The Gender Earnings Gap



Gender has been a long-standing issue since women claimed their right to the vote through the at-the-time controversial suffragette movement. Yet despite the change in legislation since these dark times the difference in pay across gender has continued to be significant.

The measurement of labour market discrimination

Average Gender Wage Gaps and Labour Market Outcomes

In the literature review of this thesis, it is necessary to note that there are broadly two types of studies which we will be making use of: the international comparison studies, and the intranational comparison studies. As will be outlined in this chapter these two types of studies each make use of different techniques, models and methodologies to derive the wage gap between different groups.

One of the first papers to be published regarding the decomposition of male-female wage gaps and into an explained and an unexplained component, enabling the evaluation of discrimination is that by Oaxaca (1973) and Blinder (1973). In this paper, data for the US is used,

International Evidence

It is essential to look initially at papers which have sought to compare wage gaps across different countries from an international perspective;

Blau and Kahn's (1992) paper attempts to explore the gender pay gap from a purely international perspective. They use microdata to evaluate international differences in the wage gap across eight industrialized countries. According to this paper, differences in the wage gaps between countries is due to two main factors: the first is the gender differences - these can be differences in qualifications between men and women as well as the difference in rewarding equally qualified workers of different genders (known as labour discrimination). The second is wage inequality, as wages represent the returns to certain labour market characteristics, and these returns do not tend to be equal across different countries. It is noted that this wage inequality may be the reason why the USA's gender pay gap is not smaller than that of other countries, even though the US has been applying equal pay legislation the longest.

A later paper by the same authors, authored more recently in 2001, attempts to analyse international differences in the gender wage gap taking account of many of the issues outlined in their previous papers. They use ISSP microdata to study the gender wage gaps in several countries over the period from 85 to 94 more or less. The hypothesis they test in this paper refers to whether the reduction of a country's gender pay gap can be due to an overall wage compression and the existence of low female supply relative to demand. (Blau and Kahn, 2001)

Following the use of this new methodology, originally proposed by Juhn etal, a paper underlining the invalid assumptions which were made use of made its appearance. According to Suen (1997), the decomposition of wage residuals into standard deviation and percentile ranks, one of the main components of the methodology, is erroneous, due to the fact that the standard deviation and percentile ranks are assumed to be independent. In reality, these are not likely to be independent, and it can be seen that as a result, discrimination is assumed to be non-existent.

As a result, ISSP data has not been analysed, as far as the author is aware, in a more appropriate context. A promising technique which has been used for many recent studies with some success is that presented by Machado and Mata (2005). In their paper they use quantile regression techniques to derive counterfactual distributions to be able to study wage gap across the distribution, rather than solely at the means. This process can be outlined over 4 steps:

  • First generate a random sample of size m from the uniform distribution.
  • Then estimate the m different quantile regression coefficients:.
  • Generate , a random sample of size m with replacement from.
  • Using the above, a random sample of size m from the unconditional distribution of can be computed, equal to.

However, this methodology uses just one observation for each of the m different quantile regression coefficients, whereas that proposed by Melly (2006) uses all the observations, as these observations are assumed to be identically and independently distributed. Furthermore, as , for all ,



In words, the probability that a certain coefficient is chosen is exactly equal to the difference . Therefore if Machado and Mata's (2005) methodology were applied using , the estimator would by equal to that derived through Melly's (2006) methodology. Melly (2006) proves this result through a Monte-Carlo simulation. This results in a lack of efficiency in the MM estimator, compared to Melly's estimator, not to mention extra computational time.

Heckman (1979) proposes an unbiased estimator for models where sample selection is an issue. In this paper, first written in 1976, he uses the concept of a mills ratio, which is a monotone decreasing function of the probability of being included in the sample. This is achieved by including this ratio as a new variable which captures the effect of sample selection on the estimates. Through this simple regression technique it is possible to remove the sample selection bias if the sample is censored. In Heckman (1976), this technique is generalized to be used in a variety of settings, such as truncation, limited dependent variables and sample selection. The fact that a common structure can be found across these statistical models enables the development of one simple estimator addressing the common issues of all of these models.

Greenhalgh (1980)

Imputation of women's labour market experience

Arrufat & Zabalza(1985) suggest a powerful alternative to the Mincer proxy for the measurement of unobserved women's labour market experience. The three main problems known to exist when using the Mincer proxy/potential experience to measure work experience are of measurement, of endogeneity and of selectivity. As a solution, a participation model is used to predict the participation rates of women in the sample over each year of their life, and then these past predicted probabilities of experience can be integrated back for each individual to end up with an estimate of market experience. A measure of “home-time” is then derived and is to be included instead of potential experience in any estimation. This home-time is the difference between potential experience and the amount of time spent in the labour market (market experience). This solution solves the first two problems outlined above, however the final problem of selectivity into participating in the labour market is dealt with through the Heckman (1979) Two-Step estimator.

Occupational attainment inclusion with imputed labour experience for women

In Miller (1987), a conventional approach to analyzing the gender pay gap and the discrimination therein is taken at first. The Arrufat and Zabalza method is used to compute a “home time” and work experience variable for female workers. Following this conventional method, a behavioral model of occupational attainment is estimated using an ordered probit function for each gender, capturing the gender differences in this type of attainment. Following this estimation, it is then possible to decompose the gender pay gap into inter and intra-occupational components, thereby measuring the discrimination arising through a women's occupational choice rather than purely from her characteristics (usually assuming there is no discrimination affecting the distribution of women over different occupations).

Wright and Ermisch(1991)

The Imputation of Women's Work Experience:

In the paper by (Wright & Ermisch 1991), the Women's Employment Survey is used for gender wage gap decompositions, which enables the use of data on actual work histories, resulting in an improved model including actual work experience for women. When using other surveys, it is necessary to predict work experience through a labour participation equation, which increases the overall error of the evaluation of the wage gap. This paper compares the discrimination results from the use of actual experience derived from the WES and those from the use of “imputed experience” derived from the General Household Survey, as this survey does not record actual work experience data. It is found that imputed work experience is in fact highly correlated with actual labour market experience, and results in similar estimates of women's earnings functions as well as similar discrimination measures. These results arise from controlling for sample selection bias using the methodology proposed by Heckman (1980). According to their conclusions, use of imputed work experience is the best solution to missing work experience data, as the alternatives such as selecting only single males and singles females, or even using potential experience, add many more potential issues needing to be corrected.

In Lambert (1993), four different measures of women's labour experience are compared through the effect these different measures of experience have on the coefficients of an estimated wage equation. These measures are: age, the Mincer proxy (age minus schooling minus five), the Mincer proxy plus adjustment for the presence of current dependent children, and the Mincer proxy plus adjustment for presence of current and past dependent children. As well as this, the results derived from using Heckman (1976,1979)'s Two-Step probit model to correct for censoring bias are also investigated in this paper, as this technique is widely used to correct for various sample structure issues. It reports that the results are very sensitive to the specification of the labour market experience measure. In fact when experience is incorrectly measured, a censoring bias occurs, due to the poorly specified experience variable itself. This misspecification of an important independent variable causes the two-step probit model to be non-robust when encountering heteroskedasticity in the data.

Kidd and Shannon(1997)

Kidd and Shannon use Australian data from the National Social Science Survey of 1984 which is devoid of work histories to evaluate the efficacy of imputing labour market experience of females using the Zabalza and Arrufat method versus using the traditional Mincer proxy. It is found that imputing for labour market experience provides a more accurate measure of actual labour market experience than the Mincer proxy, but it must be noted that the precision of this imputation method is sensitive to the identification restrictions used.

Propensity Score Matching to control for selection bias vs. Heckman selection bias correction

Nopo (2004)'s paper uses more up to date techniques and uses matching on characteristics rather than on propensity scores, as well as a non-parametric regression design. This design is an alternative to the usual Oaxaca blinder decompositions requiring no estimation of earnings equations, thereby removing many of the issues presented by choosing the appropriate covariates for the equations. The issue of difference in supports between males and females is underlined, and a matching methodology is made use of to take into account differences in the supports.

The Heckit or Heckman Selection Bias correction is used by Albrecht et al. (2009) to address the “potentially serious” issue of sample selection, as employment rates are likely to vary by gender. To achieve this they build onto the technique used by Machado & Mata to account for this sample selection, using data on full-time workers in the Netherlands. Albrecht et al. (2009) also prove that the quantile regression decomposition procedure followed by Machado and Mata generates consistent and asymptotically normal estimates of the quantiles of the counterfactual distributions they are meant to predict.

British Evidence

French Evidence

Spanish Evidence

Del Rio et. al (2006) study the level of poverty as well as wage discrimination due to gender in Spain. Using household data from the 2001 round of the “PHOGUE”, which is the European Household Panel Survey, they analyse the wage gap's effect on the Spanish household's distribution of earnings. They go from the individual towards the household, first estimating the wage gap for each working female, then building a counterfactual distribution of household's earnings should the working female not suffer wage discrimination. Then the levels of poverty and earnings inequality can be compared between the actual distribution of earnings and the counterfactual distribution.

The Distribution of the Gender Wage Gaps and Labour Market Outcomes

International Evidence

Buchinsky (1998), by focussing on changes at different points along the female wage distribution, analyse female wage structure for the USA. They use data taken from the March Current Population Survey over several years[1], and use quantile regression methods to derive their results. They find that the most significant changes occurred at the bottom of the distribution for the less-skilled women, whereas for the more skilled women, these changes occur at both top and bottom of the distribution. As a result, it can be seen that wage inequality increase for young college graduates, but decreased for high-school graduates, and the most important gain in wages was experienced by the more skilled women independently of their position in the distribution of wages.

Using US data for 1973 and 1989, Melly (2005a) applies the semi-parametric estimator of distribution functions with covariates present, which he derived in his previous paper (Melly, 2004). This estimator is based on the estimation of a conditional distribution by quantile regression, and the integration of this conditional distribution over the range of covariates. This enables the generation of counterfactual distributions, which can then be used for the decomposition of differences in distribution (in this case across time, but it can be used across gender or any other binary variable) into the three components affecting wage distribution: coefficients (the effect of the price assigned to each covariate), characteristics (the effect of the covariates themselves), and the residuals (the effect which can be explained by neither coefficients or characteristics). This method is used to evaluate the sources of change in wage inequality in the US between the 1973 and 1989. The authors find that only 20% of the explosion in wage inequality in this period can be explained by the effect of the distribution of residuals on the distribution of wages.

Melly (2005b) uses the above method to analyse the private-public sector wage differential and its distribution....

Spanish Evidence

Gardeazabal and Ugidos (2005) use quantile regression methods to evaluate discrimination in the Spanish labour market. Their motivation is due to the fact that existing discrimination measures show discrimination as equal across the distribution of wages, and two equal discrimination measures from different groups of people do not necessarily imply that it is distributed the same in these two groups. They therefore extend the Oaxaca decomposition to be applied to quantiles of the wage distribution rather than the mean. It is important to note that they still use values of the regressors which are conditional on a certain wage quantile, which may be erroneous. According to the authors this measure of discrimination can then be used to compare not only discrimination within the same population, but also between different populations. Their results using the Spanish Survey of Wage Structure indicate that gender wage discrimination increases as women move up the distribution. However the fraction of the gender pay gap explained by discrimination reaches its maximum at the 9th percentile.

Policy Evaluation

Wage gap effect on benefits

Unequal Pay Legislation

  1. BLAU, F. D. & KAHN, L. M. 1992. The Gender Earnings Gap: Learning from International Comparisons. The American Economic Review, 82, 533-538.
  2. BLAU, F. D. & KAHN, L. M. 2001. Understanding International Differences in the Gender Pay Gap. National Bureau of Economic Research Working Paper Series, No. 8200.
  3. BLINDER, A. S. 1973. Wage Discrimination: Reduced Form and Structural Estimates. The Journal of Human Resources, 8, 436-455.
  4. BUCHINSKY, M. 1998. The dynamics of changes in the female wage distribution in the USA: a quantile regression approach.
  5. GARDEAZABAL, J. & UGIDOS, A. 2005. Gender wage discrimination at quantiles. Journal of Population Economics, 18, 165-179.
  6. MACHADO, J. A. F. & MATA, J. 2005. Counterfactual decomposition of changes in wage distributions using quantile regression. Journal of Applied Econometrics, 20, 445-465.
  7. MELLY, B. 2004. Decomposition of differences in distribution using quantile regression.
  8. MELLY, B. 2005a. Decomposition of differences in distribution using quantile regression. Labour Economics, 12, 577-590.
  9. MELLY, B. 2005b. Public-private sector wage differentials in Germany: Evidence from quantile regression. Empirical Economics, 30, 505-520.
  10. MELLY, B. 2006. Estimation of counterfactual distributions using quantile regression.
  11. OAXACA, R. 1973. Male-Female Wage Differentials in Urban Labor Markets. International Economic Review, 14, 693-709.
  12. SUEN, W. 1997. Decomposing Wage Residuals: Unmeasured Skill or Statistical Artifact? Journal of Labor Economics, 15, 555-566.

Please be aware that the free essay that you were just reading was not written by us. This essay, and all of the others available to view on the website, were provided to us by students in exchange for services that we offer. This relationship helps our students to get an even better deal while also contributing to the biggest free essay resource in the UK!