The Rise in Union Wage Premia in South Africa¶
Julian F. Hofmeyr; Robert E.B. Lucas (1998). The Rise in Union Wage Premia in South Africa. University of Natal, Durban / Boston University (Institute for Development Economics).
| Type | working paper |
| Scope | Economics and economic history of apartheid |
| Method | econometric |
| Period studied | 1985–1993 |
| Regions | South Africa (urban areas, excluding Transkei, Venda, Bophuthatswana and Ciskei) |
| Themes | unions strikes, labour markets wages, inequality poverty unemployment, methods data sources |
| Citation key | hofmeyr1998 |
Argument¶
The paper asks why the mean earnings premium of African urban male union members over other regular employees rose sharply between 1985 and 1993, a period of rapid union growth after the 1979 legalisation of African unions. It applies a common specification to the 1985 BMR and 1993 SALDRU/World Bank surveys. It uses four estimators: an OLS union dummy, separate OLS regimes, an endogenous-switching model, and a multinomial logit sample-selection model. Levels of the estimated premia vary widely across methods, but all agree that the premium rose. A decomposition shows that changes in measured worker characteristics reduced the premium, and that the rise comes entirely from changes in pay structures. Wage compression within unions by 1993 dampened the rise. The authors conclude the rise reflects either higher union rents or unobserved compositional change.
Evidence and method¶
Two national household surveys: the 1985 Bureau of Market Research income and expenditure survey (about 4,600 African urban households, 12,300 individuals) and the 1993 SALDRU/World Bank survey (about 9,000 households, 24,000 individuals). The analysis is restricted to African urban males in the common urban areas. Union membership in 1985 is proxied by payment of union dues. Secondary sources include Fallon and Pereira de Silva (1994), Moll (1993) and Schultz and Mwabu (1996).
Place in the debates¶
- Size and determinants of the union wage premium / choice of estimator (exogenous dummy vs endogenous switching vs sample selection) (see Black trade unions: growth, wage effects, Wiehahn and strategy). Finds premia highly sensitive to specification, with the switching model preferred on nesting tests. Sample selection corrections are judged probably relevant but not robustly detected. All methods agree the premium rose.
- Do unions raise the wages of African workers (rents) vs composition effects (see Black trade unions: growth, wage effects, Wiehahn and strategy). Composition changes reduced the premium. The rise is attributed to changes in pay structure, i.e. higher rents or undetected composition shifts.
- Union wage compression and egalitarianism (see Black trade unions: growth, wage effects, Wiehahn and strategy). Finds wage compression across experience, education and occupation among union members by 1993, consistent with union egalitarian aims.
Findings¶
-
Legal context: the 1979 amendment brought African workers into the statutory definition of employee, and African unionisation grew rapidly afterwards. (p. 3)
the Industrial Conciliation Amendment Act of 1979 extended the definition of an employee to encompass African workers for the first time
-
Non-wage employment rose above half of the African labour force in the 1980s even as average real wages rose. (p. 3)
the fraction ofthe African labor force without wage employment grew quickly, to exceed 50 percent by the early 1990s, even as the simple average of real wages continued to rise.
-
Union density among regular urban African male employees rose from 27 to 46.7 percent between the samples, while the regular employment share fell. (p. 10)
the fraction of regular employees who are union members also rose substantially, from 27 percent in 1985 to 46.7 percent by 1993.
-
The raw premium in simple average earnings rose from 8 to 26.5 percent. (p. 23)
rose from 8 percent in 1985 to 26.5 percent in 1993.
-
In every pairwise comparison of estimators the estimated premium is higher in 1993 than in 1985. (p. 24)
In pair-wise comparisons, the premium is larger in 1993 than in 1985, in every single case.
-
Union pay profiles in 1993 are flatter with later turning points, indicating wage compression. (p. 20)
the earnings-experience profile is much flatter among union members in 1993, indicating wage compression within the unions.
-
Union and non-union pay structures differed significantly only by 1993, not in 1985. (p. 14)
the union and non-unionized sectors did not offer significantly different pay profiles in 1985 (other than a pure union differential) whereas by 1993 pay structures differed significantly.
-
Compositional change reduced the premium; the rise stems from changes in coefficients. (p. 27)
the previously noted changes in composition of the labor force — no matter whether it is union members or other regular employees in question — served to reduce the level of the average union premium
-
The estimates, especially the MNL results, are very sensitive to specification. (p. 28)
the estimates with sample selection prove very sensitive indeed to precise specification of the model, as does the estimated union premium.
-
Conclusion: the rise is not explained by measured characteristics and may reflect higher rents or undetected composition change. (p. 29)
the observed rise in union premium must either reflect an increase in rents to union members as compared to identical non-member workers or some change in the composition of union members
Evidence ledger¶
Sources used¶
| Source | Kind | Period | How used |
|---|---|---|---|
| 1985 income and expenditure survey, Bureau of Market Research (BMR), University of South Africa | household or labour survey microdata | 1985 | Supplies earnings, employment status and union membership (via dues paid) for urban African males in 1985; the baseline sample for estimating union premia. |
| 1993 household survey, SALDRU/World Bank (Project for Statistics on Living Standards and Development) | household or labour survey microdata | 1993 | Supplies 1993 earnings, employment category and direct union membership data for urban African males, compared with 1985 to measure the change in union premia. |
| Industrial Conciliation Act of 1956 and Industrial Conciliation Amendment Act of 1979 | legislation gazette | 1956-1979 | Provides the institutional setting: the 1956 Act barred multiracial unions and the 1979 amendment extended the definition of employee to Africans, enabling union growth. |
| Wiehahn Commission recommendations | commission or official report | late 1970s | Cited as the origin of the 1979 legislative change that opened collective bargaining to African workers. |
| Union-reported membership figures | union or movement records | 1980s-1990s | Used to assert that African union membership grew rapidly and that unionization of wage employees is among the highest in developing countries. |
| Fallon and Pereira de Silva (1994), World Bank Southern Africa Department Discussion Paper No. 7 | secondary literature compilation | 1980s-early 1990s | Source for the claim that the African labor force without wage employment exceeded 50 percent while average real wages rose; also cited on the decline in regular employment. |
| Hofmeyr (1994), An analysis of African wage movements in South Africa, Research Monograph No. 9 | secondary literature compilation | to 1994 | Cited for factors behind African real wage drift, including rising skills and reduced job barriers. |
| Moll (1993) estimates on the 1985 BMR survey | secondary literature compilation | 1985 data | Prior estimates of union wage gaps from the 1985 survey; the switching-model method used here follows Moll's approach. |
| Schultz and Mwabu (1996) analysis of the 1993 survey | secondary literature compilation | 1993 data | Prior quantile-based union premia from the 1993 survey, treating membership as exogenous; noted as not comparable with Moll's specification. |
Inference¶
Quantification: Econometric (causal design).
Cross-sectional comparison of two national household surveys (1985, 1993), estimating union wage premia with four nested specifications: OLS with a union dummy, separate OLS regimes, a switching model with probit-based Mills ratio, and a multinomial logit sample-selection model. Nesting tests compare the specifications, and an Oaxaca-type decomposition splits the premium change into composition and coefficient (pay structure) effects. Identification relies on functional form and few exclusion restrictions, so it is not a natural experiment.
Key claims¶
| Claim | Basis | Strength |
|---|---|---|
| The simple mean earnings premium of African urban male union members over non-union regular employees rose sharply between 1985 and 1993. (p. 23) | Sample means from the BMR 1985 and SALDRU 1993 surveys for urban African male regular employees. | Systematic data |
| Across all four estimation approaches and permutations, the estimated union premium is larger in 1993 than in 1985. (p. 24) | Regression estimates (Table 5.1) from OLS, switching and MNL models on both surveys. | Systematic data |
| Changes in measured worker characteristics tended to reduce the union premium; the rise comes entirely from changes in pay structure (coefficients). (p. 27) | Decomposition of log earnings differentials using estimated coefficients and mean characteristics (Table 5.3). | Systematic data |
| By 1993 union members' pay showed wage compression across experience, education and occupation relative to non-members. (p. 20) | Coefficients on experience, education and occupation in the switching model earnings equations (Table 4.3) and the F-tests of nesting. | Partial data |
| The rate of union membership among regular employees rose from 27 percent in 1985 to 46.7 percent in 1993, varying by sector and education. (p. 24) | Sample tabulations of union dues paid (1985) and direct union membership questions (1993) in the two surveys. | Systematic data |
| Estimates of the premium are highly sensitive to specification, and sample-selection corrections from the MNL model are unstable, so the level of the premium is poorly determined. (p. 23) | Comparison of Table 5.1 across models, nesting tests (Table 4.1), prediction tables and the omitted-variable check on 'other union member in household'. | Systematic data |
| The observed rise in the premium must reflect either increased union rents or an unobserved change in the composition of members. (p. 29) | Inference from the decomposition residual and the absence of detectable compositional effects in the survey data. | Asserted |
The author's caveats¶
The authors note that the 1985 survey has no clear formal/informal distinction and treats casual workers ambiguously. Earnings data in informal sectors in 1985 are questionable. Neither survey gives hourly earnings, so weekly earnings are used. The 1993 hours data are unreliable. 1985 union status is inferred from payment of dues. Occupation categories are not fully comparable across surveys and required judgement. Few identifying variables exist for the selection equations, and tertiary-educated union cells are small. They also acknowledge unobserved composition changes they cannot detect.
Gaps in the evidence¶
The argument lacks panel or repeated data on the same workers, or establishment-level data on firms' union coverage and bargaining outcomes, which would help separate union rents from unobserved composition shifts. Hourly wages, firm size and union density by industry would improve identification. Only two cross-sections exist, with non-comparable definitions of employment and occupation, and good exclusion restrictions are missing for the selection models. Later labour force surveys and the October Household Surveys (from 1994) now allow a longer time series, but nothing comparable existed for the 1980s.
Replicability: Data described but not public.