Gender Wage Discrimination in Pakistan

| Gender Wage Discrimination in Pakistan| Evidence from Pakistan 2008/09 and 2010/11| | | | Table of Contents Introduction2 Literature Review2 Methodology3 Variables Used – Characteristics of Workers5 Results7 Discussion7 Bibliography8 Appendix A9 Selectivity Bias Logit Regression Results:9 Introduction This paper explores the dynamics of gender wage discrimination in Pakistan for two data sets; Labour Force Survey for the year 2008/09 and 2010/11. We will explore whether or not women are discriminated against, as it has been suggested for a predominantly Islamic country like Pakistan.
Labour theory addresses many reasons for wage discrimination. For the purposes of this research we will concentrate on ‘employer wage discrimination’. Following this our research will be aimed at discovering if women are paid less than their male counter-parts especially with the same set of characteristics. For this purpose we will use the Oaxaca-Blinder method to calculate the coefficient for discrimination across genders. Literature Review The basis of this paper is the work done by Oaxaca and Blinder in 1973 about wage discrimination models.
In the paper ‘Interpreting the Decomposition of the Gender Earnings Gap’ (Giaimo R. 2007) this method has been applied to find out how different characteristics change the discriminatory behaviour of employers in Italy. Oaxaca’s method for calculating discrimination was further adapted in the paper ‘Gender Wage Discrimination at Quantiles’ (Javier Gardeazabal 2005), and was used to calculate discrimination coefficients for quintiles. In a study conducted in India (Tilak 1980), it was found that the incidence of unemployment was higher for women than for men with the same characteristics.

In this study the only characteristic that was taken was education. This is a different angle to look at discrimination from what this paper will do. Rather than looking at the unemployed, this paper will see the women in the labour force and if they face discrimination with respect to their wages. However, the underlying aim and also the hypothesis formed are the same. The paper ‘Wage Differentials and Gender Discrimination: Changes in Sweden 1981-98’ (Mats Johansson 2005) explored the wage gaps between men and women in Sweden.
They found that the wage gap was 14%-18% during the 1990’s. Their study also indicated that this difference could not be explained by applying the job requirements and qualifications to women’s wage function. The conclusion was that there is undoubtedly some other factors other than the characteristics of the workers that determined the wages in the Labour Market. Methodology This paper calculated a coefficient for Gender Wage Discrimination from the Oaxaca-Blinder decomposition. D= Xf? m-Bf+ ? m(Xm-Xf) Here ‘? ’ is a vector of characteristics of workers.
Therefore, the first part of the equation shows the wage differential between males and females on the basis of characteristics. Second part of the equation normalizes characteristics, for females in this instance, and then subtracts the wage differential based on characteristics, to give us the overall differential based on discrimination. As a control, we also work out the converse of this Oaxaca Blinder Decomposition as follows; D= Xm? m-Bf+ ? f(Xm-Xf) To control for selectivity bias, we have also used the ‘Heckman Procedure’.
A multi-variable Logit model was run and three variables (Lambda1, Lambda 2 and Lambda3) were calculated to act as control for variables missed in our model. This discrimination coefficient has been calculated for two data sets using characteristics such as age, marital status, education level, province, region, professional trainings and status in the family. These characteristics have been selected after being shown significant as the determinant of wage. Natural log of wages was the dependant variable in the following regression; Table [ 1 ]: Wage Determinants – LFS 2008/09
Table [ 2 ]: Wage Determinants – LFS 2010/11 Our results are much better for the data set of 2010/11. The signs of education are expected. For the data set of 2008/09, signs for education are positive which does not support theory. Even after efforts to remove multi-colinearity, they still show positive signs. Most of the variables in the regression are also insignificant. However, when we take the data for LFS 2010/11, and correct it for selectivity bias, we get much better results. Most of the variables are significant as well as show the correct signs. The same algorithm was applied to both the data sets, and the same variables have been taken). Results of Logit models for correcting selectivity bias are attached in Appendix A. Variables Used – Characteristics of Workers Summary tables from LFS 2010/11 1. Age * Theory suggests that this is one of the most important determinants of people’s decision to work. 2. Marital Status * This variable was taken as a dummy variable in the regression. * It is a significant variable in the decision to work, especially in developing economies like Pakistan. 3. Province This is also taken as a dummy. The Baluchistan province was omitted from this analysis. However, the calculations of the Oaxaca Blinder method take this omitted variable into account. This is because the method takes the vectors of the estimated regression equation. 4. Region * Whether a person is from a Rural or urban background has impact on the opportunities and the job growth pattern. 5. Education Level * This is linked directly with the variable wage. * This is again taken as a dummy variable, and higher education was omitted from the regression. 6. Migration (Rural-Urban) Although not a very significant variable in our regression, there are other empirical studies that have shown how the migrated families have better opportunities for work than those who do not. 7. Literacy * This is a dummy variable, and is significant in our analysis. 8. Selectivity Bias Variables * These are Lambda’s in the model. And have been calculated using the Heckman Procedure for controlling selectivity bias. Results To find the discrimination coefficient a matrix exercise was done in Stata using the data from LFS 2007/08. This presented the following equation; D= Xf? m-Bf+ ? Xm-Xf D=10. 030812+-7. 4166332 D= 2. 614212 The discrimination coefficient for LFS 2010/11 was calculated as follows: D= Xm? m-Bf+ ? fXm-Xf D=0. 11964462+0. 31341527 D= 0. 43305989 Just looking at the numbers we can say that discrimination have gone down significantly over the last two years. Whether this is actually the case, or this is just due to the problems in the data, we cannot be sure. However, we think that the result for 2010/11 is a better estimate overall. The results show that women are at a significant disadvantage in Pakistan’s Labour Force. These results are quite expected.
However, we also need to take the problems in data collection and measurement into account. Many of the cottage and small scale industries are not counted in the LFS and they are a prime source of employment for women in Pakistan. Discussion There are many limitations of this study. First of all this can be made more powerful if panel data is used, however, there are no sources of such data. Secondly, an easy method of expanding this study would be to do an inter year comparative study. There are more limitations that are related directly to the data that we have used.
Many questions have been raised about the methodology and the authenticity of the data in Labour Force Survey of Pakistan. However, this limitation is beyond our control. There have also been questions raised about the Oaxaca-Blinder method of calculating wage discrimination. While we have attempted to review paper that have used this technique and have achieved good results, there are still many questions about the technique, still. There are few policy implications that we can derive from these results, especially if we look at the significance levels in the data for 2008/09.
However, this paper does prove to some extent that there is a problem of gender wage discrimination is Pakistan. We can attribute a lot of this to social factors as well; women do not want to work in most professions, so we can also argue that there may be a case for discrimination by the employees rather than the employers. Bibliography Giaimo R. , Bono F. , Lo Magno G. L. “Interpreting the Decomposition of the Gender Earning Gap. ” University of Palermo Journal, 2007. International Standard Industrial Classification of all Economic Activities (ISIC-Rev. 2, 1968). ILO. 2012. http://laborsta. lo. org/applv8/data/isic2e. html (accessed 2012). Javier Gardeazabal, Arantza Ugidos. “Gender Wage Discrimination at Quantiles. ” Journal of Population Economics, 2005. Mats Johansson, Katarina Katz, Hakan Nyman. “Wage Differentials and Gender Discrimination: Changes in Sweden 1981-98. ” Acta Sociologica, 2005. Stat. Stata. 2012. http://www. stata. com/meeting/5german/SINNING_stata_presentation. pdf. Tilak, Jandhyala B. G. “Education and Labour Market Discrimination. ” Indian Journal of Industrial Relations , 1980. Appendix A Selectivity Bias Logit Regression Results: LFS 2008/09 LFS 2010/11