Trends in the working poverty rate (WPR) in post-apartheid South Africa, 1997–2012

ABSTRACT

As South Africa prepares to implement a national minimum wage for the first time, a number of questions about the potential benefits of a minimum wage have emerged. However, most of the South African literature, to date, has been concerned with the country's high unemployment rate and not on the quality of employment. In particular, there has been very little attention given to one widely used indicator of decent work, the working poverty rate (WPR). Using the October and General Household Surveys, we present an analysis of trends in working poverty over most of the post-apartheid period (from 1997 to 2012). The findings reveal that, while income poverty has decreased, poverty among the employed is a persistent feature of the post-apartheid period and that the contributions of earnings from employment to overall poverty reduction have been disappointing.

KEYWORDS:

• Employment
• poverty
• earnings

1. Introduction

The characteristics of the labour force in post-apartheid South Africa are markedly different from those in both developed and developing countries. In particular, the South African experience does not align with either the stylised low-wage-low-unemployment labour market epitomised by the United States or the high-wage-high-unemployment pattern that is the norm in many European countries (most notably Spain and Italy – see Brady et al. 2010). It is also difficult to compare the South African labour force with many developing countries due to the relatively small size (roughly a fifth of the total non-agricultural workforce) of its informal sector (Verick 2011; ILO 2013) and subsequent low employment-to-population (EPR) ratio. Moreover, South Africa is unusual in the developing country context in that it has a much higher unemployment rate than the norm and a ratio of low earners to unemployed of almost one to one, compared to the average ratio in developing countries of approximately six to one (Fields 2011a).

As a result, much of the literature in the post-apartheid period has been dominated by the country's persistently high rate of unemployment. The official (or narrow) unemployment rate has remained ‘stuck’ at about 25% for much of the post-apartheid period and the expanded unemployment rate consistently has been over 30% (Festus et al. 2016). One consequence has been that the focus on unemployment has tended to push into the background the issue of ‘working poverty’. Poverty among the employed can result from a combination of low earnings, under-employment, and/or a low ratio of employed members to total household size. Given the proposal of a monthly national minimum wage of R3,500, the present moment provides an important opportunity to reflect on how the poorest South African workers have fared over the post-apartheid period and whether the labour market has improved its contribution to reducing poverty.

In the South African context, an additional question is whether the structural inequalities of the labour market inherited from the pre-1994 apartheid government have been sufficiently transformed, particularly for low-wage workers who are still mostly black Africans. In this paper, we examine trends in the working poverty rate over a recent 15-year period, and how labour market earnings and transfers from the government's means tested social grant programme have contributed to changes in the incidence and depth of working poverty over the period. In order to do this, we extend the estimates of the WPR by making use of a popular poverty decomposition method to estimate the relative contributions of labour market income and transfer income to the incidence and depth of poverty among employed South Africans over the period.

The remainder of the paper is structured as follows. Section two sets the scene for an analysis of working poverty by reviewing the international working poverty literature and then identifying some of the post-apartheid trends in income, wages and earnings in South Africa. In this section we also review the handful of studies which have measured poverty among the employed in the post-apartheid period. In section three, we discuss how working poverty is defined, describe our data sources, introduce the decomposition method, and outline data and analytical limitations. Section four presents the findings in two parts. We begin by examining working poverty trends over a recent 15-year period. We then turn to possible explanations for those trends by considering changes in income sources in the households of the working poor and by presenting a decomposition of both the depth and incidence of working poverty. Section five concludes with a discussion of what employment has ‘contributed’ to poverty reduction in post-apartheid South Africa.

2. Working poverty in developed and developing countries

Renewed interest in working poverty in developed countries has emerged in the literature in the wake of the financial crisis and the ensuing period of austerity in many OECD countries (Cooke and Lawton 2008; Brady et al. 2010; Crettaz and Bonoli 2010; Gottfried and Lawton 2010; Slack 2010; Wagle 2011; De Beer 2012; Wicks-Lim 2012; Brady et al. 2013; Crettaz 2013; Leadbeater et al. 2014; Theseira 2014; Thiede et al. 2015). One of the main findings from this research is that working poverty is a fairly typical (and growing) type of poverty, particularly in the United States (Iceland and Kim 2001; Johnson 2001; Joassart-Marcelli 2005; Blank et al. 2006; Slack 2010; Wagle 2011; Wicks-Lim 2012; Brady et al. 2013; Thiede et al. 2015), but also in Canada (Fleury 2007), some countries in Europe (O’Connor 2000; Cooke and Lawton 2008; Brady et al. 2010; Crettaz and Bonoli 2010; Gottfried and Lawton 2010; Crettaz 2013), and Australia (Eardley 2000).

In one of the more recent studies, Brady et al (2010) found that the rate of working poverty was higher than ‘unemployed’ poverty in 14 out of the 18 developed countries for which data were analysed. Among the countries included in the study, the rate of working poverty varied considerably, e.g. from 1.8% (in Belgium) to 11.1% (in the United States). Given the high rates of working poverty in the United States, relative to other developed economies, a number of studies have investigated some of the reasons¹ for the high (Cormier and Craypo 2000; Ehrenreich 2001; Gleicher and Stevans 2005; Brady et al. 2013) and growing (DeFina 2007; Thiede et al. 2015) levels of working poverty in that country. Others have looked at who the working poor are (Iceland and Kim 2001; Joassart-Marcelli 2005; Slack 2010; Wagle 2011), the links between unionisation and working poverty (Brady et al. 2013), as well as minimum wages and working poverty (Wicks-Lim 2012).

The developing country literature on working poverty has been far more limited (Majid 2001; Sundaram and Tendulkar 2002), and has drawn attention to the typical situation in developing countries where poverty rates are higher than unemployment rates and the majority of the workforce is found in the informal sector. Research on working poverty outside of the OECD countries has been dominated by the ILO through its Key Indicators of the Labour Market (KILM) series. The first and one of the most widely cited of these studies (Majid 2001) found that approximately 534 million people, or about a quarter of the employed labour force, in developing countries could be classified as working poor in 1997. Kapsos (2004) extended this research and showed how the burden of working poverty (at the one US dollar a day rate) was on a downward trajectory in East Asia and on an upward trajectory in Sub-Saharan Africa, and projected that, by 2015, more than 40% of the world's working poor would be found in Sub-Saharan Africa.

Indeed, one of the key findings from the global estimates of the WPR is that sub-Saharan Africa has been, by some distance, the worst performer historically (Berger and Harasty 2002; Kapsos 2011; Combes et al. 2014). Between 1980 and 2004, the working poverty rate in this region actually increased (at both the one US dollar a day and two US dollar a day poverty lines) with the numbers of working poor people (at both lines) doubling in that period. The most recent estimates of working poverty in the region (ILO 2014) indicate that 39% of the total workforce is below the dollar a day poverty line and 62% is under the two dollar a day threshold. Looking at all developing countries, Fields (2011b), suggested that, in 2011, workers living in households below the US two dollar a day poverty line accounted for 57% of employment in developing countries.

3. South African trends in income, wages, and earnings over the post-apartheid period

Although there has been very little research interest in the WPR in South Africa, recent work has been concerned with how both income and earnings have changed since the end of apartheid. Real increases in household (per capita) income at the middle of the income distribution have been very modest over a recent 15 year period (about 1.4 per cent per annum) (Visagie 2013, 2015a). In terms of the distribution of income growth, there is evidence of a ‘middle income squeeze’ in South Africa – while income growth was positive for all deciles, the growth was higher at the tails of the distribution and particularly skewed towards the top income deciles (annualised per capita income growth was about 1.6% between 1993 and 2008 for the top percentiles and only about 0.2 per cent at the middle of the income distribution) (Visagie 2015a). Moreover, the contribution of labour market earnings to income growth was disappointing over the period and, ceteris paribus, without the expansion of government transfers, households in the middle of the income distribution would actually have gone backwards between 1993 and 2008 (Visagie 2015b).

Trends in labour market earnings have also been disappointing (Wittenberg 2014a, 2014b). Growth in real earnings has been ‘stagnant’, at the median, since 1994 accompanied by strong growth at the top end of the distribution (Wittenberg 2014a). At the same time, there is some evidence of wage growth at the 10th percentile of the earnings distribution. However, some of these gains have likely been eroded by inflation that has been anti-poor over the 2005–2010 period (Finn et al. 2014).

To the best of our knowledge, there are only a handful of post-apartheid studies which are concerned with working poverty, specifically. First, Vermaak (2010, 2012), using the bi-annual Labour Force Surveys, found that working poverty decreased between 2000 and 2006 (from about 25% to 18%). However, this study defined working poverty purely in terms of earnings, which meant that it measured poverty wages rather than household income poverty or the WPR. Second, two recent studies based on the National Income Dynamics Study provide updated estimates of working poverty based on an internationally recognised definition. Lilenstein et al. (2018) found that 17% of workers lived in households below the poverty line in 2012. They also show that, not surprisingly, the risk of working poverty is higher among black Africans, women, the self-employed in the informal sector and rural workers. They show also that the working poor are more likely to be employed in private households, mining, and agriculture. Similarly, Finn (2015) focused specifically on the characteristics of the working poor in South Africa and found that workers engaged in domestic work (95%), agriculture (90%), construction (63%) and the trade sector (60%) all face high poverty risks (Finn 2015).

None of these studies, however, were concerned with trends in working poverty or in the contribution of earnings to poverty reduction. This paper aims to fill this gap by examining changes in the WPR over an important period in terms of poverty reduction and labour regulation and by identifying how earnings and transfer income contributed to these trends.

3. Data and methods

3.1. Definitions

While there is no universal definition² of working poverty, there are several widely used approaches. For example, Brady et al (2010) use a broad definition in identifying an individual (of any age) as working poor if she/he resides in a household with at least one employed household member but in which total household income (or per capita income) is below the poverty threshold. In this paper we use the ILO's KILM definition in which a person is identified as working poor if she or he is employed and living in a household in which per-capita income or expenditure is below the (national) poverty line (Majid 2001; Ostermeier et al. 2015). Even though this is a slightly narrower definition than the one used by Brady et al (2010) in the US, it still accounts for both the level of earnings and how earnings are spread across a worker's household (Kapsos 2011).

3.2. Data

The key data requirement for measuring working poverty is a survey which captures information on both employment and total household income or expenditure. Since this paper is concerned specifically with measuring changes in the WPR over time, an additional requirement is that the survey captures this information consistently and regularly over a relatively long period of time. There are several post-apartheid data sources which could be used, but arguably the most consistent are Statistics South Africa's October Household Surveys (OHSs), conducted annually from 1993 to 1999, and the General Households Surveys (GHSs) which have been conducted annually since 2002 (see Posel and Rogan 2012). In terms of the measurement of poverty, one of the most important features of these two surveys is that they regularly collect data on earned income as well as information on individual access to social grants. This makes it possible to generate measures of income for individuals and households that can be disaggregated further by income source (i.e. labour market and transfer income).

In order to measure changes in working poverty over the post-apartheid period, seven surveys are analysed in the period between 1997 and 2012.³ The study uses the 1997 OHS as the base year because this is the first household survey to capture comprehensive information on individual access to social grant income. Since 2004, the GHSs have captured detailed information on the individual receipt of social grant income and are largely comparable with the post-1996 OHSs. In measuring total household income, we follow a number of other studies (Meth and Dias 2004; Leibbrandt et al. 2006; Posel and Casale 2006; Posel and Rogan 2012; Rogan 2013) in assigning the midpoint⁴ of the reported income bracket where point estimates are not available and income is reported in bands. In order to adjust for the underestimation of income associated with households that report zero income from transfers or the labour market, household income is imputed using information collected on household expenditure captured in the OHSs and the GHSs (see also Posel and Rogan 2012; Rogan 2013).

3.3. Analysis

In estimating trends in working poverty, the study uses the conventional Foster-Greer-Thorbecke (FGT) (1984) series of measures to identify the incidence and depth of income poverty among employed South Africans. Three poverty lines based on Statistics South Africa’s (2008) official poverty thresholds⁵ are used to measure poverty. For greater comparability with the broader poverty literature, we refer more frequently to the national upper-bound poverty line of R323 per capita monthly income (in 2000 prices) proposed by Statistics South Africa (2008). This poverty threshold is based on expected minimum food and non-food requirements⁶ and is very close to the R322 per capita poverty line (in 2000 prices) that has been used in most post-apartheid poverty studies. However, Hoogeveen & Özler (2005, 2006) originally specified Statistics South Africa's upper-bound threshold as a lower-bound limit, and identified an upper-bound poverty line of R593 per capita monthly income (again in 2000 prices). We also use the R593 per capita poverty line in our analysis, and treat that line as the point at which a household can just afford its minimum basic food and non-food needs.

While the analysis of working poverty in the post-apartheid period covers the years between 1997 and 2012, greater attention is given to the period between 2004 and 2012. Much of the well documented (Meth 2006; Leibbrandt et al. 2008; van der Berg et al. 2008; Leibbrandt et al. 2010) decrease in income poverty in South Africa occurred in the early to mid-2000s at the same time that government rapidly expanded its social transfer programme. Moreover, legislation which extended minimum wage determinations to low paid sectors (i.e. agriculture and domestic work) were also introduced in this period (Department of Labour 2002).

Given its impact on aggregate poverty (van der Berg et al. 2008), the role that social grant income may have had on the working poor is also considered in this analysis. In particular, we use a popular poverty decomposition method to estimate the average marginal effect of labour market earnings and transfer income on the reduction of working poverty rates. In order to estimate this decomposition, the analysis makes use of the Distributional Analysis STATA Package (DASP) module developed by Araar and Duclos (2007). The notation for the decomposition is given as:$\hat{P}\left(z;\alpha ;y=\sum _{k=1}^K{S}_k\right)=\frac{{\sum }_{i=1}^n{W}_i{\left(1-\frac{y}{z}\right)}^{\alpha }}{{\sum }_{i=1}^n{W}_i}$where, y is income, z is the poverty line, there are k income sources and S_k represents source k, W_i is the weight given to individual i, and n is the sample size (Duclos and Araar 2006; Araar and Duclos 2009a; Araar and Duclos 2009b). The decomposition (Araar and Duclos 2009a) estimates the contribution of each income source to the elimination of poverty by comparing what the FGT measures would have been without each particular source of income. By making use of the Shapley values, the model estimates the average marginal effect of each income source over all possible combinations of income sources.

3.4. Limitations

While the OHSs and the GHSs regularly and consistently collect information on labour market earnings, social grant income and household expenditure, the trade-off in using these data is that there is limited information on labour market status. Across the seven data points in which we measure working poverty rates, there is limited information, for example, on hours worked, occupation types or sectors of employment for the employed. A question on whether employment was in the informal or formal sector was only included in the GHSs from 2010 onwards.⁷ Therefore, the data used in this study should be seen as a consistent and regular, albeit somewhat broad, measure of the economic well-being of South African workers and their households.

An additional limitation associated with using the GHSs to measure poverty levels is the possible underestimation of income. Studies (Meth and Dias 2004; Posel and Rogan 2012; Rogan 2013) which have used the GHSs or the bi-annual Labour Force Surveys to measure poverty have had to account for the fact that these surveys only capture information on social grant income and labour market earnings. We tested the sensitivity of our working poverty estimates to this possible bias by creating an alternative household income measure which replaces household income data with expenditure data for observations for which total household expenditure is greater than total household income. In other words, if the GHS identifies some households as surviving on transfer income only, but there are other undetected sources of income (e.g. remittances or private transfers) in these households, then income will be underestimated in these households (and poverty overestimated). By substituting income data with expenditure data, this type of underestimation will be addressed. When we make this adjustment to the GHS data series, we find that the poverty and the WPR estimates are slightly (but not significantly) lower compared with the measure of household income used in the remainder of the paper.⁸

4. Empirical findings

4.1. The working poor in post-apartheid South Africa, 1997–2012

The analysis begins with an overview of trends in the poverty headcount rate, or the percentage of the population in households below the poverty line, between 1997 and 2012 (Table 1). In order to contextualise the findings, poverty rates for the population as a whole are presented alongside estimates of headcount rates for the employed (the WPR). At the two lower official poverty thresholds, the results are closely in line with the existing literature and show that poverty increased in the late 1990s (e.g. from 59.5% to 63.7% between 1997 and 1999 at the higher (Z = 323) poverty line) before decreasing in the early to mid-2000s.

Table 1. Poverty headcounts (P₀) for South Africa, 1997–2012 (per capita).

Headcount (P0)
	1997	1999	2004	2006	2008	2010	2012	Relative change 1997–2012
	Z = 323
All	59.53 (0.64)	63.65 (0.65)	60.23 (0.76)	54.84 (0.81)	52.36 (0.73)	50.85 (0.71)	48.06 (0.69)	−19.27†
Workers	28.79 (0. 61)	34.40 (0.65)	28.85 (0.69)	23.78 (0.64)	23.81 (0.62)	22.87 (0.56)	21.40 (0.53)	−25.67†
	Z = 219
All	48.85 (0.63)	54.33 (0.66)	49.35 (0.76)	42.86 (0.78)	40.19 (0.67)	38.85 (0.64)	35.46 (0.61)	−27.41†
Workers	19.44 (0.49)	25.97 (0.57)	20.02 (0.57)	14.79 (0.47)	15.04 (0.49)	14.52 (0.43)	13.76 (0.43)	−29.22†

†Denotes a significant change in the poverty headcount between 1997 and 2012 at the 95% level of confidence.

††Denotes a significant change in the poverty headcount between 1997 and 2012 at the 90% level of confidence.

Notes: The data are weighted (revised weights). Standard errors in brackets.

Poverty lines in 2000 prices.

Source: Own calculations from the OHSs (1997–1999) and the GHSs (2004–2012)

At both poverty lines, the relative decrease in working poverty was larger than for the population as a whole. For example, the poverty headcount rate decreased by about 19% between 1997 and 2012 for the population as a whole but by 26% among the employed. At the lower poverty threshold (Z = 219) the difference in the size of the decrease in poverty rates was much smaller (i.e. 27% for the population as a whole and 29% among the employed).

A first glance at the data would therefore suggest that the findings on working poverty during the post-apartheid period are mixed. On the one hand, while 29% and 19% of all workers were poor in 1997 at two official poverty lines, by 2012 these poverty rates had decreased by more for workers than for the population as a whole. On the other hand, in 2012, more than a fifth (21%) of workers were still living in households with a level of income which did not meet their minimum basic needs. Using our upper-bound poverty line (see Table 2 below), we find that, in 2012, more than a third of workers lived in households that could just cover their minimum basic needs.⁹

Table 2. Poverty headcount (P₀) rates for South Africa, 2004–2012 (per capita).

Headcount (P0)
	2004	2006	2008	2010	2012	Relative change 2004–2012
	Z = 593
All	72.19 (0.70)	68.87 (0.79)	67.35 (0.70)	65.75 (0.70)	62.75 (0.72)	−13.08†
Adults	66.29 (0.74)	62.40 (0.83)	61.00 (0.71)	59.78 (0.71)	55.81 (0.73)	−15.81†
Workers	45.10 (0.87)	40.64 (0.89)	41.11 (0.79)	39.50 (0.74)	36.33 (0.75)	−19.45†
	Z = 323
All	60.23 (0.76)	54.84 (0.81)	52.36 (0.73)	50.85 (0.71)	48.06 (0.69)	−20.21†
Adults	53.17 (0.76)	47.31 (0.80)	45.38 (0.70)	44.50 (0.68)	41.04 (0.65)	−22.81†
Workers	28.85 (0.69)	23.78 (0.64)	23.81 (0.62)	22.87 (0.56)	21.40 (0.53)	−25.82†
	Z = 219
All	49.35 (0.76)	42.86 (0.78)	40.19 (0.67)	38.85 (0.64)	35.46 (0.61)	−28.15†
Adults	42.14 (0.72)	35.73 (0.73)	33.74 (0.61)	33.15 (0.59)	29.08 (0.54)	−31.00†
Workers	20.02 (0.57)	14.79 (0.47)	15.04 (0.49)	14.52 (0.43)	13.76 (0.43)	−31.27†

†Denotes a significant change in the poverty headcount between 1997 and 2012 at the 95% level of confidence

†† Denotes a significant change in the poverty headcount between 1997 and 2012 at the 90% level of confidence

Notes: The data are weighted (revised weights)

Standard errors in brackets.

All poverty lines in 2000 prices.

Source: Own calculations from the GHSs (2004–2012).

Table 2 explores these findings further by focusing on the period during which progress in poverty reduction finally began to take hold in South Africa (i.e. the early to mid-2000s) and by looking at three different groups, namely the population as a whole, adults (18 and older – both working and non-working), and workers. At all three poverty lines, working poverty decreased between 2004 and 2012. Even after the large expansion of the social grant system in the early 2000s, the risk of poverty for workers decreased by more than for adults and the population as a whole (at the two higher poverty lines). In terms of the magnitude of the reduction in poverty, the relative decreases in poverty among workers are greater at the lower poverty thresholds.

In other words, the greatest reduction in working poverty is seen at the lower bound poverty threshold (z = 219), where the rate of working poverty decreased by 31% (compared with a decrease of about 20% at the R593 poverty line). This suggests that the greatest success in reducing working poverty during the 2000s took place among the poorest workers, albeit from a lower base. However, the fact that the relative decrease in working poverty was actually very similar to the overall decrease in poverty at this threshold (z = 219) also suggests that progress in poverty reduction among workers was not much different than progress among adults or the population as a whole (at this particular poverty line). A closer look at the 2004–2012 period also suggests that, after 2006, the headcount rate among the employed doesn't change significantly while there is a large and significant drop in poverty among adults and the population as a whole.

Since the poverty headcount identifies only one aspect of income poverty (i.e. whether an individual is above or below a specified threshold), an analysis of the depth¹⁰ of working poverty (or the working poverty gap) is presented in Table 3. The findings in the table suggest that the working poverty gap also decreased over the post-2004 period. Moreover, the decreases are sizeable at 30% and 32% at the respective poverty lines, and decreases over the period were very similar to the decreases in the depth of poverty for the population as a whole. In fact, at the lower poverty threshold (Z = 219), the decrease in the working poverty gap was actually smaller than the decrease in the overall poverty gap (32% and 35%, respectively). This again points to the conclusion that progress in reducing working poverty among the poorest workers was not very different from progress among the poorest households. Moreover, it would appear that the decreases in the working poverty gap halted, or even reversed, after 2006. On the whole, then, a conventional poverty analysis has shown that working poverty decreased over the early part of the 2000s but that not much progress has been made since about 2006 in either the incidence or depth of working poverty.

Table 3. The working poverty gap (P₁) in South Africa, 2004–2012.

Poverty Gap (P1)
	2004	2006	2008	2010	2012	Relative change 2004–2012
	Z = 323
All	35.11 (0. 55)	30.59 (0.54)	27.82 (0.45)	27.25 (0.44)	24.79 (0.41)	−29.39†
Workers	14.37 (0. 39)	10.52 (0.32)	10.43 (0.33)	10.34 (0.32)	10.04 (0.33)	−30.13†
	Z = 219
All	25.73 (0.48)	21.72 (0.46)	19.03 (0.36)	18.74 (0.35)	16.69 (0.33)	−35.13†
Workers	9.53 (0.32)	6.28 (0.24)	6.22 (0.26)	6.33 (0.26)	6.47 (0.29)	−32.11†

†Denotes a significant change in the poverty gap between 1997 and 2012 at the 95% level of confidence.

†† Denotes a significant change in the poverty gap between 1997 and 2012 at the 90% level of confidence.

Notes: The data are weighted (revised weights).

Standard errors in brackets.

All poverty lines in 2000 prices.

Household well-being is estimated as average per capita total household monthly income.

Source: Own calculations from the GHSs (2004–2012)

As identified in the previous section, there are a number of ways to measure working poverty in the literature. While estimating the working poverty rate (and the working poverty gap) is one of the more common methods, Table 4 considers the ‘in-working poverty’ definition which is preferred in some of the literature (see Peña-Casas and Lata 2004; Brady et al. 2010) from OECD countries. The table shows the percentage of poor South Africans (of all ages) that live with an employed member. At the two lower bound poverty thresholds, there were no significant changes over the crucial 2004–2012 period. At the R323 poverty line, roughly half of all poor South Africans lived with an employed household member over the period. At the upper-bound poverty line, the majority of poor South Africans live with an employed person (e.g. 58% in 2012) and this percentage actually increased significantly over the period (by about three percent). In other words, poor South Africans became increasingly likely to live with a worker between 2004 and 2012 (albeit at the upper-bound poverty line only).

Table 4. Percentage of the poor who live with an employed household member, 2004–2012.

	2004	2006	2008	2010	2012	Relative change 2004–2012
Z = 219	44.43 (0.29)	40.00 (0.31)	41.99 (0.29)	42.64 (0.30)	43.85 (0.32)	−1.31
Z = 323	49.81 (0.26)	46.69 (0.28)	49.08 (0.25)	49.28 (0.26)	49.42 (0.27)	−0.78
Z = 593	56.56 (0.24)	55.26 (0.25)	57.93 (0.22)	58.49 (0.22)	58.29 (0.23)	3.06†

†Denotes a significant change in the poverty headcount between 1997 and 2012 at the 95% level of confidence.

†† Denotes a significant change in the poverty headcount between 1997 and 2012 at the 90% level of confidence.

Source: Own calculations from the GHSs

Notes: The data are weighted.

Standard errors in brackets.

4.2. The working poor and their households, 2004–2012

This section now turns to an analysis of some of the possible reasons why the decrease in working poverty was not much greater than the decline in poverty for the population as a whole over the 2000s. Of particular interest is the finding that the level of working poverty has been relatively consistent since 2006, even though overall poverty rates have declined. Table 5 considers the (self-reported) main sources of household income among the working poor. While labour market income is the single largest source of household income for the working poor, there have been only very small changes between 2004 and 2012.¹¹ At the same time, the percentage of the working poor who reported that social grant income was the main household income source increased significantly (from 13% in 2004 to 16% in 2012). This increase in grant income, moreover, coincided with marginal (but significant) increases in the average monthly earnings of the working poor (particularly from 2010 onwards).

Table 5. Main source of household income among the working poor (Z = 323), 2004–2012.

	2004	2006	2008	2010	2012	Relative change 2004–2012
Salaries/wages	70.48 (0.69)	72.15 (0.77)	72.19 (0.66)	67.08 (0.72)	65.18 (0.74)	−7.52†
Remittances	4.64 (0.31)	4.47 (0.33)	3.82 (0.26)	3.38 (0.26)	2.79 (0.50)	−39.87†
Pensions and grants	13.39 (0.51)	14.44 (0.54)	18.85 (0.56)	18.14 (0.56)	16.26 (0.52)	21.43†
Sales of farm products and services	1.76 (0.20)	3.13 (0.43)	1.46 (0.17)	0.10 (0.04)	0.06 (0.03)	−96.59†
Other non-farm income	9.44 (0.43)	4.71 (0.36)	2.99 (0.25)	0.65 (0.12)	0.35 (0.08)	–
No income	0.30 (0.11)	0.61 (0.16)	0.68 (0.17)	0.05 (0.02)	0.02 (0.02)	−93.33†
Real monthly earnings (real 2000 prices) of the working poor	525.39 (6.89)	555.94 (7.99)	527.13 (6.88)	557.78 (7.09)	592.80 (8.07)	12.83†

†Denotes a significant change between 2004 and 2012 at the 95% level of confidence.

†† Denotes a significant change between 2004 and 2012 at the 90% level of confidence.

Notes: The data are weighted (revised weights).

Standard errors in brackets.

NB: The category of ‘other non-farm income’ was captured differently after 2008 and is not directly comparable with the earlier GHSs.

Source: Own calculations from the GHSs.

Given the low values of the social grants but, at the same time, the relative importance of these grants to the working poor, Table 6 looks more closely at grant receipts in the households of the working poor. The main conclusion is that, between 2004 and 2012, the overall coverage of social grants among the working poor increased substantially – the average number of grants received almost doubled and the percentage of the working poor living in a grant-receiving household increased from 51% to 63%.

Table 6. Social grant receipt among the working poor (Z = 323), 1997–2012.

	1997	1999	2004	2006	2008	2010	2012
Total number of grants received (by the household)	0.23 (0.01)	0.18 (0.01)	0.90 (0.02)	1.20 (0.03)	1.49 (0.03)	1.65 (0.03)	1.68 (0.03)
Percentage with at least one grant	18.79 (0.44)	15.42 (0.42)	51.16 (0.77)	54.80 (0.89)	61.72 (0.78)	65.55 (0.78)	63.12 (0.85)

Notes: The data are weighted.

Standard errors in brackets.

Source: Own calculations from the 1997 OHS and the 2006 GHS.

The results reported in the previous two tables demonstrate the important role of transfer income in the households of the working poor. The key question then is the extent to which grant income contributed to the decrease in working poverty and, by extension, whether working poverty would have decreased without it. The last two tables present an analysis of the respective contributions of social grant income and labour market income to the reduction of the incidence and depth (P₀ and P₁) of working poverty. Our analysis is based on the decomposition method described in section 3.3.

Between 2004 and 2012, the overall composition of total household income among the employed did not change appreciably. Not surprisingly, earned income was the single largest source and comprised, at the mean, 94% of total per capita household income in the households of workers in 2004, whereas grant income comprised only about one or two percent. The decomposition (Table 7) suggests, however, that grant income became slightly more important to the reduction of the poverty headcount among workers between 2004 and 2012, increasing from 2% to 3% between 2004 and 2012 in relative terms (compared to what it would have been).

Table 7. Relative contributions (%) to reducing the working poverty headcount (P₀) by income source, 1997–2012.

	1997	2004	2010	2012
Earned income	95.29 (0.15)	91.73 (0.29)	90.26 (0.31)	88.96 (0.28)
Grant income	1.62 (0.07)	1.93 (0.09)	2.99 (0.10)	2.90 (0.09)
Other income	2.46 (0.14)	6.34 (0.28)	6.75 (0.31)	8.14 (0.27)

Notes: The data are weighted.

Income sources are expressed in real monthly per capita terms (Z = R323).

Standard errors in brackets.

Source: Own calculations from the GHSs using the DASP module developed by Araar and Duclos (2007).

In the population as a whole, the relative contribution of social grants to poverty reduction increased from 7% to 9% (not shown in the table). So, in relative terms, the increase in importance of the contribution of grant income to poverty reduction was greater among the employed than for the population as a whole. This is likely to be the case, in part, because households with at least one worker (even poorly paid) are likely to be closer to the poverty line such that the addition of even a modest grant payment could push those households over the poverty line.

The same decomposition for the working poverty gap (Table 8), suggests even more strongly that, without social grant income, the gains made by the working poor over the early 2000s would have been more limited. The relative contribution of grant income towards the reduction of the working poverty gap increased from 7% to 8% between 2004 and 2010. Among the population as a whole, the relative contribution of grant income to reducing the depth of poverty was far higher but the increase was far more modest between 2004 and 2012 (not shown in the table).

Table 8. Relative contributions to reducing the working poverty gap (P₁) by income source, 1997–2012.

	1997	2004	2010	2012
Earned income	92.62 (0.15)	87.51 (0.26)	84.97 (0.28)	83.89 (0.26)
Grant income	4.54 (0.08)	6.62 (0.12)	8.32 (0.12)	8.29 (0.11)
Other income	2.85 (0.12)	5.87 (0.24)	6.71 (0.27)	7.81 (0.24)

Notes: The data are weighted.

Income sources are expressed in real monthly per capita terms (Z = R323).

Standard errors in brackets.

Source: Own calculations from the GHSs using the DASP module developed by Araar and Duclos (2007).

5. Discussion and conclusions

Perhaps the main conclusion that can be drawn from our analysis of the post-apartheid period is that, despite some clear progress in the early 2000s, in 2012 just over a fifth of all employed South Africans spread their earnings (and other income sources) so thinly across their households that there was not enough income to meet the minimum of the most basic needs of all household members. If one uses a higher poverty line that is closer to a minimum living level, more than a third of all employed South Africans live in households that earn enough income to meet only their minimum basic needs. Therefore, and in contrast with other developing or middle-income countries (e.g. Brazil), the post-apartheid labour market continues to be characterised by low earnings and high rates of unemployment.

In addition, the data from the OHSs and the GHSs show that changes in the labour market over the post-apartheid period have not added appreciably to the demonstrable income effects achieved through the expansion of the government's social grant system. Rather, it seems that progress in poverty reduction may have stalled somewhat among low paid workers over the past decade. Therefore, while access to employment is both a logical and a crucial step to avoiding poverty, it is clearly not enough in South Africa. Roughly a fifth of South African workers are poor and almost half of all poor South Africans live with at least one employed person.

The recent findings that the growth in the middle of both the income distribution (Visagie 2015b) and the earnings distribution (Wittenberg 2014a) was very modest over the same period, also provide some support to our findings on working poverty. The fact that improvements in working poverty became smaller as the poverty line was increased, suggests that changes were more modest among low earning workers closer to the middle of the income distribution. Again, this seems to support the conclusion that transfer income was an important contributor to poverty reduction, even among the employed, and that South Africans are often unable to ‘work themselves out of poverty’ (Lilenstein et al. 2018).

There are a number of policy implications associated with the findings presented in the paper. In terms of labour market policy specifically, studies of working poverty often contribute, at least tangentially, to debates on labour market flexibility (Thiede et al. 2015), including minimum wage levels (Ostermeier et al. 2015). Advocates of flexible labour markets tend to align with a preference for the quantity over the quality of jobs. While our study cannot contribute directly to estimates of an appropriate minimum wage, an analysis of the WPR has suggested that poverty reduction in South Africa, even among the employed, has increasingly been driven by transfer income and that progress in improving the returns to employment over the post-apartheid period in South Africa has been disappointing.

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

Michael Rogan http://orcid.org/0000-0003-1695-8179

Notes

1 Some of the reasons offered in the literature for the high and growing working poverty rates (particularly in the US) include: deindustrialisation, the rise of the service industry, the emergence of ‘low-wage traps’, declining labour standards, and low (and declining) union density.

2 For a comprehensive discussion of measurement issues in the working poverty literature, see Thiede et al (2015).

3 Ideally, we would have tracked working poverty trends up until 2016 but the national master sample used by Statistics South Africa was refreshed in 2013 in order to update population estimates from the 2011 Census. As a result, there is a clear break in the series after 2012 where poverty rates (and the WPR) increase dramatically over a short two year period. The datasets used in the analysis therefore include the 1997 and 1999 OHSs and the 2004, 2006, 2008, 2010 and 2012 GHSs.

4 An analysis (see Posel and Casale 2006) of earnings responses has suggested that assigning the mid-point is not necessarily any more biased than other methods of imputing income for those who respond in brackets (see also Vermaak 2010). Other work has also suggested that (von Fintel 2007: 310), ‘“Rudimentary” methods such as midpoint imputation should not be dismissed’ and that bracket responses in South African household surveys are reasonably stable.

5 These per capita monthly poverty lines include (in 2000 prices): the lower-bound poverty line (R219), Hoogeveen & Özler's R323 poverty threshold, and an upper-bound poverty line (R593).

6 See Statistics South Africa (2008) for details about how the official poverty lines were calculated.

7 Not surprisingly, the WPR is far higher among workers in the informal sector (46% in 2010) than for those in the formal sector (17% in 2010). The interesting finding, however, is that the risk of poverty decreased significantly between 2010 and 2012 in the informal sector (i.e. from 46% to 41%) but there was no change for formal sector workers.

8 The full results of the sensitivity tests are available from the authors upon request.

9 As identified earlier, the OHSs and the GHSs do not regularly capture information on the occupations and industries of workers. Where they do, caution should be exercised in making comparisons across time since the occupation and industry categories are not consistent and are, therefore, not directly comparable. However, as context, we provide descriptive analyses of the employment characteristics of the working poor based on the 1997 OHS and the 2006 GHS (both surveys in which at least some information on employment activities was provided). In 1997, about 20% of the working poor (at the R323 poverty line) worked in private households, 17% worked in agriculture, and 11% worked in worked in both manufacturing and trade. In terms of broad occupation groups, the working poor were concentrated largely in elementary occupations (25%), domestic work (20%) and artisanal occupations (14%). In 2006, 27% worked in wholesale and retail trade, 20% in private households, 13% in agriculture, 11% in community and personal services, and 10% in both manufacturing and construction. In terms of occupational groups, 36% worked in elementary occupations and 16% in domestic work. About 13% of the working poor in 2006 were service workers.

10 In line with other conventional poverty analyses, the (working) poverty gap is the average distance from the poverty line among the (working) poor, expressed as a proportion of the poverty line (Foster et al. 1984).

11 Comparisons with the GHSs after 2008 should be made with caution since the response categories for the self-reported question on household income sources changed slightly after 2008.