A Distributional Analysis of the Gender Wage Gap in Bangladesh

Abstract

This article empirically investigates the gender wage gap in Bangladesh during the period 2005–2009. Applying unconditional quantile regression models, the article demonstrates that women are paid less than men throughout the wage distribution and the gap is higher at the lower end of the distribution. Discrimination against women is the primary determinant of the wage gap. The article also demonstrates that the observed gender wage gap is likely to be underestimated if we ignore selection in full-time employment. A number of policy implications are discussed.

Acknowledgements

The authors thank the editor and two anonymous referees for their incisive comments and suggestions, which have vastly improved the article. The authors are also grateful to Mark McGillivray, participants at the Applied Econometrics brown bag at Monash University, at the University of Melbourne and participants at the Australian Conference of Economists and at the Econometrics Society of Australasian Meetings for useful comments. Data and stata codes are available upon request from Dr Salma Ahmed. Any remaining mistakes in the article are the authors’ own.

Notes

1. Wage regressions are unweighted and are estimated separately for men and women in order to allow for different rewards by gender for a set of productive characteristics or endowments. A Chow test (F-test) rejects the null hypothesis that explanatory variables have equal impacts on the wage rates of males and females for both years. The Chow test statistic is 8.82 (p = 0.000) for the survey year 2005 and 6.16 (p = 0.000) for 2009. The results remain unchanged when we use sampling weights.

2. In our analysis, we present and discuss the results corresponding to the case where the male wage rate is the reference category. In doing so, we have abstracted from an important debate: which wage structure should we use as the reference category? The Oaxaca–Blinder method applied both the male and female wage structures as the reference category. This creates an index number problem, as the estimate of the discrimination component differs depending on the choice of the reference category. Further, the resulting levels of discrimination provide a range within which the actual level of discrimination falls. Reimers (1983) hypothesises that the correct procedure is rather to take an average of the male and female wage structures. Cotton (1988) suggests improving upon the procedure by employing a weighted average of the two wage structures, which should then provide us with an exact figure rather than a range. In contrast, Neumark (1988) regards these benchmarks as unsatisfactory and argues that the choice of a non-discriminatory wage structure should be based on the OLS estimates obtained from a pooled regression of both males and females. However, Ginther and Hayes (2003) point out that a pooled wage structure (that is, an average of the male and female wage structures) is not likely to be used in legal frameworks that are concerned with equal opportunities for women and men. Rather, the authors argue that men are the usual comparison group in legal proceedings concerning gender discrimination. This assumption is also reasonable in our context, as the majority of the workforce in Bangladesh is male.

3. Particular care must be taken when interpreting the model residual as discrimination throughout the article. This is because the unexplained wage gap, which is often termed as discrimination, includes the effect of labour market discrimination, unobservable variables (for example, motivation) and omitted variables. The latter effect might mean that if an omitted variable has a positive effect on wages and if men are more highly endowed with this variable, the results obtained from the decomposition would overestimate discrimination. Alternatively, if some of the factors in the model are themselves affected by discrimination, then the analysis could underestimate discrimination. For example, if women have less access to the types of schooling that are deemed more valuable by the market, then the decomposition may underestimate discrimination.

4. A small but growing body of literature has adopted the unconditional quantile regression methodology to examine (and decompose) gender differences in wages across the wage distribution; for example, Chi and Li (2008) for China, and Kassenboehmer and Sinning (2014) and Heywood and Parent (2012) for the US.

5. This first step of the decomposition is semi-parametric because it does not assume any functional form for the wage distribution.

6. The unconditional properties of the wage function can be obtained by averaging the wage function over ; see Chi and Li (2008).

7. As discussed in Firpo et al. (2009), it is important to compute approximation errors in order to determine whether the linear model is well specified.

8. The results are similar to those presented by Al-Samarrai (2007) using the HIES 2005 datasets for Bangladesh.

9. The wage regression results based on the imputed sample are not reported, but these results are available from the authors upon request.

Additional information

Funding

This work was supported by the Australian Research Council Discovery Scheme [Grant DP120101781].