Smoke and mirrors? The science of poverty measurement and its application

Abstract

Measures of poverty are much used, but also much criticised as having limited value in debates on public resource allocation. Some argue that the measures are too conservative and do little more than complicate important issues of inequality and injustice. However, poverty measurement can be sensitive to these concerns if grounded in the field's well-developed theoretical foundation. In South Africa, poverty measures over more than 50 years have consistently taken into account distributional issues and the causes and implications of deprivation, and most South African analyses of poverty have recognised and incorporated the multi-dimensional nature of poverty. Recognising different perceptions of aggregation, time horizon and the role of states and markets is perhaps more important than methodology when assessing what poverty measures can contribute. With proper theorisation, and attention paid to the purpose of poverty diagnostics, measurement is more than sleight-of-hand and can provide both a tool for advocacy and a means to implement policies that promote greater social justice.

Keywords:

• poverty measures
• poverty lines
• inequality

JEL codes:

• I32
• A11
• B20

1. A politicised debate

Technical measures are now often the first step in ‘poverty diagnostics’ in both developing and developed countries. At best, these measures can be likened to medical tests that narrow down the possible causes of an illness so appropriate treatment can be prescribed; at worst, to a cursory examination followed by a one-size-fits-all remedy. The technical aspects may have been demystified, but the measures continue to provoke politicised debate, in which measurement is sometimes seen as obfuscation, the thresholds as unacceptably conservative, and the division of society into poor and non-poor as little more than sleight-of-hand and of questionable use (Hutchful, 1994).

Poverty measures are also associated with contentious political choices, such as what kind of support public policy should provide and how much, how minimum wages should be determined, or where to set the administrative threshold for eligibility for public funds. If such decisions are reduced to a mechanical exercise, ‘poverty measurement blues’ may follow, with policy makers and civil society eying the measures with suspicion, and researchers becoming defensive about their preferred measure and methodology (Du Toit, 2005). As Greeley (1994) aptly observes, the result is ‘poverty of measurement’ rather than measurement of poverty – a situation that is neither necessary nor desirable. In this contribution to the development debate I propose that, despite its shortcomings, poverty measurement is necessary not only for developing and implementing redistributive policies, but also as a ‘tools to theory’ heuristic that gives us a better grasp of the structural nature of poverty (Gigerenzer, 1991). Measurement not only identifies targets and target groups but also reflects our understanding of poverty and why poverty persists.

2. Conceptualising poverty

Our methodology for measuring poverty differs according to how we view poverty. Lipton (1997) argues that the consensus view is that poverty is the inability to attain an objective and absolute minimum standard of living and that this can be reflected by a quantifiable indicator applied to a constant threshold that separates the poor from the non-poor. This view includes other dimensions of poverty, such as life expectancy or adult literacy, but measures them individually. The measurement that the absolute approach uses is quantitative, relying on surveys of income and consumption, and international thresholds, such as one or two dollars per day, to allow for cross-country comparison.

However, poverty can also be seen as being relative. Here the poor are seen as lacking the resources with which to attain a socially acceptable quality of life. In this approach the indicator varies according to the standards of the society being measured, and distributional issues are taken into account. A minimum amount may be used, such as a national poverty line, but adjusted to take into account changing needs, preferences and national standards of living. Although the measurement here is usually objective and quantitative, qualitative approaches frequently play a role in setting definitions and standards. Finally, poverty can also be seen as being subjective and based upon the judgement of those who consider themselves to be poor. Measurement here can be both quantitative and qualitative and can use a consensual approach to decide on definitions and indicators (Drewnowski, 1977; Goedhart et al., 1977; Wright et al., 2007).

In their debate over absolutist and relativist approaches to defining poverty, both Townsend (1985) and Sen (1983, 1985) recognise the social determination of deprivation. Townsend stresses that the necessities of life vary over time and space and as changes occur in society and the products of society (1985:659). The definition of what commodities people need is relative because needs change as institutions, technology and social structure change. The relationship between needs and commodities changes: for example, in a society where a mobile phone has come to be seen as a necessity, a person without one is needy. This means that poverty lines should not be updated simply according to price changes, but also according to what is included in the bundle of goods that make up the minimum threshold. What constitutes well-being in one period or country may not be sufficient in another. This social constructivist view of need presumes that norms and values determine what goods and services constitute essential needs and that social structures determine how resources are allocated to meet those needs.

Sen argues further that we should define absolute deprivation not by comparing one person with another who may be more or less deprived, but rather by comparing their capabilities, i.e. what they can or cannot do or be. Being poor compared to others in a community may mean you are unable to meet the accepted norms and obligations and standards of that community. For Sen, poverty means constrained choices, unfulfilled capabilities and exclusion.

Many authors also recognise poverty as being multi-dimensional. Watts (1968) notes two dimensions of poverty: economic and cultural. Chambers (1983) suggests five: poverty proper, physical weakness, isolation, vulnerability and powerlessness. The Australian philosopher John Finnis proposes an extensive list of dimensions of well-being: life encompassing survival, health, and reproduction; knowledge including understanding, education, and aesthetic experience; meaningful work and play; friendship and other valued kinds of human relationships; self-integration (inner peace); authentic self-direction (participation, self-determination, practical reason); and transcendence, meaning ‘peace with god, or the gods, or some non-theistic but more-than-human source’ (Finnis, 1980). While much more could be said, the point is that those who attempt to measure poverty have long recognised that all our approaches have limitations, that the notion of poverty can be relative, and that we should acknowledge the many dimensions of poverty even if we cannot measure them all.

So, given that there is a good deal of agreement on many of the elements of a proper conceptualisation of poverty, we need to ask why there is so much disagreement when measurements are taken. To answer this question, Kanbur (2001) lists some differences in the development discourse – disagreements over the identification, pace and sequencing of economic reforms being the most important. He attributes these disagreements to people's different views on aggregation, time horizon and the role of states and markets. By this, he means that some analysts operate in a paradigm characterised by a high level of aggregation (global, regional or national), in which, with appropriate reforms, competitive markets will function efficiently over the medium term to produce a sustainable reduction in poverty. Others are concerned with lower levels of aggregation (urban/rural, men/women, marginalised areas) and see markets as inherently inefficient, at least for the poor, and thus unable to relieve the deprivation felt by those who are poor at critical points in their lives. From this perspective, market structure is as least as important as the role of government. As would be expected, different analysts have different priorities and expectations when trying to assess the impact of policy and broad socioeconomic trends.

In South Africa, these two kinds of analyst could be caricatured as ‘treasury types’, who want to stretch resources as far as possible to meet competing needs, and ‘social development types’, who want to use resources as effectively as possible to improve human potential. This does not necessarily mean that these two aims cannot be reconciled. Kanbur believes that this can be done, but only if the two types recognise and take into account the underlying reasons for their disagreements. Before they can do this, however, they must first look at what exactly it is that they do when they measure poverty.

3. Measuring poverty

When we move from conceptualising poverty to measuring it, we have to choose appropriate indicators, thresholds and aggregations. Financial or money-metric poverty is the most common kind considered in measures of absolute poverty. It can be measured using data on household income, or household consumption as evidenced by household expenditure. Consumption is usually considered a more reliable and consistent indicator of welfare than income, which fluctuates over time and tends to be under-reported (Deaton, 1997; Lipton & Ravallion, 1997). Most poverty lines denote the amount required to purchase a hypothetical basket of food capable of providing sufficient calories for healthy life, plus certain non-food items deemed important for a minimum standard of living, such as clothes, housing or children's education (Ravallion, 1995). The most widely used poverty lines have been based on multiples of ‘$1 per day’ and expressed in Purchasing Parity Power (PPP) dollars, which take into account variation in the cost of living between countries.

This ‘$1-a-day’ poverty line has been criticised. Reddy (2011) argues that it is inherently flawed and likely to have distorted global poverty estimates. Among other concerns, Reddy contends that the commodities in the basket are not selected according to a meaningful definition of poverty, nor are the purchasing power factors employed adequately and matched according to national currency equivalents. Chossudovsky (1998) is another who vigorously criticises this poverty line, noting that it departs from the notion of a threshold income, does not take into account actual living standards or requirements, and conceals double standards in the measurement of poverty, allowing different standards for poverty lines to be applied to developed and developing countries. Recognising this, Pritchett (2006) suggests an upper bound global poverty line that would be about ten times the current norm.

The published use of such absolute poverty lines can be traced back to 1894 when Charles Booth is credited as the first to make use of a line, fixed at between 18 and 21 shillings per week for a family of five living in London. Once adjusted for a century of inflation, this works out at about twice the median line used in most developing countries (PPP$2.50) in 2008 (Officer & Williamson, 2009). Further, Booth's line, and those subsequently proposed by Seebohm Rowntree, were about double the amount being paid as poor law relief at the time of their calculation (Gordon, 1973), which suggests that poverty lines were never intended to serve as guidelines for social grant thresholds.

Early thresholds were developed in the US in 1963 by Mollie Orshansky, whose purpose was to develop a measure that could be used to assess differences in opportunity for different demographic groups of families with children (Fisher, 1992). Although Orshansky's poverty line worked out at $3165 per annum for a family of four, the threshold of $3000 that was put in place was the outcome of political compromise rather than the direct application of her research (Fisher, 1992). After adjustment for a half century of inflation, this line is about six times the median line of most developing countries. It is also a little more than double the average amount paid through the Aid for Families with Dependent Children Grant, the most common form of social grant in the US between 1935 and 1996 (Miller, 1965:401).

Orshansky's approach, and that adopted by many researchers during the 1970s and 1980s, including those in South Africa (cf. Pillay, 1973), required that a diet and a lifestyle be determined by the researcher to be minimally adequate. In this sense the line is indeed absolute and does not take into account changing living standards or patterns of consumption. Examples in South Africa include the Household Subsistence Level (HSL) developed by the Institute of Planning Research at the University of Port Elizabeth and the Minimum Living Level (MLL) used by the Bureau of Market Research at Unisa (University of South Africa). Both of these measures identified a basket which allowed a minimum food energy intake to be satisfied. This was then priced in different areas, weighted by the size of the household, which made it possible to calculate a minimum household income that would permit the purchase of this basket. While there is a slight difference in the income level that is set as the poverty line, in terms of methodology it is difficult to distinguish any critical difference between the HSL and MLL.

These two poverty lines have had a chequered history, having been inappropriately used to set minimum wages (Budlender, 1985). While this is not problematic in itself, the lines were also racially based, thereby implying acceptance of differences in incomes that favoured whites. Nonetheless, these lines were useful in the period prior to the development of more systematic proposals for a poverty line. As examples, Leibbrandt and Woolard (1999) generated poverty profiles based on these measures, applied to different datasets, while McCord (2004) made use of the HSL to investigate the impact of South Africa's public works programme (McCord, 2004).

The researcher-determined approach has largely been abandoned in favour of using survey data to identify the diet preferred by low earning households and then estimating its caloric value (Lanjouw, 2001). The cost of a diet that delivers around 2000 Kcal, considered the minimum food energy requirement for a healthy active adult, is then calculated. The non-food component is estimated by calculating the cost of items for which households are prepared to forego food so that they can buy them. These are regarded as essential for an adequate life style from the perspective of the poor household. The poverty line that results is relative, since it includes items that are socially and culturally required by a particular society at a particular period. The implication is that the poverty line may change when household budget surveys show changing patterns of consumer behaviour. As examples, using these methods Van Rooy et al. (2006) generated a poverty line that has been adopted in Namibia, as did May and Roberts (2005) for Lesotho, and Oosthuizen (2008) for South Africa.

Once the threshold has been determined, poverty lines may be expressed as per household, per capita or per adult equivalent. The first implies that there are no economies of scale to be gained by larger families, and the second that adults and children have the same nutritional and non-food requirements. Although both are unrealistic assumptions, resolving these issues is not straightforward. Individuals in a household can be expressed in terms equivalent to a reference person, perhaps an adult female, and the food basket or the household size can be re-weighted accordingly. However, as Lanjouw (2001:14) observes, despite a considerable literature there is no agreed approach to determine a reasonable adjustment for equivalence. Further, while young children may consume less than adults, this changes as they grow up (Latham, 1965), which means that the common South African method of halving the number of children younger than 15 is questionable, since it implies an abrupt change in the nutritional requirements of adolescents.

Household economies of scale are also important but no more easily resolved (Lanjouw & Ravallion, 1995). A common practice in South Africa has been to adjust household size by some parameter, typically 0.9, that allows for a modest gain from sharing resources. Studies that estimate the fixed costs for low income households suggest, however, that these gains are in fact higher and that a parameter closer to 0.7 may be more appropriate (Potgieter, 1993; Woolard & Leibbrandt, 1999).

4. Applying poverty measurement in South Africa

When we measure poverty, we follow the same procedure as for most forms of quantification: we decide how to conceptualise the phenomenon appropriately, choose indicators that reflect it adequately, collect data that represents those indicators, and finally analyse and interpret the indicators. We agree broadly on how to take each step – it is when it comes to putting the measures to use that differences arise.

It is here that we must acknowledge the specifics of our situation, in particular the four decades of apartheid policies and legislation that were designed to extract cheap labour (Wolpe, 1972; Legassick, 1975; Terreblanche, 2002). This institutionalised discrimination, in effect state-driven underdevelopment, led to dispossession and exclusion for the majority of South Africans. As Wilson and Ramphele (1989:204, 230) say, it is because of these policies of deliberate impoverishment that poverty in South Africa is different from poverty elsewhere.

Despite this history, or perhaps because of it, South Africa has long experience of measures of poverty, dating back to the Poverty Datum Line in the 1940s and the ground-breaking work of Hilstan Watts in Durban in the 1960s (Watts, 1967). However, like many other things in this country, the data and measures used are inconsistent and often incomplete and reflect a legacy of 40 years of segregation and dispossession. One of the earliest studies, using sample surveys done in 1959, estimated that some 50 to 75% of the African population of urban South Africa were unable to afford a diet and a lifestyle determined to be minimally adequate (De Gruchy, 1960). During the mid 1970s it was estimated that African poverty rates increased to between 68 and 77%, suggesting a surge in deprivation during the era of ‘grand apartheid’ (SPRO-CAS, 1972; McGrath & Whiteford, 1994; Terreblanche, 2002). By the mid 1980s, poverty rates for the rural areas designated for African settlement were estimated at about 75%, compared to 43% for the total population (Simkins, 1984; Nattrass & May, 1986). Data collected from the first nationally representative sample survey done in 1993 showed that just over half (52%) of all African households in rural areas had a per capita expenditure that fell below the HSL (Carter & May, 1999). Using the same data, Klasen reports that in 1993 just under 24% of the population were living below the PPP$1 per day threshold (1997:56). It is worth noting that each of these studies recognised the inequality of South African society along with the political economy that caused it.

Given this history, it is therefore surprising that, although the data and the expertise for developing a survey-based poverty line have been available for at least a decade, at the time of writing South Africa remains without an official threshold. Fortunately, most of our poverty analysts seem to agree that an appropriate threshold for absolute poverty in South Africa lies somewhere between R260 and R515 in 2000 prices, or between PPP$2 and PPP$4. Using a year 2000 poverty line of R322 per person per month, Özler (2007) reports that about 58% of South Africa's population could have been categorised as poor in 1995, a situation that had not changed by 2000, although there had been a small drop in the poverty rate for Africans, down from 68% in 1995 to 67%.

Although they have not been officially endorsed, these poverty lines have been used in a series of hotly debated papers which contest the scale of change in the percentage of those categorised as poor since 2000 (Van der Berg & Burger, 2002; Van der Berg & Louw, 2003; Van der Berg et al., 2005; Meth & Dias, 2004; Meth, 2006, 2007). Drawing on several of the large sample surveys done by Stats SA between 1994 and 1998, Budlender (1999:93) was probably the first to suggest that both poverty and inequality had increased, a finding supported by a Stats SA publication using the results of the 2000 Income and Expenditure Survey (Stats SA, 2002). Certainly, much attention was given in the media to the report that the average annual per capita income in 1995 was R12 135, adjusted to 2000 prices, which at first glance was higher than the R11 755 reported in 2000: a difference that is in fact not statistically significant. Nonetheless, it is probably correct that a turning point in poverty trends was reached around the mid 2000s, though in the absence of large-scale, consistently applied and nationally representative surveys it is hard to be sure.

An important point lost during this debate is that no matter what line, data or approach is used, the actual number of poor people increased in the immediate post-apartheid era. This can be seen if we look at a recent OECD report on poverty trends in the post-apartheid era (Leibbrandt et al., 2010). The poverty rate did indeed decline by a modest amount, from 56% in 1993 to 54% in 2008, but at the same time the population increased by an estimated 8.5 million people, and as a result the number living below the poverty threshold increased by 3.8 million. It is thus not surprising that in 2007 the UNDP ranked South Africa the 14th most populous, in terms of the number of people living on less than PPP$2, out of 99 countries for which poverty measurements were reported. This is despite the country ranking 10th among these countries in terms of its GDP per capita (UNDP, 2009).

The changing nature of South African poverty is evident from the OECD report that shows the urban population increasing by 9.5 million, swelling the numbers of urban poor by 4.7 million, while the number of rural poor declined by 770 000 (Leibbrandt et al., 2010). The authors of this report suggest that the rise in urban poverty may be the result of migration by the poor from rural to urban areas. Kanbur's point about aggregation is worth recalling at this point. For those concerned with providing housing, health care or grants, these figures mean that their target group has grown bigger – in other words, there has been a visible increase in poverty. For those concerned with a broader view of South African society, the figures mean that the share of the population that is categorised as poor has shrunk – which may vindicate policies for promoting long-term economic growth.

Following Foster et al. (1984), most South African studies now report what have become known as the p-alpha poverty measures. These allow us to measure three important magnitudes for any dimension of poverty: its incidence, its depth (also referred to as the ‘poverty gap’) and its severity. An example of how this suite of measures may be applied can be seen in Leibbrandt et al.'s (2010) meticulous analysis. Taking the lower bound poverty line proposed by Özler (2007), these authors report that the poverty gap has narrowed, with the average shortfall of the total population from the poverty line decreasing from 32% to 28%, while the severity of poverty decreased between 1993 and 2000, but was unchanged in 2008. It is clear that the welfare of those below the poverty line has improved somewhat, but that this improvement has apparently not reached those we often term ‘the poorest of the poor’ (Leibbrandt et al., 2010). These may be people who cannot access pro-poor interventions such as social grants or subsidies on services.

An interesting attribute of the poverty gap is that it can be used to indicate the minimum cost of eliminating poverty – a hypothetical indication, of course, since it refers only to the money required were it possible to transfer the exact amount needed to lift each person above the poverty line. As such, it does not reflect the transaction costs of making such a transfer, including the cost of finding out by how much each person falls short of the poverty line. Using a survey done by the Eastern Cape Socio-Economic Consultative Council, May and Nzimande (2009) estimate that a minimum of R881.5 million would have been required in 2006 to eliminate poverty in the Eastern Cape through an income transfer (about R100 per person per month, and thus similar to the basic income grant being proposed at that time). This figure can be compared to the economic output of the province, or to the sum of the provincial and municipal budgets, or to other state expenditure, as a way of demonstrating priorities. We would then have a clear picture of the affordability of poverty reduction. As an example, the minimum cost of eliminating poverty in the Eastern Cape for one year would be a little less than half the cost of building the Nelson Mandela Bay football stadium.

Although my discussion thus far has focused on money-metric poverty, poverty measurement must take into account other forms of deprivation. Physical poverty reflects inadequate access to essential services and measures of such poverty are largely derived from a ‘basic needs’ approach to development. It is important to recognise, however, that changes in the quality and availability of services are not captured by measures that focus only on income.

Some recent studies have captured aspects of poverty other than basic neediness. Deprivation in the form of inadequate access to services is proxied using indicators such as the kind of house the person lives in and what facilities it has or does not have, such as piped water and indoor sanitation. An early example of this approach is a study by Klasen (1997), who generated a multi-dimensional index of deprivation, an exercise later undertaken by Stats SA using data from the 1996 Census (Stats SA, 2000). Some analysts use principal components and factor analysis to develop indices, while others prefer to identify a structural relationship between the components of each of the uni-dimensional measures, assess their inter-correlations, and then calculate reliability statistics such as Cronbach's alpha to determine which dimensions should be summed (De Vos, 2005; May & Nzimande, 2009). Some choose to combine data from censuses and household surveys to map poverty and create small area indicators of deprivation (Alderman et al., 2002; Noble et al., 2009).

Using large sample surveys for South Africa, Bhorat et al. (2006) report improvements in housing and access to water, electricity and toilets between 1993 and 2004, with access to electricity for lighting increasing from 52% of households to 80%, and access to piped water increasing from 59% of households to 68%. This finding is confirmed by the relatively positive position of South Africa in a report released in 2010 by the Oxford Poverty & Human Development Initiative on multi-dimensional poverty (Alkire & Santos, 2010). However, in addition to keeping up with population growth, migration into urban areas and further household fragmentation, an estimated R110 billion will need to be found to clear the remaining backlogs in basic service delivery (National Treasury, 2008:143). It is instructive here to recall Kanbur's observation that views of time horizons differ. Despite South Africa's impressive delivery of essential services, and the reduction in physical poverty that this may have brought about, population dynamics of growth, relocation and composition have resulted in a frustrated citizenry. We need to recognise that the target of our efforts to eliminate poverty keeps changing, and backlogs persist and may even grow worse.

Finally, a number of analysts have turned their attention to the structural dynamics of poverty in South Africa and thus to the third of Kanbur's areas where views differ: the role of states and markets (Carter & May, 2001; Roberts, 2001; Aliber, 2003; May et al., 2011). Their measurements of poverty are dynamic and asset based. They argue that this alternative typology of poverty, using a dynamic approach, provides a more accurate description of the events that households experience over time than does the conventional transition matrix approach that most analysts of chronic poverty use. They identify categories of household according to their structural position as determined by changes in assets and consumption-based well-being. They show that markets do not function adequately as a means for a poor household to escape poverty, since repeated shocks undermine any gains that are made. They suggest that transfers in the form of the Old Age Pension or the Child Support Grant help to buffer the household against such shocks, including those resulting from South Africa's high HIV/AIDS prevalence (Case & Deaton, 1998; Case et al., 2005; Carter et al., 2007).

5. Mathemagics, science or political economy?

Issues of distribution, the concentration of economic power and its relationship to political, social and cultural power, have been central to most analyses of poverty in South Africa. Indeed, most papers written over the past 50 years have been concerned with income and wealth gaps, whether between white and black, urban and rural or rich and poor. These analyses give a fairly good picture of poverty trends in South Africa. It seems that the already high levels of poverty in the 1960s peaked in the early 1980s at around 75% in the most deprived areas, declined slowly until the late 1990s, and may have declined more rapidly in the first part of this century, but are likely to have risen during the economic crisis and rapid food price increases of 2008/09. Access to assets, especially education, and to grants, especially the Old Age Pension, appears to have been a major determinant of these recent trends, while the violent actions of the apartheid state had much to do with the increase in poverty in the 25-year period from 1960 to the mid 1980s poignantly described by Wilson & Ramphele (1989). High levels of unemployment and the impact of premature adult deaths are significant contributing factors to the sluggish response to the government's redistributive policies. However, if the findings of recent analyses are correct, despite the global downturn South Africa may now be on a long-term path of gradual asset accumulation for poor households which will see a reduction in both observed and structural poverty.1

Banking on this, the National Development Plan proposes reducing the population below a poverty line of R418 (in 2009 prices) to zero by 2030 (NPC, 2011:3). This is an ambitious target that will require a real growth rate of around 3% per annum in the incomes of those below this line. Using poverty measures, progress towards this goal can readily be monitored, the contribution made by specific interventions assessed, and proposals for interventions thought to be more effective can be tested and then motivated.

My argument in this paper is that poverty measurement has been much more than the smoke and mirrors some claim it to be. It is not ‘mathemagics’ designed to obscure the real dynamics behind the distribution of benefits and costs of economic progress, nor is poverty measurement a naive attempt to quantify the unmeasurable. In South Africa at least, most researchers did not flinch when making often contentious recommendations, including recommendations for political change, wealth redistribution, universal grants and progressive taxation. Writing more 50 years ago, Joy De Gruchy described the urban poverty of Africans in South Africa as ‘an indictment of our social and economic system’ (1960:67). This could still be said of our current levels of poverty and inequality. When we measure poverty, rather than relying on a simple ‘basic needs’ measure we must rise to this challenge by combining methods, indicators and thresholds. At the same time, we must direct our attention to better theorisation of poverty and the reasons for its persistence. This is essential if we are to move beyond counting to the promotion of greater social justice.

Acknowledgements

Financial support provided by the National Research Foundation under the South African Research Chair Initiative is gratefully acknowledged.

Notes

¹ Observed poverty is what we see when we simply measure the number of people below a poverty line. Some of this might just be bad luck, the time of the month or measurement error. Structural poverty is what we see if we measure the assets that people use in order to generate their income. This is less prone to short-term fluctuations.