Gender Wage Discrimination in Rural and Urban Labour Markets of Bangladesh

We would like to thank Anu Rammohan, Rebecca Valenzuela, Mark Bryan, Paul Miller and an anonymous referee for their comments and suggestions. The usual caveat applies.

Abstract

Female wages in Bangladesh are significantly lower than male wages. This paper quantifies the extent to which discrimination can explain this gender wage gap across the rural and urban labour markets of Bangladesh, using unit record data from the 1999–2000 Labour Force Survey. The gender wage differential is decomposed into a component that can be explained by differences in productive characteristics and a component not explained by observable productive differences, which is attributed to discrimination. An attempt is also made to improve on the standard methodology by implementing a wage-gap decomposition method that accounts for selectivity bias, on top of the usual “explained” and “unexplained” components. Analytical results from this paper show that gender wage differentials are considerably larger in urban areas than in rural areas and a significant portion of this wage differential can be attributed to discrimination against women. The results also show that selectivity bias is an important component of total discrimination.

Notes

We would like to thank Anu Rammohan, Rebecca Valenzuela, Mark Bryan, Paul Miller and an anonymous referee for their comments and suggestions. The usual caveat applies.

¹ Kapsos (2008) in his analysis uses data from the Bangladesh national occupational wage survey, which was conducted by the Bangladesh Bureau of Statistics in 2007. That survey collected data on wages and hours of work from establishment surveys and establishments in the survey were not selected randomly.

² Wage equations are estimated separately for men and women in order to allow for different rewards by gender to a set of productive characteristics or endowments.

³ A similar method has been used by Rodgers (2004).

⁴ The selection equation represents the probability that an individual is employed at a given time conditional on a set of personal characteristics.

⁵ In other words, a significant estimate of θ _j , the coefficient on the selectivity term, indicates the presence of selectivity.

⁶ See, for example, Creedy et al. (2000) and Kidd & Viney (1991).

⁷ This part may be viewed as the differences in unobservables, which influence wages.

⁸ See also Duncan & Leigh (1980), Boymond et al. (1994) and Reimers (1983).

⁹ The lower boundary of the working age group is 15, that is, we do not account for child labour. The upper boundary of the working age group is higher than the conventional retirement age of 60–65 years. This is due to the inclusion of the rural areas where individuals tend to work beyond their conventional retirement age.

¹⁰ A casual worker refers to a wage worker whose services are solicited only for a periodic time intervals during the reference period (i.e. the week preceding the day of the survey).

¹¹ An unpaid family worker is a person who works at least 1 hour in the reference period (other than household work) without pay or profit in a family-operated farm or in a business owned/operated by the household head or other members of the household to whom he/she is related by kinship, marriage, adoption or dependency. Unpaid family workers who worked at least 1 hour or more during the reference period are considered as a part of the labour force.

¹² As we use an extended definition of the labour force, persons with non-market activities such as unpaid family workers are also included.

¹³ The detailed definitions of the explanatory variables are presented in the Appendix.

¹⁴ We do not have any information on actual labour market experience. Age is used as the approximate variable for general labour market experience. Moreover, as age increases, productivity and wage rates tend to rise; but further increases in age may lead to a decline in wage rates and productivity because of diminishing marginal returns. To capture the concavity of the wage profile a quadratic age term is included.

¹⁵ This variable measures the gross elasticity of hours worked per month with respect to wage rates. Ajwad & Kurukulasuraiya (2002) also employ this variable in their study on ethnic and gender wage disparities in Sri Lanka. The authors found that the log of hours worked has a negative and significant impact on wage rates.

¹⁶ It has been hypothesized that time out of the labour force can result in depreciation of human capital and depress wage rates (Mincer & Polachek, 1974).

¹⁷ Here skill refers to occupational skills.

¹⁸ One possible test is presented in Bryan (2007). To gain confidence that the variables included in Z, but not included in X are not actually incorrectly excluded from X, following Bryan (2007) we regressed the residuals from the selection-corrected wage equation on all the exogenous variables and then calculated the test statistic NR ², where N is the sample size and the R ² is from this supplementary regression. The statistic was then compared with the appropriate critical value from the x ²(k − 1) distribution, where k is the number of excluded variables. Under the null hypothesis, the instruments are uncorrelated with the error term, and excluded instruments are truly exogenous to the main (wage) equation. In both the rural and urban samples, the p-values (0.10 for the urban sample and 0.33 for the rural sample) of the test confirmed the validity of excluded instruments used. We conducted another test: we re-ran the regressions but this time we included all of the excluded variables in the wage equation and examined: (1) whether the remaining coefficients are sensitive to the inclusion of these additional variables; and (2) whether these additional variables are statistically significant in the wage equation. The results indicate that including the set of identification variables in the wage equation does not alter our initial estimates. These excluded variables are not statistically significant in the wage equations, implying that these variables do not have a direct effect on wage rates.

¹⁹ One of the major drawbacks in our data set is that we do not have information on pre-school-age children, which would not only be a deterrent to women for being employed but also influence their engagement with the labour market.

²⁰ Differences in the natural logarithm of hourly wages can be converted to percentage wage differences using the formula 100[exp (difference) − 1]. Hence, the difference between the mean of the natural logarithm of hourly wages for males and females in the urban sample in Table 1 of 0.6703 yields a 100[exp (0.6703) − 1] = 95% wage differential. It is also worth noting that the gender wage differential in our urban data is significantly larger than that found using similar methodologies in other studies: Loureiro et al. (2004) found a 20 and 61% gap in the urban labour market of Brazil in 1992 and 1998, respectively, while Ashraf & Ashraf (1993) found a 61.14% gap in Rawalpindi City, Pakistan.

²¹

²² SSC: Secondary School Certificate; HSC: Higher Secondary School Certificate.

²³ The results indicate that additional income from these assets increases the male's probability of employment compared with a situation in which he has “no asset” (the omitted category). One explanation could be that in the absence of the value of the assets (which are not provided in the sample), one cannot gauge the importance of feedback effects of these assets on the probability of employment.

²⁴ A number of studies have used the Heckman (1979) estimator, but the estimates of the coefficient on the λ variable in the male wage equation differ considerably: Miller & Rummery (1991) reported a significant, negative coefficient, Ashraf & Ashraf (1993) a positive, significant coefficient.

²⁵ Oaxaca & Ransom (1994) found similar results.

²⁶ This might indicate that the adverse consequences of a deregulated labour market are more pronounced for those women at the lower end of the wage distribution.

²⁷ In Bangladesh, the full work week is 48 hours.

²⁸ Notice that the discrimination component of column of 1 of Table 7 is negative when we use male wage structure as the non-discriminatory norm. In the Blinder decomposition, discrimination is defined as (. Now when we consider the sample of full-time workers the value for becomes large (negative), resulting in the discrimination term becoming negative. Interestingly, this is not a new result. Using a different set of data, Akter (1999) found a similar result (see table 8, p. 130) for the rural labour market of Bangladesh when she used male wage structure as the non-discriminatory norm. Unfortunately, we do not have a proper explanation for this negative discrimination component. All we can say is that the results for the rural sample are quite sensitive to the specification used.