Industry wage differentials and efficiency wages: A dissenting view with South African evidence¶
Peter G. Moll (1993). Industry wage differentials and efficiency wages: A dissenting view with South African evidence. Journal of Development Economics 41. North-Holland (Elsevier).
| Type | journal article |
| Scope | Economics and economic history of apartheid |
| Method | econometric |
| Period studied | 1960–1985 |
| Regions | South Africa, Homelands (Ciskei, Lebowa, Gazankulu, KaNgwane, KwaNdebele), Transvaal |
| Themes | labour markets wages, comparative theory, methods data sources, unions strikes, education human capital, bantustans removals influx, inequality poverty unemployment |
| Citation key | moll1993a |
Argument¶
Moll argues that inter-industry wage dispersion cannot be taken as proof of efficiency-wage, rent-sharing or gift-exchange models. Under apartheid, near-powerless African workers had industry wage premia of about the same percentage size as heavily protected, unionized white workers, and the two sets of premia were strongly correlated. Africans were not excluded from high-wage industries, and substantial premia existed where unions and union threats were virtually absent (urban Africans in 1975, homeland workers in 1980). Rent extraction, gift exchange and union-threat explanations therefore fail in this setting. Only expense-preference and internal-equity theories remain unrefuted, and Moll recommends direct survey tests of efficiency-wage hypotheses instead.
Evidence and method¶
Public-use samples of the South African population censuses of 1960, 1970 and 1980 (Mincerian log-earnings regressions with 41 two-digit industry dummies, by race and by occupation); Bureau of Market Research household surveys of urban Africans for 1975, 1980 and 1985; 1980 census data for homeland residents; BMR union density figures; Race Relations Survey; comparison with Krueger and Summers (1988) US results.
Place in the debates¶
- Do inter-industry wage differentials prove efficiency wages or rent sharing? (see Racial wage gaps, wage convergence and wage determination). Dissents. The industry wage structure cannot serve as proof of efficiency-wage or rent-sharing models, since South African African workers with no bargaining power show the same premia as whites.
- Union threat explanation of the industry wage structure (see Racial wage gaps, wage convergence and wage determination). Argues Krueger and Summers' rejection of it was unconvincing, but finds that the South African natural experiment (non-union African workers, homelands) still shows large premia, so unions plus threats cannot be the main cause.
- Racial discrimination as exclusion from high-wage industries vs wage discrimination (see Racial wage gaps, wage convergence and wage determination). Finds no systematic exclusion of Africans from high-wage industries; the 'white' dummy has a small effect on access to industry premia.
- Expense preference and internal equity as remaining explanations (see Racial wage gaps, wage convergence and wage determination). Notes the South African data do not reject expense preference (Heywood) and cannot test it for lack of firm-size data; it regards expense preference and internal equity as the only surviving candidates.
Findings¶
-
Industry wage dispersion for Africans in 1980 was comparable to that for whites, despite African powerlessness. (p. 14)
the standard deviations of industry differentials of the three races are fairly close
-
Industry premia of different races were strongly correlated in 1980. (p. 15)
In 1980 the weighted correlation between white and African industry premia was 0.70 (0.65 excluding the sector 'private household services')
-
Cross-occupation correlations of industry premia are strong, undermining sorting, compensating-differential, monitoring and turnover explanations. (p. 15)
the correlations among the industry premia of different occupations are fairly strong, varying between 0.37 (laborers and managers) and 0.79 (laborers and cleaners)
-
African workers' industry premia imply a large wage gap between industries that would have to be attributed to rents, which is implausible. (p. 17)
Machinery workers have wages approximately 43 percent higher than accommodation workers.
-
The ratio of white to African standard deviations of industry differentials is at least one, contrary to the bargaining-power prediction. (p. 18)
given the large difference in the bargaining power of these groups, one would have expected Africans to have a significantly smaller standard deviation than whites
-
Excluding private household services, there is no relationship between industry premia and African employment share. (p. 21)
when the sector private household services is removed, the correlation falls to -0.09 (t= -0.6, sig. level 0.6)
-
Being white has only a small effect on access to high-premium industries among clerks. (p. 25)
Being white raises the average clerk's industry premium by less than one-fifth of a standard deviation.
-
Urban African workers in 1975 were almost entirely non-union, yet showed substantial industry differentials comparable to US ones. (p. 26)
The adjusted weighted standard deviation of industry differentials in 1975 is 0.098, with a jackknife standard error of 0.0081.
-
Homeland workers, effectively without unions or union threat, show industry differentials as large as in the US. (p. 28)
The standard deviation for 41 industries is 0.172, with a jackknife standard error of 0.0089.
Evidence ledger¶
Sources used¶
| Source | Kind | Period | How used |
|---|---|---|---|
| Public use samples of the South African Population Census 1980 (microdata) | population census | 1980 | Main data for Mincerian earnings regressions with 41 industry dummies for white, colored and African males and for occupation-specific samples, yielding the industry wage premia and their standard deviations and correlations. |
| Public use samples of the South African Population Census 1960 and 1970 (microdata) | population census | 1960 and 1970 | Used to estimate industry premia for white and colored males (African earnings were not collected) and to track the white-colored correlation over time. |
| Bureau of Market Research (BMR) household surveys of urban Africans, 1975, 1980, 1985 | household or labour survey microdata | 1975, 1980, 1985 | Test of the union threat model: industry differentials in the non-union African sector in 1975 (and 1980, 1985), with sample sizes of 3,984, 3,294 and 3,558. |
| BMR unionization rate data for urban Africans (1975, 1980) and surveys 1984-1987 for homelands | household or labour survey microdata | 1975-1987 | Establishes that African unionization was negligible (2.6% in 1975, 5.5% in 1980) and near zero in the homelands, justifying them as non-union samples. |
| 1980 Census subsample from homelands (Gazankulu, Lebowa, Quaqua, Kangwane, Kwandebele, Ciskei) | population census | 1980 | A second 'natural experiment' on the union-threat explanation: 9,966 observations whose industry-premia standard deviation is 0.172. |
| Current Population Survey (CPS) of South Africa 1979-1982 | household or labour survey microdata | 1979-1982 | The only South African data set containing hours; unreported comparison regressions are used to argue that the lack of an hours variable in the census does not drive the results. |
| Race Relations Survey 1989/90 | secondary literature compilation | 1989/90 | Documents the persistence of job colour bars and segregation on the mines as context for African workers' lack of bargaining power. |
| Krueger and Summers (1988) U.S. industry wage differentials and Katz and Summers (1989) occupation-specific standard deviations | secondary literature compilation | 1974-1984 (U.S. CPS) | Benchmark figures (SD of 0.11-0.14, 0.097/0.069/0.094) against which South African dispersion is compared, and the template for the 41-industry specification. |
| Moll (1993) on black South African unions (union density of 19% by 1985) | secondary literature compilation | 1980s | Source for the rise in African unionization to 19 percent by 1985 after legalization. |
| Knight and McGrath (1987) and Moll (1992b) on labour market discrimination | secondary literature compilation | 1970s-1980 | Supports the statement that discrimination declined during the 1970s but was still common in 1980. |
Inference¶
Quantification: Econometric (causal design).
Mincerian log-earnings regressions with industry dummies on 1960/70/80 census microdata, separately by race and occupation; weighted adjusted standard deviations and cross-group correlations of industry premia are compared across races and occupations as a test by elimination of efficiency wage, rent sharing and gift exchange theories. Supporting regressions, a reduced-form regression of industry premia on worker characteristics (white dummy), and two 'natural experiments' using non-union samples (BMR 1975, homelands 1980). Bargaining power is treated as proportional to the standard deviation of premia, with acknowledged strong functional-form assumptions.
Key claims¶
| Claim | Basis | Strength |
|---|---|---|
| African workers, despite lacking bargaining power, had industry wage dispersion similar to that of whites (SD roughly 0.10-0.20 across races; ratio of white to African SD is unity or greater). (p. 226) | Weighted adjusted SDs from 1980 census microdata regressions for white, African and colored males (Table 1). | Systematic data |
| Industry premia are strongly correlated across races (0.70 white-African, 0.76 white-colored in 1980) and across occupations (0.37 to 0.79). (p. 227) | Employment-weighted correlations of industry coefficients from census microdata regressions (Table 2). | Systematic data |
| Africans were not systematically excluded from high-wage industries: the correlation between white premia and African employment share is insignificant once private household services is excluded, and the 'white' dummy has small effects in the clerk regressions. (p. 237) | 1980 census microdata: Figs 1-3 and Table 3 regressions of clerks' industry premia (N=15,386). | Systematic data |
| Unions and union threats cannot explain the industry wage structure: substantial differentials existed among non-union urban Africans in 1975 and in the homelands in 1980. (p. 238) | BMR 1975 sample (SD 0.098, jackknife SE 0.0081) and 1980 homelands census sample (SD 0.172, SE 0.0089), with BMR unionization rates. | Partial data |
| Machinery workers earned approximately 43 percent more than accommodation workers among Africans, a gap not attributable to unobserved ability or compensating differentials. (p. 229) | African 1980 census industry coefficients (0.17 vs -0.26) in the appendix regression. | Partial data |
| Returns to education were much higher for whites than Africans in 1980 (e.g., 0.55 vs 0.25 log points at ten years of schooling), reflecting the racial basis of education and the labour market. (p. 225) | Education coefficients in the 1980 white and African census regressions (appendix Table A.1), plus cited per-child spending ratio. | Systematic data |
The author's caveats¶
The census has no hours variable (partly checked with the CPS), no firm-size variable, and no union variable; the BMR samples are household rather than worker samples with broad one-digit industry categories; industry premia rest on strong log-linear functional-form and homogeneous-rents assumptions; unobserved ability and sorting cannot be ruled out; African earnings were not collected in 1960 and 1970; and females are excluded from the race samples.
Gaps in the evidence¶
The argument would be better tested with linked firm-level data on profits, firm size and union coverage, which would allow direct tests of rent sharing and expense-preference models; the census lacks these and they were largely unavailable then. Panel or industry-switcher data for African workers would address unobserved ability and sorting, and are scarce even now. Direct evidence on wage-setting practices (firm surveys, bargaining records) would also be needed to show that African premia really reflect rent sharing rather than, for example, labour-market segmentation or regional differences. The inference that bargaining power is proportional to SD of premia is not independently validated.
Replicability: Data described but not public.