Contact

The gender-wage gap in South Africa (blog 2 of 2)

Michele Capazario and Bianca Capazario (9 min read)

 

In a previous blog1 , we discussed unconditional wage gaps in South Africa. As a refresher, an unconditional wage gap is simply the difference in average or median wages between demographic groups (say, for example, the gap between females and males), not controlling for factors that influence wages, like experience or education. As a reminder, unconditional estimates suggest that men earned approximately 36% more than females according to the 2019 Labour Market Dynamics Survey (LMDS).

While an unconditional wage analysis is insightful to some degree, it is also quite naïve. Without controlling for factors that impact earnings, one runs the risk of comparing the wages of, for example, a veteran female lawyer to that of a junior male economist. In this example (and, if there’s any justice in the world), the lawyer would earn much more than the economist.

However, comparing their wages without controlling for their characteristics (particularly, the characteristics that impact wage levels) might lead an analyst to wrongfully deduce that female wages are higher than male wages. Comparing these two vastly different individuals’ wages without controlling for any confounding factors2 implies that one would run the risk of misrepresenting the influence of gender on wages.

Blinder (1973) and Oaxaca (1973) provide a useful framework to control for such individual characteristics and assess conditional wage gaps (i.e., conditional on labour market and social factors). Jann (2008, p. 1) describes the Blinder-Oaxaca decomposition as follows:

[The Blinder-Oaxaca descomposition] divides the wage differential between two groups into a part that is “explained” by group differences in productivity characteristics, such as education or work experience, and a residual part that cannot be accounted for by such differences in wage determinants. This “unexplained” part is often used as a measure for discrimination, but it also subsumes the effects of group differences in unobserved predictors.
 
We have, therefore, used the Blinder-Oaxaca decomposition to analyse the conditional gender-wage gap in South Africa using the LMDS. The LMDS houses nationally representative earnings and labour market characteristic data from a sample of South Africans.

In our decomposition, we control for the following factors given their likely impact on wages earned 3.

  • Demographic factors like an individual’s population group, age, age-squared, marital status, and province.
  • The number of years of experience and years of experience squared.
  • Maternity leave taken.
  • Labour market determinants like an individual’s occupation, educational attainment, degree of employment formality, contract type (verbal or written), contract duration (full-or-part-time), and employer type (private, public or non-profit organisation).

Employing this decomposition between 2012 and 2019 (the latest year for which data is available) 4, we find a significant wage differential between South African males and females that isn’t explained by the above mentioned factors.

Specifically, we find that, on average, a South African female earns a conditional wage between 18 and 23% lower than a male, even though both have the same experience and education prospects, live in the same province, are the same age, and are employed in the same line of work:

Figure 1: Size of the conditional gender-wage gap according to a Blinder-Oaxaca decomposition of the LMDS

 
Source: Own analysis of the Labour Market Dynamics Survey
 
 
This wage gap was, and still is, quite large, especially in comparison to other developing countries:

  • A decade-old analysis of 18 Latin American countries estimated the conditional gender wage gap to between 9 and 27% 5.
  • More recently, studies on small open Eastern Block economies suggest a similar conditional wage gap as in Latin America, with the wage gap shrinking the further west the European economy analysed is 6.

 

Further, a significant portion of this conditional wage gap (65% or more) has not been explained 7 by changes in the previously mentioned social and labour market factors, as evidenced in Figure 2: 
 
 

Figure 2: Proportion of the conditional gender wage gap not explained by changes in labour market characteristics.

Source: Own analysis of the Labour Market Dynamics Survey

Such a large, unexplained wage differential is often interpreted as evidence of a high degree of gender-based wage discrimination in the labour market. Of course, as Jann (2008) points out, some of this unexplained wage differential might be attributed to factors not included in the analysis. However, given the host of controls used, a portion of this wage gap is likely due to wage bias based on gender.

Nevertheless, it is encouraging that, compared to 2012, the 2019 labour market was more equitable to females, based on the size of the pay gap (both unconditional and conditional) between males and females and the degree to which the conditional wage premium was left unexplained. In light of this analysis, we’ve come full circle- in our previous blog, we raised the following question:

Has labour market progress in South Africa been sufficient over the last decade?

We ask the same question here, with less naïve evidence. Based on this evidence, our answer is probably not. But with this evidence, we can infer that the South African labour market has progressed instead of regressed. That probably counts for something.

Right?

[1] See the following link: https://www.dnaeconomics.com/pages/regional_integration/?zDispID=NewsArtDemographic_Gaps_in_South_Africa_Over_Time_Blog_1_of_as_many_as_it_takes

[2] Confounding factors are those that compete with the variable of interest (gender) in explaining an outcome (wage level)

[3] For more information on why these controls are included, see (Reddy, Rogan, Mncwango, & Chabane, 2018) and (Fisher, Biyase, Kirsten, & Rooderick, 2020)

[4]The decomposition was applied to logged earnings, although similar results manifest themselves when using earnings in Rands.

[5](Atal, Nopo, & Winder, 2009)

[6](Khitarishvili, 2019)

[7]In simple terms, the proportion of the conditional wage gap that is explained (35% in 2019) can be likened to the R-Squared in a regression analysis. Once the conditional wage gap is estimated, our control variables explain approximately 35% of the variation of the conditional wage gap in 2019.

Bibliography

Atal, J., Nopo, H., & Winder, N. (2009). New Century, Old Disparities: Gender and Ethnic Wage Gaps in Latin America. Inter-American Development Bank Working Paper Series, 1-77.

Blinder, A. (1973). Wage discrimination: reduced form and structural estimates. Journal of Human Resources, 436-455.

Fisher, B., Biyase, M., Kirsten, F., & Rooderick, F. (2020). Gender Wage Discrimination in South Africa within the Affirmative Action Framework. Economic Development and Well-Being Research Group Working Paper Series, 01-2020, 1-16.

Jann, B. (2008). The Blinder-Oaxaca Decomposition for Linear Regression Models. The Stata Journal, 8(4), 453-479.

Khitarishvili, T. (2019). Gender Pay Gaps in the Former Soviet Union: A Review of the Evidence. Journal of Economic Surveys, 1257-1284.

Oaxaca, R. (1973). Male-Female Wage Differentials in Urban Labour Markets. International Economic Review, 693-709.

Reddy, V., Rogan, M., Mncwango, B., & Chabane, S. (2018). Occupations in High Demand in South Africa. Labour Market Intelligence Partnership.