210) at the end of the nineteenth century risked poverty because of irresponsible breeding by an inferior type of ‘thriftless and irresponsible’ immigrant Catholic and Jew.98 In fact, during the nineteenth century the poorest decile of households—paupers and vagrants—lived in very small households. Unskilled labourers lived in much smaller households than richer classes.99

In Africa, larger households are conventionally believed to be more vulnerable to poverty. For example, an analysis of poverty in Uganda based on LSMSs concludes that ‘the chronic poor have relatively large households … Those that have never been poor have small households.’100 But the view that large households are more likely to be poor than other households leans towards tautology.

Households are defined as poor because household per capita expenditure/consumption is below the cut-off. If the denominator (i.e. the size of the household) is large, it is not a surprising arithmetic result that the incidence of poverty is higher in larger households. It also follows arithmetically that a small reduction in the size of the denominator as a result, for example, of the death of a new-born baby, could immediately raise per capita household expenditure above the poverty cut-off.

Whether or not a household is classified as poor is obviously influenced by the accuracy of and the methods used to count the number of household members, that is, survey estimates of the size of the denominator. The problem is that these estimates are, in many African contexts, unreliable. They continue to rely on an a priori standardized, narrow, and inappropriate definition of ‘the household’ and its ‘residents’, despite decades of vigorous criticism showing the difficulties of using conventional (residential) definitions of the household in studying rural Africa.101 There is no coherent account of how survey enumerators should delimit the boundaries of the household and the pattern of visits by enumerators that may be required: ‘little attention has been paid to the issue of what “household” means in these surveys: how it is defined for data collection purposes and what the definition implies for the analysis and interpretation of results’.102 Moreover, ‘The probability of finding no one at home in a one person household is larger than in a multiple person household. Small households are therefore likely to be underrepresented in the survey.’103

(p.211) Although poverty has often been associated with large households on the basis of household expenditure data, very different conclusions are reached when poverty is analysed using a wealth index. For example, DHS survey data show that household size in the poorer wealth (not expenditure) quintiles in 2011 was smaller than in the richer quintiles in Uganda and Ethiopia. In South Africa too, the poorest decile of households are not significantly larger than other households, if poverty is defined in terms of an asset index rather than expenditure per capita.104

Historical arguments support the view that very small households face greater risks. The most distinguished historian of poverty in Africa points out that: ‘In several African languages the common word for “poor” … implies lack of kin and friends, while the weak household, bereft of able-bodied male labour, has probably been the most common source of poverty throughout Africa’s recoverable history.’105 The intuition that successful farming may be impossible for some very small households is also supported by survey evidence from Ethiopia, Ghana, South Africa, and elsewhere: the challenges of crop production are likely to be insurmountable where small households lack sufficiently close links to adult males, or to the network of local connections necessary to acquire land, agricultural inputs, seasonal credit, and certain types of labour.106 One implication of these difficulties was noted many years ago by Iliffe, but has since been insufficiently emphasized, that is, that in these small and very poor households the main or most reliable source of income may be casual wage labour for neighbouring, larger households.

Conventional Wisdom III: Female-Headed Households are the Poorest

In UN agencies ‘there is a continued focus on comparisons between households, and especially between male-headed and female-headed units. The thorny question of how to define “female headship” is often ignored.’107 World Bank publications also continue to argue that ‘For several countries, a larger share of the multidimensional poor live … in female-headed households’, although a number of their economists have admitted that the classification of households as either male or female headed is ‘not very useful’.108

(p.212) There have been many attempts to link deprivation with ‘female-headed households’ or ‘female farm managers’ in sub-Saharan Africa (and elsewhere), but these typically reach ambiguous conclusions.109 One reason is that, like ‘rural’ households, female-headed households are a very mixed bag. A highly educated woman on a career break can choose to spend some time bringing up children alone on her own peaceful farm; her husband may very rarely leave his highly paid urban job to visit her, but he does send her a large monthly money order. The bucolic standard of living enjoyed by this woman cannot easily be compared to the survival struggles of a landless widow or divorcee bringing up children and grandchildren in a rural area without any support from an adult male. The gulf between these two women has been papered over in key documents produced by UN Women, where ‘lack of consideration of differences between women is a recurrent theme’.110

One problem is that ‘A household is likely to report a female head if the usual male head is a migrant working out of town, in which case the household may benefit from remittances that make them less likely to be poor.’111 More generally, we have already argued that most survey enumerators do not depart from simplistic tick-box rules and local social norms when they identify ‘the head’ (or list all approved ‘members’) of a ‘household’. If enumerators were sufficiently trained and supervised to collect all the data necessary to understand the fluidity of residential arrangements in rural Africa and the role of labour mobility and remittances in reproduction strategies, then the gendered disadvantages faced by individuals (as well as households) could be highlighted. A recommendation in 2018 that poverty analysts should abandon the simple classification of households by ‘headship’ and gender in favour of an array of new household typologies, for example ‘multiple adults, only female—with children’, may be a step in the right direction.112 But there will still be difficulties in probing respondents to enumerate all household members (currently resident or non-resident) and the economic relationships between them.

Domestic servants and other female (and male) migrant workers remit part of their wage earnings to other rural individuals. Unfortunately, no information is recorded in conventional surveys about the workers who remit. This information gaps make it extremely difficult to understand how the poorest rural households actually pay for children’s school expenses and food; the gendered and generational dynamics of power and labour market inequalities are obscured in a rural black box ‘household’ especially if, as is usually the case, the education levels achieved by, and the gender of, ‘non-residents’ are unrecorded.

(p.213) In our own surveys we have tried both to overcome these difficulties and to analyse gendered experiences of poverty in rural Africa—without making any attempt to identify ‘household heads’. Our methodological innovations have been discussed at length elsewhere.113 The broad aim of our research was to collect data on vulnerable wage workers, especially female workers engaged in seasonal, casual, and low-paid jobs outside major urban centres. We recognized that many of these individuals are frequently not ‘resident’ in households. They live and work for long periods in labour camps, construction sites, and illegal squatter settlements, or they have been given some space to sleep at their workplace during the harvest season, or while working as domestic servants: they are the ‘nowhere people’.114

With more information on a larger range of individuals than are usually recorded in DHS or LSMS household rosters, we had the information to develop a clearer idea of the nature and implications of economic flows between male and female individuals, wherever they were currently located. If more than 75 per cent of the whole range of listed adults were female, then we described these households as ‘female dominated’; and these households were much more likely than other households to be extremely deprived.115 The proposed definition of female domination and how we measured ‘extreme’ deprivation—using a simple Deprivation Index—did not require any technically complicated weighting, smoke or mirrors. We use an index that is transparent, easy to construct, and intuitively appealing.

8.8 Selecting Stylized Facts with an Extreme Deprivation Index

Some of the rural women we met are difficult to forget, especially their embarrassed faces when they could not even offer us a seat while we talked—not even on the simplest stool or bench. They lived in utterly bare rooms. These women helped us to understand that owning a few consumer goods could result in large absolute improvements in the quality of rural life. Without reliable access to electricity, a torch makes the night safer; sleeping on earthen floors cannot be compared to sleeping on a bed; a radio and a mobile phone can expand intellectual horizons, reduce isolation, and even help in searching for casual wage employment. Respondents may also be able to benefit in less obvious ways if they own ‘honorific’ or ‘prestige-based’ consumer goods—such as a sofa. The women we have talked to in poor rural areas of Southern and Eastern Africa are, of course, well aware of these benefits and our surveys do show rather widespread ownership of this type of consumer good. For example, about half of all our respondents in (p.214) Uganda and Ethiopia owned a radio and more than half owned a table, while 60 per cent owned a bed. It was intuitively obvious that individuals who had access to none or hardly any of these basic wage goods were extremely deprived.

Building on this intuition and rejecting statistically driven methods for constructing and weighting asset indices, we designed a new and very simple measure—the Extreme Deprivation Index (EDI). The practical relevance of the EDI is that it allows a quick, reliable and cost-effective way of identifying people who suffer from extreme deprivation and of assessing the distributional impact of policy interventions. The EDI takes out the ‘unavoidable guesswork in establishing who is poor’ that a leading proponent of RCTs seems simply to accept.116 We made a context-specific selection of 10 basic non-food wage goods, each of which would probably make a difference to the quality of rural life in many areas of rural Africa. The goods we selected were: a cupboard; a metal or wooden bed; a table; a sofa; a stove or cooker; a thermos; a torch; a mobile phone; a radio; and a cassette/CD player. We did not estimate the prices of these items or attempt to justify a complex weighting system: the EDI gives a score of 1 to ownership of each of the 10 selected basic consumer goods; the lowest scoring quintile of our respondents either scored no points or a maximum of 2. We defined these respondents recording a score in the bottom quintile of the distribution of EDI scores as living in the ‘most deprived’ households.

Many other surveys have shown that a high proportion of rural African households in the lowest expenditure/wealth quintile fail to consume the amount of food necessary to prevent child stunting,117 but these surveys also agree that relatively poor households do devote some of their expenditure to non-food consumer goods; they often do manage to acquire items very similar to the simple furniture and other basic consumer goods that we suggest should be the basis for constructing a context-specific EDI. For example, in rural Kenya, the expenditure elasticity for furniture—‘beds, chairs, tables, etc.’—has been found to be very high in a random sample of poor households.118

We have used survey data from Uganda and Ethiopia to compare the characteristics of households in the bottom quintile of the EDI (the ‘most deprived’) with other households; and we found little evidence to support conventional conclusions about the vulnerability of large households or female-headed households. But we did find that a significantly higher percentage of the ‘most deprived’ households was small and female dominated.119 In East African rural contexts where women and widows are often subjected to brutal forms of discrimination and gender inequalities are pervasive, it is perhaps not surprising that the ‘most (p.215) deprived’ households contain relatively few adult males and that relatively few of these households have regular access to any financial support from an adult male.120 For example, in tea-growing areas of Uganda, about half of the ‘most deprived’ households had no regular male support, compared to 15 per cent of other households.

We have also used the EDI to confirm the importance of other uncontroversial stylized facts about gender relations and deprivation. For example, it is widely acknowledged that adolescent pregnancies are hazardous both for the mother and the child.121 Teenage mothers in Africa and elsewhere are at greater risk than more mature mothers of mortality; and their lifetime labour incomes are likely to be significantly lower than the earnings of women who did not have children when they were teenagers.122 In the ‘most deprived’ households identified by the EDI, young women aged between 20 and 30 years are very likely to have had a child as an adolescent. In Uganda, for example, the risks of teenage pregnancy are remarkably high in the ‘most deprived’ households: only a very low proportion of young women in the ‘most deprived’ households (17%) had their first child when they were mature (20 years old or older), while a much higher proportion of women in other households (44%) were able to delay having their first child until they were mature.

A low level of female educational attainment is widely and correctly viewed as a particularly useful marker of poverty and of the adverse longer-term consequences of deprivation in Africa, because a woman’s lack of education is likely to be transmitted inter-generationally, negatively affecting the health, productivity, and lifetime earnings of her children.123 Our analysis using the EDI confirms the relevance and policy importance of the uncontroversial stylized fact that parental education and poverty are linked: we show, for example, clear associations between the ‘most deprived’ households and the absence of any adult who had graduated from secondary school, the presence of children not attending school, adult failures to complete primary school, and adult functional illiteracy.124

Another uncontroversial and policy-relevant stylized fact is that African children are vulnerable to chronic undernutrition if their diet is monotonous—if they can only achieve a low dietary diversity score.125 Again, our rural surveys were able to use the EDI to confirm the importance of this link. We simply asked respondents how frequently different types of food were eaten by anyone in the household. (p.216) The EDI proved surprisingly useful in predicting dietary diversity: only about 14 per cent of the ‘most deprived’ claimed to eat any high-value food items regularly, compared to over 45 per cent of other households.

We conclude that it is possible to identify policy-relevant stylized facts about the poorest people in Africa: they live in relatively small, female-dominated households and lack regular access to support from an adult male; adult females and children in these households probably have not received or are not receiving an adequate education; these women and children are threatened by the risks of teenage pregnancy; they also risk undernutrition because they rely on a monotonous and undiversified diet; and they have been able to acquire hardly any basic consumer wage goods. The absence of these goods in their homes can rapidly (and accurately) be confirmed by enumerators, using an EDI. This index can reliably, and at low cost, predict many of the difficulties they face.

Perhaps the least controversial and easily accepted policy implication of these stylized facts would be an agreement on new funding to reduce educational deprivation, especially investment to allow rural girls to complete or even attend secondary school. Appropriate expenditures on education would probably delay pregnancy and improve the labour market prospects for rural women, but the targeting of these interventions—for example, scholarships or conditional cash transfers—towards the ‘most deprived’ girls would involve a major reallocation of resources (see Figures 8.2 and 8.4), and an ability to resist powerful political demands for wider inclusion and/or the retention of patriarchal and other atavistic privileges.126

In some African countries, advocating policies to reduce high levels of unmet demand by girls and women for contraception (Chapter 2) will also face powerful opposition. In these countries, effective policies will not be given priority as long as the relationship between teenage pregnancy and poverty is downplayed or blamed on the moral failures of girls.

8.9 Wage Labour and Poverty: The Most Controversial Stylized Fact

It is even more unlikely that the policy implications of another, often ignored, stylized fact will be accepted with any enthusiasm by aid donors, capitalist employers, and other powerful political forces in Africa. This stylized fact, emphasized both here and in Chapter 7, is that the poorest rural people depend on wage labour. We have shown that the ‘most deprived’ households in Uganda and Ethiopia depend on the earnings of agricultural wage workers. Similarly, a (p.217) very high proportion of women in poorer Malawian households are engaged in seasonal agricultural wage labour; and an analysis of surveys covering over half of the population of sub-Saharan Africa found that ‘poorer rural households tend to have a higher rate of participation in agricultural wage employment … the share of income from agricultural wage labor is more important for poorer households’.127 Often, too little attention is paid to this evidence, or to research on the importance of hired labour inputs on smallholder maize and wheat farms in Eastern and Southern Africa by the Consultative Group on International Agricultural Research (CGIAR).128 Instead, there is a romantic stress on the resilient self-sufficiency of African rural families—in the face of growing evidence that a high proportion of rural households not only purchase their food but also engage in agricultural wage labour.

We have used the EDI to analyse the recent labour market experience not only of our principal respondents but also of all other (broadly defined) members of the ‘most deprived’ households. We began by identifying two crude categories of job: the ‘worst’ and the ‘more decent’. The ‘worst’ is a large category covering all the lowest paid and least desirable and most stigmatized types of rural wage work, especially manual labour performed in the open air. Other menial jobs in this category include working as a domestic servant for a rural private household and shining shoes. The ‘more decent’ jobs ranged over many different types of (mainly) non-agricultural wage employment, including: nursing, teaching, police, supervisory work inside processing plants and pack-houses, and so on. One of the largest groups of workers in this category were ‘guards’. Female respondents were much less likely than male respondents to have found employment in a ‘more decent’ job.

Escapes from poverty are conceivable if at least one household member can obtain more decent employment. Unfortunately, if a respondent works as a manual agricultural wage labourer, she/he is unlikely to live in a household where anyone listed on the household roster has managed to secure a more decent job in the last 12 months. This suggests that the consequences of deprivation can be cumulative; it may never be easy to escape from poverty by building on the success in the labour market of your parents or another household member. In the USA after 1945, a surprisingly similar story can be told—of cumulative disadvantage, rooted in a labour market unable to provide adequate job opportunities for people with low education.129

What policy implications follow from the stylized fact that that vulnerable rural women depend on the income they earn from wage labour and, therefore, on the number of days of wage employment they are offered (and the real daily wage rate)? We are not confident that legislative interventions to improve the wages and (p.218) working conditions of these women will achieve rapid results, partly because of the well-documented failures of a battery of progressive legislation and of a relatively strong trade union movement in South Africa.130 There is perhaps a better case for two immediate interventions: a massive increase in expenditure to monitor and publish the wage rates of poorly educated seasonal and casual workers; and a surge in investments to expand the demand (direct and indirect) for their labour in rural areas. Regular publication of these wage data may encourage long-overdue policy debates, new political demands, and even organizational successes. In Chapter 9, we focus on the untapped investment potential to increase rural wage employment and agricultural output in Africa.


(1) Labrousse (2016a).

(2) Deaton and Cartwright (2018).

(3) Bédécarrats et al. (2019a); Pritchett and Sandefur (2013) argued that the scope of appropriate application of RCTs is ‘vanishingly small’; and a group of Department for International Development (DFID) managers estimated that less than 5 per cent of development interventions are suitable for RCTs (Stern, 2012: 1, cited in Bédécarrats et al., 2019b: 10).

(5) Chandy (2017: 11).

(6) World Bank (2016d: xv–xvi, xviii).

(7) Wilson (2019b); Merridale (1996).

(8) Deaton (2015).

(9) Jerven (2014: 10); Chinsinga and Poulton (2014: s140). See also Kilic et al. (2018: 4).

(10) Jerven (2018: 468) and additional references in Serra (2018: 662). Marivoet and de Herdt (2017) discuss census data in the Democratic Republic of Congo (DRC).

(11) Human Rights Watch (2018): https://www.hrw.org/world-report/2018/country-chapters/tanzania-and-zanzibar#; Taylor (2017); The Economist, 14 March 2019.

(12) Hanlon (2019).

(13) Hoogeveen and Nguyen (2017).

(14) Glassman and Ezeh (2014: xii).

(15) Labrousse (2016b: 528).

(17) Hoogeveen and Nguyen (2017: appendix).

(18) http://roape.net/2018/11/21/the-cover-up-complicity-in-rwandas-lies/. For another view of the quality of Rwanda’s statistics see Krätke and Byiers (2014: box 3), and Wilson and Blood (2019).

(19) Allison (2013) describes the extremely defensive response to criticism by South Africa’s Statistician-General.

(20) Finn and Ranchhod (2017).

(21) Wittenberg (2017).

(22) Bamford et al. (2018: 28).

(23) Dorrington and Bourne (2008); Lodge (2015).

(24) Ravallion (2015: 7, 23).

(25) Briggs (2017: 203).

(26) Ilie and Rose (2018: 637).

(27) Beegle, Coudouel & Monsalve (2018: 73–5).

(28) For examples of this evaluation literature see: Favara et al. (2019) and Desalegn and Ali (2018).

(29) World Bank (2015: 12–14).

(30) Rajkumar et al. (2011: 134).

(31) World Bank (2015: 12, 49).

(32) Gebrehiwot and Castilla (2018).

(33) Sharp, Brown, and Teshome (2006: 21).

(34) Cochrane and Tamiru (2016: 657); Roelen, Devereux, and Kebede (2017: 22). See also Elias et al. (2013: 177).

(35) Lefort (2012); Cochrane and Tamiru (2016: 655).

(36) Hickey and Bukenya (2016: 18); Kjær and Joughin (2019); Jacob and Pedersen (2018: 23).

(37) Simpson and Hawkins (2018: 338).

(38) Chinyoka (2017: 17); Marongwe (2012: 144–5); Munyani (2005: 69–73).

(39) Simpson and Hawkins (2018: 365).

(40) Cameron (2018: 42).

(41) Jayne et al. (2018); Sender (2016).

(42) De La O Campos et al. (2018: 38).

(43) Hazell (2013: 20).

(44) Mellor and Malik (2017: 2–3). In one Ethiopian simulation it is even assumed that ‘all the additional income generated from increased production of these large farms is spent on urban goods, spent on imports, saved or sent abroad’ (Dorosh and Mellor, 2014: 429).

(45) Sen and Ghosh (1993). For later years see: Himanshu and Kundu (2016).

(46) Cramer, Di John, and Sender (2018).

(47) Van den Broeck and Maertens (2017).

(48) Lazaro et al. (2017: 24).

(49) Cramer et al. (2014a); Ragasa, Mazunda, and Kadzamira (2016: 22); Gray, Dowd-Uribe, and Kaminski (2018).

(50) Wineman and Jayne (2018: 24); Benali, Brümmer, and Afari-Sefa (2018: table 5).

(51) Bardhan and Rudra (1980a, 1980b).

(52) Cramer et al. (20167).

(53) Sender, Cramer, and Oya (2018: 597).

(54) Lowder et al. (2016: 14).

(55) On Pakistan, see Mellor and Malik (2017: 3); on Ethiopia, see Dorosh and Mellor (2013: 423).

(56) Hazell (2019: 150).

(57) Deaton also argues against another supposed ‘gold standard’, that of RCTs as a means of finding out ‘what works’: Deaton and Cartwright (2018: 2) argue that ‘any special status for RCTs is unwarranted’.

(58) Deaton (2016: 1223). Drèze and Deaton (2017: 66).

(59) Deaton (2013: 228).

(60) Ferreira et al. (2016: 161).

(61) Deaton (2013: 223); Drèze (2019: 55); https://www.youtube.com/watch?v=zfxH6qL9_ik.

(62) Edward and Sumner (2016: 10).

(63) World Bank (2018: 27).

(64) Pape and Parisotto (2019: 3).

(65) World Bank (2018: 75).

(66) Aguilar and Sumner (2019: 2).

(67) World Bank (2018: 4).

(68) Daniels and Minot (2014); ROAPE (2018).

(69) http://iresearch.worldbank.org/PovcalNet/povOnDemand.aspx; see also Clementi, Fabiani, and Molini (2019).

(70) Ngo and Christiaensen (2018).

(71) Stifel et al. (2018).

(72) Kilic et al. (2017: 21).

(73) Lesnoff (2015); Himelein, Eckman, and Murray (2014).

(75) Kaiser, Hruschka, and Hadley (2017: 2).

(76) Tusting et al. (2016: 651); Ngo and Christiaensen (2018: 12); Rich, Desmond, and Makusha (2019: 494).

(77) Sender, Cramer, and Oya (2018): Liu, Esteve, and Treviño (2017); Kabudula et al. (2017). There is, of course, no such thing as an unweighted index; it simply means that the components of an index are given equal weights.

(78) Some of these statistical techniques are outlined in Vollmer and Alkire (2018).

(79) Vaz, Pratley, and Alkire (2016); Mahajan (2014).

(80) Ravallion (2012); Beegle et al. (2016: 108).

(81) Alkire et al. (2017: table 1).

(82) World Bank (2018: 94).

(83) Brunori, Ferreira, and Peragine (2013: 17); Aguilar and Sumner (2019: 22).

(84) Lawson (1989: 76).

(85) Beegle et al. (2016: 10); De La O Campos et al. (2018: 9).

(86) Aguilar and Sumner (2019: table 5); World Bank (2018: table 4C.1).

(87) Jayne et al. (2016); Lay, Nolte, and Sipangule (2018); Whitfield (2017); Greco (2015); Schaefer (2016).

(88) Van Noorloos and Kloosterboer (2018).

(89) Pincus and Sender (2008).

(90) De Magalhães and Santaeulàlia-Llopis (2018: table 2).

(91) Beatriz et al. (2018: table 1).

(92) Marivoet, Ulimwengu, and Sedano (2019); Oxford Poverty and Human Development Initiative (2018: 71); Alkire (2018: 11).

(93) Beatriz et al. (2018).

(94) Stifel et al. (2018).

(95) The high priests of the movement to reduce ‘urban bias’ are Bates (1980) and Lipton (2012b).

(96) United Nations (2017: 10).

(97) Shepard (2007).

(98) Cited in Aldrich (2019: 19).

(99) Schürer et al. (2018: table 9).

(100) Van Campenhout, Sekabira and Aduayom (2016: 150). For similar views on Ethiopia see: Abebaw and Admassie (2013: 127). More generally, see Beegle, Coudouel, and Monsalve (2018: 50). See also: Beegle et al. (2016: 130) and World Bank (2018: 38).

(101) Guyer and Peters (1987); O’Laughlin (1995); Adato, Lund, and Mhlongo (2007); Akresh and Edmonds (2010); Cramer et al. (2014b: 178–9).

(102) Randall, Coast, and Leone (2011: 217). Estimating ‘household size’ in Victorian Britain is also difficult because of shifts in instructions to census enumerators (Schürer et al., 2018: tables 1 and 3).

(103) Hoogeveen and Schipper (2006: 77–8). Poor widows are also likely to be under-represented (Randall and Coast, 2016: 150).

(104) Rich, Desmond, and Makusha (2018: 8). In Vietnam, LSMS results linking poverty with large households have been criticized in detail by Dinh Vu Trang Ngan, Pincus, and Sender (2012).

(105) Iliffe (1987: 7, 238). The poor in England during the Industrial Revolution lived in much smaller households than other social classes (Allen, 2019: 92).

(106) Siyoum (2012: 50ff.); Skalidou (2018: table 9); Palmer and Sender (2006).

(107) Bradshaw, Chant, and Linneker (2017: 16).

(108) World Bank (2018: 109); Boudet et al. (2018: 4). The emphasis on female-headed households is widespread among African leaders of NGOs (ACPF, 2019: 33).

(109) Milazzo and van de Walle (2015); Liu, Esteve, and Treviño (2017); Djurfeldt, Dzanku, and Isinika (2018); Fransman and Yu (2019).

(110) Bradshaw, Chant, and Linneker (2017: 13).

(111) Casteñada et al. (2018: 257).

(112) Boudet et al. (2018: 25ff.).

(113) Cramer et al. (2014b).

(114) Breman (2010: 135).

(115) Sender, Cramer, and Oya (2018: table 4).

(116) Banerjee (2015: 8).

(117) Black et al. (2013).

(118) Haushofer and Shapiro (2013: 30). For similar results in rural South Africa, see Browne, Ortmann, and Hendriks (2007: 571).

(119) Sender, Cramer, and Oya (2018).

(120) Marshall, Lyytikainen, and Jones (2016); Semahegn and Mengistie (2015); Bantebya, Muhanguzi, and Watson (2014); Sharp, Devereux, and Amare (2003: 56).

(121) Saloojee and Coovadia (2015: e342).

(122) Pradhan and Canning (2016: 1).

(123) Castañeda et al. (2018: 258); ICF International (http://www.statcompiler.com); Alderman and Headey (2017: 456); Bado and Sathiya Susuman (2016); Ambel and Huang (2014: 14); Keats (2018: 155). See also: Tusting et al. (2016: 653).

(124) Sender, Cramer, and Oya (2018: 7–8).

(125) Hernández (2012); Herrador et al. (2015); Hirvonen, Taffesse, and Worku Hassen (2016); Hirvonen et al. (2017); Muhoozi et al. (2016).

(126) The strength of political demands for wide inclusion in Ghana is discussed in Abdulai (2019).

(127) Koolwal (2019: 8); Davis, Di Giuseppe, and Zezza (2017: 161 and table A2).

(128) Baudron et al. (2019).

(129) Case and Deaton (2017: 29ff.).

(130) Devereux (2019).