Gender Wage Gap: Do Occupational Differences Play A Role?

Sing Rou Lee Mengfei Li Hui Wang September 9, 2012

Introduction The war for gender equality in regards to having equal pay has been ongoing since before the Equal Pay Act was passed in 1963. John F. Kennedy passed the Act in hopes that it will at least reduce if not abolish wage disparities based on gender. Unfortunately, the gender wage gap still continue to exist even though it has reduced signicantly since the Act was passed. In 2009, the ratio of women to mens average wage is 77%, a slight increase from approximately 74% in 2000. Time invested in education and additional working hours also contributes to the decrease in gender wage gap [7]. We hope with this paper, we are able to identify additional factors affecting the gender wage gap by incorporating occupational attainment into the existing models which will be discussed in the subsequent sections. Over the years, there has been numerous research investigating the factors behind wage differentials between men and women. Both working experience and education have appeared dominantly in most papers investigating the wage gap between men and women (see [6] [2]) amongst many others. In this paper, we are going to investigate the determinants of the gender wage gap within and across different occupations. In our previous work, we included dummy variables which account for the various occupational categories in our regression. This clearly operates under the assumption that the difference in the occupational distribution between men and women is negligible. We seek to improve our work from last semester by applying a different method to estimate the wage differential by incorporating occupatinal attainment in the model. The remainder of the paper is organized in the following way: Section 1 gives a literature review on the literature that has guided our work in this paper and also mentions the various methods and improvements that the authors have discovered and made throughout 1

the years. Section 2 develops the theoretical model that we will apply empirical data on and mentions the basic Mincerian equation which expands to the wage discrimination and decomposition model that has been introduced by Oaxaca and Blinder [6] [2] and further improved by Brown, Neumark and Appleton [3] [5] [1]. Section 3 gives the description of the dataset and explanatory variables and demonstrates the link with the theoretical model in Section 2. Section 4 gives the empirical application with the results obtained and interpretations on the results. Finally, Section 5 concludes the paper.

Literature Review

It has been long known that the gender wage gap is no longer only attributable to the fact that there exists unequal pay for equal work [6]. In the early 1970s, women are more concentrated in lower paying jobs, [6] but with increased higher education opportunities, we are beginning to see an increase in womens participation in the labor force. The most basic method employed is the decomposition method proposed by Blinder-Oaxaca. [6][2]. The wage gap is estimated by considering the difference between the coefcients of a separately regressed wage equation [4] for both men and women sample. [6][5] This method has proven to be extremely useful in studying gender discrimination in pay rates but most wage determination models only take into account personal characteristics as the explanatory variables with the unexplained portion indicating wage differences due to discrimination. [3] Brown [3] incorporates the existing models for gender wage gap and models that studies the effect of occupational barriers to develop a better rened model to explain men and women wage differentials. In his paper, Brown introduced a separate model to account for occupational attainment to be included in the decomposition of the men and women wage equations. By using a multinomial logit model, Brown simulated the occupational distribution for women had discrimination not exist. This is done rst by estimating the occupational attainment for men and applying them to womens personal characteristics to produce the estimation. Naturally, this is under the assumption that both men and women possess the same structure in their occupation options. Both the models proposed by Oaxaca and Brown has been improved by ways of reestimating an independent set of coefcients that can be used to determine both the occupational distributions of men and women without being subject to the previous assumptions. Later, in Neumark and Appletons papers, they addressed a new way which add a weight to estimate the nondiscriminatory parameters to solve the index number problem.[1] [5] The Mincers equation and the aforementioned models are the main ideas that this paper is working off to investigate the determinants behind the wage differentials across and within

occupations.

The Model

The Mincers wage equation will be the basic model that will be used to break down the wage differentials between the men and women sample. The classic Mincers wage equation measures the relationship between log wages, working experience, working experience square, education and other explanatory variables as follows:
2 ln wi = 0 + 1 Xiex + 2 Xiex + 3 Xied + t Xit + i t

(1)

where Xt is the list of individual characteristics variables, i is a disturbance term. By the popular wage decomposition method introuced by Oaxaca-Blinder [6] [2], we can write the wage differential as: ln wm ln ww = Xm m Xw w = (Xm Xw )m + Xw (m w ) (2)

whereby both Xm and Xw contains the means of observable personal characteristics for the men and women sample with m and w being vectors of the OLS estimated coefcients. Notice that the equation above does not take into account differences in the occupation distributions between men and women. Here we would like to introduce the idea proposed by Brown [3] to account for the occupational distributions for both the men and women sample. Brown proposed the following wage decomposition model: ln wm ln ww = Pjm lnw jm Pjw lnw jw
j j

(3) (4) (5)

= Pjw (X jm X jw ) jm + Pjw X jm ( jm jw )
j j + (Pjm Pjw ) ln w jm + (Pjw Pjw ) ln w jm j j

such that Pi j is the probability of person i being employed in occupation j calculated using a multinomial logit model: Pi j = eXi j (6)

eXi j
j

whereby Xi is a vector of given personal characteristics for individual i and the probability Pi j is calculated conditional on Xi . We can then use equation (6) to calculate the occupational distributions for women assuming that the occupational distribution structure for the men sample is non-discriminatory and justied. We would then have:
Pjw

eXw jm

eXw jm
j

(7)

whereby the occupational distribution for the women sample is estimated for when women face the same occupational structure for men given their observable personal characteristics. Equation (4) represents the wage gap within occupations, i.e. the wage gap in the jth occupation whereby the rst term can be explained due to differences in observable personal characteristics between the men and women sample. The differential from the second term is the remaining differential that is not explained by the differences in observable personal characteristics and can therefore be attributed as a discriminatory factor. The wage gap across different occupations is represented by equation (5) whereby the rst term is explained by the differences in observable personal characteristics that affect occupational attainment and analogously, the second term represents the residual differences between the occupational structures estimated for the men and women sample which can be considered also as the discriminatory portion. The Browns decomposition model clearly operates under the assumption the wage structure for men is non-discriminatory and estimates what women would receive given that they face the same wage structure as men. However, the reverse can also be estimated but yielding very different results. [1] That is to say, we can also estimate what the men sample would receive given that they face the same wage structure as women (i.e. wage structure for women ) but would give us results that are far from close. In light of that, Neumark [5] proposed a way to estimate, , the non-discriminatory wage structure without being subject to the assumption that either the men or women sample possess this property. In his paper, Neumark demonstrated that under the assumption whereby employers preference towards men and women labor is homogenous of degree zero, employers are only affected by the proportions of men and women labor they hire. [5] [1] Hence, the non-discriminatory wage structure, , can be estimated by using the combined men and women samples which yields the weighted average of both the men and women samples as follow [5]: = m + (I )w (8)

with the weights [6] being dened as: = (X X)1 (Xm Xm ) (9)

given that X is a matrix of observable personal characteristics for both the men and women sample and Xm is naturally the observable personal characteristics for the men sample.I is an identity matrix. This leads to the decomposition proposed by Neumark whereby the rst term is the portion explained by differences in observable personal characteristics for both men and women. Both the second and third term represents the differences between the nondiscriminatory wage structure and the wage structure of the men and women samples respectively: ln wm ln ww = (Xm Xw ) + Xm (m ) + Xw ( w ) (10)

By adding this newly estimated wage structure to Browns decomposition model, we have the following: ln wm ln ww = Pjw (X jm X jw ) j +
j

(11) (12) (13)

+ Pjw X jm ( jm j ) + Pjw X jw ( j jw )
j j + (Pjm Pjw ) ln w jm + (Pjw Pjw ) ln w jm j j

Equation (11) represents the wage gap decomposition within the jth occupation that can be explained by differences in observable personal characteristics between the men and women sample. Equation (12) represents the wage gap that exists but cannot be explained by observable personal characteristics and can be considered as arising from the discrimination between the men and women sample within the jth occupation. Finally, for equation (13), they represent the wage gap across occupations and is similar to the explanation before. Notice again, that we run into the same problem as before with estimating the nondiscriminatory wage structure for the men and women sample from our estimation of the occupational structure for the men and women sample. We estimate the occupational attainment for the women sample based on the occupational structure for men and just as before we can estimate the reverse and get completely different results. As proposed by Appleton [1] we can estimate the non-discriminatory occupation structure for both the men and women sample by adding weights to the combined men and women sample just as we did for the estimation of the non-discriminatory wage structure: = jm + (I ) jw j 5 (14)

whereby using the same weighting matrix as above, and I is an identity matrix. Given the new estimated non-discriminatory wage structure, the occupational attainment model becomes: Pij = eXi j
j

eXi j

(15)

Let Pj be the mean of Pij for all i, Pjm be the mean of Pij for male only, and Pjw be the for female only. We then have our full model as follows: mean of Pi j

ln wm ln ww = Pj (X jm X jw ) j +
j

(16) (17) (18) (19)

+ Pj X jm ( jm j ) + Pj X jw ( j jw )
j j + (Pjm Pj ) ln w jm + (Pj Pjw ) ln w jm j j

+
j

(Pjm Pjm ) ln w jm +

(Pjw Pjw) ln w jw j

Equation (16) and (17) represents the same decomposition as equation (10) and (11). Equation (18) takes into account the wage gap due to differences in observable personal characteristics that affects their occupational attainment. Equation (19) represents the wage differential from the actual and predicted occupational attainment that is not explained by observable personal characteristics and can be considered as the discriminatory portion for both the men and women samples.

Dataset

The dataset that we employ to perform analysis on the wage differential and occupational attainment between gender is taken from the March Supplement of the Current Population Survey(CPS) for both year 2000 and 2010. The CPS data is the primary source for labor force statistics in the United States and provides a detailed list of the occupational categories that we are interested in. The CPS data also contains information on demographics, wage, household characterictics and education level for an individual worker. For our empirical results, we have selected 10 different occupational categories for our analysis and they are as listed in Table 1. To facilitate our analysis, weekly wage data are computed for those with a full time labor force participation and individuals who are between the ages 6

Table 1: Denition of Occupation Categories Occupational Categories Occ1 Occ2 Occ3 Occ4 Occ5 Occ6 Occ7 Occ8 Occ9 Occ10 Management, Business, and Finance Professional Service Sales Ofce and Administrative Support Farming, Fishing, and Forestry Construction and Extraction Installlation, Maintenance, and Repair Production Transportation and Matertial Moving

Variables ln w HourPWk EducYr WkExp NumofFamHH MrtalSt Race Region

Table 2: Denition of Variables Description

Log weekly wages which are computed by diving yearly earnings and the amount of week Hours worked per week Years spent in education estimated according to their educational level Labor market experience computed by Age-Years of education - 6 (assuming that the scho Number of family in household Marital status (1 = Married, 0 = Others) Race (1=White, 0=Black) Region (Northeast, Middlewest, South, West) (Each region was recoded as a dummy varia

Figure 1: Wage gap across different occupations (2010) of 22 and 65 with more than 35 working hours in a week. Table 2 gives a full description of the variables that we used in our estimations. In Figure1, we can see that there is a signicant wage gap between men and women across all the occupations.

4 5

Empirical Analysis Conclusions

References
[1] Simon Appleton, John Hoddinott, and Pramila Krishnan. The gender wage gap in three African countries. In: Economic Development and Cultural Change 47.2 (1999), pp. 289313. [2] Alan S. Blinder. Wage Discrimination: Reduced Form and Structural Estimates. In: The Journal of Human Resources 8.4 (1973), pp. 436455. [3] Randall S. Brown, Marilyn Moon, and Barbara S. Zoloth. Incorporating Occupational Attainment in Studies of Male-Female Earnings Differentials. In: The Journal of Human Resources 15.1 (1980), pp. 328. [4] Jacob Mincer. Schooling, Experience and Earnings. Ed. by Jacob Mincer. Columbia University Press, 1974. 8

[5] David Neumark. Employers discriminatory behavior and the estimation of wage discrimination. In: Journal of Human Resources 23.3 (1988), pp. 279295. [6] Ronald Oaxaca. Male-Female Wage Differentials in Urban Labor Markets. In: International Economic Review 14.3 (1973), pp. 693709. [7] Jingyo Suh. Decomposition of the Change in the Gender Wage Gap. In: Research in Business and Economics Journal 1 (2010), pp. 118.