Unemployment and wages in South Africa: A spatial approach¶
Geeta Kingdon; John Knight (1999). Unemployment and wages in South Africa: A spatial approach. Centre for the Study of African Economies Working Paper Series, WPS/99-12 (Institute of Economics and Statistics, University of Oxford). Centre for the Study of African Economies / Institute of Economics and Statistics, University of Oxford.
| Type | working paper |
| Scope | Economics and economic history of apartheid |
| Method | econometric |
| Period studied | 1993–1993 |
| Regions | South Africa (national), former homelands vs non-homeland areas, rural and urban clusters |
| Themes | labour markets wages, inequality poverty unemployment, bantustans removals influx, unions strikes, methods data sources, comparative theory, transition post apartheid |
| Citation key | kingdon1999 |
Argument¶
Using the 1993 SALDRU/World Bank household survey, the paper tests whether the OECD 'wage curve' (a negative relationship between local unemployment and wages, elasticity about -0.1) holds at South Africa's very high unemployment rates. It finds a wage curve with elasticity of about -0.11 using broad unemployment, and it holds over a wide unemployment range (10-30%) but flattens at very high rates. Union and public sector wages are insulated from local unemployment, which the authors attribute to non-local bargaining via Industrial Councils. The former homelands show no wage curve, which makes the homeland/non-homeland divide the key fault-line of labour market segmentation. The wage curve is stronger for broad than narrow unemployment, which supports the broad definition including discouraged workers. The authors apply robustness checks (robust standard errors, cluster-mean regressions, 2SLS) and compare sub-groups to discriminate between efficiency-wage and bargaining explanations.
Evidence and method¶
1993 South African Living Standards Survey (SALDRU/World Bank): 8848 households in 360 clusters; estimation sample of 6498 wage employees aged 16-64 resident in their cluster. Cluster-level unemployment rates, cluster food prices, and cluster characteristics; official labour statistics (CSS 1994) and the ILO (1996) report for institutional coverage.
Place in the debates¶
- Wage curve (negative wage-unemployment relation) vs Harris-Todaro compensating differentials (see Unemployment: causes, measurement and labour market institutions). Finds a negative wage curve in South Africa, with a convex or U-shape whose upward part appears only at very high unemployment rates. It treats Harris-Todaro as a long-run relationship compatible with a short-run wage curve.
- Narrow vs broad definition of unemployment in South Africa (ILO 1996 vs broad-definition proponents) (see Unemployment: causes, measurement and labour market institutions). Supports the broad definition: broad unemployment significantly affects wages while narrow unemployment does not, so discouraged workers act as genuine labour supply.
- Efficiency-wage vs local bargaining vs institutional explanations of the wage curve (see Unemployment: causes, measurement and labour market institutions). Sub-group results are consistent with efficiency-wage behaviour and local bargaining. Union and public sector wages are flat because non-local bargaining, notably through Industrial Councils, insulates them.
- Labour market segmentation: rural/urban vs homeland/non-homeland (see Unemployment: causes, measurement and labour market institutions). The homeland/non-homeland division is the clearest fault-line; the rural-urban division does not capture segmentation well.
- Wage flexibility and unemployment over time vs across space (Fallon and Lucas time-series) (see Unemployment: causes, measurement and labour market institutions). Reconciles a positive wage-unemployment relation over time with a negative one across space as an outward shift of the wage curve, probably from increased bargaining power.
Findings¶
-
The wage curve elasticity in South Africa matches the OECD value but persists over a much larger range of unemployment rates. (p. 1)
The wage curve elasticity in South Africa is similar to that in OECD countries (-0.1) but persists over a much larger range of unemployment rates
-
Only broadly measured unemployment has a significant wage effect; the narrow measure has none. (p. 8)
The table shows that there is no significant relationship between narrowly measured unemployment and wages but that there is a significant quadratic relationship between broadly measured unemployment and wages
-
Headline elasticity at the mean for broad unemployment is about -0.11, whereas the narrow elasticity is near zero. (p. 9)
the broad unemployment elasticity of the wage evaluated at the mean is nearly -0.11. In other words, a doubling of broad unemployment generates an 11% reduction in wages. However, the narrow unemployment elasticity of wage is close to zero.
-
The wage curve supports the broad definition of unemployment, contrary to the ILO view that discouraged workers are outside the labour force. (p. 9)
discouraged workers are taken into account by wage-setters and their numbers in a locality do influence local wage determination.
-
Union wages are insulated from local unemployment, which the authors attribute to non-local bargaining. (p. 10)
unionised workers’ wages do not respond negatively to local unemployment rate at all
-
The homeland/non-homeland split is the main labour market fault-line and unemployment is catastrophically high in the homelands. (p. 11)
the clearest labour market fault-line occurs along ‘homeland’-non homeland lines. Broadly measured unemployment is catastrophically high in the (former) homeland regions, and it is much lower in non-homeland regions.
-
Homeland wages show no relationship with local unemployment, while non-homeland areas show a significant wage curve. (p. 11)
homeland South Africa - with its very high unemployment rates - exhibits no significant relationship between local unemployment and wages.
-
Non-homeland wage curve elasticity is -0.111 with a minimum at 36% broad unemployment. (p. 11)
non-homeland South Africa displays a significant convex relationship between unemployment and wage, with a WUE of -0.111: a doubling of the unemployment rate there leads to a 11.1% reduction in wages.
-
A significant minority of South African wage employees live in clusters beyond the U-shape minimum, so the upward part cannot be ignored. (p. 11)
one cannot dismiss the upward-sloping part in the South African wage curve because a significant minority (27%) of wage employees live in clusters in which the broad unemployment rate is greater than 36%.
-
In non-homeland areas, the elasticity stays near -0.1 or larger when unemployment triples from 10% to 30%. (p. 12)
in non-homelands, the unemployment rate can be tripled from 10% to 30% without reducing the wage curve elasticity below about -0.1.
-
Instrumenting for endogeneity strengthens rather than removes the wage curve, consistent with causation from unemployment to wages. (p. 13)
the elasticity estimated from the 2SLS model is larger (a bigger negative coefficient) than that from the corresponding OLS model.
-
Using the narrow definition for international comparability, urban South Africa's elasticity is close to OECD and Côte d'Ivoire values. (p. 14)
using the narrow definition of unemployment, the wage-unemployment elasticity in urban areas is - 0.08, that is, close to the -0.10 found for OECD countries and the -0.12 found for urban Cote d’Ivoire.
Evidence ledger¶
Sources used¶
| Source | Kind | Period | How used |
|---|---|---|---|
| South African Living Standards Survey (SALDRU/World Bank integrated household survey), 1993 | household or labour survey microdata | Second half of 1993 | Sole micro dataset: supplies individual wages, hours, characteristics, cluster unemployment rates and cluster characteristics for the wage-curve regressions (6498 wage workers). |
| SALDRU survey cluster food prices combined with CPI weights | household or labour survey microdata | 1993 | Builds a regional price index to deflate wages, assuming uniform non-food prices. |
| SALDRU survey cluster characteristics and community data (roads, transport, facilities, religion) | household or labour survey microdata | 1993 | Cluster characteristics are wage-equation controls (TARROAD, DISFACI) and the source of instruments for 2SLS. |
| SALDRU survey cognitive skill test scores | household or labour survey microdata | 1993 | Considered as a proxy for schooling quality but rejected because of selective test administration. |
| South African Labour Statistics (CSS, July 1994) | official statistical series | c.1993 | Source for the share of formal-sector employees covered by Industrial Councils, Wage Boards and unions in 1993. |
| ILO (1996) Restructuring the Labour Market: The South African Challenge, ILO Country Review | commission or official report | 1996 | Source of the claim that the broad unemployment definition exaggerates unemployment, which the paper tests; also source of institutional coverage data. |
| Klasen and Woolard (1998) and StatsSA (1998) unemployment estimates | secondary literature compilation | 1998 | Establishes the 31-37% average unemployment rates that motivate testing the wage curve under very high unemployment. |
| Schultz and Mwabu (1997) union wage premium estimate (1993 data) | secondary literature compilation | 1993 data | Cited for the 19% union wage premium among African workers to show unions' importance in wage determination. |
| Fallon and Lucas (1998) time-series employment elasticity of wages | secondary literature compilation | 1998 | Time-series result (-0.7 employment elasticity) is contrasted with the spatial wage curve in the conclusion. |
| Blanchflower and Oswald (1994) wage curve evidence for OECD countries; Hoddinott (1996) Côte d'Ivoire | secondary literature compilation | various | Benchmark elasticities (-0.1 OECD, -0.12 Côte d'Ivoire) for comparing South African wage flexibility. |
| Industrial Council and Wage Board minimum wage determinations (clothing industry, 1996) | legislation gazette | 1996 | Illustrates the sub-national, local nature of institutional minimum wages. |
Inference¶
Quantification: Econometric (causal design).
Cross-sectional wage-curve regressions (OLS Mincer-type wage equations, log hourly wage on cluster-level unemployment, linear and quadratic) on 6498 wage workers in 360 clusters, with robust (Moulton-corrected) standard errors, sub-group comparisons, aggregate cluster-mean regressions, and 2SLS with cluster characteristics as instruments. Competing theories are tested by comparing elasticities across worker groups.
Key claims¶
| Claim | Basis | Strength |
|---|---|---|
| A wage curve exists in South Africa with broad-unemployment elasticity of about -0.11, but near zero for narrow unemployment. (p. 9) | 1993 SALDRU survey cross-section wage regressions (Tables 1-2), with robustness specifications. | Systematic data |
| Broad unemployment is the more appropriate definition for South Africa because wage-setters take discouraged workers into account. (p. 9) | Comparison of narrow vs broad unemployment coefficients in the same SALDRU regressions. | Partial data |
| Union wages are insulated from local unemployment, and public-sector wages respond less than private-sector wages. (p. 9) | Sub-group elasticities (Table 3) by union status and sector from SALDRU data; interpretation draws on institutional information (CSS 1994, ILO 1996). | Partial data |
| The clearest labour market segmentation is homeland vs non-homeland; homelands show no significant wage curve while non-homelands show elasticity of -0.111. (p. 11) | Cluster unemployment rates by region (Tables 4a/4b), a Chow test, and separate regressions (Table 5, Appendix Table 1). | Systematic data |
| The elasticity stays around -0.1 over a wide range of unemployment (10% to 30%) and falls to zero above 30%. (p. 12) | Elasticities computed at different unemployment rates from the quadratic estimates (Table 7). | Partial data |
| The negative relationship is robust to endogeneity and is consistent with causality running from unemployment to wages. (p. 12) | 2SLS with cluster characteristics as instruments (roads, transport, facilities, religion), with partial R2, F and overidentification tests (Table 10). | Partial data |
| Using the narrow definition in urban areas, South Africa's wage elasticity (-0.08) is close to OECD (-0.10) and urban Côte d'Ivoire (-0.12). (p. 14) | Table 11 urban/rural narrow-definition regressions on SALDRU data, compared with published benchmarks from other studies. | Partial data |
The author's caveats¶
Cross-section data permit only provincial dummies, not cluster fixed effects; local labour market boundaries are uncertain, and there is no information on commuting or distance to work. Only food prices are available by cluster. Earnings and hours are missing for some individuals, and no correction for sample selection is made in the basic specification. School-quality test data are unusable. The elasticity is sensitive to the value imputed for zero-unemployment clusters. Results lose significance at the aggregate (cluster-mean) level for all South Africa. Union status is unavailable for casual workers.
Gaps in the evidence¶
Panel or repeated cross-section data on local unemployment and wages would allow cluster fixed effects and lagged-unemployment instruments, and the paper had only a single 1993 cross-section. Information on place of work and commuting, and on whether workers are covered by specific Industrial Council or Wage Board agreements, would test the bargaining and segmentation mechanisms directly. Later household and labour force surveys exist and could replicate the results over time.
Replicability: Data described but not public.