chapter  5
16 Pages

The Making of the Middle-Class in Africa: Evidence from DHS Data

ByAbebe Shimeles & Mthuli Ncube

Recent interest in understanding the evolution of the middle class in the context of developing countries is inspired partly by economic successes in Asia and Latin America, where the emerging middle class has played a major role in driving growth (Chun, 2010; Desgoigts & Jaramillo, 2009; Easterly, 2001). Studies also highlighted the role of the middle class as an agent of change reforming institutions (Loyza, Rigolini, & Llorente, 2012), a catalyst for the realisation of inclusive growth (Birdsall, 2010; Ravallion, 2012), and innovation and entrepreneurial drive (Banerjee & Duflo, 2008). Despite these positive contributions of the middle class to development, very little is known about its size and attributes in Africa until recently (African Development Bank [AfDB], 2011; McKensie Global Institute, 2010). This study builds on recent work to provide evidence on the making of the middle class in Africa,

noting the challenges African statistics face in terms of reliability, comparability and consistency. Recent studies highlighted severe faults in the national accounts of African statistics that pervade cross-country comparisons both spatially and temporally (Jerven, 2013a, 2013b). Statistics on household welfare are even murkier where such basic concepts as ‘wealth’, ‘poverty’, ‘middle class’ or units of analysis such as a ‘household’ vary considerably across livelihood systems and data collection methods (for example, Randall & Coast [2013]). Available evidence on the making of middle class so far is based on per capita consumption

expenditure collected through large budget surveys. Such data are collected infrequently and in irregular time intervals in many countries in Africa, making contemporaneous comparisons difficult (Deverajan, 2013). It is not also unusual for survey methods and procedures to change between surveys without due regard to comparability issues. Furthermore, systematic measurement errors in the

construction of consumption expenditure abound due to complex livelihood systems. In rural areas, subsistence farming means non-marketable goods dominate the consumption basket, making valuation an intricate and complex task, while in urban areas consumption data have to rely on reliability of respondents’ memory (Deaton, 1997). For example, recently Beegle, De Weerdt, Friedman, & Gibson (2012) reported that the measurement error arising from different methods of collecting consumption data has significant bias on per capita consumption expenditure aggregates in Tanzania. In addition, consumption surveys in rural areas are complicated by the existence of large spatial and temporal price differences and diverse measurement units across villages (Howe et al., 2012). This article proposes to use asset or wealth status reported in the Demographic and Health Surveys

(DHS) instead of income or consumption expenditure as a key indicator to identify the population of the middle class. This approach has several advantages. First, it provides an opportunity to compare existing estimates of the size of the middle class based on consumption expenditure or income. Second, asset or wealth status reported in DHS is not only better measured than consumption (Moser & Felton, 2007), but also designed carefully to be comparable over time and across countries. Comparatively, it is also available for many countries in multiple waves. In addition, most of the countries covered in this study do not have well-developed financial systems or borrowing against future earnings are available to smooth consumption. Assets are built often by drawing down cash savings. Household assets are accumulated in part as a buffer against shocks and also as a way of improving the standard of living (tap water, better roofs, floors, bedrooms, and so forth). Given the predominance of food in total expenditure in a number of African countries, many of the factors commonly believed to constitute a middle class status are easily missed. Furthermore, the data on asset ownership are based on a wide range of indicators offering an opportunity to capture a broader meaning of the middle class. While ownership of assets in the DHS data set provides a reasonable source of information to

measure the size and trend of the middle class, it is not without caveats. The most serious refers to the arbitrariness in the treatment of ownership of a variety of durable goods and the inherent difficulty in using the asset index to make inter-temporal comparisons (Johnston & Abreu, 2013). However, to redress some of the deficiencies, our measure of a middle class based on the asset index was compared with other approaches such as consumption-based measures, multi-dimensional measure using assets and subjective measures of social class. This study goes further in examining the factors driving cross-country differences, as well as in

providing household and individual characteristics correlated with the middle-class characteristics using unit record data covering 37 African countries with multiple data points. The total sample size used for such analysis consisted of the history of over 790,000 households. It also utilises carefully constructed synthetic or pseudo panel from this sample to examine the dynamics of the middle class across cohorts. Institutional and policy aspects are presented as correlates of the size of the middle class. In particular, the role of ethnic fractionalisation and level of trust among citizens in assisting social mobility is examined. With respect to policy, we focus on governance, education and health as the main pathways into nurturing the middle class using unit record data and a synthetic panel constructed from the country surveys. Our result is informative and revealing. The trend in the size of the middle class in the last decade

depends mainly on how the middle class is defined. We present four definitions of the middle class that reflect the different approaches in the recent literature, but also allows comparisons across countries and surveys (see Section 2 for details). Based on our preferred definition, the size of the middle class has been growing in several African countries, though very slowly compared with the rapid economic growth witnessed, particularly in the last decade. The factors that led to the changes could be broadly classified into those of asset shrinking (average asset index declined), or rising asset inequality, or rapid asset accumulation. Country cases are provided to illustrate these patterns. The cross-country variation in the size of the middle class is explained largely by differences in

initial level of development, social structure (mainly ethnic fractionalisation and the degree of mutual trust among citizens), and most of all the degree of asset inequality. Generally, countries that have homogenous ethnic groups, stronger social capital as captured by mutual trust, and high initial level of

development tend to have higher size of the middle class. These results very much echo previous studies for other regions (Easterly, 2001; Knack & Keffer, 1997) for developing world. We also recognise the debate on the use of ethnic fractionalisation as a proxy to capture fragility in institutions and other political economy dynamics (Jerven, 2011). Findings from analysis of multiple country-level household surveys indicate that individuals in the

middle class tend to be well educated and the ‘returns’ to education in terms of higher asset accumulation are consistently higher for all level of education than no education at all. What would be the implications of expanding education opportunities as a policy to promote the middle class? This has to be seen cautiously, since our results indicate that the ‘returns’ to additional level of education declines with stock of educated people at secondary and tertiary levels, while it remains unchanged for primary education. But it should also be welcome, as it could lead to a decline in overall inequality by narrowing relative differences across the education spectrum. The rest of the article is organised as follows. Section 2 describes data and methodology, Section 3 presents the key results and Section 4 concludes the article.