MATCHING CONCEPTS WITH MEASUREMENT: WHO BELONGS TO AFRICA S MIDDLE CLASS?

Transcription

1 Journal of International Development Published online in Wiley Online Library (wileyonlinelibrary.com).3105 MATCHING CONCEPTS WITH MEASUREMENT: WHO BELONGS TO AFRICA S MIDDLE CLASS? JAMES THURLOW 1, DANIELLE RESNICK 1 * and DUMEBI UBOGU 2 1 International Food Policy Research Institute, Washington DC, USA 2 Columbia University, New York, USA Abstract: This paper advocates for a conceptualization of the African middle class that would be universally valid: security from economic vulnerability and prospects for social mobility. We suggest three minimum criteria that are jointly necessary and sufficient to reflect this conceptualization: secondary schooling completion, decent housing amenities and skilled employment. Using household surveys for nine African countries, we then demonstrate how measures of the middle class based on expenditure thresholds refer to households with vastly different abilities to meet these three criteria. This cautions against overreliance on expenditure-based definitions, which may obscure qualitative differences among the middle class across African countries The Authors. Journal of International Development published by Keywords: Africa; middle class; measurement; conceptualization 1 INTRODUCTION Recent excitement in the business community about the growth in Africa smiddleclasshas been accompanied by an equal amount of skepticism in the popular press about how the middle class is conceptualized and quantified. 1 At the root of the debate lies the now wellcited African Development Bank (AfDB) report that suggested Africa s middle class grew from approximately 80 million to over 120 million people between 1990 and 2010 (see Mubila & Ben Aissa, 2011). In the AfDB calculations, the lower and upper middle classes consist of those people with daily per capita consumption of $4 10 and $10 20, respectively, *Correspondence to: Danielle Resnick, International Food Policy Research Institute, Washington DC, USA. d.resnick@cgiar.org 1 Throughout this paper, Africa refers solely to the sub-saharan region unless stated otherwise. 2 As is standard, this paper uses United States dollars in order to make comparisons across countries. PPP-adjusted exchange rates include the relative cost of goods and services that are not internationally traded. They are derived from price surveys that may not accurately capture differences in the quality of items consumed by households throughout the expenditure distribution (see Deaton, 2010). This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.

2 Who Belongs to Africa s Middle Class? 589 after adjusting for purchasing power parity (PPP). 2 Skeptics wonder if the $4 per day lower threshold is meaningful and note that the numbers at the higher end of the range are inflated by the inclusion of North African countries (see Enedau, 2013; Fletcher, 2013; Simmons, 2013). Others have promoted different figures, suggesting a smaller middle class, ranging from 32 million (Kharas, 2010) to 50 million (cited in Melik, 2012). Still others have argued that multidimensional measures more accurately reflect the vast heterogeneity in the region (see Burger & McAravey, 2014; Shimeles & Ncube, 2014). Given the potential implications for such a broad range of behaviours related to economic growth, political activism, government accountability and social transformation, the measurement of the middle class is understandably a contested topic. However, there has been minimal empirical analysis of how and why existing approaches are flawed or detailing the exact type of heterogeneity that they fail to capture. Moreover, because of data constraints, most scholars proposing alternative measures focus exclusively on South Africa (Burger & McAravey, 2014; Burger et al., 2004; Rivero et al., 2003; Southall, 2004; Visagie & Posel, 2013). 3 Consequently, little cumulative knowledge is being generated about the size and characteristics of this phenomenon across multiple African countries. This paper aims to address this gap by analysing national household surveys for nine African countries that collectively represent roughly half of the region s population and gross domestic product (GDP): Ghana, Kenya, Malawi, Mozambique, Nigeria, Rwanda, South Africa, Uganda and Zambia. 4 Although Africa is on average poorer than other regions of the world, we aspire to avoid lowering the bar for defining Africa s middle class. Therefore, we present a classical conceptualization of the middle class that builds on existing sociological and economic literature about the minimum criteria, anywhere in the world, that would justify belonging to this constituency. These criteria include whether a household survey respondent: (i) lives in a dwelling with piped water, flush toilet and electricity; (ii) completed secondary schooling; and (iii) works in relatively skilled employment outside the agricultural sector. We then highlight the profile of the middle class who simultaneously meet all three criteria at the various absolute expenditure cut-offs adopted by the AfDB. This cross-country analysis is complemented by a temporal examination of Zambia s middle class between 1996 and 2006, and a racial disaggregation of South Africa s middle class. A number of key findings emerge. First, there is vast heterogeneity in the qualitative meaning of middle class status in Africa. In some countries, a very small share of the total population spends more than $4 per day but among those who do, a high share simultaneously meet our three middle class attainment criteria. In other countries, the reverse is true. Relatedly, we show that where one sets the lower absolute expenditure threshold for belonging to the middle class has qualitatively different implications in some countries compared to others. Second, there is substantial variation across countries in whether secondary education or adequate housing represents the main binding constraint for meeting the minimal criteria of being qualitatively middle class. In some cases, we suggest that this may reflect the effects of government policies related to abolishing school fees or subsidizing housing services. Third, the findings emphasize that per capita expenditure measures clearly move faster than the basic assets associated with middle class 3 However, an important exception to this characterization is Shimeles and Ncube (2014). 4 This is calculated using the World Development Indicators (World Bank 2014) for the year 2005, which is the period when most of our included surveys were conducted. The GDP share adjusts for differences in purchasing power.

3 590 J. Thurlow et al. mobilization, particularly education. This suggests that while the private sector s excitement about Africa s middle class as a growing consumer class might be warranted, considerations of the political and social implications of this constituency might be much more premature. The next section reviews the literature on measuring the middle class in the developing world. The broad debate about these measures becomes even more complicated in the African region, which has such a large share of its population still living below the poverty line and which suffers from a lack of broadly comparable data. Subsequently, we draw on existing theoretical literature to justify our three criteria for inclusion in the middle class, arguing that they collectively constitute a basic universal minimum requirement. This is then followed by a discussion of our data and a cross-country analysis indicating the degree of variation across Africa in terms of meeting the three criteria at the $4 a day or lower threshold. The empirical fact that the same expenditure brackets have very different qualitative meanings is further illustrated through more detailed analysis for South Africa and Zambia. 2 MEASUREMENTS OF THE MIDDLE CLASS IN AFRICA AND BEYOND Determining how to measure the middle-class is critical not only for understanding empirical trends in Africa but also for gauging this constituency s potential to engage in economic and political transformation within the region. Such measures, however, are highly contested and often limited by data availability. Typically, there has been sizable variation between those who rely on sociological approaches compared to economic ones. For instance, in the Marxian view, classes are ultimately defined by individuals common positions within the social relations of production in the capitalist system (Wright, 1979: 326). Ownership of particular assets, including money, physical capital and labour, translates into a hierarchy of power relations that advantages some over others. In the Marxian view, individuals who might collectively comprise a middle class, including professional, technical and managerial employees, were collectively classified as the petty bourgeoisie. Max Weber viewed adherents to a particular class as those who shared similar life chances as a result of not only their ownership of material assets but also their skills and education (Breen, 2005). He further distinguished between class and status with the latter referring more to the consumption patterns, lifestyles, values and ideas of those in a particular group. Erikson and Goldthorpe (1992) focus more on occupational categories as identifiers of class positions, noting that labour differentiation is a defining feature of societies that have become increasingly industrialized. They delineate 11 different occupational categories, ranging from high-level professionals to semiskilled and unskilled manual workers to smallholder farmers. Focusing on occupation categories and especially labour contract status highlights that economic well-being is very much associated with employment security (see Goldthorpe & McKnight, 2004). Because comparable data to operationalize and apply the sociological approaches is often missing, particularly for developing countries, economists have in turn focused more on income and expenditure measures. Some focus on relative measures, which define the middle class according to a threshold corresponding with the nature of the income distribution. For instance, scholars have suggested that the middle class consists of those within a certain range relative to the median income (e.g. Birdsall et al., 2000; Davis & Huston, 1992). Others have defined the middle class as those falling within a particular segment of the income distribution. For example, Barro (2000) and Easterly (2001) focus

4 Who Belongs to Africa s Middle Class? 591 on those within the middle three income quintiles, while Alesina and Perotti (1996) define the middle class as those within the third and fourth quintile. There are advantages and disadvantages of the relative approach. On the one hand, it offers the advantage of being more congruent with the idea that the middle-class represents a less extremist and more temperate constituency that advances progressive views on governance (see López-Calva et al., 2011). On the other hand, a relative measure fails to capture the degree of economic security that middle class status should otherwise signify, especially when an economic shock occurs that would shift the overall income distribution and therefore alter those who are considered part of this class group. The relative measure is especially problematic in Africa given the large share of individuals still living in vulnerable conditions. Consequently, applying relative measures in Africa typically captures those who are actually living very close to the poverty line. Countries with higher inequality also demonstrate a higher disparity across different measurements of the middle class. For instance, based on South Africa s 2008 National Income Dynamics Study, Visagie and Posel (2013) find that using the relative definition, almost 32 per cent of the population is middle class but only 20 per cent if the absolute definition is employed. In contrast to the relative approach, absolute measures use thresholds based on a particular level of income or expenditure. Many different cut-offs have been suggested. Banerjee and Duflo (2008) propose that the middle class consists of those with per capita expenditures between $2 and $10 per day, while Ravallion (2010) suggests the same minimum threshold but argues for $13 per day as the upper threshold because that is equivalent to the poverty line in the United States. Milanovic and Yitzhaki (2002) defined the middle class as those living with a per capita income of $12 50 per day (PPP) even as Kharas (2010) decided that $ per day expenditures in PPP best captures this class. López-Calva and Ortiz-Juarez (2014) also use $10 per day (PPP) as their lower threshold, but they arrive at this cut-off through regression analysis of Latin American data that identifies this as the level at which individuals have only a 10 per cent probability of falling into poverty. The justification for their underlying approach is that those who are firmly in the middle class should resist falling back into poverty even when the macroeconomy performs poorly. Birdsall (2010) combines the absolute and relative approaches, proposing that the middle class refers to those with an income falling between $10 per day (PPP) and less than 95 per cent of the income distribution in their own country. The use of absolute cut-offs, instead of relative ones, implicitly suggests that greater affluence (i.e. income) results in a shift in priorities from basic survival to more outward engagement in one s community and a change in values (see Inglehart, 1997). In addition, it may more accurately reflect one s ability to consume goods beyond basic needs. Yet, the disadvantage is that much of the data used to identify absolute thresholds in developing countries, especially in Africa, is based on consumption or expenditure measures. This is because consumption can be smoothed over time and fluctuates less than income, and because income measures in countries with large agricultural or informal sectors are often inaccurate (see Deaton, 1997). 5 But, using consumption rather than income measures does not capture the savings, investment and taxation dynamics that theoretically underpin the middle class expected economic and political behaviours (see van de Walle, 2012). 5 Income from self-employment is difficult to measure. Even when households accurately report revenues from their farm and non-farm enterprises, they struggle to recall and attribute input or production costs. Net income from self-employment, which is dominant in Africa, is therefore measured with considerable error. Households ability to recall their expenditures is considered more reliable and therefore used as a proxy for income.

5 592 J. Thurlow et al. Table 1. Sub-Saharan Africa s expenditure distribution, Daily per capita expenditure bracket (2005 PPP dollars) <$1.25 $1.25 $2 $2 $4 $4 $10 >$10 All Population (millions) Population share (%) Source: Own calculations using World Bank (2014). In addition, while the absolute approach can facilitate international comparisons, it may fail to capture regionally specific dynamics. In other words, a $10 minimum threshold may be appropriate in the Latin American context but far fewer Africans would qualify. 6 For instance, Table 1 below draws on the World Bank s PovcalNet database and presents the overall population distribution in sub-saharan Africa that falls within particular absolute per capita expenditure thresholds that correspond to those used by the AfDB. 7 A sizeable share of the population still subsists below the international poverty line of $1.25 per day (in 2005 PPP terms). The largest shift in expenditure patterns between 1993 and 2010 was in the $2 4 per day category, which corresponds to the AfDB s floating middle class. In addition, those in both the lower and upper middle class category more than doubled in absolute terms. However, as a share, the $10 cut-off is applicable to less than 3 per cent of the region s total population. Given the limits of existing measures, others have suggested alternative approaches for the African context. Shimeles and Ncube (2014) develop an absolute measure based on an assets index for 37 African countries. Relying on the Demographic and Health Surveys (DHS) over the period , the authors construct an index based on 11 indicators, including source of water for household, condition of housing and ownership of a radio, television, telephone, refrigerator and car. 8 They then define the middle class as those falling between 50 per cent and 125 per cent of the median asset index value for the entire sample of African countries. They find that based on this index, the middle class comprised 15 per cent of the population over the 2000s, compared with 11 per cent in the 1990s. The importance of employment status is not examined in Shimeles and Ncube (2014). Indeed, this is the downside to utilizing the sociological approach in Africa, where there are few surveys that capture this dynamic. Most of the work in the region that applies this approach focuses exclusively on South Africa. Rivero et al. (2003) rely on the sociological approach to understand the evolution of South Africa s middle class because the transition to democracy from apartheid in Using the Omnibus Surveys of the Human Science Research Council for 1994 and 2000, they distinguish between an old middle class that 6 Even within regions, the heterogeneity across rich and poor countries may render such thresholds questionable. 7 See: 8 The authors do not elaborate on the full list of 11 indicators that are included in the index.

6 Who Belongs to Africa s Middle Class? 593 encompasses those who are self-employed and propertied (e.g. own a small shop) and a new middle class that refers to professionals, managers and clerks. They find that the new middle class has expanded from 8.8 to 11.9 per cent of the adult population since The professional category of the new middle class is the category into which most middle class, black South Africans have transitioned. Southall (2004: 522) elaborates on this categorization. He defines the middle class as a group that derives its primary income (directly or indirectly) from non-manual employment, as white-collar employees, managers, self-employed business persons, or professionals. Drawing on Nzimande (1990), he further disaggregates this distinction into four categories of petty bourgeoisies: bureaucrats (e.g. state managers), traders, civil servants (e.g. nurses, teachers and clerks) and corporate managers. Based on various data sources, he finds that the share of black South Africans in all four categories has expanded since the end of apartheid in 1994, with much of this growth aided by equity policies. Burger and McAravey (2014) begin to combine these sociological approaches with the asset-based method used by Shimeles and Ncube (2014). Focusing again on South Africa and drawing on Sen s (1989) theory of capabilities, they propose that instead of looking at the middle class per se, the more accurate distinction is between the empowered and the disempowered. In their view, empowerment is derived from the combination of material assets (stove and refrigerator), key services (electricity, water and sanitation), access to media information, being employed or living in a household with an employed member and having at least 7 years of education. To be middle class, or empowered, requires attaining all of these criteria simultaneously. Using the Project for Statistics on Living Standards and Development Survey and the National Income Dynamics Surveys they find that South Africa s empowered class increased from 28 to 48 per cent of all adults between 1993 and ADVOCATING A CLASSIC CONCEPTUALIZATION OF THE MIDDLE CLASS Drawing on Sartori s (1984) concern with conceptual stretching, we argue that measurements should follow conceptual precision to the extent that data allows, and we suggest that the concept of the middle class should be viewed in classical terms. Classical concepts identify the minimal conditions that are necessary and jointly sufficient in order for an instance of the concept to be present or not. Such concepts involve identifying a basic core and then the representative indicators that capture that core (Goertz, 2006). They require defining categories based on attributes that are variously stressed by different scholars, joined together by a logical and, thereby requiring all attributes to be met to qualify as an instance of that category. Thus, they set a high standard for inclusion and help reduce the chance of false positives (Sartori, 1984). This can be especially important when measuring the middle class in developing country contexts where high education, for example, does not automatically translate into employment (Filmer & Fox, 2014) or where income may be spent on non-essential goods even when basic food security or educational needs have not yet been met (see Christiaensen & Pan, 2012). Indeed, many of the absolute and relative measures discussed earlier fail to approximate classical concepts because certain income or expenditure levels may be necessary, but certainly not sufficient, to qualify as middle class. Classical conceptualizations also help to avoid conflating behavioural attributes and preferences with definitional ones (Weyland, 2001). For instance, the middle class should

7 594 J. Thurlow et al. not be defined according to whether its members have fewer children, decide to live in urban areas, pursue more leisure activities or buy more consumer durables. Classical concepts can be differentiated from radial concepts that focus only on sufficient conditions and brought together with a logical or. In this approach, a case that has just one defining attribute can be classified as belonging to the concept under investigation. In other words, claiming that the middle class is characterized by a certain income threshold or a particular educational attainment is a radial conceptualization. Although radial concepts may help foster consensus across contentious areas and incorporate multiple interpretations of the same concept, they can also result in a plethora of diminished subtypes rather than a core definition (Collier & McMahon, 1993). In other words, they fail to distinguish the single theoretically relevant element that unites different interpretations. For instance, Rivero et al. s (2003) old and new middle class are subtypes of the middle class, but they do not define how they view the essence of that constituency, only the occupational categories to which they belong. A similar conclusion can be drawn for the various types of bourgeoisies, including petty, administrative, bureaucratic and managerial, which historically have been used to discuss Africa s middle class (see Nzimande, 1990; Sklar, 1979). Likewise, Shimeles and Ncube s (2014) approach is also based on a radial conceptualization because an individual or household does not need to possess all 11 assets in their index but only must have enough assets to fit within the median thresholds they establish. Burger and McAravey (2014) come closer to a classical concept by defining the middle class as linked to empowerment but some of their indicators (e.g. access to media) verge on conflating behaviours to the core of being middle class. A similar critique can be levelled at those studies that rely on subjective measures of the middle class, such as Schlemmer s (2005) in South Africa, because they assume that being middle class requires a middle class consciousness and a recognized, common identity. The importance of conceptual clarity is not simply an intellectual exercise. In fact, Burger et al. (2014b) show how different concepts, and in turn, measurements, result in very different conclusions, at least in the South African context, about how large the middle class has become. Consequently, we conceive of the middle class as being, at the basic level, about security from economic vulnerability and prospects for social mobility. This conceptualization is relevant across a wide range of countries and regions and does not pre-suppose a particular ideological or behavioural orientation of this constituency. We argue that three, necessary and sufficient conditions are jointly required at a minimum for a household to be considered middle class. The first is the completion of secondary education. Over the decades, secondary school attainment has been found to be a major driver of occupational stratification and economic outcomes in developing countries, particularly for women (see review by Buchmann & Hannum, 2001). Easterly (2001) also finds a strong correlation between secondary (and tertiary) schooling with middle class status, but not for primary schooling. The corresponding indicator is whether a household head, or his/her spouse, finished all levels of secondary school in a particular country that would allow continuation onto tertiary education in that country. In most countries, this was 12 years, but in the case of Uganda, this was 13 years. We recognize that there are can be important differences in the quality of education such that completing secondary schooling in one context may not be equivalent to the same achievement in another country. Moreover, secondary school completion certainly does not guarantee full-time employment. However, it is a reasonable minimum condition for finding a skilled job and having prospects for social mobility. In this way,

8 Who Belongs to Africa s Middle Class? 595 we are approximating the Weberian perspective that those belonging to the middle class should share equivalent life chances because of their skills and education (Breen, 2005). A second condition for inclusion in the middle class is residence in housing with decent amenities. This was measured by whether a household had, within the dwelling, piped water for drinking, a flush toilet and electricity from a main grid, solar panels and/or private generator. Households that live in such circumstances are less likely to be vulnerable to water-borne illnesses caused from using unprotected water sources and shared, public sanitation facilities. Moreover, access to these services allows for greater productivity and does not require household members to spend additional time collecting, for example, water from public taps and other sources (see UN-Habitat, 2014). These indicators often suggest that a residence is connected to a utilities network and that its inhabitants possess secure property rights, resulting in less vulnerability to eviction or demolition. The absence of such amenities has undergirded multidimensional poverty measures, such as the Lived Poverty Index (see Mattes, 2008). We recognize that having basic amenities present within the household is an imperfect indicator of actual access to services. Being connected to piped water systems and electricity grids does not guarantee a reliable supply of these services from public utility companies. Rolling electricity black-outs and occasional water shortages in many African cities are evidence of this. But, existing household surveys do not capture differences in the quality of services. Accordingly, our housing criteria assume that households assets (e.g. piped water facilities) is an adequate proxy for access to services (e.g. reliable supply of clean water), and that the relationship between assets and service delivery is reasonably similar across African countries. The third condition for inclusion in the middle class requires that the head of household or his/her spouse is employed and his/her occupation is within the skilled, non-farm sector. Undoubtedly, if no one in a household is employed, it places that household in a very precarious position, even more so the longer this condition continues. Moreover, Goldthorpe and McKnight (2004) note that those in skilled employment in particular demonstrate greater earnings potential over their lifetimes and more opportunities for advancement than those in unskilled, or working class, jobs. 9 They are also more likely to have greater stability in their earnings flows by being employed on a contract, and they may also enjoy some social security benefits, such as pensions. As with housing, sub-indicators were used to compile this indicator. Accordingly, the head of household or spouse needs to be (i) an employee or running his/her own business, and (ii) engaged in non-agricultural work other than elementary occupations. 10 In this way, while informal vendors may run their own business, street vending is considered an elementary occupation, and therefore the individual is not considered middle class according to our employment criteria. The exclusion of agricultural work is because of the overwhelming concentration of approximately 60 per cent of Africans in the small-scale, 9 Frey and Stutzer (2002) also report that based on happiness research, those who are unemployed report a lower level of subjective well-being. 10 In identifying those engaged in skilled, non-farm work, we rely on four-digit codes from the International Standard Industrial Classification (ISIC version 3.1) and the International Standard Classification of Occupations (ISCO version 2008). We exclude workers in the agricultural, forestry and fishing industrial sectors (ISIC 1xxx, 2xxx and 3xxx) and those doing manual labour (ISCO 9xxx), street vendors and market salespersons (ISCO 521x), and skilled agricultural, forestry and fishery workers (ISCO 6xxx). The latter category of workers were excluded because the ISCO coding classifies skilled agricultural workers as including cash crop smallholder farmers and subsistence farmers. A majority of Africans are smallholder and subsistence farmers so including them would render the formal employment designation meaningless.

9 596 J. Thurlow et al. often subsistence, agricultural sector (Diao et al., 2007). Although such an exclusion may omit from our sample a handful of large-scale, commercial farmers, particularly in Southern Africa, it also avoids the inclusion of a very large number of vulnerable smallholder farmers and thereby, as Sartori (1984) suggests with classical concepts, prevents false positives. Our conceptualization therefore combines both López-Calva et al. s (2011) view that the middle classes are those who are not vulnerable to falling into poverty with Burger and McAravey s (2014) notion of being empowered. Certainly, a more expansive set of conditions correlates with these factors, but they are not necessary in and of themselves. For example, Shimeles and Ncube (2014) also include possession of durable goods, such as a car. But, this may reflect consumption preferences rather than class position per se and begins to undermine broader comparability. Indeed, many middle class residents in developed countries do not own cars because they live within close access to public transport. Moreover, these consumption patterns reflect Weber snotionof status more than class per se. Furthermore, durables such as a television or a computer may correlate with a middle class lifestyle, but having access to electricity is the necessary condition to use such goods. 4 APPLICATIONS TO THE AFRICAN CONTEXT A major challenge in Africa is the lack of data to make comparisons across countries in a manner that takes into account both economic and sociological approaches to the middle class. We draw on recent household income and expenditure surveys often quite similar to the World Bank s Living Standard Measurements Surveys (LSMS) for nine African countries that collectively represent more than half of the region s population and twothirds of its GDP. 11 These nationally representative surveys include high-quality data, and their similar sets of questions allow for reasonable comparisons across countries (see Deaton, 1997). The advantage of these surveys is that we can combine different measurement approaches discussed above in an attempt to get at the essence of being middle class. By including major regional economies, such as Kenya, Nigeria and South Africa, smaller resource-rich countries, like Ghana and Zambia, and largely rural and agrarian countries, like Malawi, we capture a diverse array of demographic and socioeconomic characteristics. In order to facilitate cross-country comparisons, the surveys used were all conducted during a similar period in the mid-2000s, from 2004 to Table 2 below provides a full list of surveys used and their sample sizes and populations. For each survey, we use the official per capita daily consumption measure computed by the national statistical agency. This combines the value of home produced and purchased foods, expenditure on nonfood items and the use-value of household assets, including the imputed value of owner-occupied dwellings. Consumption values within countries are also typically adjusted for price differentials across internal regions (and time periods if households were not surveyed at the same time in the year). We convert consumption expenditures in each country to constant 2005 PPP-adjusted dollars. We will not pool our country data, and so we do not attempt to adjust consumption values to a common base year. Based on the household survey data broken down by country, Figure 1 largely confirms that those living at $10 per day or above constitute a very small share of the population. 11 Banerjee and Duflo (2008) also rely heavily on the LSMS surveys. However, they only included three African countries (Côte d Ivoire, South Africa and Tanzania).

10 Who Belongs to Africa s Middle Class? 597 Table 2. Description of survey data Country Survey year Household sample size Weighted population (millions) Survey name Ghana 2005/06 8, Living Standards Survey Kenya 2005/06 13, Kenya Integrated Household Budget Survey Malawi 2004/05 11, Integrated Household Survey 2 Mozambique , Inquerito Sobre Orcamento Familiar Nigeria 2003/04 21, Living Standards Survey Rwanda 2005/06 6, Enquete Integrale sur les Conditions de Vie de Menages South Africa 2005/06 25, Income and Expenditure Survey Uganda 2005/06 7, Uganda National Household Survey Zambia , Living Conditions Monitoring Survey Zambia 2002/03 9, Living Conditions Monitoring Survey Zambia , Living Conditions Monitoring Survey Source: Own calculations based on national household surveys. 100 Cumulative population share (%) Kenya Malawi South Africa Ghana At or above daily per capita expenditure level (2005 PPP dollars) Source: Own calculations using national household survey data (see Table 1). Notes: Grey lines indicate unlabeled countries in our nine country sample. Figure 1. Population expenditure distribution, mid-2000s. Source: Own calculations using national household survey data (see Table 1). Notes: Grey lines indicate unlabeled countries in our nine country sample In fact, the category of $10 20 per day is really only meaningful in South Africa where about 20 per cent of the population spends at that threshold. Otherwise, the concentration of the population in most countries largely remains at the lower end of the expenditure distribution. Before proceeding, it is important to note that our analysis of the middle class focuses on a subsample of surveyed households, and that smaller subsamples imply larger margins of error for our estimated statistics. That being said, our surveys oversampled urban areas, which is where higher-income households are more likely to reside. As such, the share of sample households falling into the higher expenditure groups exceeds these groups share of the weighted population. On average, our national surveys include a sample of about 2300 households with $4 or more per capita expenditure per day, and 1300

11 598 J. Thurlow et al. Cumulative population share (%) Electricity? Piped water? Flush toilet? All three? 0 Population At or above daily per capita expenditure level (2005 PPP dollars) Source: Own calculations using Kenya s Integrated Household Budget Survey (see Table 1). Notes: Electricity is for lighting and includes main grid connection, solar panels and private generators. Piped water is for drinking. Flush toilet includes only private toilets within the dwelling. Figure 2. Housing Characteristics in Kenya, 2005/06. Source: Own calculations using Kenya s Integrated Household Budget Survey (see Table 1). Notes: Electricity is for lighting and includes main grid connection, solar panels and private generators. Piped water is for drinking. Flush toilet includes only private toilets within the dwelling excluding South Africa. 12 Given the small sample of households with more than $10 per capita expenditure per day (outside of South Africa), we do not impose upper bounds on our expenditure-based classifications of population groups, e.g. we refer to people spending $4 or more rather than $4 10. We apply the three middle class attainment criteria that were outlined above, i.e. secondary education completion, decent housing and skilled employment, and focus on daily per capita expenditure levels that correspond to the AfDB s (2011) cut-off levels for the poor, floating-, lower- and upper-middle classes. We then identify which conditions are more binding than others across different countries and calculate the share of the population in our sample that meets all three conditions. We begin by looking at our housing indicator, which in turn consists of three sub-indicators. For the sake of simplicity, Figure 2 focuses only on one country, Kenya, which is frequently cited as having a sizeable and growing middle class (see Juma, 2011; Mubila & Ben Aissa, 2011). The solid downward-sloping line on the right-hand side of the figure highlights the declining share of the population with daily per capita expenditure at, or above, each expenditure threshold. In other words, approximately 10 per cent of the population lives on $4 or more per day. The upward-sloping lines then indicate the corresponding share of the population who meet the three components of our decent housing indicator. Thus, while almost 60 per cent of Kenyans at $4 or more per day have electricity, less than 40 per cent have electricity, piped water and a flush toilet. The binding constraint across the expenditure distribution is a flush toilet. Indeed, even for the small share of the population at $10 or more per day, only approximately 60 per cent have the minimum conditions necessary that constitute decent housing. Figure 3 combines the housing indicators with the other two middle class attainment criteria, i.e. education and employment. From about $3 upwards, a sizeable share of the 12 The number of households in our surveys with $4 or more per capita expenditure per day is as follows: Ghana (1942 or 22.3 per cent of the full sample); Kenya (2574 or 19.2 per cent); Malawi (361 or 3.2 per cent); Mozambique (1093 or 10.1 per cent); Nigeria (892 or 4.1 per cent); Rwanda (516 or 7.5 per cent); South Africa (9316 or 37 per cent); Uganda (707 or 9.5 per cent) and Zambia (2273 or 12.2 per cent).

12 Who Belongs to Africa s Middle Class? 599 Cumulative population share (%) At or above daily per capita expenditure level (2005 PPP dollars) Work Housing Education Composite Source: Own calculations using Kenya s Integrated Household Budget Survey (see Table 1). Notes: Work is if head or spouse are employed outside of farm, trading or manual laborer occupations. Housing includes piped water, electricity and flush toilet. Education is if head or spouse have completed secondary school. Composite is if meet all three of the work, housing and education criteria. Figure 3. Composite Middle Class Attainment Rate for Kenya, 2005/06. Source: Own calculations using Kenya s Integrated Household Budget Survey (see Table 1). Notes: Work is if head or spouse is employed outside of farm, trading or manual labourer occupations. Housing includes piped water, electricity and flush toilet. Education is if head or spouse has completed secondary school. Composite is if household meets all of the work, housing and education criteria Kenyan population is working in skilled jobs outside of the agricultural sector. There is more inequality with regards to the housing and education indicators, which increase almost monotonically as one moves along the expenditure distribution. Interestingly however, even fewer Kenyans have completed secondary education than have decent housing. Consequently, the solid line representing the composite of the three indicators highlights that only about 10 per cent of those living at $4 or more per day meet all three conditions, while about 30 per cent do so at $10 or more per day. In other words, even in a country that is viewed as having a booming middle class because of its larger consumer potential, only a very small percentage of the middle class in absolute expenditure terms actually has the combined housing, employment and education profiles that would constitute the basic minimum for meeting a middle class lifestyle. How does Kenya compare to other African countries? Figure 4 illustrates that Uganda s middle class is the worst off among the countries included in our sample. Less than 20 per cent of the population at $10 or more per day meets our three conditions. By contrast, a much higher share of those at higher levels of the expenditure distribution adhere to these criteria in Southern Africa, particularly in Malawi, South Africa and Zambia. Particularly in Malawi and Zambia, there is a larger gap in the share of the population who meet all three conditions at, for example, $3 or more per day than at $4 or more per day. This suggests that the thresholds for distinguishing between the floating and lower middle class, in the AfDB s terms, might be more meaningful in these countries than in others. Table 3 provides more details about these cross-country differences by focusing only on the population who spend $4 or more a day or higher. Not surprisingly, South Africa has the largest share of the population above the expenditure threshold, with a little over a third meeting all three attainment criteria. It is also not unexpected that a majority of those who meet this expenditure cut-off reside in urban areas, although this trend is notably lower in our sample s more agrarian countries. Significantly, the table highlights how relying on absolute cut-offs alone is deceptive and obscures the vast heterogeneity among Africans

13 600 J. Thurlow et al. 70 South Africa Zambia Cumulative population share (%) Malawi Mozambique Kenya Ghana Uganda At or above daily per capita expenditure level (2005 PPP dollars) Source: Own calculations using national household survey data (see Table 1). Figure 4. Composite Middle Class Attainment Rates, mid-2000s. Source: Own calculations using national household survey data (see Table 1) Table 3. Population characteristics at $4-a-day or higher per capita expenditure, mid-2000s Share of this population group who (%) Share of population (%) Live in urban areas Meet the following middle class criteria Housing Work Education Composite Zambia South Africa Malawi Mozambique Nigeria Ghana Kenya Rwanda Uganda Source: Own calculations using national household survey data (see Table 1). Notes: Work is attained if head or spouse is employed outside of farm, trading or manual labourer occupations. Housing requires piped water, electricity and flush toilet. Education is attained if head or spouse has completed secondary school. Composite is if household meets all of the work, housing and education criteria. who may collectively demonstrate similar consumption behaviour. For instance, Ghana is a particularly striking case in that from an expenditure perspective, it has our sample s second largest middle class. However, only 14 per cent of those who fall into the $4 or more per day expenditure group adhere to our classical conceptualization of the middle class. By contrast, only a very small share of Zambians consume at or above the $4 a day threshold but among those who do, approximately 53 per cent meet all three of our criteria for being middle class. A similar share of Ugandans live at the $4 a day or above expenditure threshold but there are substantial qualitative differences in their lifestyles compared with their Zambian counterparts. Table 3 further illustrates which of our three conditions is most binding across which countries. Notably, either housing or education is more restrictive than the employment

14 Who Belongs to Africa s Middle Class? 601 criterion. In Nigeria, for instance, the small share of those above the $4 threshold who meet the housing condition reduces the overall share of those who fall into the composite category. By contrast, in South Africa, education is the clear binding factor, mostly because the $4 threshold is quite low for that country; in other words, if the sample was only focused on a higher threshold, a much larger proportion of the South African sample would have reached secondary school completion. This reflects the lasting legacies of apartheid for lower income South Africans, who are disproportionately black. More broadly, variations in public policies, such as those related to the timing and extent of school abolition fees (see Stasavage, 2005) or the provision of government subsidies for water provision and flush toilets (see Muller, 2008), might play some role in explaining these differences. Table A1 in the appendix provides a fuller elaboration for all the countries in our sample by examining attainment rates across the three main indicators and their corresponding subindicators for those who spend $2, $4 and $10 or more per day. The advantage of our composite measure is that it not only illustrates differentiation across countries during a relatively similar period ( ) but also reveals temporal changes within the same country that may be obscured by income or consumption measures alone. Zambia is particularly well-suited to this purpose for two reasons. First, the Living Conditions Monitoring Survey has been repeated there throughout the 1990s and 2000s, with relative consistency over time. Second, Zambia s economy has experienced well-documented volatility over this period (e.g. Mwanawina & Mulungushi, 2008; Rakner, 2003; Thurlow & Wobst, 2006). Table 4 demonstrates the implications of these macroeconomic shifts for Zambia s middle class. In 1996, almost 6 per cent of the population was living on $4 or more per day in expenditure terms. Yet, only a third of those in that category lived in decent housing, completed secondary schooling, and worked in relatively skilled non-farm jobs. Half-hearted structural adjustment reforms, high levels of corruption and poor macroeconomic management contributed to a declining economy by the late 1990s, reaching its nadir around Thus, the share of population at the lower middle class expenditure cut-off declined to 3.5 per cent and those living in urban areas declined because of reverse migration (see Potts, 2005). Debt forgiveness, lower inflation, a rebound in the price of copper and a change in government administration collectively began to re-energize the economy by the mid- 2000s. Importantly though, the share of the middle class did not just increase in expenditure terms by 2006 but also improved substantially in qualitative terms. Relying on Table 4. Zambian population characteristics at $4-a-day or higher per capita expenditure, Share of population in this per capita expenditure group who (%) Survey year Share of population (%) Live in urban areas Meet the following middle class criteria Housing Work Education Composite / Source: Own calculations using Zambia s Living Standards Monitoring Surveys (see Table 1). Notes: Work is attained if head or spouse is an employee or run their own business, and is employed outside of farm, trading or manual labourer occupations. Housing requires electricity, flush toilet and piped water. Education is attained if head or spouse has completed secondary school. Composite is if household meets all of the work, housing and education criteria.

15 602 J. Thurlow et al. consumption/expenditure measures alone would obscure how Zambia s middleclass became less vulnerable than just a decade before. In other words, the same growth in expenditure resulted in a qualitatively different middle class over time. Finally, by focusing on South Africa, we consider the value of our composite measure by examining variation among different groups of populations within the same country. As noted earlier, South Africa is the only country in our sample where a $10 20 per day upper middle class category is valid and meaningful because in other countries very few of the population spend more than $10 per day. As such Table 5 below presents profiles of those in both the $4 10 and $10 20 expenditure groups. The population groups are further disaggregated by the four racial groups that are recognized by the South African Constitution: African, coloured, Asian and white. Most of the African and coloured middle class is distributed in the lower middle class category while a majority of the white middle class is located in the upper middle class strata. The Asian community demonstrates sizeable representation in both categories. Critically, the table highlights that even among those within the same expenditure groups, a higher share of those belonging to the Asian and white communities meet the composite measure of the middle class than the African or coloured communities. By disaggregating the composite indicators, education appears to be the binding constraint across all these racial groups but is especially problematic for the coloured community. At the same expenditure level, the African community lags behind the other communities with respect to decent housing. Indeed, for all the media and market industry excitement about the expansion of South Africa s black middle class (e.g. Brulliard, 2010; Goyal, 2010; Van Eyk, 2008), Table 5 provides a sobering reminder that higher consumer power cannot be automatically equated with security from economic vulnerability and prospects for social mobility. This is particularly important for policymakers, who may overlook the fact that an expenditure-based middle class can simultaneously lack access to basic amenities and services, and that this deprivation varies substantially across population sub-groups. Table 5. South African population characteristics by race and per capita expenditure groups, 2005 Share of population in this per capita expenditure group who (%) Racial group Share of population (%) Live in urban areas Meet the following middle class criteria Housing Work Education Composite $4 10, all races African Coloured Asian White $10 20, all races African Coloured Asian White Source: Own calculations using South Africa s Income and Expenditure Survey (see Table 1). Notes: Work is attained if head or spouse is an employee or run their own business. Housing requires electricity and flush toilet. Education is attained if head or spouse has completed secondary school. Composite is if household meets all of the work, housing and education criteria.