Poverty Dynamics in Rural Zimbabwe: The 30 Years (Lost) ‘War against Poverty’ Bill H. Kinsey (Senior Research Fellow, Ruzivo Trust, Harare) Paper prepared for the conference: Ten Years of ‘War against Poverty’: What Have We Learned since 2000 and What Should We Do 2010-2020? Chronic Poverty Research Centre and the Brooks World Poverty Institute The University of Manchester 8-10 September 2010 (Keywords: multidimensional poverty measures, politics and poverty dynamics, mobility and poverty traps, intergenerational poverty, child poverty) Contact details: 7a Belfast Close Emerald Hill Harare, Zimbabwe Email: [email protected] or [email protected]Landline: +263-4-302 812 Mobile: +263 912-782 493
Transcript
Poverty Dynamics in Rural Zimbabwe:
The 30 Years (Lost) ‘War against Poverty’
Bill H. Kinsey
(Senior Research Fellow, Ruzivo Trust, Harare)
Paper prepared for the conference: Ten Years of ‘War against Poverty’: What Have We Learned since 2000 and What Should We Do 2010-2020?
Chronic Poverty Research Centre and the Brooks World Poverty Institute
The University of Manchester 8-10 September 2010
(Keywords: multidimensional poverty measures, politics and poverty dynamics,
mobility and poverty traps, intergenerational poverty, child poverty) Contact details: 7a Belfast Close Emerald Hill Harare, Zimbabwe Email: [email protected] or [email protected] Landline: +263-4-302 812 Mobile: +263 912-782 493
Poverty Dynamics in Rural Zimbabwe:
The 30 Years (Lost) ‘War against Poverty’
Bill H. Kinsey1
Introduction At the time Millennium Development Goals (MDGs) become a focal concern for development practitioners, per-capita income in Zimbabwe was only fractionally higher than at independence 20 years earlier. Thus, in income terms, Zimbabwe’s first two decades were characterized by stagnation. The absence of changes in income was not however mirrored in other spheres. On the contrary, several developments initiated dramatic changes. This paper addresses several of these: a land redistribution programme launched only six months following independence; drought (in the 20 years following 1980, at least six droughts were experienced); and economic reform (an economic adjustment programme began in 1991). The paper does not however address the economic and social consequences of the turmoil that ushered in the millennium decade and has continued ever since. For each of these three factors the question is asked: what can be said about its impact on rural poverty? In the attempt to untangle the various consequences of land redistribution, drought and economic reform, use is made of a unique data set comprising longitudinal information on two groups of households: those who benefited from the earliest phase of Zimbabwe’s land reform programme and those who did not.2 In attempting to understand the impact of change-stimulating factors, a distinction is made between these two groups. This distinction is of significant policy relevance because land reform was a key instrument in Zimbabwe’s arsenal of anti-poverty measures for the first decade of independence. During the second decade of independence—1991 to 2000—political interest in both land reform and poverty alleviation waned. Beginning in 2000, the forced seizure of thousands of commercial farms—in the name of land reform—has been a major contributor to the dramatic worsening of poverty levels nationally. For several reasons, the paper focuses primarily on the middle decade. The first reason is that this period was supposed to have seen the earliest resettlement schemes, where the data for this paper have been collected, reach their full economic maturity. Second, severe droughts, including the worst of the century, punctuated this period. And, third, an IMF-World-Bank-inspired structural adjustment programme was launched in 1991. The paper is an empirically based analysis of policy outcomes and employs several approaches. Using expenditure information collected during the 1990s, an update of national poverty estimates for Zimbabwe is provided for these rural households. In an attempt to disentangle the effects of drought and economic reform, both expenditure and income data are used to assess poverty indicators and their response to adjustment. The analysis is then extended through the use of non-monetary indicators of poverty, in this case simultaneous consideration of nutritional indicators for both children and adults within a household. The validity of nutritional indicators of poverty is assessed against an array of more conventional indicators. Finally, because a paper of this scope cannot possibly address all the poverty-related issues in a data set spanning 28 years and covering 500+ households—with thousands of variables per household annually, an explicit purpose of the paper is to provide a flavour of what is possible in order to inform potential collaborators in future work.
1 I am deeply indebted to my former colleagues at the Free University Amsterdam—Karin Bouwmeester, Kees Burger, Hans Hoogeveen and Robert Sparrow—not only for their contribution to the ideas in this paper but also for years of congenial collegiality. I am also appreciative of the critical reading given to parts or all the material by Brian Thompson, John Hoddinott, Lionel Demery and Gareth James. Surviving errors are mine. 2 This data set—the longest panel study every undertaken in Africsa—covers the same households over 28 years, 1982 to 2010. Some 550 rural households are covered, and fieldwork continues.
3
Background The investigation upon which this paper is based was launched to answer one ‘simple’ question: what are the effects of land redistribution on the welfare of rural families. Starting with a baseline data set established through interviews in 1983 and 1984, data have been collected over a 28-year period on some 400 households from 22 randomly selected communities in three of Zimbabwe's earliest resettlement schemes. These schemes were chosen so as to ensure representation of each of the three major agro-ecological zones in the country suited to cropping. The households were re-interviewed in detail in 1987, 1992, and every year from 1992 through 2001. Less-detailed data have also been collected in every year from 1984 to 2000, and also in 2002 and 2007-10. There has been remarkably little sample attrition. Some 82 per cent of households from 1983/84 were re-interviewed in 2001, and there is no systematic pattern to the few households that drop out. Beginning with the 1997 round of the survey, coverage was extended to include 150 additional households in villages in the communal areas (CAs) from which the resettled households originated in the early 1980s. This supplemental data permits explicit comparisons between the resettlement and communal experiences and between current living conditions in the communal and resettlement areas (CAs and RAs). The original objectives of the resettlement programme were the enhancement of the socio-economic well-being of low-income households (that is, the reduction of rural poverty), including their ability to feed themselves adequately (that is, achievement of food security) while at the same time earning a reasonable income from the sale of crops and livestock. To achieve these objectives the amount of arable land allocated to beneficiary families was more than double the area of the average family's holding prior to resettlement, the land is generally of higher quality, and a whole range of supporting services and facilities—health, markets, agricultural credit, veterinarians, housing loans, schools, etc.—were provided. In contrast, the CAs are typified by small holdings on poor soils in remote areas with poor infrastructure and support services. Analytical work has until now focused on three themes: i) the determinants of food supply and childhood nutritional status; ii) the processes governing income generation and asset accumulation; and iii) the factors that determine households' ability to withstand income shocks, most notably those caused by drought and economic reforms. The paper is organized as follows. Section 2 provides a brief overview of the major economic developments in Zimbabwe during the period of interest. Section 3 presents an overview of poverty and inequality in Zimbabwe and contrasts the supposed benefits of structural adjustment with the realities found in Zimbabwe. Section 4 begins the examination of poverty dynamics among resettled households with a discussion of the data set available for analysis, after which rural poverty and welfare estimates are presented for the mid-1990s. Section 5 focuses on the validity of nutritional status as an indicator of poverty. Section 6 concludes. A Snapshot of Economic Performance since Independence Zimbabwe became independent in 1980. Immediately thereafter there was an economic boom, led by a remarkable but short-lived growth in marketed output by smallholder farmers. This boom was accompanied by major distributional advances, particularly in the extension of social services to an indigenous population that had received little attention from the previous political regime. Another distributional advance was the land reform programme, which began within six months of independence and has continued at a varying pace since. The boom soon ended, however, so that by the mid-1980s, the economy had slowed, and there was increasing concern with macroeconomic imbalances and a worsening in the already large inequalities of wealth and income. From 1991, the government (hereafter GOZ) pursued an adjustment programme (Economic Structural Adjustment Program–ESAP), but the response of the economy was erratic and, overall, sluggish. With the population growing at more than three per cent per annum, real per-capita income was estimated to be perhaps a little lower at the end of the 1990s than in 1980 (Killick, Carlsson & Kierkegaard 1998). A major feature of Zimbabwe’s economy, that explains much of the variation that occurs in its economic growth, is reliance on an uncertain rainfall, with particularly severe droughts in the early 1980s, in 1992, and again in 1995. There is evidence that the frequency of droughts worsened in the post-independence period and the substantial economic consequences of the erratic and declining rainfall reflect the still dominant role of agriculture, particularly the vulnerability of the so-called communal farming areas, where most of the rural population live.
4
Since the mid-1980s, reducing the chronic tendency for GOZ to run large budget deficits has been the primary focus of concern for macroeconomic management. The problem here has been on the side of expenditures, for by international standards Zimbabwe used to have a high ratio of tax revenues to GDP. Despite cutbacks in public services and subsidies, and the reductions in the size of the civil service, government expenditure remains a major source of difficulty, with the Ministry of Finance and Parliament exerting weak control over state spending. In consequence, budget deficits regularly exceed estimates by large margins. Despite the efforts that were made to strengthen public finances, the deficits of the public sector remained large, in the range of 10-15 per cent of GDP in the 1990s. As a consequence, there was a crowding-out of the credit needs of the private sector, chiefly through the effects of high interest rates (with a nominal rate of well over 50 per cent for borrowers at the end of the 1990s), a factor which depressed investment in the private sector. Reduced investment levels contributed to the slow growth of productivity and undermined the objectives of ESAP. There was also substantial inflation, generally around 20-25 per cent per annum during the 1990s—although the year-to-year inflation figure fluctuated in the range 50-70 per cent. High inflation also discouraged productive investments by increasing uncertainties, as well as by reducing levels of living of those groups unable to protect themselves from the effects of rising prices. Although there were large fluctuations around the trend, average incomes fell in the period 1990-96 at more than a one per cent a year and, despite several years of good rainfall following 1995, there was no reversal of this trend. Every adverse circumstance mentioned in the few paragraphs above was monumentally amplified from 2000 onwards, but this paper does not address the post-2000 economic pathology. Inequalities and the Poverty Problem in Zimbabwe Against this backdrop of economic stagnation, what can be said about the distribution of income and wealth and about the incidence of poverty? The preferred approach is to take a broad view of poverty, as multi-dimensional deprivation, referring not merely to income and/or consumption levels but also to people’s access to public services and to productive assets (including skills), and autonomy. In practice, however, virtually all reliable information available at national level relates to incomes and/or consumption levels. Although the data for Zimbabwe are better than for many other African countries, the detailed information on poverty remains pieces of a jigsaw puzzle with many gaps left to be filled. In an important attempt to remedy this situation, in 1995 the Ministry of Public Service, Labour and Social Welfare (MPSLSW) undertook a national poverty assessment survey—PASS (GOZ 1996, 1997a). Unfortunately, the survey was concentrated in a few months that coincided with the aftermath of a serious drought. The results are therefore distorted, and there are also problems with the reliability and representativeness of the data collected. The resulting estimates of the incidence of poverty are regarded by many as serious overestimates, apparently indicating that nearly two-thirds of the population is either very poor (16 per cent) or poor (46 per cent). These figures greatly exceed earlier, although more partial, estimates and should probably be discounted.1 Notwithstanding the methodological problems of the PASS, there was still an unambiguous increase in poverty in Zimbabwe between 1990 and 1996. A higher percentage of people in 1996 were below each level of real consumption expenditure than people in 1991 (GOZ 1998). This finding that poverty has increased is insensitive to the choice of the poverty line, and it is suggested that within a range of ‘reasonable’ total consumption poverty lines there was an increase in household poverty of around 23 per cent (GOZ 1998). Even without the PASS, sufficient pieces of the puzzle are available for there to be a substantial consensus on the main features of poverty, at least as it relates to income and/or consumption. The following set of stylized generalizations summarizes what appear to be the main features on which consensus exists about poverty in Zimbabwe in the 1990s and sets the stage for the analysis that follows:2
• Although substantial, income poverty in Zimbabwe is less severe than in other African countries. In a comparison of the proportions of the population living on less than US $1 per day in 11 African
1 A new poverty assessment study (GOZ 2006) was conducted in 2003 and indicated a dramatically worsened situation regarding the extent and depth of poverty 2 The material here is based chiefly on the information contained in DANIDA (1996), Killick, Carlsson & Kierkegaard (1998), the World Bank (1996) and GOZ (1998)
5
countries, the value for Zimbabwe was the median value—an estimated 39 per cent (Ravallion & Chen 1996), while UNDP rankings of countries according to their scores on a ‘human poverty index’ (which, however, excludes income as an indicator) placed Zimbabwe top of all continental African countries (UNDP 1997).
• A large proportion of the population is, nevertheless, poor, and substantial numbers are extremely poor. According to the analysis of the 1995/96 Income, Consumption and Expenditure Survey–ICES (GOZ 1998a), more than three-quarters of the entire population is classified as living in poverty, including 47 per cent in extreme poverty.
• Poverty is overwhelmingly rural. Although nearly a third (31 per cent) of the population is classified as urbanized, only 23 per cent of those described as poor are urban dwellers. Rural areas, which accommodate a little over two-thirds of the population, contained 77 per cent of the poor and 90 per cent of the very poor (GOZ 1998). Urban poverty is growing, however, with evidence of declining nutritional levels in the towns, particularly among women and children (Raftopolous & Jazdowska 1997). This increase is closely related to the unrelenting spread of HIV/AIDS.
• Within the rural population, there is a particularly severe concentration of poverty in communal farming areas. These contain half of Zimbabwe’s total population but three-quarters of the poor and over 80 per cent of the very poor (GOZ 1998). There is also a serious degree of poverty in the resettlement areas—former commercial farming areas that were purchased for the relocation of peasant farmers from the communal areas (although there is now evidence that poverty in the longer-established of these areas may be diminishing (Kinsey, Burger & Gunning 1998, Kinsey 1999). Prior to 2000, there were also serious symptoms of poverty among commercial farm workers and their dependants, who suffered from low job and food security, high levels of child malnutrition, poor housing and poor access to water, sanitation, health and educational services (World Bank 1996).1
• There are very large inequalities of income and wealth. It was estimated for the early 1990s that 50 per cent of the population received less than 15 per cent of total incomes, while the richest three per cent received 30 per cent of the total (Stenflo 1993). The Gini coefficient for the country as a whole was estimated to be 0.57 at that time (Killick, Carlsson & Kierkegaard 1998). Half a decade later the national Gini coefficient—based on mean consumption per person—had risen to 0.63 (GOZ 1998). Underlying the prevalent inequality at the time was a highly skewed distribution of ownership of agricultural land. About 4000 large-scale commercial farms, largely owned by white farmers, occupied large areas of the country’s most fertile and well-watered land. For this reason the Gini coefficient for inequality among people was actually slightly worse in rural areas than urban areas; the urban Gini coefficient was 0.58 while it was 0.60 in rural areas, including communal, commercial and resettlement farming areas (GOZ 1998).
• There are also large inequalities within the black population. While many of the values reported above are a reflection of surviving differences between the black and white populations, there are also considerable disparities in incomes within the black population. For example, analysis of the 1990/91 ICES showed that levels of inequality within the communal areas were greater than for the country as a whole (Jenkins & Prinsloo 1995). In contrast, in the absence of shocks, incomes among households resettled in the early, poverty-focused phase of the resettlement programme appeared to be both rising and converging (Gunning, Hoddinott, Kinsey & Owens 2000). Subsequent analysis, however, shows that while there has been rapid growth in per-capita assets, inequality has remained high throughout the period (Elbers, Gunning & Kinsey 2002). Economic Policy Reform and Poverty in Zimbabwe
Where is poverty situated? Identifications and descriptions of the poor vary widely in the literature available for Zimbabwe, but most of the research reviewed typically attempts either to depict the conditions experienced by the poor, or more narrowly, simply to identify affected groups in society. The following groups have been explicitly identified by one or more research analyses as being among those most adversely affected by poverty during the period under consideration:
• Farm workers (Jamali 1997b, World Bank 1996) • Domestic workers - urban (Jamali 1997b, Matshalaga 1997a & b) • The homeless (GOZ 1997a, ZimRights 1996) • The unemployed, the young (Jamali 1997a, ZimRights 1996) • Resettlement dwellers (Jenkins & Prinsloo 1995, ZimRights 1996) • Squatters (ZimRights 1996) • Refugees, victims of war (ZimRights 1996)
1 The situation for this group has become dramatically worse since 2000.
6
• Those retrenched without pensions (Jamali 1997a) • Informal traders, vendors and hawkers (ZimRights 1996) • Women (Matshalaga 1997a & b) • Those living on poor soils or in drought-prone areas (World Bank 1996) • The aged (over 60) (Madzingira 1997) • Small-scale farmers (Jamali 1997a, Kwaramba 1998) The above cataloguing suggests that there are few segments of Zimbabwean society where poverty had not been discovered in the 1990s. Some researchers argue that in fact everyone is affected by poverty in the country—both the poor and the rich. The poor are most affected because they fail to satisfy their basic needs and the rich are affected by the fact that, if a nation is full of poor people, then the rich do not enjoy security (Jamali 1997a). This concept is particularly significant where the differences between the rich and the poor are wide (and growing) and where a middle class is just a tiny proportion of the population. Disparities in the already high level of inequality in Zimbabwe have worsened since 1990; the effects of the distributional reforms of the early 1980s have been reversed, the gap between rich and poor has widened, and the latter have become more numerous. Causes of poverty. The causes of poverty are described in various household surveys,1 however—of the surveys available—only those used in the analysis below permit judgments on changes in the extent of poverty over time in a specific setting. Some major changes at national level are commonly said to have accelerated the process of impoverishment in Zimbabwe between 1990 and 2000. These include:
• The implementation of ESAP since 1991 • The drought seasons of 1991/92 and 1994/95 • The collapse of the trade agreement with South Africa in the mid-1990s • The growing but thinly documented impact of HIV/AIDS at the household level • Unbudgeted spending and fiscal mismanagement by GOZ and excessive consumption-focused
borrowing on the local money market, leading to severe pressures on the Zimbabwe dollar beginning in late 1997, accelerating inflation, and a contraction of investment
• Unsustainable de facto fixed exchange rates and persistently high rates of inflation • Lack of transparency in land reform policies by GOZ beginning in 1997 and the further reduction
in investor confidence • Lack of a shared vision and common sense of national purpose on which to build a development
strategy with any genuine prospect of success In addition, the national economy in the 1990s was affected by:
• A sharp decline in the price of gold on international markets, affecting one of Zimbabwe’s major mining activities, which provides employment and income in both large- and small-scale mining operations, and
• A decline in world prices for tobacco and cotton, Zimbabwe’s major export crops. Information on how people perceived the causes of poverty in the mid-1990s derives from the Poverty Assessment Study Survey (GOZ 1996), work on the gender dimensions of urban poverty (Matshalaga 1997a & b), and the resettlement and communal areas longitudinal household study in 1997 (Kinsey 1998a). Other work, such as that under the farm and domestic workers’ programme, addresses these themes but does not quantify them. The perceptions of the causes of poverty are summarized by sector in Table 1. Due to their different characteristics and methodologies, the results from surveys can be strikingly different. One point is that the PASS (GOZ 1997a) is a stratified national random sample, while the urban study and the resettlement and communal area studies draw their samples from groups which are, according to the PASS, widely affected by poverty: about 84 per cent of the total rural population is poor. On small-scale commercial farms and resettlement areas, some 70 per cent of the population was judged to be poor (GOZ 1995). Definitions and perceptions of poverty. The Poverty Assessment Study Survey (GOZ 1995, 1997a), like other poverty surveys (Mundy 1995), uses poverty lines as criteria to assess the extent of poverty. Two poverty lines each are drawn for rural and urban areas. An upper one relates to basic consumption needs inclusive of food and other basic needs, while a lower line relates to food needs only. Those below the upper poverty line only are considered poor, while those below the
1 A select bibliography on poverty in Zimbabwe is available from the author on request.
Formatted: Space Before: 6pt, Tabs: 0.63 cm, Left
Formatted: Tabs: 0.63 cm,Left
Formatted: Indent: Left: 0cm, Hanging: 0.63 cm, Tabs: 0.63 cm, Left
Formatted: Tabs: 0.63 cm,Left
Formatted: Indent: Left: 0cm, Hanging: 0.63 cm, Tabs: 0.63 cm, Left
lower poverty line are regarded as very poor. The assessment of the poverty lines is based on a combination of consumption and income, with income being defined in cash and kind. On this basis, the PASS concluded that 61 per cent of the Zimbabwean population was poor, with the majority of these living in rural areas (GOZ 1995). Although consumption and income indicators are merged in the calculations, no consideration is given in the PASS to the availability of assets. This is a particularly important omission for rural areas. Nor does the PASS take account of polygyny in defining household size or calculating per-capita income and consumption. Instead it defines a polygynous household simply as the husband, the eldest wife and her children; excluded are other wives and their children. This treatment is presumably justified on the grounds of simplifying survey design and because the extent of income and consumption pooling is difficult to ascertain in multi-wife households. However, based on the 28 years of experience in conducting the panel study for the ZRHDS, it is felt that this procedure is likely to distort badly certain calculated indicators of welfare, particularly for rural households. Some caution is thus advised in considering the proportion of poor households reported in 1995. Based on a nationally representative survey of 18,797 households, the rural poverty lines for households from the PASS are defined as follows:
Total consumption poverty line Z$ 1,924.20* Food poverty line Z$ 1,180.40*
* Relates to the value of the Zimbabwe dollar in Sept/Oct 1995. In contrast, the 1995 income, consumption and expenditure survey (ICES)—using better data and more meticulous procedures1 reports a mean monthly consumption value of Z$ 178.05 per person for rural areas, implying an annual figure of Z$ 2,136.60 in 1995 prices (GOZ 1998a). The 1995 ICES concludes that 76.2 per cent of Zimbabwe’s poor and 89.5 per cent of the very poor households are found in rural areas. As poverty is experienced as a set of circumstances or states at individual and household level, people's perceptions and perspectives of these experiences are helpful in interpreting more aggregative data. People in Zimbabwe employ various descriptions of poverty and concepts of who is poor. Central issues associated with poverty relate to food supply, employment, access to land, clothing, housing, lack of knowledge of human, legal and political rights, and absence of improvement (GOZ 1995). Other surveys report largely the same findings but formulate them in different ways, such as: ‘no shelter, no clothing and nothing to eat’, ‘not having enough money to survive’, ‘have no implements for farming’, ‘having poor soil’, and ‘when one lives without having the basic needs in life’. There are notable differences in perceptions between rural and urban dwellers, as shown in Table 1. The direction of poverty in Zimbabwe. Aside from the data used in the analysis later in this paper, there is no single set of data that allows poverty to be systematically tracked over time. In some cases, such as the ICES, it is possible to compare two different points in time, but such instances are rare. Moreover, the conclusions of many of the independent studies are coloured by the perspective of the researchers and make it difficult to attribute causality to the changes observed. Thus, for example, it is possible to find two parallel studies of a particular sector, both of which report an increase in poverty or a decline in welfare. But one study will attribute the observed changes to adjustment effects, while the other will attribute them to the effects of drought. Nevertheless, the growing body of literature that addresses poverty-related issues now allows some broad overall assessment of trends in poverty.2 And the picture that emerges is a strong consensus that poverty became much more pervasive and intractable during the period covered here than it was in the pre-reform period (see, for example, IMF 1998). Before turning to the detailed analysis of poverty among the panel households, it is useful to attempt to disentangle some of the polemical issues through examining what the panel data can tell us about the effects of adjustment-related policy reforms. This is the task of the following section.
1 Confusingly, however, this source does not report actual annual figures, and multiplying mean monthly data from one of its annexes gives an annual figure of Z$ 2,763.48 per person for the total consumption poverty line in rural areas.
8
Evidence of Policy-induced Impacts In 1990, in recognition of the fact that the patterns of economic growth could not sustain the post-independence improvements in social welfare, Zimbabwe agreed to embark on an Economic Structural Adjustment Programme (ESAP).1 Confronted by weak growth for the main economic indicators, low investment, stagnant exports, low levels of formal sector employment and deteriorating external accounts, Zimbabwe launched ESAP in January 1991—supported by the World Bank and the IMF. ESAP was a broad package of measures around a tripartite core of monetary and fiscal reforms, trade liberalization and domestic deregulation. The measures embodied in ESAP were supported by various sectoral initiatives and selected actions to assist the poor. ESAP was to be implemented in phases, the first of which spanned the years 1991-95, a period particularly well captured in the data utilized in this paper.
Table 1.—Perceptions of the causes of poverty as indicated by urban and rural dwellers Identified cause As identified by: Urban dwellers Rural dwellers (Proportion of responses) Little produce Shortage of food Poorly nourished children Little money Failure to pay fees for school/health No employment/non-farm job Shortage of clothing Drought Low paying jobs High prices "ESAP" Rural/urban migration Retrenchment Laziness Poor parents Have no/not enough cattle Have no/not enough farm equipment No/not enough assets/little property Poor accommodation No supporting family Short of labour Poor soil/land quality No improvement in livelihood Family large, many dependants Lack of gardens Lack of background/education Ill health Shortage of farming land
-- 69c 1c ? 3c
44 – 75b,c 11c
3 – 10b,c 14 – 55b,c 13 – 30b,c
20c 0b 10c 5b,c
3 – 5b,c ? -- ? 6c ? -- 0b ?
6 - 18b,c ? 2b 0d 1c
4.5a,b 75c 0c
11.9a,b 1c
4.5a,b 10c
27.5 - 40b,c 11c 5c ? 0c ? 8c 3c
6.0 - 27.1a,b,c 15.8a,b 7.3a,b
2c 6.8a,b 2.8a,b
1.0 - 2.8a,b,c 2.8a,b
4.0 - 2.3a,b,c 1.7a,b
2.0 - 1.1a,b,c 1 - 3c
1c a Refers to how people themselves judge their own household to be poor. b Kinsey (1998a). c GOZ (1995 & 1998b). d Matshalaga (1997a & b).
The key structural reforms likely to impact directly or indirectly on rural poverty included the following:
• Domestic marketing of all agricultural commodities, most importantly maize, was liberalized and marketing boards for meat, dairy and cotton commercialized and their monopsony powers ended.
• Restrictions on the movement of commodities were lifted, and domestic trading was allowed. • Virtually all domestic prices were decontrolled and consumer and producer subsidies phased out. • Entry into small-scale food processing, particularly the milling of maize, was facilitated by
removing non-capital barriers to entry.
1 ESAP very quickly came to be referred to as ‘Eternally Suffering African People.’
Deleted: _.
Deleted:
Deleted: ¶
Deleted:
9
• The foreign exchange allocation system was dismantled and current account transactions and most capital account transactions freed from control. The exchange rate was largely market determined.
• Government’s role in wage-setting was reduced, and retrenchment became more flexible and no longer required ministerial approval.
Several features of ESAP could be expected to translate into impacts detectable at the level of the farming households in the panel study. First is the decontrol of the prices of agricultural products—both inputs and outputs. Government was to withdraw from direct price intervention and marketing subsidies to allow private sector initiatives to compete. Thus statutory marketing bodies were to be deregulated and a multiple-channel marketing system encouraged. In the same vein, subsidies were to be removed on fertilizer, seeds and agro-chemicals at the same time as credit and financial markets were to be deregulated. While the need was recognized for government to withdraw from direct intervention through price control and subsidies, the government acknowledged the need to support programmes necessary to guarantee food security and development of the smallholder subsector. The sections below look briefly at specific aspects of the panel data set which are able to indicate directions of impact from these policy changes. The impact of marketing reforms Reforms. The chief liberalization measure, which proceeded much more slowly than the other measures, dismantled the monopsonistic commodity marketing system, removed restrictions on internal commodity movements, and replaced comprehensive statutory regulation of prices with a much simpler system which set only a floor price for maize. Starting in 1991, Zimbabwe progressively moved away from a single-channel system of marketing to multiple-channel arrangements in which the commercialized marketing boards (such as the Grain Marketing Board (GMB) and the Cotton Marketing Board (CMB)—now Cottco)) compete against private buyers. Liberalization of the maize market was completed only in the 1994/95 season. A parallel fiscal adjustment measure resulted in the closure of large parts of the previous network of parastatal marketing facilities, particularly in marginal producing areas.
Anticipated outcomes. The expected impacts were the rapid growth of multi-market channels, with increased competition leading to higher prices. At the individual farm level, it would also be expected that changes in the amounts of land allocated to various crops in response to improved price incentives would be visible. Recall, however, that RA households have access to fixed, equal-sized plots of arable land, so that any response to increased prices must come either a reallocation of land among crops or increases in factor intensity and productivity. Evidence of impact. Of all the ESAP-instigated reforms, that most likely to have an impact on households represented in the panel data set were the changes made to the domestic marketing system. For smallholders, one of the primary advantages of selling to private buyers is immediate cash payment.1 Farmers selling to GMB may wait four to six weeks for payment by cheque—which must then be banked in an urban centre, and in many cases the delay is measured in months. Such delays create consequential delays in paying school fees, in buying seasonal inputs and contracting for tillage services to pay for the next season’s planting, and in repayment of loans to the Agricultural Finance Corporation (now Agribank). Evidence that the farmers in the panel study responded to market reforms is shown in Figure 1, which plots the outlets used in every recorded market transaction since the early 1980s. At first glance, the figure suggests that there is a trade-off between sales to the GMB and the CMB, but this is purely a seasonal effect. In dry years—such as 1984, 1992 and 1995, farmers sell little maize, so that sales of drought-tolerant cotton assume a relatively greater share of a smaller total number of transactions. What is striking in the figure is that in every year prior to 1996, the combined proportion of market transactions claimed by GMB and CMB is at least 70 per cent. In 1996, however, this figure falls to below 35 per cent, and in the following year it is below 30 per cent. In contrast, the proportion of sales transacted through private traders increased tenfold from 1984, and sales through other outlets rose some fourfold. Sales to neighbours and other local markets,
1 In the 1995 crop-selling year, several private buyers operating in the most remote of the three areas represented in the panel bartered for their maize supplies using sugar, salt, tea, cooking oil and other consumer commodities.
Deleted:
10
which typically accounted for between 10 and 20 per cent of transactions until the mid-1990s, also rose after 1995. This evidence is consistent with—but does not establish—the proposition that a recently liberalized marketing system may be having a positive impact on the rural poor. For one thing, part of the reform package was the closure by GMB of loss-making collection depots beginning in 1994 (Addison 1997). One explanation therefore could be simply that farmers had less access to GMB facilities than in the past and thus had little option but to rely on other market outlets. In the mid-1990s, much concern was expressed by farmers, their organizations and NGOs about the impact on smallholder incomes of closing GMB depots (Balleis 1993, MacGarry 1994, ZFU 1994). The fear was that farmers in remote rural areas would be forced to rely on private traders who would use their market power to offer low prices. The analysis here of the producer price index for the panel farmers (See Figure 2) does not suggest that they benefited significantly from higher prices under market liberalization after 1995.
Figure 1. Changes in the marketing outlets utilized by RA farmers in the period 1984-2000
Evidence that the conclusion above regarding lack of a price impact is incorrect might be found, for example, if panel households changed their cropping patterns consistent with changing relative prices. The relative farmgate prices for the four main crops—maize, cotton, groundnuts and sunflower—grown by panel households are plotted in Figure 2. Over the full period under review, real farmgate prices for maize and groundnuts rose marginally, that for sunflowers declined, and the price for cotton increased substantially. At the end of the period, however, all prices were lower than they were in the 1992 selling season—following the major drought. The total area planted in each year and the proportion of planted area allocated to the major crops are plotted in Figure 3. The staple maize always occupies over half the planted area, and this proportion rises steadily from 51 per cent in 1990/91 to 65 per cent in 1994/95. Part of this increase will have been a lagged adjustment to the large price increase in 1991/92, but part also will reflect attempts by farmers to replace stocks exhausted after the poor 1992 harvest. Between 1994/95 and 1996/97, the proportion of land allocated to maize declines by 11 per cent—just under one-third the decline in real prices over the same period. Despite a favourable price trend, the proportion of land planted to cotton peaks at 25 per cent in 1990/91 and 1991/92 and declines to 20 per cent in 1996/97. Over the period shown, sunflower virtually disappears from the cropping pattern, as it declines from seven to only one per cent of cropped area, while the proportion of land planted to groundnuts moves in fairly close accord with real prices—first declining and then rising again.
11
There is some suggestion in Figure 3 that farmers in the panel diversified their crop mix after 1993/94, since there is steady growth in the proportion of area planted to other crops. It is impossible to construct a meaningful price index for this aggregate category, however the finding is consistent with the interpretation that farmers were beginning to grow crop—such as tobacco and paprika—that yielded better returns than the most common crops.
Figure 2. Relative mean farmgate prices for the four major crops grown
Total cropped area (acres)7.98 8.96 7.93 7.89 8.13 8.38 8.14
Figure 3. Relative area planted to the major crops, 1990/91 – 1996/97
12
Decontrol of the foreign exchange market and devaluation Reforms. Dismantling of the system of nonmarket allocation of foreign exchange was substantially accomplished by 1996 and was accompanied by a gradual devaluation of the Zimbabwe dollar. Anticipated outcomes. The expected results were higher producer prices, especially for export crops, and a reallocation of productive resources toward export crops. Evidence of impact. As discussed above, the major tradable crop produced by farmers in the panel is cotton. Over the period under review, the real farmgate price of cotton rose by 62 per cent from the initial to the terminal year and, for the entire period after the initial year, averaged 71 per cent higher. The same pattern of higher average prices over the period as a whole compared to the terminal year prevails for all four major crops shown in Figure 2. For maize, the period average real price was 83 per cent above the initial year, but in 1996/97 it was only 14 per cent higher. For groundnuts, the period average price was 38 per cent above that of 1990/91 but in the terminal year only nine per cent. Sunflower, which had a price at the end of the period 19 per cent lower than at the beginning, actually had a period-average price 24 per cent above that in 1990/91. As noted earlier, there is no evidence of a strong short-term response to price changes in terms of a reallocation of the land resource. As also noted, however, any such response would have been dampened by the necessity of farmers to allocate a major proportion of their land to maize, particularly given that two of the seven years considered experienced severe drought. The impact of deregulation of input prices Reforms. Fertilizer, for many years regarded as a strategic commodity, was among the last commodities to have price controls and subsidies removed. Anticipated outcomes. The removal of subsidies, together with the freeing of the foreign exchange market, was expected to encourage the expansion of fertilizer production and increased imports. At the farm level, the main impacts were expected to be increased availability, higher prices and increased fertilizer use. Liberalization of foreign exchange should also lead to increased imports and production of farm machinery, resulting in wider availability, lower prices and increased purchases. Evidence of impact. One of the criticisms levelled at ESAP is that it incorrectly assumed that the agricultural sector was homogeneous and that, in the absence of market distortions, liberalization would improve the efficiency of production through adoption of improved technology, thus stimulating diversification and improving net returns to farmers (ZCTU 1996). The panel data set contains information that allows examination of long-term trends in two dimensions of agricultural technology adoption by smallholders. These are use of chemical fertilizers and investment in agricultural capital equipment. Figure 4 plots the average quantity applied per hectare cultivated of two types of fertilizer: the quantity of fertilizer used is the sum of both initial basal fertilizer applied prior to planting, or at planting, and top-dressing applied later in the season. Even though the trend is clearly one of declining levels of fertilizer use, there are two years—1992 and 1995—when levels are particularly low. The reason for this is that use of top-dressing is discretionary; in years of adverse weather, such as these drought years, farmers will not apply top-dressing even though they may have already purchased it. Overall the panel farmers appear to be using roughly a quarter less fertilizer per unit of land planted than they did in the mid-1980s.1 Since fertilizer is sold in discrete units of 50kg bags, Figure 4 suggests that farmers are applying about one less bag per hectare now than they did in the past. Policy considerations aside, there may be good technical reasons for this trend. Every year between 1988 and 1996 experienced below average rainfall, and the returns to chemical fertilizer are reduced in the absence of adequate soil moisture. A more likely reason for the decline, however, is that the maize-fertilizer price ratio declined during the 1990s. Farmgate maize prices in real terms were static or declining after 1992, while decontrolled fertilizer prices rose rapidly and were being adjusted almost continuously.1
1 Oni (1997) also reports a decline in fertilizer use in the communal areas following the removal of the subsidy on the price of fertilizer.
Figure 4. Mean quantity (kg) of mineral fertilizer used annually per hectare planted,
1985/86 – 1999/00
As noted elsewhere, agricultural terms of trade have deteriorated since 1992. One concrete piece of evidence of the response to this deterioration is the declining rates of investment and reinvestment in agricultural capital equipment in both the resettlement areas and in the communal areas (Figure 5). Agricultural capital equipment is defined here as all tools and equipment—whether hand-operated or animal- or machine-powered. The two fitted curves in Figure 5 show that investment in farming capital declined markedly in both these areas since the early 1980s, when the policy environment so strongly favoured smallholder farming. On a trend basis, annual investment in equipment has declined by some two-thirds since the early 1980s.2 In the late 1990s it appeared that investment levels may have been merely enough to maintain the existing stock of capital—if that.3 Encouragement of the private sector Reforms. As one of the measures in support of employment creation, some of the regulations restricting small-scale activities (such as licensing of vendors, zoning of markets, and transport controls) were reduced and/or their enforcement relaxed. Anticipated outcomes. The anticipated outcome of these measures would be an expansion of both off-farm and nonfarm activities, particularly in CAs where farming may provide only a marginal income. It is not clear a priori what the effect would be in the comparatively land-abundant RAs, where the returns to farming are relatively more attractive. If the net effect of all adjustment measures is to leave the terms of trade for smallholder farming unchanged or enhanced, then one would expect to find no change—or signs of an increased commitment to farming. If, on the other hand, the net effect of reforms was to make farming less attractive, one would expect to find indications of a faltering commitment to agriculture in the RAs.
1 Fertilizer prices were adjusted upwards no fewer than six times during the 1998/99 season, and even commercial farmers who had prepaid at the old price prior to delivery had to pay the difference in order to have fertilizer delivered. 2 Elbers, Gunning and Kinsey (2007) have recently shown that risk has a very substantial effect on capital accumulation (and hence on poverty). The average (across households) expected long-run capital stock is estimated to be 46 per cent lower than in the absence of risk. 3 It would be erroneous however to assume that the equipment typically used in small-scale farming in Zimbabwe has a short economic life. Farmers report that they utilize the prevalent ox-drawn equipment sometimes for periods of 30-40 years and purchase only the occasional spare part to repair what wears out or breaks.
Deleted: ¶
14
Figure 5. Comparative rates of investment in agricultural capital equipment,
RAs (top lines) and CAs (lower lines), 1980 – 1997
Evidence of impact. A detailed analysis was carried out of the involvement of RA households in nonfarm income-earning activities over the period 1992 to 1997.1 The general trends can be clearly seen in Figure 6. The proportion of RA households earning income from nonfarm activities more than doubled in the four years from 1992 to 1996. At the same time, the average number of nonfarm activities undertaken by the average household rose six-fold.
0
10
20
30
40
50
60
70
80
90
100
1992 1993 1884 1995 1996 1997
Year
Prop
ortio
n of
hou
seho
lds
0
1
2
3
4
Mea
n nu
mbe
r of a
ctiv
ities
/hou
seho
ld
Proportion of households with positive income from nonfarm
activities(left scale)
Mean no. of activities(right scale)
Figure 6. Indicators of participation rates in nonfarm activities, 1992-97
1 These are the calendar years which compare with the growing seasons used in the analysis of cropping incomes. The details of the analysis are too extensive for inclusion here, but tables reporting the characteristics of households making transitions into and out of nonfarm-income-earning categories are available upon request from the author.
Deleted: ¶
15
Over the same period, mean nonfarm earnings per RA household also rose continuously, as shown in Figure 7. There are indications however that the rapid expansion of nonfarm activities has nearly exhausted the earning possibilities from those activities that have low barriers to entry in the form of capital. In real terms, the earnings from each activity undertaken in 1996 were less than half what they were in 1992 (Figure 7). And, by 1997, the effects of the rapidly escalating inflation were to reduce the real nonfarm earnings per household and per activity to only slightly above zero when compared to 1992.
0
100
200
300
400
500
600
700
1992 1993 1884 1995 1996 1997
Year
Inco
me
(Z$
1990
) Earnings per household
Earnings per activity
Figure 7. Real household mean total and average earnings from nonfarm activities,
1992-97 It has to be assumed that households diversified out of agriculture into nonfarm activities for one or more of several reasons. First, farming may simply have been no longer as profitable as it was in the 1980s, when smallholder production expanded so rapidly. The gains from the rapid adoption of new technology in the first five post-independence years had been exhausted.1 Second, farming may still have offered levels of profitability just as attractive as in the past, but more difficult access to credit may have made securing the inputs to earn those profits much more difficult. Thus, in the face of imperfect financial markets, households have turned to nonfarm activities to finance farming operations. Finally, the reforms may have had the intended effect and stimulated a strong response in resettlement areas in the form of diversification of the economic base.2
The major policy change that has affected households in the panel is not the adjustment reforms however but Zimbabwe’s land reform programme, which provided them—in 1980 and 1981—with access to the land they now cultivate. While there is no doubt that real incomes have grown for resettled households—and grown in ways that tended to reduce inequality,3 the detailed analysis of growth rates has thus far been extended only as far as 1995/96, the first good season in the 1990s. The evidence in this section however points at a weakening of the policy environment that favoured growth in the early 1980s. While some of the outcomes expected from economic reforms have materialized, these changes have occurred in a setting where they appear to have been offset by other, countervailing forces.
1 This is Rohrbach’s (1989) contention. He concludes that—for maize—the dramatic growth of output in the first half of the 1980s represented once-for-all gains arising from freeing smallholders from a set of previously inhibiting restrictions. 2 Although the data are available to explore each of these propositions in detail, such an investigation is beyond the scope of this paper. 3 See Gunning, Hoddinott, Kinsey & Owens (2000) and Elbers, Gunning & Kinsey (2002 & 2007).
16
Poverty Dynamics among Resettlement Households in Zimbabwe To this point inequalities and poverty—two measures of changes in welfare over time—and economic policy and poverty reform in Zimbabwe have been examined. Here attention is turned to an analysis of the dynamics of poverty among households in the resettlement panel. The data for the poverty analysis. Exploring the impact of changing government policies on poverty over time requires a panel data set that extends over a long period. The analysis here draws upon the same panel data set utilized earlier. The data used cover 1991/92, a year characterized by a severe country-wide drought, and 1994/95, another very dry year. The data also include long periods both prior and subsequent to the introduction of adjustment in 1991. Here primarily the portion of the data set covering the resettlement areas is utilized. Specifically, extensive use is made of a subset of 354 households for which the same set of variables exists for each household over a period of six years in the 1990s.1 Various definitions of income are possible. The one used here is based on the sum of the following components of the annual income stream:
• the value of crops actually sold (net of credit repayment costs) plus the value of crops grown but not sold and valued at the median selling price for that crop;
• the gross value of sales of livestock; • the gross value of sales of any livestock products (eggs, milk, hides, etc) and livestock
services (plowing, transport, etc); • the gross value of off-farm earnings (petty wages, etc.) and income from nonfarm business
activities; • the gross value of any female-controlled income-generating activity not counted elsewhere
(market gardening, craftwork, brewing, etc); • remittances in cash (but no value is attributed to remittances in kind); and • the value of aid or transfers received in cash (but not those in kind). In the late 1990s, six nationally calculated poverty lines existed in Zimbabwe. As Cavendish (1999) has shown, there is no clear consensus at all among these surveys on the definition of a rural poverty line for Zimbabwe. In order to make a comparison, Cavendish deflates (and inflates) the various poverty lines. His results indicate, taking the poverty lines calculated for 1993/94 for example, that the ICES 1990/91 defines a poverty line that is about half that of the later ICES 1995/96, and just under two-thirds the PASS 1995 poverty lines. He notes that even poverty lines produced in the same year (PASS 1995 & ICES 1995/96) differ by at least 25 per cent. These surveys therefore provide no clear lead as to where an acceptable poverty line for rural Zimbabwe might lie. Thus, analysts are forced into the position of calculating poverty measures specific to their own studies. Below the ICES 1995 (GOZ 1998) poverty line is used. The common practice is followed here of using three measures from the Foster-Greer-Thorbecke (FGT ) class of poverty measures (P). The general class of P poverty indices is calculated based on a poverty line and a population density function of income. The indices used here are P0, P1 and P2. These measures are extensively described in the poverty literature but, in simple descriptive terms, P0 is the headcount index, a measure of the incidence of poverty, P1 is the poverty gap index, a measure of the depth of poverty, and P2 is the FGT index, or an indicator of poverty severity. The headcount measure is regarded as purely an informative index, indicating the percentage of households in a sample that are poor, while P1 and P2 are more in accord with normative perceptions concerning the measurement of poverty. Inequality among households. At the start of the resettlement programme, virtually all households were poor or near-poor (Kinsey 2004). Poverty was one of the more important criteria for inclusion in the programme. Other criteria were landlessness or effective landlessness, and economic disadvantages. Each household was allocated an identical amount of arable land, while grazing land was provided on a communal basis and varied by agro-ecological zone. At the start of the programme, therefore, differences in incomes of the households may be due to differential wealth taken from their previous homes, but only to the extent that this wealth—perhaps in the form of draft animals or agricultural equipment—had a pay-off in income terms. Other sources of inequality can be traced to differences in skills, in luck or the timing of the arrival at the new location, and to household composition.1
1 When the data for 1998-2001 have been fully edited, it will be possible to extend the analysis here by a further four years.
Deleted: ·
Deleted:
Deleted: ·
Deleted:
Deleted: ·
Deleted:
Deleted: ·
Deleted:
Deleted: ·
Deleted:
Deleted: ·
Deleted:
Deleted: ·
Deleted: ¶
17
On the basis of the ICES poverty line for 1995, adjusted to 1983 prices, the initial situation of those resettled can be characterized as uniformly income poor.2 In the three areas together, more than 99 per cent of the people in the households were below the poverty line in income terms. The Gini coefficient over all regions was 0.62, reflecting a fairly high degree of inequality in that year. This range of per-capita incomes among poor households is also reflected in the measures used to define the poverty gap: the P1, which amounted to 0.86, and the P2 that stood at a high 0.78. To check the effects of household composition, these measures were also calculated on the basis of the household as a unit, irrespective of the number of household members.3 In this case, the Gini amounts to 0.63, hardly different from the 0.62 calculated earlier. Hence, the initial inequality was not due to differences in household composition among resettled households. In later years, poverty has decreased, and so has inequality.4 While the headcount index declined to 0.73 in 1995/96, the Gini fell to 0.41 (and to 0.47 on a household basis). Figure 8 shows the evolution of the P0 and the Gini over the period 1991/92 to 1995/96, with 1982/83 added as a
benchmark. Both poverty and inequality came down from the early 1980s to the mid-1990s. It is interesting to see that rainfall matters enormously, not just for the headcount index, but also for inequality. Rainfall obviously affects income positively and appears to affect equality the same way, although it is far from obvious why the latter effect should obtain.
Figure 8. Gini coefficient and headcount over time, with rainfall deviation
1 Analysis by Hoddinott, Gunning, Kinsey & Owens (2000) shows that the effects of initial conditions on incomes tend to weaken with time after households are resettled. 2 Kinsey, Burger & Gunning (1998) show that poverty measured in terms of assets was less uniform at the outset. Because the programme was designed on the basis of small family farms, it was intended that all those who were resettled should possess draft animals—a curious requirement for a poverty-focused programme in a post-war setting. Some 40 per cent however did not own any cattle when they were resettled. 3 The size of the household in the panel surveys is defined as those members who normally stay and eat at the family’s musha (homestead). Excluded, therefore, are family members working away or away looking for work, students at boarding school, children who are temporarily being fostered by others, and so on. 4 This theme is the particular focus of the analysis in Hoddinott, Gunning, Kinsey & Owens (2000) which shows not only that incomes have risen but also that they have risen in such a way as to reduce income disparities.
18
Household composition is highly likely to exert an important influence on the poverty measures. When the weight of the children below 12 years of age is adjusted from 1.0 to 0.5, the headcount index of course declines. Table 2 includes the available measures with this adjustment made for all years from 1992/93.1
If 30 per cent of all persons had a per-capita income above the poverty line in 1992/93 (i.e., 70 per cent were below the line—see Table 2), this percentage rises to 39 (i.e., 61 per cent were below the cut-off) if children receive a weight half that of an adult. The trend in the measures over time remains the same: in the drought years 1991/92 and 1994/95, poverty was most severe.
Table 2.—Poverty measures for resettled households*
*Based on the ICES poverty line of Z$ 1,760 in 1995 (GOZ 1998)
Comparing the three natural regions across the years, the least-favoured region shows the highest incidence of poverty—100 per cent in the early years, and 84 per cent in 1995/96—while the best region—NR 2—shows a score for P0 of 98 per cent in 1982/83 against 70 per cent in the final year.
Inequality is less affected: in NR 2, the Gini coefficient for the 1994/95 drought year, based on all persons, is 0.49, the same as in NR 4, while in NR 3 it is only 0.38. This is not a structural feature however. The coefficients change rather strongly over time, and the relative positions of the three regions change as well.2 On the basis of nationally defined criteria, the measures show that the incidence of poverty is very high in these resettlement areas. They also indicate a gradual improvement over time and a strong link to rainfall. Nevertheless, almost 42 per cent of all households find themselves below the poverty line in all five years 1991/92 to 1995/96. In relatively favourable region NR 2, this percentage is 35, in NR 4 it is 63. No more than 0.3 per cent escaped the experience of poverty in these years, and 95 per cent had at least one year in which income fell below the poverty line. Some 72 per cent of the households were in the poor group in two years or more. Characteristics of income groups. What are the characteristics of these two groups defined in terms of incomes? Here the differences are examined between the poor and non-poor across selected years, as are the characteristics of the groups that move from poor to non-poor and vice versa. Selected characteristics are income, household size and composition, age and education of the head of household, cultivated area and land use and livestock assets. The analysis here is based on the 354 households that are in the sample in each year. The proportions of this number of households in the poor group differ, of course, as incomes vary from year to year. (See Table 3.) The definition of the two groups in Table 3 is based on the 1995 ICES poverty line, with weights for children as indicated.
1 Although some sensitivity analysis has been carried out on weights assigned to children, without extensive reworking of the original data the comparison cannot be made for all years. Nor can it be made for the earlier years in terms of full adult-equivalents using WHO weights because ages were recorded not in years but as codes indicating age brackets. 2 Again, the detailed tables are available from the author.
Deleted: l
Deleted: non-poor
19
Table 3.—Proportion of households below and above the poverty line
Season Group and child weights 1991/92 1992/93 1993/94 1994/95 1995/96
(proportion of 354 households)
Poor (weight 1.0) 99.9 59.0 71.2 91.5 67.5
Poor (weight 0.5) 48.9 61.9 85.6 56.5
Non-poor (weight 1.0) 0.1 41.0 28.8 8.5 32.5
Non-poor (weight 0.5) 51.1 38.1 14.4 43.5
The more detailed analysis underlying the results reported here make it clear that poor and non-poor groups differ in many respects. The most obvious difference is household size, which is on average over all years 19 per cent smaller in non-poor households.1 Years-of-education is 19 per cent higher, and the emphasis on cash cropping, such as cotton, is also much higher in the non-poor group. Cultivated area is 11 per cent greater and cattle numbers are 19 per cent higher, while overall herd value is 29 per cent higher in the richer group. (These averages are calculated with the number of households taken as weights.) It should be noted however that these differences in characteristics are small compared with the enormous difference in per-capita incomes in the two groups—per-capita income in the poor group is merely one-fifth that of the non-poor group. Looking at the mean characteristics of households that change from one wealth category to another helps shed more light on what drives transitions into and out of poverty. Here the focus is on some transitions that are relatively well populated with households, namely those between 1991/92 and 1992/93 and between 1994/95 and 1995/96 for the poor group. This latter transition may show why some move out of poverty after a drought year and others do not. The transition between 1992/93 and 1993/94 for the non-poor group is examined specifically to see why some became poor (again) in a moderate year after a relatively good year. Distinguishing features of the income-poor households that become non-poor in the two periods after a drought year are: they were less poor in the first place (40 per cent less poor), had smaller families (by 18 per cent), slightly younger heads with more (20 per cent) years of education, cultivated more land (15 per cent more in 1991/92, 25 per cent more in 1994/95), grew less maize and more cotton, and possessed more wealth in the form of cattle (30 per cent more in 1991/92 and 48 per cent more in 1994/95).2 Distinguishing features of the non-poor households in 1992/93 that become poor in 1993/94 are that they were relatively less well-off in the initial year, especially where non-crop income is concerned, cultivated more land and had smaller household sizes in the initial year, but quite the reverse in the terminal year. They were more oriented towards food production and less to cash crop production. Herd sizes were larger. It cannot be established with certainty whether having a larger number of cattle helps in moving up the income scale, or whether it is a consequence of higher incomes. As can be seen from Figure 9, over the entire period covered by the panel data set, herd sizes have grown remarkably steadily from one year to the next at the same time that the proportion of families without cattle has declined from some 40 per cent to under 10 per cent (Kinsey, Gunning & Burger 1998). Thus, while the accumulation of cattle is treated here as capital gains rather than investments out of income, it is clear that this approach is not entirely correct. For many households, the acquisition of cattle has clearly been an investment. It will also be clear, whether growing cattle numbers reflect investments or capital gains, that the ability to realize returns from a given herd size differs
1 It should be noted however that household size among the panel households is remarkably unstable. It is not uncommon for the number of persons resident in a household to more than double from one year to the next, a phenomenon related both to the loss of jobs in the formal sector and to the growing numbers of AIDS-related deaths. It is equally common for the number of resident persons to decline by 50 per cent or more. Thus all per-capita comparisons of income or wealth on a year-to-year basis are bedeviled by profound shifts in the underlying demography. 2 These features do not change substantially if household income measure is recalculated on the basis of adjusted per-capita income with lower weights for children.
Deleted: non-poor
Deleted: ¶
20
dramatically from season to season. Nevertheless, households in the panel are in a far better position to smooth consumption through livestock sales now than they were in the early 1980s.
Figure 9. Size and value of domestic cattle herds, 1983 – 2001
Figure 10. Crop income per capita, by land-man ratio (Z$ 1990)
The characteristics discussed above apply to the households in the first year. They may not fully explain therefore why some households fared better than others, as changes that have taken place from one year to the next are not taken into account in this treatment. Such changes may occur particularly in household composition relative to the fixed land resource. As shown in Figure 10, per-capita crop incomes are related to per-capita land-use. In broad terms one observes that
21
households that cultivated more than one acre per person fall into the non-poor category, while those below this level tend to be classified as poor. This finding does not apply to the drought year 1991/92, when all households had very low crop incomes and when the classification by poverty category is attributable to nonfarm sources of income. These are shown as the lowest two points in Figure 10. The other point below the fitted line refers to the other drought year 1994/95. The group means of the households that changed from the non-poor group to the group below the poverty line all show a decrease in land-man ratios. And the land-man ratios of the households that were non-poor are almost all greater than one, as shown in Figure 11. Figure 11 shows that, again, the land asset cannot adequately proxy for incomes in drought years such as 1994/95. But in other years the point is clear: those that become or remain poor tend to cultivate little land per resident household member.1
Households that cannot manage the land they have access to may turn to other solutions.2 One is to alter household composition. This can be done by sending household members elsewhere, or by hosting additional family members or other persons. As shown in Figure 12, households classified as non-poor show more variation in size than those in the poor group. The same holds for the areas they cultivate—as Figure 13 shows. These findings do not necessarily imply that households make adjustments as part of a consciously selected coping strategy. It might simply be that the criteria utilized to define the groups mean that small households are relatively non-poor in some years.
More concrete evidence comes from the groups that change. Here the selection of households is kept constant, but the variation in household characteristics from one year to the next is checked. In other words, the focus is on selected economic and social dynamics of households that make a transition from one poverty category to another. It is a drawback however that only one year-to-year change is considered here.
1 This point is critical, particularly in view of the very rapid and politically motivated redistribution of land that began in early 2000. 2 The land market, however, is not one of these. Both because of restrictions and the physical distances created by the layout of resettlement areas, renting of land is not a managerial option for most RA farmers.
Deleted: l
22
Figure 12. Household size, 1991/92 – 1995/96, by poverty status
It was mentioned above that, among poor households, those that became non-poor were smaller and managed to cultivate larger areas initially. Detailed analysis shows that these households adjusted their cultivated areas and household sizes in the course of this transition: between 1994/95 and 1995/96 poor households that became non-poor reduced household size by 6 per cent and increased cultivated area by 7 per cent, compared to poor-households-remaining-poor, which increased both by 2 per cent. The richer households that remained non-poor between 1992/93 and 1993/94 grew in size by 4 per cent but increased cultivated area by 17 per cent, while those that fell into the poor group in 1993/94 increased in size by 32 per cent and reduced cropped area by 9 per cent. Obviously, not all changes in household size are aimed at achieving higher income per capita; distress migration undoubtedly explains some of the changes observed. These changes are however relatively large compared to other differences between the two sub-groups.1 The changes from one year to the next indicate that knowing the point of departure of a household is not enough—changes within households between years are often as large as the initial differences between the households that moved from poor to non-poor and those that did not. Income dynamics. As is shown in Figure 14 below, rainfall has an enormous impact on poverty among resettlement households. Rainfall affects not only the level of poverty but also equality among households. This section looks in more detail at how changeable incomes are from one year to the next. Here an examination is made of mean household incomes rather than per-capita incomes. Table 4 shows the levels of real income and its constituent components over the growing seasons 1991/92 to 1996/97. Incomes clearly vary drastically over the years. The decomposition details show that in drought years—such as 1991/92 and 1994/95—sources of income other than agriculture become relatively much more important, notably off-farm income and income from businesses. The value of food aid in these years is far from negligible as well. Incomes also depend on the agro-ecological zone in which the farmers operate. Natural region (NR) 2 is the area with the highest agricultural potential; NR 4 has the lowest potential; and NR 3 is in an intermediate position.
1 The changes in household size discussed here are relatively modest compared to those observed between 2000 and 2001, when CA households doubled in size and RA households shrank by a fifth—to their average size 18 years earlier (Kinsey 2010).
23
Figure 13. Cultivated acres per household, 1991/92 – 1995/96
It has been noted that 1991/92 was a severe drought year. In this year incomes across the three areas vary relatively little (Table 5). The 1994/95 season was another drought year, but the effects are less noticeable in NR 3 than in the other two areas.1 Table 6 presents correlations between average rainfall and the different components of total household income. The table illustrates two points. First, there is a strong correlation between rainfall—even when expressed as a national average—and total income. Second, the table suggests avenues that these agricultural households may have to pursue to attain more stable income levels. Crop income is the most important part of total income in all but the very worst years. Income from crops can differ substantially from one year to another, however, both because of yield changes and also because changes in cropping patterns may occur. Prices may vary as well. Each of these possibilities is now examined in more detail.
Table 4.—Breakdown of main sources of income, by year
Real incomes Harvest year 1992 1993 1994 1995 1996 1997 Total gross income (Z$ 1990) 1,091 3,315 2,381 1,338 2,919 2,202 Proportion of gross income from: (per cent) Agriculture 0.19 0.81 0.86 0.53 0.80 0.71 Livestock 0 0.02 0.01 0.05 0.01 0.02 Off-farm income 0.23 0.01 0.01 0 0.05 0.04 Business revenue 0 0.04 0.05 0.20 0.08 0.12 Female income 0.17 0.02 0.03 0.08 0.04 0.06 Remittances 0.16 0.01 0.03 0.04 0.02 0.06 Value of aid received 0.25 0.09 0.01 0.11 0 0 Average rainfall (mm) 335 630 519 419 701 750
1 There is a temporal effect at work in NR 3. Between the 1992 and 1995 droughts, one of the six villages in NR 3 was equipped with an irrigation scheme. In addition, households in this area are increasingly turning to market gardening to supply a nearby and rapidly growing administrative and business centre. It is said that, as a consequence, farmers are neglecting their rainfed field crops—which may help explain why NR 3 performed relatively badly in the good years 1995/96 and 1996/97.
24
Table 5.—Total household income by natural region and growing season
Growing season Agro-ecological zone
1991/92 1992/93 1993/94 1994/95 1995/96 1996/97
(Z$ 1990)
NR 2 (n=174) 1,090 3,950 3,035 1,426 3,622 2,583 NR 3 (n=81) 1,146 2,636 1,494 1,423 1,907 1,285 NR 4 (n=45) 1,000 2,081 1,449 846 2,025 2,382
Table 6.—The correlation between average annual rainfall and components of total household income (300 households over six years)
Correlation coefficient using yearly averages
Correlation coefficient at household level
Average annual rainfall (mm) 1.00 1.00 Gross total income 0.77 0.26 Gross income from agriculture 0.75 0.31 Gross income from livestock products -0.31 0.07 Gross off-farm income -0.26 -0.03 Gross business revenue 0.14 0.13 Gross female income - 0.30 -0.02 Remittances in cash - 0.32 -0.03 Value of food aid - 0.04 -0.10
Yield variations. Yields are, of course, highly sensitive to weather in rainfed agriculture. The correlation coefficient between maize yields and rainfall is high—0.77, while those for yields of groundnuts and cotton are 0.76 and 0.65 respectively. Figure 14 shows the rainfall-yield relationship graphically for maize.
Figure 14. Average rainfall (mm) and average maize yield per acre (kg/acre)
Source: Department of Meteorological Services and Famine Early Warning System (based on 1200 observations on maize yield)
Deleted: ¶
25
Cropping patterns. It is not an easy matter to adjust cropping patterns to unreliable patterns of rainfall. Although cropping patterns differ in each natural region and have been modified to suit average circumstances, ex ante adjustments to expected weather are very difficult to achieve for both small- and large-scale farmers, not least because of extreme uncertainty regarding the availability and price of purchased inputs. (Figure 3 gives an idea of the variability in RA cropping patterns for the six years following the introduction of adjustment.) Price patterns. Although rainfall is the dominant factor causing instability in incomes, prices fluctuate as well, as shown in Table 7.
Table 7.—Nominal median farmgate crop prices (per kg)
Price fluctuations may have a stabilizing effect on agricultural income however, as can be inferred from the negative correlation between prices and yield, illustrated in Table 8 below for the three most important crops. It must be noted however that any such stabilizing effect is restricted because of different cropping patterns in the three areas. While maize is universally grown, cotton is grown widely only in NR 2 and groundnuts are more extensively grown in NRs 3 and 4.
Table 8.—The correlation between farmgate prices and yields, 1990/91 - 1996/97
with yield Real median farmgate prices per kg (Z$ 1990) Maize 0.20 0.45 0.36 0.30 0.29 0.20 0.17 -0.63 Groundnuts 0.30 0.45 0.36 0.27 0.33 0.32 0.24 -0.85 Cotton 0.78 2.87 0.87 0.94 1.04 1.09 0.93 -0.79
The price variability in conjunction with varying cropping patterns is not sufficient to have a significant impact however, as the coefficient of variation of income is still 66 per cent. Prices do however have a substantial effect on incomes in the three areas. As farmgate prices are more closely related to the economic policy pursued by the government, this fact provides an important link to the overall economic environment. All prices heretofore have been deflated by the consumer price index, which better represents urban realities. It is possible however to define a producer price index, and this is done in Table 9 and shown in Figure 15.
Table 9.—Derivation of a producer price index equivalent to the CPI in 1990/91
Source: CPI is the mean of CSO monthly data (1990 = 100). The PPI is based on the average weight of the different crops grown (1990/91-1996/97) and valued at median crop prices.
Deleted: ¶
Deleted: 6
26
Figure 15. Consumer and producer price indices, 1990/91 – 1996/97
Given that rainfall and incomes are so volatile, households must utilize a number of smoothing mechanisms. In addition to the sale of animals discussed earlier, another important mechanism is to hold maize stocks (Figure 16). Liquid money savings balances are also employed for this purpose (Figure 17).
Figure 16. Maize retentions (bars) and average rainfall (line), 1990/91 – 1996/97
27
Figure 17. Household liquid savings balances (bars) and average rainfall (line),
1990/91 – 1996/97 The earning of off-farm or nonagricultural income appears to be gaining in importance. In the past, earning from nonfarm activities in the resettlement areas appeared to be done typically in response to adverse weather (and reduced agricultural incomes). With the introduction of adjustment reforms, however, it became less sporadic and grew rapidly, as shown earlier. Incomes from remittances also appear to increase in difficult times, as does income from female-controlled activities (mainly small-scale market gardening and craft activities). Livestock sales are a typical means of coping with negative income shocks. Figure 18 illustrates the relationship between livestock sales and rainfall. While changes in livestock numbers respond to specific climatic conditions and the accumulation of livestock can be temporarily disturbed by adverse conditions, households have nevertheless been able to accumulate more and more livestock over the years. Kinsey, Gunning and Burger (1998) show that the average household moved from a mean herd size of 4 bovines in 1983 to almost 10 in 1995. Mean herd sizes tended to continue to grow in the immediate years since—in three out of four years—as Figure 9 shows. The mean herd size in 1999 was almost 20 per cent larger than four years earlier. Changes in Welfare over Time: the Dynamics of Child Nutrition Money-metric indicators of welfare and poverty are employed throughout this paper. The panel data set however is rich in possibilities to construct nonmoney-metric measures as well. This section reports on one such variable: changes in the nutritional status of children over time. From the outset, the ZRHDS collected anthropometric data in order to be able to document objectively broader changes in household welfare. It is contended that, if child nutrition declines over time, there has likewise been a decline in household welfare even if money-metric indicators move in the opposite direction. The nutrition data from the panel study can be best understood if it is appreciated that they come from what is a moving cohort sampled across many years. In 1983 and 1984, all children aged between six months and five years and resident in the household on the day of the visit were weighed and measured. The same procedure was followed in all subsequent years except that the age cut-off point was moved to six years in order to include as many children as possible from
Deleted: 9
28
previous survey rounds.1 Thus the pool of children included in any given year will contain new children who have attained an age of at least six months at the time of the visit to the household and will drop older children who are then above the age of six years. To the extent that there are secular influences from incomes or poverty on long-term child nutrition, these will be manifested as each year's recruits to the cohort grow to the age of six years and then exit the cohort.
Figure 18. Value of livestock sales (bars) and average rainfall (line),
1991/92 – 1996/97 Another feature of the nutritional data needs to be borne in mind as well. With a panel extending over 28 years, the supply of new entrants to the cohort comes less and less from the children of mothers who were bearing children in the early 1980s. Indeed, increasingly the panel includes the children of children who were themselves assessed in the early 1980s, or the grandchildren of the original heads of household. This fact means inevitably that genetic influences join environmental and economic influences as determinants of children's biometric indicators. With these comments, two of the conventional anthropometric indicators—height-for-age (HA) and weight-for-height (WH) are plotted as median z-scores (Dibley et al. 1987) in Figure 19a for each year in which anthropometric data have been collected.2 Low height-for-age is considered an indicator of chronic undernutrition (shortness or stunting), which is frequently associated with poor overall economic conditions or repeated exposure to adverse conditions, or both. Looking first at height-for-age, there was an improvement from 1984 to 1987. Underlying this improvement was undoubtedly the provision of improved health services to resettled areas as well as recovery from the 3-year drought of the early 1980s. This improvement in HA was however reversed in the late
1 The number of valid assessments obtained in each year averaged 682, with a low of 205 in 1992 and a high of 910 in 1993. In the early 1990s, anthropometric data from adults began to be recorded to broaden the basis for analysis. Results from early analysis of the combined adult and child nutritional data sets are reported in Hoddinott and Kinsey (1998a and b) and Kinsey (1998a, b and c). See also Alderman, Hoddinott & Kinsey (2006). 2 The hashmarks across the data series in Figures 19a and 19b serve as a reminder that coverage before 1992 was not continuous and that there is imperfect consistency between observations prior to 1993 and those from 1993 onward. These inconsistencies arise from a change in the techniques of weighing and measuring children and obtaining their date of birth and some variation in the time of year assessments were made. A third common indicator, weight-for-age, is not used here as it is primarily a composite of the other two and fails to distinguish tall, thin children from short, well-proportioned ones.
29
1980s or early 1990s, and by 1992 stunting was back at the level of eight years earlier, while the following year—1993—was the worst ever recorded. The 1993 outcomes were largely the result of the severe drought of the 1991/92 season, but they may embody also the early signs of the cutbacks in public health services. Following 1993, there was one year of marked improvement, but this was then succeeded by a resumption of the worsening trend. This trend may not be explained solely by variables related to food consumption, as it is likely that health-related factors—particularly the relentlessly spreading effects of HIV/AIDS—are likely to be involved also.1 Thus the linear trend shows that children in the three resettlement areas have tended to become more stunted, or chronically undernourished, over time, with the period 1995-2000 displaying consistently adverse outcomes. The worst outcome, in 1993, can in large measure be attributed to the severe drought in the 1992 season because children were assessed some 6-9 months after the failure of the 1992 harvest and before the 1993 harvest. Low weight-for-height is regarded as an indicator of acute undernutrition (thinness or wasting) and is usually associated with failure to gain weight or a loss of weight. Paradoxically, this indicator of acute undernutrition exhibits a slight improving trend over time, and it is a puzzle that this indication of dietary improvement is not reflected in a lagged improvement in height-for-age. Instead, the much more positive z-scores for weight-for-height, and the upward trend, simply tell us that children’s weight is proportional to their height, and increasingly so over time. In summary, Figure 19a shows a somewhat mixed picture, but one in which chronic undernutrition as assessed by height-for-age has worsened over time. The resettlement experience has not, therefore, led to general improvements in food security sufficient to reduce this dimension of chronic undernutrition.
Figure 19a. Changes in children's nutritional status as measured by z-scores,
1983/84 - 2001 1 A source of bias would exist if AIDS-affected children were dying between the annual rounds of the survey and thus were not being recorded as badly ill or undernourished, but there is no evidence that this is the case. There is instead evidence that the relatively isolated setting of many of the panel villages, together with the prohibition—effective until 1992—on men taking urban jobs, delayed the onset of the full range of AIDS-related miseries.
30
Figure 19b presents a different perspective. The plots in each case represent the proportion of the children assessed lying below two standard deviations below the mean—a level commonly defined as severe undernutrition. HA—or stunting, as noted above is an indicator of chronic nutritional status and underlying child health, and exhibits the most dramatic changes. In 1983/84 some 34 per cent of children were severely stunted. In the following assessment period—1987—the extent of severe stunting dropped by more than a third. This remarkable improvement was a consequence of several factors. Among them were cost-effective community-based health interventions that resulted in rapidly improving access to health services and even more striking improvements in child immunization rates. And although the early 1980s experienced three consecutive years of drought, an effective drought relief programme meant that crop failures were not experienced in the form of pronounced checks to child growth.1
Figure 19b. Changes in the proportion of children severely undernourished,
1983/84-2001 Over time, the proportion of children who are severely stunted (HA) has ranged from 23 to 37 per cent, but the trend is for a growing proportion to be severely stunted. Ten per cent more children were likely to be seriously stunted at the end of the period than at the beginning. The second indicator in Figure 19b also shows a worsening trend over time. WH, which consistently had levels of severity below 10 per cent of children prior to 1994, exhibits higher mean levels of severe undernutrition in the subperiod beginning in 1994, and the likelihood of a child being acutely undernourished increases by 100 per cent over the entire period. Taken together, Figures 19a and 19b suggest that two related processes are taking place together. The z-scores of Figure 19a are median values, so we know half the children assessed will have better scores than those plotted. And the other half will have worse scores. It is this latter group that generates the consistently worsening results shown in Figure 19b. What appears to be happening is that serious child undernutrition is becoming increasingly concentrated in one group of households and, moreover, children in this particular group of households are becoming increasingly badly nourished.
1 The children in the panel households were also changing over this period. There had been a post-war baby boom, which increasingly saw children conceived, carried and born after independence replacing those who went through early childhood in the stressful war years of the late 1970s.
31
While it is far from obvious what is driving the changes observed, they are consistent with two possible but very different explanatory variables. First, they match well the timing of the reversal of other healthcare indicators—infant, child and maternal mortality—at the national level. This pattern is explained in part at least by the fact that real per-capita health spending—which increased more than 60 per cent between 1980 and reached its peak in fiscal 1990—and the launch of ESAP in 1990/91—was in the late 1990s marginally lower than at independence (GOZ 1998). A second factor that may help explain worsening nutritional outcomes during the 1990s is the fact that every growing season between 1988 and 1996—seven consecutive years—experienced below long-term-average annual rainfall, including the two serious drought years of 1992 and 1995. Then when heavy rains came, as they did in 1996 and 1999, they brought with them national epidemics of malaria, which is particularly serious in the case of already undernourished children. Nutritional Status as an Indicator of Poverty If policy-makers are to assist the poor and vulnerable efficiently, they must be able to differentiate them from others in society. In order to do so, they require particular criteria which are easily and inexpensively employed and which are not prone to the risk of moral hazard. In market economies, means-testing and categorical indicators of various sorts are commonly employed for this purpose. The use of such measures however requires a large amount of information. There are practical advantages therefore to defining poverty in ways that can be measured with small data sets which do not rely on income-based definitions of poverty. Particularly useful would be multi-dimensional concepts of poverty in which a relatively small set of indicators can encapsulate the much broader set of underlying concepts. A basic question therefore is the extent to which households identified as being poor using conventional criteria, such as income and expenditure, are also identified as poor when nutritional criteria for children and adults are used as well. At the inception of the research, a decision was made to use children's nutritional status as a key indicator of rural welfare. The research design thus made the use of anthropometric measurements for children a feature from the outset. The anthropometric data are complemented by data on food production and expenditure, child-feeding patterns, use of wild foods, drought relief, morbidity and mortality, and a wide range of additional data. From 1994, the data for children have been supplemented by measurements on adults. By taking weight and height measurements of adults, their body mass index (BMI) may be calculated and used as an indicator of chronic energy deficiency. A review of the application of the BMI (Shetty & James 1994) concludes that BMI in adults is a responsive index, sensitive to changes in nutritional status which are influenced by socio-economic status, seasonal fluctuations in food availability and level of physical activity. On this basis, BMI has been judged to be useful as an indicator for monitoring nutritional status. Further, it is argued that measures of the prevalence of adult undernutrition are "likely to be a better indicator of and reflect more truly the nutritional status of the community than estimates of childhood undernutrition alone" (Shetty & James 1994, v). This section is based primarily on the analysis of a single year's data—1997—from the longitudinal survey including observations on the nutritional status of both children and adults, household resources, incomes and expenditures, and other socio-economic variables. The nutritional status of children is compared with that of adults in the same household using standard anthropometric measures of children and the BMI of adults. How useful are nutritional indicators as potential tools of the analyst addressing food security and poverty issues in rural Zimbabwe? It would be useful to know, for example, how good the indicators of nutritional status are in agreeing with more traditional poverty indicators such as household resources, incomes and expenditures, and other socio-economic variables. In particular, what additional contribution does the collection of nutritional information on all household members—adults as well as on children—make in this regard? Do different distributions of adult and child nutritional status within households reflect fundamentally different socio-economic situations? And, finally, do the data available support the hypothesis that, when adults are thin, food insecurity rather than factors such as health or sanitation is more likely to be the dominant causative factor? By correlating adult BMI to child anthropometry and to the socio-economic data, the strength and validity of the use of body mass index of adults is assessed both as an indicator of nutritional status of a wider population and as a proxy for indicators that are more difficult to collect.
32
Preliminary analysis of the nutritional data revealed that, contrary to all expectations, children’s’ nutritional levels in RAs are lower than virtually anywhere else in the country. It was originally thought that this outcome might have been a consequence of the experience of relocating at a time of great environmental stress—the 3-year drought of the early 1980s. Subsequent analysis suggests however that the relatively poor nutritional status of children in RAs reflects persistent structural causes (Kinsey 1997). Not only has poor nutritional status been reflected in every round of research, but it has also been confirmed nationally by the Demographic and Health Surveys of 1988 and 1994 (GOZ 1989, 1995). The finding is also corroborated by the findings of the 1995 and 2003 poverty assessment studies (GOZ 1996, 1997, 2006). Here an examination is made of the proposition that undernutrition is structural in the RAs by combining indicators of the nutritional status of children and comparing these with the BMIs of adults from the same household. The point of departure is an assessment of the extent to which households identified as being poor using conventional criteria, such as income and expenditure, are also identified as poor using nutritional criteria. Although detailed explanatory analysis is impossible in a paper of this scope, hypotheses will then be suggested for cases where the outcomes diverge. These outcomes will be discussed according to i) the difference in mix between adult and childhood nutritional status and ii) location—differences between RAs and CAs and across agro-ecological zones. Data. In the 1997 survey round, fieldwork took place between late January and early April. Anthropometric data were collected for all children resident in the household on the day of the interview and aged between six months and six years. If present, the parents of the children were also weighed and measured. A household is included in the analysis if data were collected for at least one child-parent pair and excluded if data exist only for children. A total of 357 households is included, just over 65 per cent of the total of 547 households covered in the 1997 survey round. Of the total number of households included, three-quarters reside in RAs and one-quarter in CAs. The socio-economic data were collected at the same time as the anthropometric data and using the same format employed over many years. Although collected in early 1997, the data set captures the outcomes of the harvest from the 1995/96 season—a good harvest, livestock and nonfood consumption outcomes for the year preceding the interview, and food consumption and expenditure levels, as well as health indicators, for the month preceding the interview. Procedure. Two different procedures have been employed to capture the extent of undernutrition at the household level. The first (not reported here) attempted to derive a measure of the depth of undernutrition in households. A single indicator was calculated by taking the mean of all BMIs for all adults in the household. Similarly, three indicators were calculated for children by calculating separately the mean for each of the three z-score indicators: weight-for-age, weight-for-height and height-for-age. There were two difficulties with this approach. First, the use of means meant that inclusion of one individual with a score at the high end of the distribution could result in a household mean above the cutoff point even though two other individuals in the household might lie somewhat below the cutoff point. Second, the approach of using the three z-scores separately, combined with the averaging described above, resulted in very small numbers of households in some of the categories of interest for analysis. The second procedure used, and the basis for the reporting here, attempts only to identify those households where the phenomenon of undernutrition is present in adults, children, or both. Undernutrition is considered to exist among adults in a household if any one adult has a BMI below 18.5.1 Similarly, undernutrition among children is considered to exist if any one of the three z-scores for any child in the household lies below two standard deviations below the mean. Two new binary household-level variables—BMI and Z—were created and added to the set of socio-economic indicators. The variable BMI is set to 0 in cases where no adult in the household has a BMI score below the cutoff point and to 1 where any adult has a BMI score below 18.5. An identical procedure was followed in creating the second variable; Z is assigned a value of 0 if all of the children in the household have weight-for-age, height-for-age and weight-for-height z-scores above two standard deviations below the mean. If any one child has any one of the three z-scores below two standard deviations below the mean, Z is assigned a value of 1. Values of 0 for BMI and Z thus
1 An upper limit for the diagnosis of chronic energy deficiency using BMI has been defined as less than 18.5 (Shetty & James 1994).
33
indicate an absence of undernutrition—asymptomatic households, while values of 1 indicate the presence of undernutrition—symptomatic households. A simple way of assessing the potential usefulness of combining the nutritional status of adults with that of other family members in the household is to create a two-by-two matrix with undernourished and non-undernourished adults on one axis and undernourished and non-undernourished children from the same household on the other. The four cells of the matrix will contain: i) households with no undernourished children or adults (referred to below as 0/0 households); ii) households with both undernourished children and undernourished adults (referred to as 1/1 households); iii) households with one or more undernourished adults but no undernourished children (referred to as 1/0 households); and, finally, iv) households with one or more undernourished children but no undernourished adults (referred to as 0/1 households). If this approach is valid, each cell of the resulting matrix may be thought of as representing households that differ in significant ways. The 0/0 households exhibit no adverse nutritional phenomena and are therefore not regarded as impoverished or vulnerable. If both adults and children from the same household are undernourished—the 1/1 households, then this suggests the same pathways are affecting nutrition among the young and the old and that food availability—perhaps as a result of poverty—is most likely to be a major contributing factor. This finding suggests that policy measures are needed to address poverty and food security directly. The mixed cases present greater challenges to interpretation. If adults are well-nourished and children poorly nourished (the 0/1 households), one can conclude that the primary cause is not likely to be so much a lack of food—resulting from poverty—as poor intrahousehold distribution of food, poor child-feeding practices, or complications of nutritional status caused by child-specific health-related factors. This finding would provide evidence that attention should be paid to education and health aspects in the family. In the other mixed case, where adults are undernourished and children well-nourished (the 1/0 households), it may be concluded that the intrahousehold allocation of food is likely to be satisfactory but that adults are experiencing a situation in which physical activity levels are high in relation to the supply of food or the time available to prepare and eat nutritionally satisfactory meals is inadequate. An alternative or additional explanation is that adult-specific health-related factors, such as HIV/AIDS, are at work. Poverty, however, cannot be ruled out for these households. Thus the 4-way array of nutritional data described above can suggest something about the relative influences of the three basic determining factors of nutritional status: food availability, health and care. The crosstabulation procedure outlined above was applied first to the 1997 data set. To provide a comparison, the identical procedures were applied also to the 2001 data. The distribution of households in the two categories obtained by applying this approach to the 1997 data is summarized in Table 10a. Households are regarded as symptomatic if undernutrition exists and asymptomatic if it does not. Some 36 per cent of households in 1997 exhibited no sign of undernutrition on the basis of the z-scores of the children in the household, while nearly 80 per cent of resident parents had BMIs in the normal range.1 The significance level for the crosstabulation indicates that the hypothesis that intrahousehold child and adult nutrition levels are independent can be rejected. Examining the crosstabulations, the most common outcome—48.5 per cent of all cases—is a household in which the z-score for at least one child is below the cutoff point while the BMI for all assessed parents is above the cutoff point. The rarest outcome is a household where an undernourished adult exists but no undernourished child does; there are only 15 such cases—4.2 per cent of the total—in this category. The 'mixed cases' therefore represent over half of all outcomes.
1 Many instances exist where children were assessed anthropometrically but where no parent was resident to be weighed and measured. These cases are excluded here. It should be recognized however that this exclusion may bias the results since children who are being fostered, either as the result of the death of parents or because of a broken marriage, may be particularly prone to failure to thrive.
34
Table 10a.—Distribution of households by nutritional category, 1997
Children's z-scores
Symptomatic
(1)
Asymptomatic
(0)
Totals
Adults' BMIs
(per cent (n))
Symptomatic (1)
16.0 (57)
4.2 (15)
20.2 (72)
Asymptomatic (0)
48.5 (173)
31.4 (112)
79.8 (285)
Totals
64.4 (230)
35.6 (127)
100.0 (357)
X2 = 7.76 (df = 2; p = 0.0053)
Looking only at the 'pure' outcomes, i.e. those where neither adults nor children are undernourished, or both are, in 16 per cent of households undernourishment exists in both groups while it exists in neither group in 31.4 per cent of all cases. Comparing the distribution of households for 1997 (Table 10a) with that for 2001 (Table 10b) reveals striking similarities despite the lapse of four years. Despite a smaller sample in 20011 and thus somewhat lower significance levels, the patterns are almost identical. The fact that most of the children assessed in 2001 had not been born in 1997 lends support to the idea that patterns of undernutrition in RAs are structural in nature. And this point comparison provides no evidence of either dramatic worsening or improvement in families’ nutritional status. While there is a small increase in the proportion of asymptomatic (0/0) households in 2001, there is also a slight increase in the proportion of symptomatic (1/1) households. The largest single change (down by 8.7 per cent) is the reduction in the proportion of 0/1 households, suggesting that the nutritional status of children relative to that of adults in these households had improved, or vice versa.
Table 10b.—Distribution of households by nutritional category, 2001
Children's z-scores
Symptomatic
(1)
Asymptomatic
(0)
Totals
Adults' BMIs
(per cent (n))
Symptomatic (1)
16.2 (48)
5.7 (17)
22.0 (65)
Asymptomatic (0)
44.3 (131)
33.8 (100)
78.0 (231)
Totals
60.5 (179)
39.5 (117)
100.0 (296)
X2 = 6.23 (df = 2; p = 0.0130)
From this point the relationship between nutritional indicators and a set of poverty measures is examined for 1997 in two different ways. The first approach taken is to treat all the poverty measures as household-level means. Although analyses of poverty which treat the household as a single entity suffer from a number of theoretical and practical shortcomings, this approach is justified here because it provides comparability with most other studies of rural households conducted in Zimbabwe.2 The second approach, which is more defensible theoretically, is to focus on the household but define those poverty measures that are based on continuously distributed variables in per-capita terms. This approach adjusts for the considerable heterogeneity in household size in the population.3 Both approaches include comparisons between the set of poverty indicators and the combined nutritional indicators, and include as well a geo-tenurial stratification of households.
1 The smaller sample was caused in large part because many parents were away in early 2001 ‘occupying’ commercial farms. 2 See Kinsey, McQuie & Rukuni (1995). 3 The number of resident household members has ranged from 2 to 76 in some years.
35
Nutritional Indicators and Poverty Measures: Household-level Outcomes As noted earlier, the data available from the panel study allow construction of a wide range of indicators of poverty. Prior analysis, both of the panel data and of national-level data sets such as the 1990/91 Income, Consumption and Expenditure Survey (Zimbabwe 1994) and the Poverty Assessment Study Survey—PASS (Zimbabwe 1997) suggested that a set of variables highly likely to include valid indicators pointing to the extent of poverty in households should focus on consumption—both of food and nonfood items; food security; income—both from agricultural and nonagricultural sources; assets—land and livestock; the extent of cash-cropping; and health.1 The set of variables characterized in Table 11 was thus defined from the panel data. The indicators in Table 11 were calculated for the entire population of 357 households in 1997 and separately for each of the four nutritionally defined groups resulting from the crosstabulation in Table 10a. The mean level of each variable is set out in Table 12 according to the status of the nutritional indicator, and the findings are summarized below. Except where noted, the discussion points below refer only to instances where the differences in Table 12 are statistically significant. It should be noted that the dummy variables in Table 12 can be read as incidence, e.g. the coefficient of 0.67 for illness means that 67 per cent of all households experienced an illness in the month before the survey.
Table 11.—Poverty Indicators Constructed from the 1997 Data Set Variable
Purchases of grain or maize meal in the month preceding the interview (January-March)—an indicator that own grain supplies are exhausted
Total food purchases in the month preceding the interview—an ambiguous indicator since higher values may be associated with either poverty or wealth
1 if the family had home-produced food in storage at the time of the visit; 0 otherwise—an indicator of household food security
Kilograms of grain in storage at the time of the interview—an indicator of household food security
Kilograms of legumes in storage at the time of the interview—an indicator of both food security and dietary adequacy
1 if the family had a grain loan in 1995; 0 otherwise—an indicator of household food security two years previously
1 if the family repaid the 1995 grain loan in full; 0 if not; two if the family had no loan—an indicator of recovery from the 1995 drought
Nonfood consumption: total annual household expenditure excluding education—an indicator of the household's level of living
Nonfood consumption: total annual household expenditure including education—an indicator of the household's level of living
Market value of all crops harvested whether or not sold—a composite indicator of production levels
Total revenue from all crops grown and sold in the 1996/97 season—an indicator of disposable income
Market value of the household's total holdings of livestock—an indicator of wealth and—indirectly—of agricultural technology
Total revenue from sale of livestock products and services—an indicator of disposable income
Total revenue from the sale of animals—an indicator of disposable income Total cash remittances from nonresident household members or others—an
indicator of disposable income and/or the inadequacy of household income Total income from nonagricultural sources (excluding remittances)—an indicator
of disposable income, the inadequacy of household income or the level of diversification
Total household income from all sources—identical to disposable income Total income earned/controlled exclusively by women in the household Total area cropped in the preceding season (1995/96) Total area planted to cash crops in the preceding season (1995/96) The ratio of cash-crop area to food-crop area (1995/96 season) 1 if any family member was ill in the month prior to the interview; 0 otherwise—
an indicator of the health status of the family 1 if any family member was too ill to work in the month prior to the interview; 0
otherwise—an indicator of the impact of ill health on family labour supply
Note: Variable names ending in D are zero-one dummies while those ending in a dollar sign represent a continuous monetary value.
1 The PASS (Zimbabwe 1997) reports incidence of diarrhoea, fever and respiratory illnesses three times higher in rural than in urban areas, with most cases occurring among the very poor in rural areas but among the non-poor in urban areas.
36
Relative to the four nutritional categories, Table 12 provides findings that fall into two clusters:
Findings from Table 12 that accord with expectations: • The group with no undernutrition—0/0—has a significantly lower incidence of illness and a
significantly higher income from the sale of livestock.
Findings from Table 12 that confound expectations: • Households in the worst-nourished category—1/1—had the lowest expenditure on grain despite
no evidence of a purchasing power constraint.
Table 12.—Mean levels of poverty measures with combined nutritional indicators
*Significantly different from the mean value at P=0.05.
• Households in the best-nourished category—0/0—have the lowest total food expenditure. • There is no significant difference among nutrition categories arising from nonfood consumption
expenditure with the single exception of the 1/0 cases. • There are no significant differences among groups arising from the total market value of all
crops grown. • The highest-income group in terms of revenue from crop sales has the worst nutrition. • All livestock-related outcomes for the 1/0 group differ significantly from the mean despite that
fact that two of the three indicators are below the mean value and one (LsPrdRev$) is above—and the highest of the four groups.
• Above-average cash remittances exist for both groups containing undernourished adults. • Higher levels of female-controlled income are associated with all undernourished households
(although not significantly so) and significantly for the 1/0 case. • The worst-nourished group of households (the 1/1 group) cultivates on average the largest
acreage and the 1/0 group cultivates the smallest. • The worst-nourished group and the 1/0 group cultivate the largest areas of cash crops, but
only the latter is significant. • There is no significant difference among any groups in terms of the ratio between areas
planted to cash crops and food crops. • Perhaps most surprisingly, neither total off-farm income nor total income is significantly
associated with any nutritional category. • Group 1/0 has significantly the highest incidence of illness overall but the lowest incidence of
illness affecting work because illness is concentrated among children.
37
One reason for the large number of apparently perverse findings noted above is the very high level of variability in almost all the indicators. In an attempt to go beyond a simple comparison of means, the variable for total income was broken down into quartiles and the distribution of households was plotted across the income quartiles on the basis of their nutritional group. The results are shown in Figure 20a.
Figure 20a. Distribution of household nutritional groupings by total income quartiles Figure 20a shows few consistent patterns when nutrition-identified groups are arrayed against income-identified groups. If there were a consistent positive association between total income and the way households are classified nutritionally, one would expect the 0/0 and 1/1 groups respectively to look like a descending staircase and an ascending staircase from left to right. The appearance of the mixed cases is less predictable except that, because undernutrition is present in these households, there should be relatively fewer households in the higher quartiles. In fact, the distribution of the 0/1 group of households is fairly uniform across all quartiles, suggesting—in conformity with Table 10a—that the norm in rural Zimbabwe is a household with well-nourished adults but at least one poorly nourished child. Households with poorly nourished adults (where BMI = 1) are consistently more common in the second income quartile, while the lowest quartile actually does comparatively well and always contains fewer households than the top quartile for this group. The highest income quartile displays an unambiguous advantage only for the 0/0 households, where there are no undernourished adults or children. The results in Table 12 and Figures 19a and b suggest that, except in the 1/0 case—where the small sample size indicates the results could be entirely idiosyncratic, that the four nutritional categories are not particularly good proxies for traditional socio-economic indicators of poverty. Before dismissing the 1/0 outcomes altogether, however, it is appropriate to recall the earlier comments regarding cases where adults are undernourished and children well-nourished. Since one of the more plausible explanations for this phenomenon is the effects of HIV/AIDS, it may be hopeful that relatively few—20 per cent—of households fall into this group, even when the anthropometry is done at a time of considerable labour stress, because the proportion of 20 per cent is less than the estimated mid-1990s prevalence rate of 30 per cent for HIV/AIDS among the sexually active population in Zimbabwe (World Bank 1996). If health is the underlying cause, it would weaken the argument that adults experience a situation in which work levels are high relative to food supply or the time available for meal preparation. But Table 12 does indicate that child rather than adult health is the major problem in this group.
38
By examining the anthropometric categories for adults and children separately, it may be possible to shed more light on the rather enigmatic results of Table 12. Thus Table 13 sets out the mean values for the socio-economic indicators and, as before, indicates which of the group means differ significantly from the population mean. In this case however the discussion is structured according to the columns in Table 13 rather than along the lines as in Table 12. The major findings that emerge are set out as bulleted points below.
Table 13.—Mean levels of poverty measures according to single nutrition indicators
*Significantly different from the mean value for all cases at P=0.05. Relative to the four nutritional categories, the data in Table 13 indicate that:
Households with only well-nourished adults: • Differ in no significant way from the population mean for any indicator.
Households with one or more poorly nourished adults: • Are significantly more likely to contain poorly nourished children as well. • Spend significantly less on grain—less than a quarter of the mean, have significantly more
grain in storage and are more likely to have had a grain loan. • Possess herds with significantly lower market values. • Earn significantly less from sales of livestock. • Receive significantly more remittances in cash—80 per cent more than the mean. • Plant a significantly larger acreage to cash crops and have the highest ratio of cash crops to
food crops. • Report significantly lower rates of debilitating illnesses.
Households with only well-nourished children: • Are significantly less likely to contain poorly nourished adults. • Earn significantly more from sales of livestock.
Households with one or more poorly nourished children: • Are significantly more likely to contain poorly nourished adults. • Have significantly higher than average total food expenditure. • Report significantly lower incidence of all illnesses.
39
Of the four groups based on nutritional status—two for adults and two for children, only the group for poorly nourished adults appears at all well-differentiated according to the set of indicators used, even though some of the differentiating factors are counterintuitive. The presence of a poorly nourished adult is a good 'predictor' that there will also be a poorly nourished child; just under 80 per cent of households with a poorly nourished adult will also contain a poorly nourished child. These households also have very low expenditure on staple grain and possess low-valued livestock holdings from which they earn relatively little in sales. They do however plant the largest acreage of cash crops, both in absolute terms, and in relation to the area of food crops. These households also have mean total incomes more than eight per cent above those of households with only well-nourished adults and have the lowest incidence of incapacitating illness. The findings discussed above raise the question as to whether the nutritional criteria do a bad job of identifying poor households or whether the socio-economic poverty indicators used here adequately differentiate households by level of poverty. Comparing Tables 12 and 13, it is possible to construct a simple test of explanatory power of each approach by counting the number of significant differences identified in each case. Combining indicators of adult and child nutrition, as in Table 12, yields 25 significant differences while treating the nutrition indicators separately, as in Table 13, yields only 15 for the same set of poverty indicators, an improvement of 66 per cent for the combined approach. Over half of the significantly different indicators in Table 12 however identify a relatively small group of households, but it may be that this group is one that would need to be targeted in poverty alleviation efforts. The influence of ecology and tenurial regime. In order to test the proposition that the poverty indicators themselves are valid measures, the data have been restratified using two criteria for Zimbabwe which we already know a good deal about: land tenure regime—resettlement and communal areas—and by agro-ecological zone—Natural Regions. The results are set out in Table 14. If the indicators used are generally valid, we would expect to find two strong patterns. First, since 84 per cent of households in communal areas (CAs) are poor in total consumption terms1 (Zimbabwe 1997) and resettlement areas (RAs) have been provided access to a superior resource base, we would expect to find systematically stronger indicators of poverty in CAs than in RAs. Second, common sense suggests that rural households attempting to make a living from agriculture will achieve more positive results in areas physically better suited to farming. Thus it would be expected that poverty indicators, at least the agriculture-related ones, will generally indicate a progressive worsening as one moves from the better areas—NR 2—to areas of lower inherent potential—NR 3 and NR 4. And, since the RA-CA comparisons incorporate the NRs, and vice versa, the figures in Table 14 provide an even stronger test of the ability of the chosen indicators to identify distinctly different socio-economic groups. How well do the indicators fit with these prior expectations? In the case of the RA-CA comparison, all indicators with the exception of four have the expected relationship.2 The first two exceptions are the nutritional indicators, which show that the probability of a household containing either an undernourished child or adult is less in the CAs than in the RAs. The second two exceptions are the indicators related to health, both of which show the CAs to be healthier places to live—especially for children—despite the fact that all the RAs were provided with new clinics in the early 1980s. The agriculture and livestock income variables and the consumption indicators show the advantage of living in a resettlement area, while the remittance and off-farm income variables are indicative of some of the disadvantages of living in CAs. The different nutritional and health outcomes for RAs and CAs suggest sets of influences operating at different levels. Why are nutritional and health status worse in RAs, where households have generous land holdings and preferential access to health and agricultural services? One possible explanation may lie in settlement patterns and the time allocations of women. Villages in RAs have been laid out in a consolidated pattern to facilitate provision of services. Travel time to fields is therefore long in RAs, as are the hours spent in the field. Busy mothers may leave young children at home in the care of
1 On the basis of the income required to purchase a basket of basic food needed by an average person per annum and meet non-food needs (clothing, housing, education, health, transport, etc.). 2 See Kinsey (1998b) for a more complete discussion of the nature of and reasons for the relationships.
40
older children or take them to the fields. In neither case are they likely to be well-fed. In contrast, in the CAs, the fields surround the homestead, travel times are short, and mid-day meals can be easily managed. A further explanation may be that official exhortations to be productive have propelled RA households in the direction of cash cropping of commodities such as cotton and tobacco, leading to high ratios between the area planted to cash crops and food crops and/or reductions in diversity in the mix of food crops grown. Because of their small land-holdings, CA households tend to market surplus food crops, if they have any, rather than growing crops for market which cannot be consumed by the household.1 Incomes from agriculture and livestock are generally much higher in RAs than in CAs. Conventional wisdom on the effect of commercialization of agriculture on nutrition of farm families holds that there should be minimal if any adverse effects on nutrition because of the compensating effects of higher cash incomes. Why is this not the case here? The study sites span zones of agricultural potential ranging from fairly high—NR 2—to quite low—NR 4. In the area of best potential, farming appears dynamic and cash incomes are high as a result of widespread cultivation of cash crops such as cotton and tobacco and novel crops such as paprika. Across all the years surveyed, however, this area has consistently displayed the lowest nutritional outcomes. In the area of lowest natural potential, agriculture appears stagnant; and no farming system yet identified produces reliable incomes in this uncertain environment. Yet it is in this weakly commercialized area that the best nutritional outcomes for children have consistently been found.2 The data in Table 14 do not illustrate this finding, but they do show that the probability of an undernourished adult in a household is three times higher in the best agro-ecological zone than in the intermediate and low-potential zones. How good a job does the set of poverty indicators do in distinctly identifying population groups in Table 14? The following patterns emerge:
• Seven of the indicators in Table 14 appear useful for accurately identifying important differences among the five geo-tenurial groupings in that they show each group to be distinctly different from the population as a whole. These are: the amount of grain stored, the two measures of consumption, the values of crops grown and crops sold, total acreage planted, and the number of acres of cash crops planted.
• Another five indicators also do a reasonable job of differentiating the groups in that they show four of the five groups to be distinctly different from the population as a whole. These are: BMI, expenditures on both grain and total food, total income and the ratio of cash crops to food crops.
• The mean number of significant indicators in each subpopulation column in Table 14 is 14.4 compared to 6.3 in Table 12 and 3.8 in Table 13, suggesting that geography/tenurial status is more than two times as powerful as the combined nutritional indicators in identifying groups relevant for poverty analysis and that combined nutritional indicators are more than twice as powerful as single nutritional indicators for this same purpose.
The weak explanatory power of the nutrition indicators suggests that the relationship between nutrition and traditional poverty indicators is not as straightforward as intuition might suggest.3 Evidence for this contention can be found simply from correlating the entire set of poverty indicators with the separate nutritional indicators for adults and children. This is done in Table 15, which reveals some quite startling relationships.
Turning first to the unambiguous results in Table 15, the crop- and livestock-related indicators are unequivocally correlated with nutritional outcomes of both adults and children. An increase in a crop-related indicator always worsens nutritional status, while an increase in a livestock-related indicator always enhances nutritional outcomes. Why should this pattern occur so clearly?
1 But 40 per cent of CA households marketed nothing at all following the relatively good 1996 harvest. 2 The higher potential areas have higher rainfall, and high levels of rain provide beneficial conditions for certain disease vectors. The relationship between nutritional status and disease will be investigated further in work underway. 3 See Behrman & Deolilakar (1987) for an analysis of rural panel data which concludes that “increases in income will not result in substantial improvements in nutrient intakes” (p505).
41
Table 14.—Mean levels of poverty measures according to tenure regime and natural region
*Significantly different from the mean value for all areas at P=0.05. Answers to this question may come from a deeper appreciation of both the data and the farming systems from which they come. It should be borne in mind that the data are collected annually at, and immediately following, the period of peak labour stress and when food supplies are at their lowest point in the season.1 Collectively, an increase in the crop-related indicators can be interpreted as an increase in the seasonal demand for labour for field operations. This increase implies in turn two other associated shifts: an increase in the demand for caloric energy to sustain the labour inputs and a reduction in the amount of time available for women to care for children. Thus, greater commitments to cropping (and especially to cash-cropping) are associated with poorer nutritional outcomes. Nor do higher crop incomes from the previous harvest compensate during the current season. Why do the livestock-related indicators have consistently the opposite effect? There are likely to be at least four effects at work. First, livestock are probably the best single indicator of wealth for rural households and of their ability to cope with cash shortfalls.2 Second, the labour demands for livestock-keeping are nonseasonal in nature and do not require high levels of caloric expenditure; moreover, cheap, unskilled labour is often hired for herding during the busy period for cropping, and cattle are often herded collectively, thereby saving labour. Third, the value of the herd is positively associated with possession of draft oxen, which can significantly substitute for human labour in the demanding tasks of land preparation and weeding. Finally, revenue from sales of livestock products is indicative that households have surpluses of milk and eggs, suggesting that the family is consuming all of these valuable food sources it wishes to.
1 Studies in Zimbabwe remarking on the seasonality of nutritional status draw contrasting conclusions on the period of maximum stress. Kizita (1982), Sanders (1982) and Unicef (1985) argue that the rainy season (November through March) is the most critical time. A more recent empirical study (Wright et al 1997) narrows the period of peak stress to January-March. Allart (1983), however, argues that the dry season is the period of greatest stress. Wilson (1990) notes that nutritional stress reflects not only seasonal changes in diet but also seasonal changes in the profile of disease risk; thus the peak period of stress will vary from one ecological zone to another. 2 See Kinsey, Burger & Gunning (1998a).
42
Table 15.—Correlation between poverty indicators and nutritional outcomes (n=357)
An increase in [...] has the indicated effect on nutritional status
Nutritional outcomes of adults
Nutritional outcomes of children
Livestock-related indicators Market value of livestock Revenue from livestock products/services Revenue from sale of animals Agriculture-related indicators Area cropped Area planted to cash crops Cash-food crop-area ratio Market value of crops harvested Total crop revenue Consumption-related indicators Purchases of grain or maize meal Foodstuffs in storage Amount of grain in storage Amount of legumes in storage Total food purchases Consumption excl education Consumption incl education Income-related indicators Remittances Income from nonagricultural sources Had a grain loan after the 1995 harvest Repaid the 1995 grain loan in 1996 Total household income Female-controlled income Health-related indicators Any illness in the family in the previous month Serious adult illness in the previous month Tenurial regime and agro-ecology Live in a resettlement area Live in a communal area Live in NR 2 Live in NR 3 Live in NR 4
+ + + – – – – –
+ + + – – – – – + – + – – – – - + - + +
+ + + – – – – – – + – – – + +
+ + – + + –
– – - + 0 - 0
A more simplified explanation is also possible. Households with large livestock holdings are the wealthy; they have made it, and they have decreased their vulnerability to the vicissitudes of rainfed farming. Households with many positive crop-related indicators aspire to make it in a similar fashion and are working extremely hard to do so. Much of their income from crops may therefore be used to increase investment rather than improve consumption. The two health-related indicators are also in accord for adults and children and display the expected relationships. In contrast to the crop, livestock and health indicators, however, the 13 income and consumption indicators in Table 15 exhibit highly ambiguous outcomes. Seven of the 13 variables are positively associated with improvements in child nutritional status, but only 5 of 13 display the same association with adult nutritional status. Moreover, in almost half the cases (6 of the 13), the indicators exhibit opposite signs for adults and children, suggesting that the pathways to better household nutrition are more complex than is sometimes suspected. The patterns for consumption indicators in Table 15 are difficult to explain satisfactorily. It feels intuitively correct that total household nonfood consumption would be positively associated with child nutrition, since there has to be a strong association with household income, but why should it be negatively associated with adult nutrition? And why should food purchases be negatively associated with nutrition for both children and adults, while grain purchases and the amount of grain in storage have positive effects for adults and negative ones for children? It is possible however to
43
hazard a guess as to why the quantity of legumes in storage is negatively associated with nutrition. Legume-growing in Zimbabwe is usually a woman's activity, and the values in Table 15 are thought to arise because in this case legume-storage is a proxy for female-headedness.1 The income variables are, if anything, even more paradoxical. Why should all sources of cash income—aside from crop income and income controlled by women—be positively correlated with child nutrition and yet only one—nonagricultural income—correlates positively with adult nutrition? If grain loans are thought of as income in kind, then the effect associated with these indicators is at least what would be expected. Receipt of a grain loan in 1995 suggests that the household had inadequate savings, food in storage and/or other coping mechanisms to be able to cope with the failure of the 1995 harvest; repayment of a 1995 grain loan in 1996 is associated with a rapid recovery from the previous bad year and is positively associated with nutritional state in 1997.2 The following reasoning might be invoked to explain the outcomes for adults. Grain purchases will have supplemented the supply of starchy staples for households in the month prior to making the anthropometric measurements, however it will only be households that had a poor harvest the preceding year and whose supplies are exhausted that will be forced to purchase grain. If they in fact had a poor harvest, the implication is that income from crop sales was low, perhaps explaining why income from nonagricultural sources has a positive effect. And if poor harvests are structural for these households, they may not experience the effects (discussed above) that cause the negative correlations for the set of crop-related variables. Finally, and unsurprisingly, both the health-associated indicators display the expected association with nutrition. Nutritional Indicators and Poverty Measures: Per-capita Outcomes Examination of the relationship between nutritional indicators and the mean of poverty measures at the household level revealed some puzzles and suggested that the ability of the combined nutritional indicators to proxy for poverty measures was generally weak. This section replicates the previous analysis but transforms the poverty measures to a per-capita basis. A number of the poverty measures are dropped at this point because they cannot be transformed; these include all the dummy variables from Table 11 and the variable for the ratio of cash crops to food crops. A new continuous variable—HHsize—is added to represent the number of persons resident in the household.3 The outcomes using this approach are set out in Tables 16, 17 and 18, which compare with Tables 12, 13 and 14. Comparing Tables 12 and 16 reveals clearly the greater validity of an approach based on poverty measures defined in per-capita terms. The reason why can most easily be seen from the results for the new variable representing household size. Households with no undernutrition (0/0) and no undernourished children (1/0) are significantly smaller than the average, whereas those with undernutrition among both adults and children (1/1) are significantly larger than the average. The most common group—households with undernourished children but well-nourished adults—are still the norm and differ in no significant way from the mean. The changes observed are most striking for the entirely well-nourished (0/0) group. Whereas only 3 poverty measures are significantly differentiated in Table 12, 13 are in Table 16—a more than four-fold increase.4 Expenditure on food is no longer significant in explaining the absence of undernutrition, but the measures that have become significant include: grain storage, nonfood consumption expenditure, all the crop-related measures (except area planted to cash crops), two additional livestock-related measures, and the two measures of nonagricultural income. Eight of the
1 This explanation is supported by a strong positive correlation coefficient between storage of legumes and receipt of both remittance income and off-farm income. It is contradicted however by the strong positive correlation coefficient between storage of legumes and total income and the absence of a significant correlation with female-controlled income. 2 All failures to repay grain loans were not due to inability to repay however; in many cases, the responsible authorities simply failed to collect the grain that had been set aside for the loan. In many such cases, families either sold or consumed the maize; in other cases, the maize was ruined by weather while awaiting collection. 3 The variable includes nuclear and extended family members as well as, in some cases, unrelated persons (some of whom work for the household) who live and eat with the family. Excluded are family members away at school, working elsewhere and absent looking for work elsewhere. 4 The comparisons here include the significant dummy variables from Tables 12, 13 and 14 even though they are not reported again in Tables 16, 17 and 18.
44
poverty measures that were below the mean values for this group in Table 12 are above the mean when treated in per-capita terms as in Table 16, and no measure drops below the mean with the transformation to per-capita terms.
Table 16—Mean levels of poverty measures on a per capita basis according to combined nutritional indicators
*Significantly different from the mean value at P=0.05. Turning to the worst-nourished (1/1) group of households, the number of poverty measures that is significant has risen from 9 to 16, a 78 per cent improvement. Nine of the poverty measures that were above the mean for this group in Table 12 are below the mean in Table 16, and no measure rises above the mean with the transformation to per-capita terms. The measures that lose their discriminatory power on a per-capita basis are crop sales and remittances. The measures that acquire power are: food purchases, grain storage, both consumption measures, value of livestock and sales of livestock products, total income, and the area planted to cash crops. For the small mixed (1/0) group, there has been an increase in significant measures from 13 to 15, a 15 per cent improvement. No measure for this group that was above the mean on a household basis drops below the mean on a per-capita basis, but five measures rise above the mean with the transformation: both the nonfood consumption measures, both the crop value measures, and the total income measure. The only poverty measures that successfully differentiate all four groups in Table 16 are the two nonfood consumption measures. Overall, compared to Table 12, 38 per cent of the 72 poverty measures for the 4 groups in Table 16 reverse their position relative to the mean with the transformation of the measures to per-capita terms. With transformed values two groups—those with undernourished children—tend to drop below the population means of the poverty measures, while the two with well-nourished children tend to rise above the mean. Thus the procedure of transforming the values is picking up the same thing that inclusion of the household size variable does: larger households are far more likely to contain poorly nourished children. Comparing the outcomes in Tables 13 and 17, in which only a single nutritional indicator is used, there is again a dramatic improvement in the ability to discriminate among groups on the basis of the poverty measures. The biggest shifts are associated with the Z indicator. For households where Z equals 0, the number of significantly differentiated poverty measures rises from one to eleven and, where Z equals 1, the rise is from 2 to 5. In the first of these cases, the transformation of
45
poverty measures to a per-capita basis raises six of the group means from below to above the population mean; in the second case, the same transformation reduces the group means for the same six measures from above to below the population mean. Whatever the state of the Z indicator, food expenditures lose their significance while nonfood consumption and total income acquire significance. Further, all the crop- and livestock-related measures are significant for the group of households with well-nourished children. It is worth noting in passing that the measure representing female-controlled income is not a significant poverty measure according to either grouped or individual nutritional indicators. This finding flies in the face of conventional wisdom and compels further research. Turning finally to the comparison between Tables 14 and 18, the entire set of indicators displays relatively little change in discriminatory power. There is a sizeable decrease in the number of poverty measures that display significance for RAs, a marginal decrease for NRs 2 and 3, a marginal gain for CAs, and no change for NR 4. Of the 85 grouped outcomes in Table 18, only 8 reverse position relative to the population means with the per-capita transformation. This pattern of relative stability is a confirmation that geo-tenurial differences capture a good deal of the essence of poverty in Zimbabwe.
Table 17—Mean levels of poverty measures on a per-capita basis according to single nutrition indicators
*Significantly different from the mean value for all cases at P=0.05. To summarize:
• Every poverty measure in Table 18 is useful for accurately identifying important differences for at least one of the five geo-tenurial groupings.
• Five of the measures in Table 18 are useful for showing each group to be distinctly different from the population as a whole. These are: the amount of grain stored, the value of crops grown and crop revenue, the area planted to cash crops, and household size. The two measures of consumption and the total area planted no longer universally discriminate.
• The mean number of significant indicators in each subpopulation column in Table 18 is 13.4 compared to 11.8 in Table 16 and 7.5 in Table 17. Whereas geography/tenurial status was previously more than twice as powerful as the combined nutritional indicators in identifying distinctly different groups, with a per-capita approach it is only 14 per cent better. Using per-capita measures, however, the combined nutritional indicators perform somewhat less well vis-à-vis single nutritional indicators for this same purpose; the advantage with household measures was 110 per cent whereas with per-capita measures it is only 96 per cent.
46
As was the case for the poverty measures defined as household means, there is also a very high level of variability in the measures defined in per-capita terms.1 To ascertain whether the per-capita treatment provides outcomes that accord better with the expectations suggested by theory—and better than the use of household means, the earlier treatment is replicated by dividing the variable for total income into quartiles and plotting the distribution of households across the income quartiles on the basis of their nutritional group. The results are shown in Figure 20b.
Table 18—Mean levels of poverty measures on a per-capita basis according to tenure regime and natural region
*Significantly different from the mean value for all areas at P=0.05. Compared with Figure 20a, Figure 20b shows quite marked changes for the two groups with well-nourished adults but very little change for the two containing poorly nourished adults. The 0/0 group displays the expected 'descending staircase' pattern for the top three income quartiles, but the percentage of well-nourished households in the bottom income quartile remains unchanged. Using a per-capita measure of total income has differentiated households that were formerly in the third quartile and relocated them in the top quartile. In the 0/1 group, the shift has been in two directions: households that were formerly in the top and bottom income quartiles have been relocated to the third quartile. Households with poorly nourished adults (where BMI = 1) remain more common in the second income quartile, while the lowest quartile continues to do comparatively well in relation to the other three quartiles. The highest income quartile now displays an unambiguous advantage in households where there are no undernourished adults or children. With the earlier caveat about the small size of the 1/0 group, the results in Table 16 and Figure 20b suggest that the ability of the four nutritional categories to provide results consistent with prior expectations are improved when a set of commonly used socio-economic indicators of poverty is defined in per-capita terms.
1 It might be thought that transforming the poverty measures to a per-capita basis would reduce their statistical variability by correcting for household size. This is not the case because of the huge variability in household size: from 2 to 76 persons. While the transformation does reduce the coefficient of variation for 11 out of 17 continuous variables, it actually increases marginally the mean coefficient of variation across all variables.
Figure 20b. Distribution of household nutritional groupings by per-capita income quartiles Discussion. This section set out to examine how indicators of nutritional status agree with other socio-economic indicators commonly used to identify poverty at the level of the household. With the approach taken and the data used, the conclusion has to be that agreement is relatively weak. This conclusion is most valid when mean values are reported only at the level of the household, as is so common in Zimbabwe, and less true when values for poverty measures are reported in per-capita terms. While there are clearly fundamentally different socio-economic situations represented in the data set, these are not well delineated by the distribution of the nutritional indicators. Although the binary BMI and Z variables are positively correlated in a statistically significant way, the correlation coefficient is small (0.15), indicating only a weak linear association. Nor is there any consistent evidence that households with adults identified as being thin by the BMI are any worse off in terms of food security or health status than households without thin adults. Moreover, the combined nutritional indicators lack discriminatory power when a group in which undernutrition exists is the norm, as is the case in the population used here, where households with undernourished children and well-nourished adults are the expected outcome. Several factors could help to explain this weakness in discriminatory power. First, much has been lost by converting the anthropometric scales to simple 0-1 dummy variables depending upon the position of an observation relative to a defined cutoff point. This procedure fails to differentiate between degrees of undernutrition, and it may be the severity of undernutrition that accords better with the socio-economic indicators. Moreover, the indicators test simply for the presence or absence of undernutrition in households, while it could be that the extent of undernutrition is more significant. A basic problem however is that, because BMI does not correlate well with the anthropometric indicators for children, many of the most promising cause-and-effect variables operate in opposite directions for children and for adults. This makes generalizations about the household unit extremely difficult. Of the three limitations noted by Nubé, Asenso-Okyere and van den Boom (1997) to the use of BMI as an indicator of levels of living, one seems particularly pertinent here. This limitation is that seasonal fluctuations in food availability or labour demands may affect BMI. The adults assessed for this study were all examined during or immediately following periods of peak labour demands and at
48
a time when food supplies from the previous harvest would normally have been running low.1 The outcomes suggest that, in this setting, the BMIs may be better at identifying stress in terms of arduous farm labour than at differentiating poor rural households from other rural households. The assertion by Shetty and James (1994) that “The BMI is sensitive to socio-economic status and to seasonal fluctuations in food consumption relative to the level of physical activity” (pvii) seems somewhat paradoxical in the context of the results discussed here. Socio-economic status is a phenomenon which can normally be expected to change only relatively gradually over time, whereas seasonal changes in food consumption and physical activity are likely to be very pronounced for rural households. Precisely because of its sensitivity to both sets of factors, identical outcomes can arise with the approach used here because a wealthy household experiences labour stress, a poor household has inadequate food or a moderately well-off household has experienced one case of illness. The analysis here indicates that much more needs to be known about changes in BMI in environments where multiple causal agents operate. Finally, while structuring analysis on the basis of per-capita rather than household values clearly yields results more indicative of underlying poverty relationships, improvements can still be made. The difficulty with a per-capita approach is that it weights adults and children equally and thus masks significant differences in household composition. A logical next step therefore is to repeat and extend the analysis while weighting household members in equivalent consumption terms. Concluding Remarks and Discussion One clear result is that the incidence of poverty among land reform beneficiaries is as high as that among non-beneficiaries. But this outcome is not a consequence of the failure of early land reform to be economically successful; rather it is a reflection of redistribution within extended families. Redistribution takes place, not by transferring resources from well-off households to poorer ones—as village insurance models in the vein of Townsend (1994) suggest, but via the movement of individuals from poor to better-off households. Mobility is thus critical, and the implication is that improvement of rural welfare hinges upon improvements in the economy as a whole. Product price increases or improved off-farm, income-generating opportunities are also shown to have potential to assist in reducing rural poverty, but the most significant reduction in poverty is brought about by reductions in household size, an outcome that could be achieved through income-earning opportunities elsewhere in the economy so that rural households no longer have to act as safety nets. Increases in crop prices associated with the economic adjustment programme of the first half of the 1990s helped initially to reduce rural poverty, while rural poverty at the time was relatively insensitive to the formal sector contraction that accompanied the same programme. Variation in rainfall, of course, has an enormous impact on rural poverty. Poverty is generally measured by per-capita consumption, which is not easily possible with the panel data set discussed here. Income has therefore been employed instead. A disadvantage of this approach may be that income is used for purposes other than just consumption. In the case here, this does not cause too much bias as investments in agricultural capital have declined over the years and do not amount to much2 (See Figure 5). The accumulation of cattle over time is regarded here as a capital gain rather than investment out of income. Income is also much more volatile than consumption. The major cause of income volatility is the erratic rainfall experienced over the period reviewed, coupled with a correlation coefficient between the annual means of rainfall and income of 0.77. Attention has been drawn to a number of smoothing mechanisms by which households cope with this problem: adjustment of maize sales, sales of cattle and depletion of money balances. Price changes also help offset potential income shortfalls to some extent, as prices are higher in years with low production. Receipt of food aid has also helped compensate for reduced incomes, but the political handling of food aid may have helped to create a dependency syndrome. The trend in incomes over the period reviewed is positive but not spectacular, and the large majority of the households—even after the maturity of a major public programme aimed at alleviating rural poverty—still live below a nationally defined poverty line (which is believed to overstate the extent
1 It should be recalled that the 1996 harvest was a relatively good one; indeed it was one of the best observed over the 28 years of the study. 2 One possible caveat to this conclusion arises among the small but growing number of RA farmers who are beginning to grow flue-cured tobacco. This group is making very substantial investments in curing barns and equipment.
49
of poverty). The positive trend can be ascribed to more land being taken into cultivation, acquisition of more cattle, and slightly higher yields.1 Agricultural terms of trade however deteriorated after 1992/93, as consumer prices rose by more than producer prices. And it is alarming that underlying the modest average gains in income appears to be persistent and worsening inequalities. At the individual level, incomes per capita show some slow improvement over time. Households cultivating more than one acre per capita manage, in general, to achieve incomes above the poverty line. Yet, the number of household members can and does change drastically from one year to the next (Kinsey 2010). Demographic changes at the household level—driven by retrenchments in the formal sector, HIV/AIDS morbidity and mortality, and household cycle stages—may be influencing measures of per-capita income as much as, if not more than, underlying economic realities. The longer term trend in the size of rural households was also upward, lending support to the conclusion that—despite worsening rural-urban terms of trade—former urban dwellers were returning to their previous rural homes.2 Based on the analysis reported here and the contents of the data set, several possible extensions to the analysis are possible. Most importantly would be to calculate the components of income from the 1998 to 2001 rounds and incorporate these into an analysis covering a longer period. A related step would be to examine more carefully the relationship over time between poverty measured in terms of income and poverty measured in terms of assets. This would require further detailed work on the way in which assets are utilized—both in coping with stressful events and in more ordinary times. In addition to data on livestock, the panel study data set contains a high level of detail on the stock of housing and other fixed capital, agricultural capital equipment, and consumer durables. In extending the decomposition of incomes, it would be informative to analyze remittances and social transfers in more detail. Policy reforms should promote expansion of the nonagricultural sector, particularly manufacturing. Flows of remittances from urban workers can be a powerful means for reducing rural poverty, for they not only finance consumption but also provide investment and working capital to purchase farm inputs, livestock and working equipment for small-scale enterprises. While all cash income is fungible, remittances to rural households typically come in a mixture of cash and in-kind forms. This second strand of remittance income needs to be quantified and valued so that some idea of the overall multiplier effects of remittances may be obtained. With the decline in urban real wages in recent years, it is likely that the trend in remittances has not been favourable for the rural poor. Indeed, there are indications in the panel data set that the flow may have reversed, as rural households transfer resources to urban areas to support a family member in his or her search for paid employment. It is thought that the analysis characterized above—supplemented by additional analyses—could usefully be carried out at two different levels. At the first, aggregative level, the analysis would simply be carried out as described but would incorporate the data from more recent years. At the second level, the analysis would be conducted separately for each of the three agro-ecological zones. Although in some senses the three zones represent a continuum of changing potential, in many important ways each area is characterized by a unique poverty profile and set of challenges for poverty alleviation. One of the richest parts of the panel data set is that relating nutritional status and health to a wide range of socioeconomic variables. Preliminary work has been done to ascertain the extent to which nutritional status—as measured by the anthropometric status of both adults and children—can proxy for more-difficult-to-obtain measures of poverty (Kinsey 1998b & c). Additional work could usefully extend the work reported in Alderman, Hoddinott and Kinsey (2006) by focusing on the long-term human capital costs of episodes of poor nutrition in childhood and on the delineation of the relationship between poor nutritional status and poverty in other dimensions. Second, a more systematic analysis could be undertaken of the abrupt shifts in household size to ascertain their nature, causes and permanence. It is clear from a long-term study such as that utilized here that shifts in demographic characteristics drive much of what is observed about poverty
1 The sources of higher yields need further investigation. One of the effects of ESAP was the removal of the subsidy on fertilizer, resulting in a decline in its use in the communal areas (Oni 1997). Analysis in this paper shows the same is true for the resettlement areas as well. During the mid-1990s, fertilizer prices went up several times during each season; and supplies of the right type of fertilizer at the right time of the season have been irregular also. 2 The survey work in 2010 is showing that this trend was strikingly reversed over the preceding decade.
Deleted: i
50
from survey work. The analysis here shows the critical significance of land-man ratios in shaping the profile of poverty of rural households. It would be well worth analysing therefore whether household size and composition are altered deliberately—as a response to economic pressures—or whether economic pressures arise as the consequence of unanticipated alterations in household size or composition. Finally, the data set contains detailed food consumption data decomposed by source—own production, purchases and transfers—for comparable periods of food stress each year for most of the 1990s. While analysis of this data could not be extrapolated to give annual consumption, investigation would be worthwhile in terms of indicating differentials among socioeconomic groups and changes in the patterns of seasonal stress in food consumption. The final word must be that land redistribution as a blanket policy for reducing rural poverty is a failure. To be sure, the average household appears to have benefited, but there is no such thing as the average household. Even after so many years, poverty remains a lived experience for the majority of the households studied. And there are disturbing indications that the poverty gap is widening as inequalities in outcomes become more extensive. It would appear that improving the effectiveness of land reform as a poverty-reducing instrument now hinges critically upon delineating more carefully the demographic, social and economic profiles of rural households and abandoning blanket policies in favour of far more carefully tailored programmes catering more explicitly to the needs and abilities of multiple categories of households.
-o0o-
References Addison, Tony. 1997. Zimbabwe: The Impact of Economic Reform on Livelihoods and Poverty (Draft,
University of Warwick, May 1997). Alderman, Harold, John Hoddinott and Bill Kinsey. Long-term consequences of early childhood
malnutrition Oxford Economic Papers 58, 3, 2006. Allart, L. 1983. Nutritional education in Zimbabwe. Central African Journal of Medicine 29, 3: 21-
34. Balleis, Peter. 1993. A Critical Guide to ESAP. Silveira House Social Series No. 2. Gweru: Mambo
Press. Cavendish, William. 1999. Incomes and Poverty in Rural Zimbabwe during Adjustment: The Case of
DANIDA. 1996. Evaluation of Poverty Reduction in Danish Development Assistance: Country Study—Zimbabwe. Two volumes: I: Main Report; II: Assessment of Selected Interventions. Copenhagen: DANIDA—Ministry of Foreign Affairs.
Dibley, M. J., N. Staehling, P. Nieburg and F. L. Trowbridge. 1987. Interpretation of Z-score anthropometric indicators derived from the international growth reference. American Journal of Clinical Nutrition 46: 749-62.
Elbers, Chris, Jan Willem Gunning and Bill Kinsey. 2002. Convergence, Shocks and Poverty. Tinbergen Institute Discussion Paper TI 2002-035/2. Amsterdam: The Tinbergen Institute.
Government of Zimbabwe. 1989. Zimbabwe Demographic and Health Survey 1988. Harare: Central Statistical Office.
Government of Zimbabwe. 1994. Income, Consumption and Expenditure Survey Report, 1990/91 [ICES 1991]. Harare: Central Statistical Office.
Government of Zimbabwe. 1998. Income, Consumption and Expenditure Survey Report, 1995/96. [ICES 1996]. Harare: Central Statistical Office.
Government of Zimbabwe. 1996. 1995 Poverty Assessment Study Survey Preliminary Report. Harare: Ministry of Public Service, Labour and Social Welfare, Social Development Fund.
Government of Zimbabwe. 1997. Poverty Assessment Study Survey (1995) [PASS 1995]. Harare: Ministry of Public Service, Labour and Social Welfare.
Government of Zimbabwe. 1998. Poverty in Zimbabwe. Harare: Central Statistical Office. Government of Zimbabwe. 2006. 2003 Poverty Assessment Study Survey, Final Main Report [PASS
2003]. Harare: Ministry of Public Service, Labour and Social Welfare. Gunning, Jan Willem, John Hoddinott, Bill Kinsey and Trudy Owens. 2000. Revisiting forever gained:
Income dynamics in the resettlement areas of Zimbabwe, 1983-1996. Journal of Development Studies 36, 6.
IMF. 1998. External Evaluation of the ESAP: Report by a Group of Independent Experts. Washington, D.C.: International Monetary Fund.
51
Jamali, D. 1997a. Economic and social conditions in independent Zimbabwe: A human rights perspective. Harare: ZimRights.
Jamali, D. et al. 1997b. Report of the farm and domestic workers programme. Harare: ZimRights. Jenkins, C. and Prinsloo, E. 1995. Inequality among households in Zimbabwe: An assessment using
the 1990/91 ICES. Oxford: University of Oxford (ms. August 1995). Killick, Tony, Jerker Carlsson and Ana Kierkegaard. 1998. European Aid and the Reduction of Poverty
in Zimbabwe. Working Paper 109. London: Overseas Development Institute. Kinsey, B. H., H. McQuie and M. Rukuni. 1995. Rural Household Studies in Zimbabwe: A Review.
Working Paper AEE 2/95, Department of Agricultural Economics and Extension, University of Zimbabwe.
Kinsey, B. H. 1997. Dancing with El Niño: Drought, the state and the nutritional welfare of rural children in Zimbabwe, In Helen O'Neill and John Toye (eds) A World without Famine?: New Approaches to Aid and Development (Macmillan, 1997).
Kinsey, B. H. 1998a. Determinants of Rural Household Incomes and their Impact on Poverty and Food Security in Zimbabwe. Resource paper for the Workshop on Rural Household Dynamics. 15-16 June 1998. Harare, Zimbabwe.
Kinsey, B. H. 1998b. Multiple Pathways from Animal Husbandry to Better Nutrition: Evidence from a Study of Household Economies in Zimbabwe. In Proceedings of the Symposium on Human Nutrition and Livestock in the Developing World. Heifer Project International. Little Rock, Arkansas, October 14, 1998.
Kinsey, B. H. 1998c. Nutritional status as an indicator of poverty: Does it support or contradict traditional poverty indicators? Resource paper for the Workshop on Rural Household Dynamics. 15-16 June 1998. Harare, Zimbabwe.
Kinsey, B. H., J. W. Gunning and Kees Burger. 1998. Coping with drought in Zimbabwe: Survey evidence on responses of rural households to risk. World Development 26, 1.
Kinsey, B. H. 1999. Growth-friendly poverty reduction? Long-term household-level outcomes of land reform in Zimbabwe. Journal of Southern African Studies 25, 2, 173-96.
Kinsey, B. H. 2004. Zimbabwe’s land reform program: Under-investment in post-conflict transformation. World Development 32, 10.
Kinsey, B. H. 2010. Who Went Where ... and Why: Patterns and Consequences of Displacement in Rural Zimbabwe after February 2000. Journal of Southern African Studies 38.
Kizita, M. 1982. Development of nutrition centres in Musami and Chamachinda Wards. Central African Journal of Medicine 28, 195-6.
Kwaramba, Philemon. 1998. The Socio-economic Impact of HIV/AIDS on Communal Agricultural Systems in Zimbabwe. Working Paper No. 19. Harare: Friedrich Ebert Stiftung.
MacGarry, Brian. 1994. Double Damage: Rural People and Economic Structural Adjustment in Time of Drought. Silveira House Social Series No. 7. Gweru: Mambo Press.
Madzingira, N. 1997. Poverty and ageing in Zimbabwe. Journal of Social Development in Africa 12, 2. Matshalaga, N. 1997a. The Gender Dimensions of Urban Poverty, The Case of Tafara. Harare:
University of Zimbabwe, Institute for Development Studies. Matshalaga, N. 1997b. The Gender Dimensions of Urban Poverty, The Case of Dzivarasekwa. Harare:
University of Zimbabwe, Institute for Development Studies. Mundy, Verity S. 1995. The Urban Poverty Datum Line in Zimbabwe. Harare: Catholic Commission
for Justice and Peace in Zimbabwe. Nubé, M., W. K. Asenso-Okyere and G. J. M. van den Boom. 1997. Body mass index: Indicator of
Health and Wealth. Staff Working Paper WP-97-01. Amsterdam: Centre for World Food Studies, Free University.
Raftopolous, Brian and Jazdowska, Niki. 1997. The political and institutional environment for poverty-reducing aid to Zimbabwe. Background paper submitted to the Overseas Development Institute. Harare: Institute of Development Studies (mimeo.).
Ravallion, Martin and Shoahua Chen. 1996. What Can New Survey Data Tell us About Recent Changes in Living Standards in Developing and Transitional Economies? Background Paper to World Development Report 1996. Poverty and Human Resources Division, Policy Research Department. Washington, DC: The World Bank.
Sanders, D. 1982. The origins of malnutrition in Zimbabwe. Central African Journal of Medicine 28, 168-91.
Townsend, R. M. 1994. Risk and Insurance in Village India. Econometrica 62: 539-591. Unicef. 1985. Women and Children in Zimbabwe: A Situation Analysis. Harare: Unicef. Shetty, P. S. and W. P. T. James. 1994. Body Mass Index: A Measure of Chronic Energy Deficiency in
Adults. FAO Food and Nutrition Paper 56. Rome: Food and Agriculture Organization. Stenflo, Gun Alm. 1993. Poverty in Zimbabwe 1990-91. Report from a Short Term Mission to the
Central Statistical Office, Harare. Washington, D.C.: The World Bank. UNDP. 1997. Human Development Report 1997. Oxford and New York: Oxford University Press.
52
Wilson, K. B. 1990. Ecological dynamics and human welfare: A case study of population, health and nutrition in southern Zimbabwe. Unpublished PhD thesis, University College, London, UK.
Wright, J., S. Hendry, P. Hoyles, J. Nyatsanza and P. Vazé. 1997. Seasonal aspects of under-nutrition in young children in Zimbabwe. Paper presented at the Annual Conference of the Development Studies Association, University of East Anglia, Norwich, UK, 11-13 September.
ZCTU. 1996. Beyond ESAP. A Framework for a Long-term Development Strategy beyond the Economic Structural Adjustment Programme (ESAP). Harare: Zimbabwe Congress of Trade Unions.
ZFU (Zimbabwe Farmers’ Union). 1994. The Future of Smallholder Farming under ESAP. Harare: Zimbabwe Farmers’ Union in association with the Friedrich Ebert Stiftung: Proceedings of the conference, 2-3 November 1994.
ZimRights. 1996. Final Report of the Conscientization Capacity Building Workshop 29-31 May 1996. Harare: ZimRights.
World Bank. 1996. Understanding Poverty in Zimbabwe: Changes in the 1990s and Directions for the Future. Washington, DC: The World Bank.
