Winners and losers: South African labour-market dynamics between 2008 and 2010

Abstract

There is little empirical work in South Africa using panel data to describe employment and earnings dynamics. This paper contributes by describing labour-market transitions in South Africa between 2008 and 2010 for individuals aged 20 to 55 across two waves of nationally representative panel data. We find that women had more mobility than men into and out of the labour market and employment. However, women were less mobile between kinds of employment and across occupations. Casual employment was very unstable, with very few individuals remaining in this state. In contrast, there was little mobility out of regular employment into other kinds of employment. Average real earnings changes were positive for movements into regular employment and negative for movements into self-employment or casual employment. Strong movement out of semi-skilled occupations was striking, especially for males. In multivariate models, being the mother of young children was important in leaving regular employment.

Keywords:

• labour-market dynamics
• panel data
• National Income Dynamics Study

1. Introduction

The use of rich labour-market data from South Africa's cross-sectional surveys has added to our understanding of the evolving nature of the labour market over the post-apartheid period. Good examples are Van der Westhuizen (2007), Burger & von Flintel (2009), Banerjee et al. (2008) and Verick (2012). However, there are inherent difficulties associated with using a series of cross-sections to explore labour-market dynamics. Even if employment rates or unemployment rates are stable over time, there is no way of knowing whether this is because the labour market has operated in a stable fashion between the surveys or whether there have been changes in earnings and employment for certain individuals and groups but these changes have netted out to stable aggregate snapshots. Generally, repeated cross-sections cannot deal with the movement of people between labour-market segments or between jobs within sectors or with related real earnings changes over time. This is a particular concern if policy-makers are interested in knowing which specific individuals or groups are experiencing movement in the labour market and, in particular, who are the winners and losers from the current operation of the labour market.

To study these important dynamic issues more thoroughly we need longitudinal data from panel studies. The importance of such work has been widely recognised in South Africa since the late 1990 s. Early work (Cichello et al., 2005) made use of panel data from a regional survey, the KwaZulu-Natal Income Dynamics Study. This study showed that there was enough churning in the formal and informal sector labour markets to flag the importance of the description and understanding of these dynamics. Ideally, however, a national perspective on these dynamics is preferable. Statistics South Africa's Labour Force Survey (LFS) was initiated in 2000 as a biannual survey of the labour market. It has been used as the basis for much of the cross-sectional work cited above. However, in order for it to present accurate employment and unemployment trends, it is designed as a rotating panel of dwelling units (Stats SA, 2001). Therefore, it has some potential to illuminate labour-market dynamics. Early work in this regard is presented by Ranchhod & Dinkelman (2008), who derived an algorithm to track individuals for the waves in which they remain in the LFS and then used this to describe key transitions in the South African labour market from 2001 to 2003. Banerjee et al. (2008) used the same waves of LFS data to spell out some broad labour-market dynamics as part of their report on the labour market. In 2008, the LFS became the Quarterly Labour Force Survey (QLFS), reflecting the fact that its statistics would now be based on four rounds of interviews each year (Stats SA, 2008). Verick (2012) offers a preliminary exploration of the QLFS panel as part of a broader review of the South African labour market over the period of the financial crisis.

While the rotating design of the QLFS is appropriate for its purpose, it does not generate data that track specific individuals into, out of and through the labour market over the medium to long term. Thus, it has limitations in terms of the picture of labour-market dynamics that it can offer. The National Income Dynamics Study (NIDS) is a national panel study that has been tracking over 28 000 individuals since 2008 (SALDRU, 2012a, 2012b). It is therefore well-suited to describing and explaining South Africa's medium to longer-term labour-market dynamics. This paper uses the first two waves of NIDS data to investigate the dynamics of the South African labour market between 2008 and 2010. This period was marked by recession and significant job shedding. According to Stats SA (2010), there were 900 000 job losses between the fourth quarter of 2008 and the fourth quarter of 2010. Men were affected more strongly than women, with the labour absorption ratio (the percentage of working-age people that were employed) falling by 5.4 percentage points for men compared with a fall of 3.2 percentage points for women (Stats SA, 2010). The official unemployment rate over the same period rose less than might have been expected with a rise from 18.8% to 21.8% for men and from 25.4% to 26.6% for women. This can be explained by people withdrawing from the labour market either because of discouragement or a decision to stay in school longer while waiting out the crisis.

The longitudinal nature of the NIDS data allows us to study which groups (by gender, age and job type) were worst affected by the crisis. In this paper we describe labour-market transitions in South Africa between 2008 and 2010 by examining changes in labour-market behaviour of individuals aged 20 to 55 in 2008 across two waves of nationally representative panel data. The data show extensive mobility across employment status and significant mobility across the type of employment. We present evidence that women exhibit much greater mobility into and out of the workforce and employment, while men exhibit more mobility across types of employment. Among those employed in regular employment in both periods, there is considerable mobility across industry and occupational groupings. Flows out of manufacturing and into services and out of semi-skilled occupations into elementary occupations are particularly noteworthy for men. The benefit individuals derive from working in regular employment in comparison with self-employment or casual employment is demonstrated using some basic summary statistics regarding real earnings changes over time. The same individuals typically earned much more after moving from self-employment or casual employment to regular employment and earned much less after moving from regular employment to self-employment or casual employment. Our multivariate results show for females that movement out of regular employment is strongly correlated with demographic factors such as the number of young children. It is also correlated with being a recent hire. For males, demographic factors are important but being a new hire has no impact on moving out of regular employment. In addition, males in lower skilled occupations were more likely to move out of regular employment than skilled males. In contrast, one's location is an important predictor of whether an individual will move into regular employment.

2. Data and methods

Our analysis in this paper takes advantage of Wave 2 of NIDS. The data are nationally representative and interview the same individuals that were interviewed in Wave 1.

2.1 Sample size and non-response

Wave 1 of the NIDS data (collected in 2008) had 7301 unique households, with a total of 28 247 household residents. In the adult dataset, there were 15 633 respondents (aged 15 or older). Wave 2 of NIDS (in 2010/11) achieved successful interviews in 6809 unique households, with a total of 28 641 household residents successfully completing interviews. However, some of these individuals were new to the survey in Wave 2 and others who were in Wave 1 were not interviewed in Wave 2. The reasons vary, as will be shown below. In the 2010 adult dataset, there were 17 682 respondents who participated, 11 388 of whom had successful interviews in the adult questionnaire in 2008. An additional 587 had proxy responses in 2008.

Of the 15 633 respondents aged 15 or older in 2008, 77% had successful interviews again in 2010; 6.5% refused or were unavailable for the household-level interview; 2.2% had successful household interviews but refused or were unavailable for the individual interview in 2010; 10.3% came from households that could not be relocated or were not tracked; and 4.5% were deceased or had moved outside South Africa.

Baigrie & Eyal (2013) show that attritors are more likely to be employed and have higher labour-market income. They find ‘moderate’ evidence of attrition bias in estimated coefficients based on the balanced (non-attriting) sample, although one should note that the actual attrition indicator is not individually significant. We would thus caution that our findings are limited to the balanced panel and are not necessarily generalisable to the whole population.

2.2 Data concerns

Users of NIDS Wave 2 should be aware of a few data irregularities that may have a significant impact on labour-market analysis using these data. We highlight three examples of problematic data that we have identified. In these cases, responses were within the valid range and thus were not identified during standard edit checks during the survey collection phase. However, the distribution of results suggests that certain members of the fieldwork team had a poor understanding of the intent of the relevant questions.

The first concern is the large reduction in the number of unemployed, particularly searching unemployed in 2010. This is surprising and we cannot identify a reason for such a dramatic change. Importantly, while there is not an exact comparison available in published Stats SA documents, their statistics do not show a large decline in the percentage of searching unemployed during this time period. Thus, one should use caution in interpreting the sharp decline in unemployment rates between 2008 and 2010 found among the NIDS panel respondents.

The second concern is the number of individuals working in subsistence agriculture: there was a significant decrease in the number of individuals recorded as being employed in subsistence agriculture. Seasonality, while potentially a factor in explaining some of the decreased agricultural employment, cannot fully explain the dramatic decline from 6.5% of employment in 2008 to 2% in 2010. It seems likely that there was some problem in the way this question was asked in the field. Given this, when analysing changes in the types of employment and earnings, we exclude those employed in the subsistence agriculture sector in either year. We focus exclusively on those employed in regular employment, self-employment, or casual employment. Assuming that those working in subsistence agriculture in 2010 were placed in the non-employed categories rather than in regular employment, self-employment or casual employment, the reported changes are not driven by any changes in our ability to identify those working in subsistence agriculture. Given that employment in casual employment and self-employment were declining and the non-employment category was growing more than expected, the authors believe this is a reasonable assumption.⁴ We would recommend similar exclusions if analysing changes in industry or occupation. In this paper, we limit our analysis of industry and employment to those in regular employment and are thus unaffected.

The third data concern is the ‘hours worked’ variable. There is a dramatic increase in the number of respondents reporting that they work less than 10 hours per week, from 6.4% in 2008 to 16.4% in 2010. Additional examination of the data suggests that a number of field staff misinterpreted the question and were asking for the hours worked per day rather than per week. Thus, this variable is not used in our analysis. We recommend that others use considerable caution with this variable.

3. Aggregate outcomes, trends and transitions

Following Ranchhod (2010) we categorise each adult into one of four mutually exclusive categories. ‘Employed’ is composed of people who are engaged in some type of productive activity, generally for the purpose of earning money. ‘Searching unemployed’ are people who are not employed and have actively searched for employment in the past four weeks. ‘Discouraged unemployed’ are unemployed people who would have liked to have worked in the past four weeks, but have not actively searched for employment in that same time period. ‘Not economically active’ (NEA) are people who are not employed and do not want to find employment (e.g. scholars/students, home-makers, the disabled and retired persons). Unless explicitly stated, analysis in this paper is limited to those individuals who were 20 to 55 years old in 2008 and gave valid responses in both interviews. The age restriction is intended to keep our analysis focused on the progression of individuals who are of working age throughout the entire two-year period. In other words, we do not want large in-flows from being NEA to employment among school-leavers or from employment to NEA among retirees to overwhelm our story of transitions across employment status. Similarly, we do not want changes across employment/occupation types for individuals just entering the workforce or preparing for retirement to dominate our analysis. These flows are also worthy of study, but should be examined separately.

We refer to this group who were 20 to 55 years old in 2008 and had successful interviews in both years as the ‘panel group’. Accordingly, we weight all analysis using the panel weights provided with the Wave 2 data. Table 1 shows the 2010 labour-market categories on the left-hand side and the changes in categories between 2008 and 2010 on the right-hand side. In 2010, 32% of the panel group (who were aged 20 to 55 in 2008) were NEA. This represents a 9.8 percentage point increase from 2008. The trend is consistent across race groups, with the exception of Indian/Asians who have a much smaller sample size. It is not driven exclusively by females, although their rate and change is considerably higher. It is also not driven by retirement as the declines in economic activity are observed across all age categories. The large increase in the percentage of individuals out of the labour force (i.e. the NEA category) is driven by a decline in the percentage employed and even greater declines in the percentages that are classified as unemployed under both narrow and broad definitions of unemployment. This combination leads to large declines in the unemployment rates. Of course, from the perspective of social well-being, one would prefer to see declines in unemployment being driven by increases in employment rather than these increases in the NEA.

Table 1: Employment status: levels and changes from the cross-sectional view

		Percentage in each category in 2010						Percentage change, 2008 to 2010
		i	ii	iii	iv	v	vi
			Unemployed			Unemployment rate
	Obs.	NEA	Disc.	Search	Employed	(Strict)	(Broad)	NEA	Disc.	Search	Emp.
Aggregate	7 889	32.0	4.9	11.8	51.3	18.7	24.6	9.8	–1.6	–6.5	–1.7
African	6 383	33.7	5.3	13.2	47.8	21.7	28.0	10.5	–1.5	–7.4	–1.7
Coloured	1 143	27.7	4.4	8.1	59.8	11.9	17.3	11.4	–3.6	–3.2	–4.7
Indian	96	24.4	6.7	8.4	60.4	12.3	20.1	–0.8	1.9	2.0	–3.1
White	267	21.2	0.4	2.8	75.6	3.6	4.1	4.4	–1.6	–2.2	–0.6
Male	2 573	22.3	3.9	10.4	63.3	14.1	18.5	7.1	–0.3	–4.6	–2.1
Female	4 616	39.4	5.7	13.0	41.9	23.7	30.8	12.3	–2.4	–7.7	–2.2
Age 20 to 25	1 824	35.7	6.3	18.4	39.6	31.8	38.5	3.3	–1.5	–6.3	4.5
Age 26 to 35	2 205	26.5	4.9	13.7	54.9	20.0	25.3	11.0	–1.9	–9.3	0.3
Age 36 to 45	2 072	28.4	3.9	8.6	59.1	12.6	17.5	12.2	–2.5	–4.6	–5.1
Age 46 to 55	1 788	41.4	4.4	5.0	49.2	9.3	16.1	12.5	0.1	–4.3	–8.3

Notes: Obs., observations; Disc., discouraged unemployed; Search, searching unemployed; Emp., employed.

These changes do not entirely comport with changes reported in Stats SA's LFSs. These surveys have a much larger sample than NIDS and are designed specifically to obtain accurate estimates of employment and unemployment rates and trends. Therefore, we would caution against readers drawing broad policy conclusions from these snapshots of the labour market. The contribution of NIDS lies in giving us a unique window into the dynamism of the labour market, and this is our focus in this paper.

Table 2 provides a longitudinal view of the employment status changes using a transition matrix. Here, we are clearly taking advantage of the panel data by linking an individual's employment status in 2008 with their employment status in 2010. Each row sums to 100%. Of those people who were NEA in 2008, 56.8% were NEA again when interviewed in 2010, while 6.1% were discouraged job-seekers, 15% were strictly unemployed, and 22% were employed. This matrix shows that the majority of those who were NEA in 2008 were NEA again in 2010.

Table 2: Employment status, longitudinal perspective

	Employment status in 2010
Employment status in 2008	NEA	Discouraged	Searching	Employed
NEA	56.8	6.1	15.0	22.0
Discouraged	43.1	10.8	18.1	28.0
Searching	39.7	6.5	21.6	32.3
Employed	18.5	3.2	6.7	71.6

Those who were discouraged job-seekers in 2008 had outcomes in 2010 that looked more like those of the searching unemployed than the NEA. This might be considered to be more circumstantial evidence in favour of Kingdon & Knight's (2004, 2006) assertions that the broad unemployment rate is the best measure in South Africa. Similar conclusions with more recent data have been found by Ranchhod & Dinkelman (2008) and Verick (2012). The transition matrix results are consistent with these messages, although a more detailed analysis would be required to flesh this out.⁵

Table 2 also demonstrates that, while NEA and employment categories might be considered relatively stable states, they are not overly stable. Almost 30% of those employed in 2008 were not employed in 2010 and over 40% of those that were NEA in 2008 were in the labour force in 2010.

Tables 3 and 4 show that the employment participation rate among the panel members is much lower for women than men (63.3% versus 41.9% in 2010) while women have a much greater proportion of individuals classified as unemployed (using the broad definition) and NEA. This does not change over time. We also see that women experience much greater mobility across employment status than men. One-half of women changed employment status as compared with 38% of men.

Table 3: Augmented transition matrix: male employment status, 2008 to 2010

Employment status in 2008		Employment status in 2010
		22.3	3.9	10.4	63.3
		NEA	Discouraged	Searching	Employed
15.2	NEA	50.0	5.9	14.9	29.2
4.3	Discouraged	30.3	11.9	13.6	44.1
15.0	Searching	35.9	3.2	22.3	38.7
65.4	Employed	12.4	3.3	6.7	77.5

Table 4: Augmented transition matrix: female employment status, 2008 to 2010

Employment status in 2008		Employment status in 2010
		39.4	5.7	13.0	41.9
		NEA	Discouraged	Searching	Employed
27.1	NEA	59.5	6.2	15.1	19.2
8.0	Discouraged	47.4	10.4	19.5	22.7
20.7	Searching	41.5	8.2	21.2	29.1
44.2	Employed	24.7	3.2	6.7	65.4

3.1 Changes in employment type

Table 5 provides a cross-sectional view of changes in the labour-market outcomes among those who were employed in regular employment, self-employment or casual employment in both 2008 and 2010. Among this panel of people that were employed at both time periods, there was a shift out of self-employment and casual employment and into regular employment. The shift out of self-employment was strongest for females and for workers aged 46 to 55. Changes for those aged 20 to 25 are distinct, as individuals in this age group were leaving casual employment in much greater percentages. This is not unexpected if the young are finding their way into a more appropriate position within the labour market.

Table 5: Type of employment: levels and changes from the cross-sectional view

		Percentage in each employment category in 2010			Change in percentage between 2008 and 2010
		i	ii	iii	i	ii	iii
	Number ofobservations	Regular	Self	Casual	Regular	Self	Casual
Aggregate	3 452	81.1	11.7	7.2	6.4	–3.0	–3.4
African	2 574	80.3	11.1	8.6	7.7	–4.7	–3.1
Coloured	635	86.9	6.9	6.2	–0.6	3.9	–3.3
Indian	50	73.1	24.7	2.2	8.2	5.7	–13.9
White	193	83.6	16.1	0.4	3.2	–0.7	–2.5
Male	1 461	80.4	11.2	8.4	3.7	–0.6	–3.0
Female	1 653	80.5	13.4	6.1	8.1	–4.6	–3.5
Age 20 to 25	617	80.1	10.2	9.8	8.3	2.4	–10.7
Age 26 to 35	1 045	83.7	9.4	6.9	6.4	–3.5	–2.9
Age 36 to 45	1 072	80.9	13.4	5.7	4.5	–1.7	–2.8
Age 46 to 55	718	77.7	15.0	7.5	7.4	–6.9	–0.4

Table 6 gives us a dynamic view of the transitions experienced by those that were initially employed at baseline. It is clear that casual employment is a transitory state, with just 6.5% of those who were casually employed in 2008 being casually employed in 2010. Self-employment is also much less stable than regular employment. Few people leave regular employment for self-employment or casual employment, but 14.8% of the self-employed and one-third of the casually employed in 2008 were in regular employment in 2010. To the extent that the choice of their original type of employment was still available, this strongly suggests that individuals choose regular employment over self-employment and casual employment.

Table 6: Type of employment, longitudinal perspective

Type of employment in 2008		Type of employment in 2010
		40.4	5.8	3.6	50.2
		Regularemployment	Self-employment	Casualemployment	Other/non- employment
36.5	Regular employment	76.2	3.0	2.9	17.9
7.2	Self-employment	14.8	34.6	4.1	46.5
5.2	Casual employment	32.1	9.1	6.5	52.3
51.2	Other/non-employment	17.5	3.7	3.6	75.2

In contrast to employment status mobility, women are less mobile than men when it comes to changing their type of employment. When women move out of regular employment it is much more likely to be into non-employment than it is for men. As shown in Tables 7 and 8, while the self-employment status is slightly more stable for women than men, just 1.9% of women in casual employment in 2008 were still casually employed in 2010. For both groups, regular employment is a relatively stable position and regular employment comprises a little more than 80% of the employment for the group that are employed at both waves.

Table 7: Augmented transition matrix: type of employment among males

Type of employment in 2008		Type of employment in 2010
		49.5	6.9	5.2	38.5
		Regular employment	Self-employment	Casual employment	Other/non-employment
47.5	Regular employment	77.6	3.8	3.9	14.7
7.3	Self-employment	19.0	32.4	6.3	42.2
7.1	Casual employment	29.9	10.2	9.9	50.0
38.1	Other/non-employment	22.6	4.5	6.0	66.9

Notes: Sample restricted to adults aged 20 to 55 in 2008. Other/non-employment includes those out of the labor force, those broadly defined as unemployed and those individuals who are employed in subsistence agriculture or other employment. All proportions have been weighted using the post-stratification weights that account for attrition. Row percentages sum to 100%.

Table 8: Augmented transition matrix: type of employment among females

Type of employment in 2008		Type of employment in 2010
		32.6	5.4	2.5	59.5
		Regular employment	Self-employment	CasualEmployment	Other/non-employment
28.6	Regular employment	74.4	2.0	1.9	21.7
7.1	Self-Employment	11.9	36.1	2.6	49.4
3.8	Casual employment	35.0	7.6	1.9	55.5
60.5	Other/non-employment	15.3	3.4	2.6	78.7

Notes: Sample restricted to adults aged 20 to 55 in 2008. Other/non-employment includes those out of the labor force, those broadly defined as unemployed and those individuals who are employed in subsistence agriculture or other employment. All proportions have been weighted using the post-stratification weights that account for attrition. Row percentages sum to 100%.

4. Industry and occupation transitions

We examine employment transitions for those who were employed in regular waged employment in both periods. We begin by classifying regular workers into industry categories. The primary sector consists of agriculture (including hunting, forestry and fishing) and mining/quarrying. The secondary sector consists of manufacturing, utilities and construction. The tertiary sector consists of wholesale and retail trade; transport, storage and communication; financial intermediation, insurance, real-estate and business services; and community, social and personal services. The final category is private households, exterritorial organisations, and other activities not adequately defined.

Approximately one-quarter (22.4%) of this group changed industry category between 2008 and 2010 (see Table 9). The exodus from secondary employment is particularly prominent, with less than one-half of those who were in the sector in 2008 found there again in 2010. In contrast, just 10% of those in the tertiary sector in 2008 were not in the tertiary sector in 2010.

Table 9: Augmented industry transition matrix

Industry in 2008		Industry in 2010
		12.6	15.5	64.9	6.9
		Primary	Secondary	Tertiary	Private household/other
12.7	Primary	68.2	12.3	17.5	2.0
21.1	Secondary	10.3	48.9	38.0	2.9
59.4	Tertiary	2.1	5.7	90.2	2.0
6.8	Private household/other	5.6	2.5	18.0	73.9

Tables 10 and 11 show that there is greater mobility across industry for males than females. The exodus from the secondary sector for those who were in the secondary sector in 2008 is significant for both males and females. However, this is particularly stark for males for two reasons. First, just 46.3% are there again in 2010 (as compared with 55.8% for women). Secondly, this group represented 28.6% of 2008 employment for males in this sample as compared with 12.4% of the 2008 employment for females in the sample. Similarly, while women in the primary sector had a high propensity to move to other sectors in 2010, there were relatively few women engaged in the primary sector in 2008.

Table 10: Augmented industry transition matrix for males

Industry in 2008		Industry in 2010
		18.9	20.8	58.9	1.3
		Primary	Secondary	Tertiary	Private household/other
18.8	Primary	73.3	11.2	14.1	1.4
28.6	Secondary	11.3	46.3	40.7	1.6
51.5	Tertiary	3.4	9.6	86.6	0.4
1.1	Private household/other	17.7	19.9	25.7	36.8

Table 11: Augmented industry transition matrix for females

Industry in 2008		Industry in 2010
		5.0	9.1	72.1	13.7
		Primary	Secondary	Tertiary	Private household/other
5.6	Primary	48.0	16.8	31.1	4.2
12.4	Secondary	7.5	55.8	30.4	6.3
68.6	Tertiary	1.0	2.3	93.3	3.4
13.4	Private household/other	4.5	0.8	17.2	77.5

We can also investigate changes in occupation for those who were employed in regular employment, self-employment or casual employment in both periods. The managerial/professional category includes managers; professionals; and technicians and associate professionals. The semi-skilled group includes clerical support workers; service and sales workers; skilled agricultural, forestry and fishery workers; craft and related trades workers; and plant and machine operators, and assemblers. Lastly, there is a group working in elementary occupations.

Two results are apparent from Table 12. First, there is movement out of the semi-skilled occupations. This is apparent in both the transition and the cross-sectional results. Second, there is less movement across elementary occupations and managerial/professional positions.

Table 12: Augmented occupation transition matrix

Occupation in 2008		Occupation in 2010
		32.3	44.4	23.2
		Managerial / professional	Semi-skilled	Elementary occupations
27.8	Managerial / professional	77.0	18.3	4.7
54.4	Semi-skilled	16.7	65.5	17.8
17.8	Elementary occupations	6.3	22.5	71.2

Notes: Sample restricted to adults aged 20 to 55 in 2008 who were regular wage workers in both periods. All proportions have been weighted using the post-stratification weights that account for attrition. Row percentages sum to 100%.

Tables 13 and 14 show the augmented occupation transition matrices separately by gender. The divide between elementary occupations and managerial/professional positions is most stark among females. The overall mobility is similar but slightly greater for males (31.6%) than females (28.8%) despite the fact that females exhibit greater mobility out of semi-skilled positions, the most common occupation type for both groups.

Table 13: Augmented occupational transition matrix for males

Occupation in 2008		Occupation in 2010
		25.4	52.3	22.3
		Managerial/ professional	Semi-skilled	Elementary occupations
20.5	Managerial / professional	73.7	17.3	9.0
66.7	Semi-skilled	12.2	67.9	19.9
12.8	Elementary occupations	9.9	27.6	62.4

Notes: Sample restricted to adults aged 20 to 55 in 2008 who were regular wage workers in both periods. All proportions have been weighted using the post-stratification weights that account for attrition. Row percentages sum to 100%.

Table 14: Augmented occupation transition matrix for females

Occupation in 2008		Occupation in 2010
		40.7	35.1	24.2
		Managerial / professional	Semi-skilled	Elementary occupations
36.5	Managerial / professional	79.2	18.9	1.9
39.7	Semi-skilled	25.7	60.7	13.5
23.8	Elementary occupations	4.1	19.3	76.5

Notes: Sample restricted to adults aged 20 to 55 in 2008 who were regular wage workers in both periods. All proportions have been weighted using the post-stratification weights that account for attrition. Row percentages sum to 100%.

Finally, the cross-sectional decline in semi-skilled employment is clearly driven by males. A look at more detailed occupation classifications available in NIDS (not shown) shows that the decline in semi-skilled occupations was driven by declines in (predominantly female) clerical support workers; (predominantly male) skilled agricultural, forestry and fishery workers; and (predominantly male) craft and related trades workers.

5. Earnings changes

Table 15 examines earnings changes across employment-type transition experiences for those who were employed in regular employment, self-employment, and casual employment in both periods. The mean change in real monthly earnings, its standard error, the median change in real monthly earnings and the percentage of positive changes in real monthly earnings are presented for each cell in the transition matrix.⁶

Table 15: Changes in monthly earnings by type of employment transition

		Employment type in 2010
Employment type in 2008		Regular employment	Self-employment	Casual employment
Regular employment	Mean	135	–3 886	–640
	Mean SE	(185)	(1 205)	(219)
	Median	383	–1 131	–139
	Percent positive	63%	29%	35%
Self-employment	Mean	3 011	–2 286	617
	Mean SE	(884)	(862)	(790)
	Median	1 061	75	195
	Percent positive	80%	53%	56%
Casual employment	Mean	909	158	–121
	Mean SE	(120)	(184)	(155)
	Median	808	264	–7
	Percent positive	84%	67%	50%

Results from Table 15 demonstrate the benefits of regular employment: 71% and 65% of those moving from regular employment to self-employment or casual employment, respectively, experienced losses in earnings. Average losses were sizeable. An even greater percentage of those moving from self-employment (80%) or casual employment (84%) into regular employment experienced earnings gains, with large mean and median gains in earnings. Those remaining in self-employment and casual employment appeared to have rather equal earnings gains and losses, although those remaining in self-employment experienced losses on average. Two-thirds of those moving from casual employment to self-employment experienced earnings gains, although the average gain was not statistically different from zero. Surprisingly, the median gain for those moving from self-employment to casual employment was also positive, although the average was also not statistically different from zero.

Table 16 examines earnings changes across industry transition experiences for those who were in regular employment in both periods. Most of those moving out of primary employment experienced earnings gains, but, on average in this sample, they experienced earnings losses. The median gain for those moving from secondary to tertiary sectors was R496 per month.

Table 16: Changes in monthly earnings by industry transition

		Industry type in 2010
Industry type in 2008		Primary	Secondary	Tertiary	Private household/other
Primary	Mean	225	–5 046	–1 077	15
	Mean SE	(183)	(2 999)	(974)	(181)
	Median	180	384	547	85
	Percent positive	62%	69%	73%	57%
Secondary	Mean	–79	828	1 354	70
	Mean SE	(378)	(506)	(494)	(141)
	Median	248	274	496	227
	Percent positive	61%	64%	66%	67%
Tertiary	Mean	1 168	507	21	–51
	Mean SE	(517)	(630)	(359)	(63)
	Median	1 289	985	562	–74
	Percent positive	61%	66%	65%	40%
Private household/other	Mean	–13	205	177	257
	Mean SE	(164)	(394)	(268)	(77)
	Median	122	–269	174	103
	Percent positive	52%	42%	72%	64%

Table 17 examines earnings changes across occupation transition experiences for those who were in regular employment in both periods. The benefit of moving to managerial and professional occupations is readily apparent, with three-quarters of those moving into these occupations experiencing gains and large, statistically significant average gains. In contrast, those moving out seem just to hold steady with a small, positive median gain in earnings but an average loss that was not statistically different from zero. These earnings changes are well below the norm for this select sample of individuals who were in regular employment in both periods (see Table 15).

Table 17: Changes in monthly earnings by occupation transition

		Occupation in 2010
Occupation in 2008		Managerial / professional	Semi-skilled	Elementary occupations
Managerial / professional	Mean	–858	–509	–334
	Mean SE	(622)	(815)	(560)
	Median	605	126	100
	Percent positive	62%	50%	52%
Semi-skilled	Mean	1 983	611	–107
	Mean SE	(553)	(266)	(182)
	Median	1 125	405	139
	Percent positive	75%	63%	61%
Elementary occupations	Mean	687	352	506
	Mean SE	(278)	(117)	(613)
	Median	380	488	132
	Percent positive	78%	68%	63%

6. Employment winners and losers

In this section of the paper, we first look at the correlates associated with moving from regular employment in 2008 to either non-employment or subsistence agriculture in 2010. We want to analyse these changes bearing in mind the general transition trends described previously. Twenty-four per cent of those in regular employment in 2008 were not in a similar job in 2010, with 18% of those in regular employment now in either non-employment or subsistence agriculture. We also observed that 18% of those in non-employment or subsistence agriculture ended up in regular employment. Thus, significant churning of workers across employment categories and status is observed in just a two-year span.

There are many reasons for such churning of individuals across employment categories and status. Four broad explanations come to mind immediately: structural change, resulting in a change in demand in particular industries, occupations, or skill levels; individual churning driven by choice, where individuals voluntarily leave regular employment for a short spell either to care for family needs or for other reasons; individual churning driven by firm death, where particular firms go out of business leaving employees to try to get another job; and worker productivity mismatches, where the churning of jobs is driven by firms and/or workers learning the value of their productivity and responding to it. We review the correlates of transitioning from regular employment to non-employment or subsistence agriculture with these ideas in mind.

Table 18 presents the results of this multivariate analysis. In this analysis, we restrict our analysis to African individuals. The first column presents the marginal effects from a probit model with demographic variables, location variables, broad industry and occupation variables, and other initial employment characteristics. We then run a similar analysis separately by gender. As the results differ considerably by gender, we focus on these gender-specific results.

Table 18: Determinants of transitions across regular employment and non-employment/subsistence agriculture among Africans

	Probit (marginal effects)
	Regular employment to non-employment/subsistence agriculture			Subsistence agriculture/non-employment to regular employment
Demographics (2008 values)	All	Male	Female	All	Male	Female
Male	–0.02			0.09***
Age	–0.02*	–0.04**	–0.01	0.02***	0.04***	0.02**
Age squared	0.00	0.00*	0.00	–0.00***	–0.00***	–0.00**
Married	0.03	0.07*	0.00	–0.01	0.00	–0.01
Number of children under six years	0.04***	0.03*	0.06***	0.01	–0.04*	0.02**
Education (years)	–0.00	0.00	–0.01*	0.01***	0.01	0.01***
Industry
Primary (O)
Secondary	0.03	–0.01	0.06
Tertiary	–0.01	–0.05	0.01
Private household	0.06	0.08	0.06
Occupation
Managerial/professional (O)
Semi-skilled	–0.06	–0.10*	–0.01
Elementary occupations	–0.06	–0.14**	–0.02
Initial employment characteristics
Recent hire (tenure <2 years)	0.03	–0.05	0.12**
Tenure at job, 2008 (years)	–0.00	–0.00	0.00
Union member	0.03	0.04	–0.02
Working full-time	–0.03	0.00	–0.06
Permanent contract	–0.05*	–0.11***	–0.00
Log (monthly pay)	0.28*	0.49*	0.24
Log (monthly pay) squared	–0.03**	–0.04*	–0.02
Months since Wave 1 interview	–0.00	–0.01	0.00	0.01*	0.02***	0.01*
Location type
Rural formal (O)
Traditional authority area	0.04	0.05	0.01	–0.06**	0.03	–0.09***
Urban formal	0.00	0.04	–0.08	–0.02	0.03	–0.02
Urban informal	–0.03	0.03	–0.12*	–0.04	0.12	–0.09**
Province
Western Cape (O)
Eastern Cape	0.00	0.01	0.00	–0.12	0.18	–0.16***
Northern Cape	–0.01	–0.19	0.11	–0.12	0.29**	–0.21***
Free State	0.17*	0.11	0.24*	–0.11	0.14	–0.12**
KwaZulu-Natal	0.08	0.11	0.05	–0.06	0.20	–0.08
North-West	0.06	0.04	0.08	–0.05	0.29**	–0.11*
Gauteng	0.03	–0.00	0.04	–0.03	0.24*	–0.07
Mpumalanga	0.11	0.07	0.14	–0.02	0.33**	–0.06
Limpopo	0.11	0.12	0.05	–0.07	0.17	–0.08
n	1338	642	696	3455	961	2494

For women, variables associated with individual ‘choice’ seem to play a significant role in the labour-market moves out of regular employment. Holding all other variables constant, for each additional child under six years old a woman is six percentage points more likely to move out of regular employment.

In contrast, the structural change argument does not appear compelling. Industry and occupational variables, and education (a proxy for skill level), are not jointly or individually significant. Even locational category variables are not jointly significant. An important caveat to these results is that we are using only very broad industry and occupational groupings.

Evidence is consistent with a model where, due to imperfect information, employees and/or employers become aware of the firm–worker productivity match only after employment begins. In this type of model, movements out of regular employment would disproportionately occur early in one's tenure. As expected under this theory, a woman's tenure negatively impacts her likelihood of leaving regular employment. A one standard deviation change in tenure (7.6 years) would result in a 4.3 percentage point increase in the probability of leaving regular employment. To test this, a dummy variable was introduced for those with two years or less in tenure. For women, such recent hires were 12.3 percentage points more likely to leave regular employment than others, holding all else constant. The remaining tenure effect (two years or more) was no longer statistically significant. While other stories may explain these results, it fits well with a story of ex-post learning about employee–employer productivity match.

Our estimation approach is relatively silent on the possibility of individual firm deaths motivating these transitions. We do not include any variables that would provide a reasonable correlate for these events. In some sense, the residual can be thought of as capturing some of these changes. The low predictive power of the regression leaves plenty of room for these impacts to play a significant role.

Thus, for women, the correlative results appear more in line with stories of individuals voluntarily leaving the labour market and with workers and firms gaining knowledge about the firm–employee match only after employment begins. They do not seem consistent with women being driven out of the labour market due to structural changes.

Alternative explanations are also possible. For example, women who are pregnant may face labour-market discrimination and be forced out of employment, rather than choosing to leave the labour market. Also, the terms ‘voluntary’ and ‘choice’ are positive but the individual may well not feel they have a good choice and they may feel compelled to stay home if a young child is in the home.

The evidence for men is also consistent with choice being a major driver of churning. Being married and having children under six years old are statistically significant predictors of an increased likelihood of a move out of regular employment. Additionally, the age profile suggests decreasing probability of movement from regular employment until approximately 45 years old and then an increasing probability of movement. While a choice argument would suggest that the work is no longer the optimal choice for the older individual (possibly due to health issues or early retirement), there could also be labour-market discrimination against older workers. If these are indeed drivers of this labour-market churning, policy-makers might consider subsidising childcare for workers (or possibly eldercare support for workers).

Other multivariate results for males are quite different to those found for females. There is no support for labour-market churning among males being driven by revealed worker–firm productivity matches as the recent hire dummy variable is not significant.

Additionally, unlike females, males show some evidence of a general structural change driving the employment transitions. Men in semi-skilled and elementary occupations were much less likely to leave regular employment, holding all else constant, than their managerial and profession male counterparts. The unexplained part of the model leaves plenty of room for firm deaths to play a critical role as well.

The next three columns of Table 18 identify the winners in labour-market transitions, predicting which individuals moved from subsistence agriculture or non-employment into regular employment. As before, the results differ significantly by gender.

For men, additional children are again correlated with negative labour-market transitions as each additional child is associated with a four percentage point decline in the likelihood of moving into regular employment. For women, however, it surprisingly is associated with an increased likelihood of moving into regular employment. For both genders, the likelihood of getting a regular employment job increases until 35 years old (men) or 38 years old (women) and then decreases.

An individual's location has a significant impact on the probability of moving into regular employment but the effects differ significantly by gender. For women, those in informal areas had significantly lower probability of moving into regular employment than those in rural formal areas. No such effect was found for males. For women, the Western Cape offered increased likelihood of regular employment compared with the Eastern Cape, Northern Cape, Free State and North West provinces. For men, the Western Cape offered a decreased likelihood of regular employment in these areas, although only the North West province was statistically significant among this group. For men, Gauteng and Mpumalanga offered significantly higher probability of employment.

7. Conclusion

By allowing us to follow the same individuals over time, NIDS Waves 1 and 2 allow us to see dynamic changes taking place in the South African labour market between 2008 and 2010 that may not be apparent when looking at the changes over time using cross-sectional data.

The data show extensive mobility across employment status and significant mobility across the type of employment. We present evidence that women exhibit much greater mobility into and out of the workforce and employment, while men exhibit more mobility across employment types among those employed in both periods. Among those employed in regular employment in both periods, there is considerable mobility across industry and occupational groupings. Flows out of manufacturing and into services and out of semi-skilled into elementary occupations are particularly noteworthy for men.

The benefit individuals derive from working in regular employment as compared with self-employment or casual employment was demonstrated using some basic summary statistics regarding real earnings changes over time. The same individuals were typically earning much more when moving from self-employment or casual employment to regular employment and earning much less if moving from regular employment to self-employment or casual employment.

The multivariate results that we present are consistent with individual choice and imperfect information associated with worker–firm productivity matches being the primary drivers of the churning observed in the female labour market in South Africa. These results were not indicative of female job displacement being driven by structural changes in the economy. Males showed no sign of churning driven by imperfect information associated with worker–firm productivity matches but did have evidence consistent with individual choice and/or structural change being significant drivers of male labour-market churning. The differences across gender are surprising and were also dramatically present when examining which individuals were moving into regular employment. Future research in this area would be welcome.

Acknowledgements

Murray Leibbrandt acknowledges funding from the Department of Science and Technology and National Research Foundation through the Research Chairs Initiative (Research Chair in Poverty and Inequality) and the Human and Social Dynamics Grand Challenge.

Notes

⁴Also, there is no resulting increase in the proportion of casual employment engaged in occupations that would be associated with subsistence agriculture (elementary occupations/skilled agriculture). Unfortunately, the occupational data for self-employment activities are not present for more than one-half of the self-employed enterprises. Therefore, we cannot perform similar analysis.

⁵See Posel et al. (2013) for a more complete, although still initial, treatment of this issue using NIDS.

⁶Earnings are measured in September 2008 Rand using CPI (Consumer Price Index) data from STATS SA.