Does Having Children Make Cents? An Economic Analysis of the Gender Wage Gap in Nevada

Jeffrey Wheble and Dr. Djeto Assané

University of Nevada, Las Vegas. Department of Economics.

DOI: http://dx.doi.org/10.15629/6.7.8.7.5_4-1_S-2018_1

Citation: Wheble, Jeffrey, Assané, Djeto. “Does Having Children Make Cents? An Economic Analysis of the Gender Wage Gap in Nevada .” Nevada State Undergraduate Research Journal. V4:I1 Spring-2018. (2018). http://dx.doi.org/10.15629/6.7.8.7.5_4-1_S-2018_1

How to Cite: 

Author last name, Author first name first initial. “The name of the article in parenthesis.” Nevada State Undergraduate Research Journal. V(volume):I(issue) Semester (Fall or Spring)-Year. (2018). http://dx.doi.org/[Insert DOI here].

Abstract:

I examine Jane Waldfogel’s hypothesis regarding the family pay gap in Nevada. Women have made progress in human capital equality with men, yet there remains an unexplained gap. The salient summary statistics illuminate why there remains a pay gap before moving on a pair of ordinary least squares (OLS) models split by gender. A Blinder-Oaxaca decomposition was performed to uncover how much of the gap in this sample is due to traits differences and how much comes from the treatment of traits (i.e. discrimination). A total wage gap of 31% is found, about two-fifths of which is due to discrimination. The family pay gap hypothesis finds some evidence in this analysis via marriage. Human capital and occupational choices are also found to be important in explaining the wage gap in Nevada. The implications of these results suggest further research into the differences in labor market choices, maternity leave policy, and differing returns to education are offered.

Introduction

Despite the skills convergence discussed by Goldin (2014) which brought the gender pay gap from 60% to 76% (Blau & Kahn, 2000), the difference between the wages of men and women calls out for explanation. This paper focuses on testing the hypothesis offered by Jane Waldfogel (1998) and several labor economists worldwide: children and the family structure are behind the remaining wage gap. Waldfogel (1998) argues that a lack of support in the US for mothers is the cause of the wage gap. Mothers often bear the cost of children, which deters from their careers. Intuitively, the idea of unequal burden-sharing makes sense. If the mother and father both work and have an equal interest in raising those children, there are still more ways to unequally divide labor than to do so equally. The distribution of hours spent with children might be thought of as standard-normal, where the mean is equal division and the tails are skewed towards one parent or the other spending more hours with children. If that distribution is shifted by social pressures on women to take on the responsibilities of children, then one can easily expect that more women will be stuck with an unequal burden than fathers. Time is the scarcest resource of all, and when one partner in a family must make the tradeoff between time with kids and time at work, it follows that the partner saddled with the extra responsibility will lose out. Without maternity leave, Waldfogel (1998) argues that mothers also experience a delay in their careers while fathers continue with theirs because women are, as mentioned earlier, usually burdened with childcare. I will refer to Waldfogel’s explanation broadly as the “family pay gap” in this paper.

Since the family pay gap research was initially completed, there have been other analyses focused on the family and its influence on wages. A decade after the Waldfogel work, a paper by Miller (2009) addressed the possibility that when one has children matters. Implicitly, this means that children matter to wages. This later work offers some support to Waldfogel’s hypothesis via unequal burden effects; it is not the number of children, but the mere fact of having them that impacts women’s careers. In other words, each child after the first has a diminishing marginal cost to the mother in terms of lifetime wages. Miller (2009) points out that having a gap in one’s career without maternity leave and legal protections (which are not protected under the law in the United States) stunts one’s capital accumulation, which has long-term consequences, such as capital compounds. The earlier the stunting, the greater the effect (Miller, 2009).

Another analysis by Budig and England (2001) details four main routes by which motherhood may hurt women’s wages. Mothers may, by taking time off to care for children, be trading job experience for child-rearing. This results in lowered wages, since experience is strongly linked to pay (Budig and England, 2001). One route which echoes Goldin’s (2014) flexibility preference explanation is that mothers prefer jobs that are suitable for their need for a schedule which might allow them to work and raise children (Budig and England, 2001). The remaining two routes are lower productivity due to motherhood and outright discrimination.

Recent research by the Department of Labor also refers to the family to explain part of the gap. Goldin’s convergence of human capital leaves much to be explained since the gap is widest among those with higher levels of education (“Women’s Earnings and the Wage Gap”, 2016). Furthermore, the gap widens as people grow older, which would complement the idea that children disrupt the careers of women; if there is an interruption early in life, perhaps between 25 and 35 years of age, then that effect will affect those older than 25-35 much more than those younger on average (“Women’s Earnings and the Wage Gap”, 2016).

Goldin’s (2014) oft-referenced explanation compliments Waldfogel’s explanation: job preference may differ between genders via differing preferences for flexibility, which would impel women to seek part-time work in more flexible fields at a higher rate than do men. That explanation may strongly relate to children because the possibility of having children and the social expectations of handling childcare could impose extra costs on women, which men are not concerned with. That difference in costs, which generally involves the time-commitment of children and the potential loss of career advancement, could explain why women make different occupational choices than men, if they do.

The preceding analyses of the gender wage gap will be applied to Nevada, a frontier state in terms of demographics and as a service-oriented economy. Focusing on a single state will allow this research to address the wage gap without distortion from unobserved factors and the methods may be later extended to other states for comparison in later research. I first illustrate the pair of models with which I attempt to analyze the wage gap in Nevada. I use the Nevada subset of the American Community Survey (ACS) from 2014, a sample which describes 12,430 workers. The data set has been cleaned to eliminate outliers. Summary statistics will be provided to contextualize these results in Nevada. I proceed to perform several regressions and analyze the results with respect to the hypotheses in question. I compare means and OLS regression coefficients of female and male subsets of the data set, examine joint significance for each category of variable in the model, and perform a Blinder-Oaxaca decomposition by the female variable.

Summary Statistics

By 2060, some estimates assert that the US census will reflect the results from Nevada (Kolko, 2017). Some selected summary statistics are presented in Table 1.

Variable Mean S. d. Min Max
Wage ($/year) 41738.29 46608.79 520 355000
Part-time (<35 hours/week) .2656476 .4416952 0 1
Bachelor’s degree .2501207 .4330998 0 1
Married .5267900 .4993019 0 1
Age 42.91022 14.35152 16 91
Black .0625101 .2420894 0 1
Hispanic .2180209 .4129183 0 1
Asian .0817377 .2739758 0 1
Female .4769107 .4994867 0 1
Food and Bevarage .0938858 .2916815 0 1
Entertainment .0179405 .1327404 0 1
Obs = 12430

Table 1: Summary Statistic. Each variable has been statistically calculated with ranges in cost as min and max.

The average wage in the sample was about $41,000 per year, with a standard deviation of nearly $47,000, i.e. the variation of wages across workers is high among Nevadans. Roughly one-in-four workers was a part-time employee, which will prove salient in the latter parts of this analysis, wherein part-time work is shown to be a major determinant of wages. A fourth of individuals in the sample had a bachelor’s degree. Education is relevant to one’s wages; Nevadans are slightly less educated than the national average of 32.5% of people over 25 years of age, even if one only considers members of the sample 25 and older (Ryan & Bauman, 2016). The youngest sampled individual was 16 years of age while the oldest was 91 years old. The average age among Nevada workers was approximately 43 years-old. Nevadans worked in numerous industries in this sample; 9% of them were in food and beverage jobs, a staple of the Las Vegas economy, while just under 2% worked in entertainment.

It is important to note the racial and ethnic demography of Nevada since this still plays a role in wages. Just under 22% of the sample was Hispanic, which is the largest non-white population. 8% were Asian and 6% were black; combined, these numbers show a diverse racial composition which lends itself to analyzing wages in a way that can potentially be applied in a broader context. Perhaps most importantly, approximately half (47%) of workers in the sample were female.

Model

Both the female and male OLS wage models I will use can be specified as the following:

ln(wage) = f(Human Capital, Traits, Children, Industry) + ε

The female model will only include women and the male model will include the rest of the observations. Wage represents income in the last 12 months. Human capital contains variables like age (an instrument for experience) and level of education. Traits are facts about a worker which do not benefit or harm one’s capacity to be productive, such as race or marital status. Children contains the variables for having children and their interactions with the female variable. Occupation contains all variables which specify the industry of which the worker is a part. The symbol ε represents the error term with the usual assumptions. The definitions of some selected variables are listed in Table 2. Due to the scope of this model, not all variables will be specifically listed in this table.

Human capital variables should all have positive coefficients, except for the quadratic age variable. Age is a proxy for experience, since age and experience are highly correlated. The education variables, which include level of education and years of school, capture the remaining human capital of the worker. The degree variables will capture the sheepskin effect, the additional wage premium of completing a degree on top of spending years in school. The base group for the degree binary variables is the population which has not graduated high school.

Trait variables vary widely in their impact and are included to account for factors which are important in existing research, but these do not constitute the focus of this research.

Goldin’s work, among numerous others, points to a persistent wage gap, which will be partially captured by the female binary variable (2014). The difference between estimates on married between models should favor men, coinciding with Waldfogel’s (1998) hypothesis; women, when married, are frequently expected to settle down and raise children. Marriage, thus, may allow men to work comparatively more and thus benefit from the marriage more than women. Both genders should benefit from marriage in general.

The part-time and hours variables control for the possibility that men and women work different amounts. I expect that working more hours will bring higher yearly wages, simply because each hour worked represents an hourly wage (de facto in salaried positions, yet still applicable). In turn, parttime work can be expected to have a wage penalty because part-time workers are legally entitled to fewer benefits than full-time workers. Furthermore, the flexibility of part-time work lowers the opportunity cost of working for the worker, so the worker does not need as much direct monetary compensation to justify spending his or her time at the job.

Variables Definition Expected Signs
Dependent Variable
wage Wages in last 12 months n/a
Independent Variables
Human capital
age Age of worker +
age2 Age of worker, squared
school Years of schooling of worker +
assoc 1 if worker has an associate’s, 0 if not +
colgrad 1 if worker has a bachelor’s, 0 if not +
masters 1 if worker has a master’s, 0 if not +
phd 1 if worker has a PhD, 0 if not +
Traits
black 1 if worker is Black, 0 if not
asian 1 if worker is Asian, 0 if not
hispanic 1 if worker is Hispanic, 0 if not
othernw 1 if worker is another nonwhite race, 0 if not
married 1 if worker is married, 0 if not +
female 1 if worker is a woman, 0 if not
fmarr interaction of female and married
parttime 1 if worker worked <35 hours usually
hours Mean number of hours worker worked per work +
Children
children05 1 if worker only has children 0-5 years old, 0 if not
children617 1 if worker only has children 6-17 years old, 0 if not
children017 1 if worker has children 0-17
femch5 interaction of female and children05
femch617 interaction of female and children617
femch017 interaction of female and children017
Industry
eat 1 if food and beverage worker, 0 if not
ent 1 if entertainment industry worker, 0 if not +

Table 2: Variables, definitions, and expected signs. This model includes 24 independent variables and one dependent variable. The prior expectations for these variables are presented and grouped according to the model equation.

The children binary variables could have positive or negative coefficients in a model and it would not speak to the pay gap per se; their interpretations must be sensitive to the context. If the children binary variables are positive for a model containing only men and negative for a model containing only women, that would provide evidence that having children contributes to the pay gap. The children hypothesis would also be evidenced if the variables differ in magnitude or sign between the male and female models. The base group for the children variables is a person who currently has no children aged 17 or under whatsoever.

By controlling for industry, I can further isolate the effects of gender discrimination; in this model, the effect of working in a higher-paying industry will be included, which means that the resultant gap will not be explained strictly by the fact that there are fewer women in one field or another. I use industry to account for the different sectors of the economy; it is reasonable to assume that people who work in mining might make more money on average than those who work in food and beverage. This part of the model will provide ample material to analyze how much of the gender pay gap is due to differences in industry and how much is due to differences within that industry. Of special interest are entertainment with food and beverage industries, which comprise a significant amount of the economy in Nevada. As for signs, I expect extractive industries, business, finance, entertainment, and food and beverage to have a wage premium. The first three listed are well-known for their high wages; there is a lot of money to be had in oil and minerals, business, and on Wall Street. The latter two are highly valuable in Nevada, so the corresponding wages ought to reflect that value. Despite the small number of workers, entertainment is worth including for its sheer importance to the Las Vegas economy.

Discussion and Conclusion

Coefficients for the variables from both OLS models are listed alongside each other in Table 3. The averages of each variable are also given to clarify the subsequent Blinder-Oaxaca Decomposition results. Due to the scope of the model, not all variables will be included in the table. The regression which had only female workers is denoted by a sub- or superscript ‘F’ and the one which had only male workers is denoted by a sub- or superscript ‘M.’

The dependent variable is the natural log of the worker’s wages in the last 12 months. All coefficients are reported to four decimal places. To determine the percent effects of each variable on wage based on the coefficients in the table, one must enter each coefficient into this formula:

%Δwage/%Δxi = 100(eβi – 1)

where βi is the coefficient on the ith variable and xi is the ith variable.

Variable βF XF βM XM
Observations 5928 5928 6502 6502
Constant 6.7140*** n/a 7.0590*** n/a
Part-time -0.4267*** 0.3339 -0.5646*** 0.2126
Bachelor’s Degree 0.1922*** 0.2667 0.2900*** 0.2350
Married 0.0634*** 0.4992 0.2256*** 0.4973
Age 0.0908*** 42.93 0.0849*** 42.89
Children (<6 yrs) 0.0482 0.0476 0.0147 0.0441
Children (6-17) -0.0197 0.1510 -0.0201 0.1484
Children (0-17) 0.0617 0.0435 -0.0395 0.0417
Food and Beverage -0.0652 .0900 -0.0830** .0974
Entertain 0.0519 .0170 0.0704 .0188
***p<.01,
**p<.05,
*p<.10

Table 3: Regressions of the variables. Each value represents the p-value for the corresponding variable. Entertain stands for entertainment.

It is immediately notable that none of the children variables’ coefficients are significant in either model. While not shown in Table 3, the children variables are also jointly insignificant at conventional levels. This result indicates that whatever effects children have on wages are captured elsewhere. While this does not dismiss the possibility that children make up part of the wage gap, the lack of any direct effects of children on wages whatsoever in this sample does limit the avenues for the family pay gap. Perhaps the most important row in Table 3 is that of the married binary variable. Plainly, the returns to marriage are far greater for men than women. This is important because Waldfogel’s (1998) family pay gap predicts that marriage should be the route by which children affect women’s wages. It is also important that marriage, and not the mere presence of children, should matter since it points to which theoretical cause of the family pay gap is more likely to be true. It is largely the unequal burden of the cost of children, not the presence of children, that counts.

Men and women have almost the same incidence of bachelor’s degrees at 23.50% and 26.67% respectively, but the coefficients are starkly different. Men receive an almost 50% greater wage premium than women do, even controlling for type of degree via the industry variables.

Table 4 shows the results of a Blinder-Oaxaca decomposition by the female binary variable. This analysis will show the magnitude of the gap, broken down by differences in traits and by discrimination.

Blinder Oaxaca Ln wage gap Percent difference (rounded)
Total gap .2710871 34.14%
Due to skills and traits .1616092 17.54%
Due to discrimination .122221 13.00%

Table 4: Decomposition of average gender wage gap. Blinder-Oaxaca Decomposition is present as the different with percentages.

The Nevada gender pay gap in 2014 was for each dollar a man earned, a woman earned about 69 cents. The gap due to observed traits was 17.54%, which means that over half of the magnitude of the wage gap is due to differences in factors such as hours worked and industry. 13%, or two-fifths of the wage gap, however, results from what is termed discrimination in the foundational papers upon which this research is founded (Cotton, 1988). Roughly half of the wage gap found comes from differing treatment of those characteristics between men and women, such as marriage and education differentials (see Table 3).

The family wage gap hypothesis is evidenced in Nevada by the difference in the married variable between male and female models, which points to the family pay gap. The children variables do not appear significant individually or jointly. This is surprising and might indicate something akin to the sheepskin effect, but for children. The number of years of education are not as important as the degree milestones; the presence of children is more important than the ages of the children. The effects of children and family-building could also be captured, as suggested by Goldin (2014), in occupational traits such as part-time work or flexibility of hours. This result contrasts with Budig and England’s (2001) work at the national level, where children do have a statistically significant effect on the margin. Thus, further research shall examine whether Nevada does differ from the nation in the aggregate. One cannot assume that something true in the aggregate is true of its constituent parts. To address the wage gap effectively, one must understand at each level where the wage gap occurs. Furthermore, the very reason to focus on a single state is to account for unmeasured interstate variation, making it imperative to consider how that variation could affect these results in other environments.

Based on these findings, Waldfogel’s family pay gap does seem evident in the wage gap in Nevada (1998). Occupational choices and differing returns to education are also powerful for explaining why there persists such a wide gender wage gap in the Nevada labor market, yet from Budig and England’s (2001) work, one could posit that the occupational factors are related to choices stemming from motherhood. The reasoning behind men and women making different occupation choices (if they are choices) is a matter for future research. It would be important to start by examining children, since they figure greatly in the wages of human beings. It is not yet clear why women face differing returns to education, nor why the fields that women choose are often paid less than those chosen by men. Further analysis could also examine the causal links between children and the factors that have been shown by previous research to have caused the gender wage gap. Lastly, Waldfogel’s (1998) policy suggests expanding maternity leave laws must be further analyzed for effectiveness in closing the family pay gap.

Nevada achieving gender pay equity is a test for whether the United States may also someday achieve this goal. The burden of children seems to weigh disproportionately on one gender, and if that is the case, current and future generations must approach mothers and fathers as equals in parenthood.