The Gender Wage Differential: A Result of Discrimination and Personal Traits
Dr. Eric Dodge
December 5, 2003
Literature Review 6
Data and Model Specifications 11
Econometric Issues 14
Data and Empirical Results 16
Variable Definitions 16
Data Problems 18
Empirical Results 19
Men typically earn a higher labor wage than women. The causes of this wage
differential have been the focus of study and scrutiny by economists, politicians, social
scientists, and many others. The debate has been over whether the differential is due to
differences in skill and personal characteristics or due to discrimination. If the wage
differential is the result of differences in skill between men and women, the gap is
explained; however, when the wage differential is the result of discrimination, the gap is
unexplained and unjustified. This study tests whether the gender wage differential is a
result of discrimination. Furthermore, this study reports on which factors are thought to
be relevant when determining a person’s wage. The results show that there remains an
unexplained gender wage differential of approximately 11%; that is, if a person is male,
he is likely to receive wages 11% higher than if that person were female. Therefore,
according to this study, an unexplained gender wage differential does exist and is
attributed to discrimination in the labor market.
The factors that contribute to a person’s labor wage intrigue and puzzle many
economists, employers, and employees. Because most people earn a labor wage,
determining the factors that maximize potential earnings is imperative. Moreover,
determining why one person may make more money than another while doing the same
task is crucial in understanding the labor market. A plethora of authors, both within the
United States and internationally have studied the gender wage differential in hopes of
uncovering the factors that set men’s wages above women’s wages. Historically, men
tend to earn more money than women, even while doing equal work; though recently, the
implementation of federal laws such as the Equal Pay Act of 1963 and increased women
opportunities have helped to reduce the gender wage gap. However, even with added
political and social action, many researchers conclude that a gender wage gap still
remains. The gender wage gap can be split into two explanations: one, that males,
typically, are more skillfully qualified (ex. education, experience) than women, making
the gender wage gap justified. Second, that after adjusting and accounting for various
personal characteristics and skills that contribute to a person’s wage, the gap is due to
discrimination. The main focus of this paper is to test whether a gender wage differential
exists between full-time, hourly waged individuals based upon discrimination.
In this paper, I report the results of the gender wage gap due to discrimination as
well as the effects certain personal traits have on a person’s wage. All data is gathered
from a monthly survey conducted by the Bureau of Labor Statistics’ Current Population
Survey. Though complex, highly detailed wage decompositions are given in the literary
readings, a more comprehendible decomposition found in Studenmund (2001) is used.
The results indicate that, holding all personal traits constant, there remains a wage
differential of approximately 11%, which is attributed to discrimination. Furthermore,
findings show that acquiring certain skills and characteristics influence a person’s wage
The gender wage differential has long been the object of empirical study and
scrutiny. Many argue that accounting for all explained personal traits that contribute to a
person’s wage, there still remains an unexplained wage gap. In other words, there is
unequal pay for equal work. Others argue that the gender wage gap is a result of
occupational crowding and societal gender roles. That is, women are more heavily
populated in low waged industries as compared to men, and this type of discrimination is
what causes the wage gap. As stated by Fields et al. (1995), there are some important
sub-categories that need to be kept in mind while conducting a gender wage differential
study. The first is that the wage differential could be caused by inter-industry wages.
Second, the wage differential could be caused by the gender distribution among different
industries. Third, the wage differential could be caused by productivity related factors.
These sub-categories, according to Fields et al., are the various causes of the gender wage
The factors causing the gender wage differential, both explained and unexplained,
are difficult to measure, leaving the model highly susceptible to error. Gunderson (1989)
lists some common problems that typically arise in a gender wage differential study. In
all cases, there will be omitted variables; capturing all relevant variables is often times
exhaustive, unattainable, or un-measurable (ex: Each person has different personal
characteristics that may have some determination of their specific wage). Omitted
variables result in biased coefficients that may over or understate the variables effect on
the differential. Another common problem is that the occupational variables are too
broad; significant differentials may actually lie within the industry, therefore, causing the
gender wage differential to be understated. Correlation between two or more
independent variables is also a potential threat. Many variables that determine a person’s
wage are also determinant of each other.
Oaxaca (1973) studies the gender wage differentials in the urban labor market.
The unexplained gender wage gap, discrimination, is measured as the male-female wage
ratio without discrimination ((WM/WF)0) subtracted from the observed male-female ratio
(WM/WF) which is then divided by the male-female ratio without discrimination.
D = WM/WF – (WM/WF)0
The male-female wage differential is classified as:
ln Wm-WF = ∑ biM (XiM-XiF) + ∑ XiF (biM-biF)
The first term on the right represents the differences in characteristics evaluated at
the male return (explained). The last term on the right represents differences in return for
the same work, discrimination (unexplained). Gunderson (1989), Montgomery et al.
(2003), and O’Neill et al. (1993) all use this estimator of the gender wage gap.
Oaxaca includes these independent variables:
Years of schooling completed
Class of worker: government employed, and self-employed with non-union
private wage and salary workers as the reference group
Size of Urban Area
Region: North East, North Central, and West with South as the reference group;
Oaxaca conducts regression analysis by dividing male from female and further
dividing whites from blacks. He then runs two regressions, full-scale wage regressions
(includes occupation and industry) and one including only personal characteristics. Split
regressions are used, because only a broad range of occupations and industries are
included; thus, wage differentials are underestimated in the full-scale model. The results
for the full-scale regression model show that discrimination accounts for 58.4% of the
wage differential for whites and 55.6% for blacks. The results for the personal
characteristics regression show that discrimination accounts for 77.7% of the wage
differential for whites and 93.6% for blacks. Oaxaca concludes that unequal pay for
equal work is not a large part of the gender wage differential, but rather, large numbers of
women tend to be employed in low-waged industries, thus causing the large wage
Cotton’s (1988) cites a flaw within the Oaxaca (1973) study. Oaxaca divides
whites from blacks when conducting the gender wage differentials. However, Oaxaca
attempts to measure the differential with “demand-side sources of discrimination” even
when the sources originate on the supply-side (Cotton 237). For example, in the past,
blacks were discriminated against in education and other “skill acquiring opportunities,”
leaving them at an automatic skill disadvantage even without discrimination (Cotton
238). The same education and skill disadvantage can hold true for women, though as
time goes on, the disadvantage is diminishing. Cotton constructs a revision to the Oaxaca
wage differential equation:
ln WW – ln WB = ∑Bi* (XiW-XiB) + ∑XiW (BiW-Bi*) + ∑XiB (Bi*-BiB)
B* is the non-discriminatory wage structure. The first term on the right represents
the “difference in the current white and black average productivity characteristics
evaluated as the market would in the absence of discrimination. It is therefore the ‘true’
value of the skill component of the wage differential” (Cotton 238). The second and third
terms are discriminatory wage factors based first on whites and then on blacks. The
regression results show that around 49% of the wage differential is due to white males’
skill advantages or 71 cents of the $1.44 wage gap.
Fields et al. (1995) compares the inter-industry gender wage differentials. They
use independent variables of years of education, potential experience (age-yrs of
schooling-6), urban residence (central city vs. other), size of city, region (Northeast,
South, West), marital status, race (non-white vs. white), occupation type, industry. The
results show that 13-19% of the wage gap is due to women in low paying industries, and
0-22% of the overall wage gap is explained by differences in inter-industry wage
coefficients. They also find that female hourly wages differ significantly across
Horrace et al. (2001) cite two shortcomings of Fields et al. (1995). First, Fields et
al. does not state any of the standard errors for the regressions. The standard errors are
critical in determining whether the results are significant. After looking back at Fields et
al., standard errors are not mentioned but the level of significance and type of tests done
are noted . Second, the gender wage gap does not remain constant or unchanged to
“omitted reference groups of the binary variables in the model…the intercept term
catches omitted industry and also the omitted category for any other binary variables in
the specification (ex. race, occupation, marriage)” (Horrace et al. 612). Horrace and
Oaxaca construct an alternative equation:
Gender Wage Gap = (βiF – βiM) + (αF – αM) + xF (θF – θM)
- 10 -
The first term on the right refers to the unexplained difference in wage earnings.
The second term represents the changes in the intercepts of male and female wage
equations. The third term is the changes in the slope parameters. The last term is used to
offset the changes in the intercept changes (second term). Because the gender wage gap
“varies with the mean characteristics of the female workers in each industry,” a vector
average (xF) is used rather than xiF for each individual industry.
Studenmund (2001) does not conduct a gender wage differential study but rather
explains and justifies the methods in setting up such a model. In Studenmund’s example,
an earnings equation is set up based on various factors that influence a person’s wage,
Wagei = B0 + B1Di + B2Xi + Ei
Di = A binomial variable representing whether a person was male or female
Xi = A vector of factors influencing a person’s wage
Observing Di will reveal if a discriminatory wage differential exists between men and
women. When Di is being observed, all other independent variables are accounted for
and held constant, leaving only the unexplained wage gap.
- 11 -
Data and Model Specifications
The main purpose of this study is to test whether a gender wage differential exists
between full-time, hourly waged individuals based upon various personal traits as well as
discrimination. Oaxaca (1973), Fields et al. (1995), and O’Neil et al. (1993) have all
concluded that more women than men are crowded into the low skilled part-time work
sector, which contributes greatly to the appearance of the large gender wage differential.
This study focuses on the gender wage differential between full-time workers, therefore
hoping to capture a more accurate wage differential. The dependent variable is the ith
person’s hourly wage. The independent variables reflect those chosen by past studies.
Gender: Binomial variable ‘1’ given for male and ‘0’ for female.
Potential experience: Classified as age minus years of schooling minus 6.
Marital status: Binomial variable ‘1’ given for married and ‘0’ for single or divorced.
Number of hours worked (week): The number of hours the ith person works in a typical
Worker Class: Binomial variable of ‘1’ given for government worker (state, local,
federal) and ‘0’ for private worker (profit and non-profit).
Geographical region: Binomial variables given for Northeast, Midwest, West with South
as the reference group.
Union member: Binomial variable of ‘1’ given if member of a union, ‘0’ otherwise.
Size of urban area : The city size of the ith person.
Children: The number of children under the age of 18, for the ith person.
Industry and occupation: Binomial variables given for industry occupation groups
consisting of: agriculture, mining, construction, manufacturing, transportation,
- 12 -
communication, utilities and sanitary services, trade, finance-insurance-real estate,
private households, business-auto-repair services, entertainment-recreation services,
medical services, educational services, other professional services, forestry and fisheries,
public administration, and armed forces, with trade held as the reference group.
Level of education completed: The highest level of schooling completed (in years).
The variables noted in the previous paragraph are hypothesized to influence the
gender wage gap, either because of theory or past evidence. Throughout history, women
tend to earn lower wages than men, either because they spend less time in the workforce
due to family duties and childbirth, or because they traditionally have less years of
education than men, though this is changing.
Theoretically, the more experience workers have, the higher their wages. Because
the data on actual experience for each individual is difficult to obtain, an equation for
potential experience is used. As potential experience increases, wages are expected to
increase. In many instances, potential experience may be overstated for women, because
it does not take into consideration the time women leave the labor force due to
pregnancies and/or child rearing. The number of children the ith person has is included
to help adjust this overstatement. As the number of children a woman has increases, the
gender wage gap is expected to be greater due to less work experience
The marital status of a woman sends employers various signals. For most, a
married woman tends to have lower wages, either because of the increased probability
that she will eventually leave the labor force in order to start a family or that she has left
the labor force in the past in order to start a family, decreasing her years of experience.
- 13 -
Furthermore, men who are married tend to send a higher signal than single men to
employers, resulting in higher wages for married men.
The number of hours worked in a typical week is expected to have a positive
influence on a person’s wage, because theoretically, the more hours a person works, the
more he/she gets paid.
In worker class (government versus private), the effects on a person’s wage is
ambiguous. There are not many literary sources that favor one sector over the other.
Both private and government positions offer low and high waged jobs.
The geography or region a person works at is likely to increase or decrease the
wage as seen from past studies. Oaxaca (1973) found that both men and women living in
the South earn less than those living in the Northeast, West, and Midwest.
Union membership is expected to have a positive effect on a person’s wage.
Unions push for higher wages and better working conditions as well as equality among
As years of schooling completed increases, a person’s wage is likely to increase
also. Theoretically, the more schooling, the more prepared and knowledgeable a person
is, making he/she more valuable to the employer.
Though many authors choose to use the Oaxaca wage decomposition, this study is
modeled after Studenmund (2001). Studenmund (2001) uses an understandable linear
wage equation to deduce unexplained differences between male and female wages.
Furthermore, Studenmund thoroughly explains and simplifies how the model is set up
- 14 -
The wage equation:
lnWagei = B0 + B1Di + B2Xi + Ei
Di = A binomial variable representing whether a person was male or female
Xi = A vector of factors influencing a person’s wage (independent variables)
Although Studenmund does not log the wage in his decomposition, many of the other
authors do in order to show a percent change. Observing Di will reveal if a
discriminatory wage differential exists between men and women. When Di is being
observed, all other independent variables are accounted for and held constant, leaving
only the unexplained wage gap.
Because many of the independent variables mentioned above correlate to one
another, there are some econometric problems that are likely to arise in this model. Even
if perfect multicollinearity does not exist, imperfect multicollinearity may exist. The
years of schooling completed affects a number of independent variables both directly and
indirectly such as: potential experience (age-years of schooling-6) and type of job
(physical labor is associated with those who have less years of schooling). Marital status
is also correlated to children and number of hours worked a week. Women who are
married have a higher probability of having children and working fewer hours a week due
to household responsibilities.
Heteroskedasticity may pose a problem in this model. Because this is a cross
sectional model, the possibility of error term variances fluctuating is highly probable.
The variance for potential experience is most likely not constant. Those individuals who
are younger could either have minimal potential experience because of extended
schooling or could have a higher amount of experience, if they start work right out of
- 15 -
high school. This variance in potential experience would also affect the wage variance
- 16 -
Data and Empirical Results
The data used in this study comes from the Bureau of Labor Statistics (BLS)
Current Population Survey (CPS). The CPS gathers monthly information from a sample
of 60,000 households and uses this information to report on the nation’s unemployment
and employment “classified by age (over 16), sex, race, and a variety of other
characteristics” (United States). The data for this study is drawn from the February 2002
CPS, and individuals are filtered out based on certain characteristics, resulting in a
sample size of 4,394 people. Individual observations filtered according to specific
variables or characteristics may be downloaded through the United States Census
Bureau’s and The Department of Labor’s Data Ferrett1 system.
Again, the purpose of this study is to see whether a gender wage differential exists
in the labor market, assuming various demographics and characteristics that influence a
person’s wage are held constant. The sample includes 4,394 observations, and because
the data is a comparison between various individuals, the data is cross-sectional. Only
full-time, hourly waged workers are included in this model. The dependent variable is
the log hourly wage of the ith person. The independent variables listed below are those
suspected to have an influence on a person’s wage.
Gender: Binomial variable of 1 given for male and 0 for female.
Potential experience: Because potential experience is hard to calculate for each
individual person surveyed, a standard estimation equation is used. It is classified
as age minus years of schooling minus 6.
Marital status: Binomial variable of 1 given for a person who is married
(including spouse present, spouse absent but not separated), and a 0 is given for a
The Data Ferrett is an electronic “search system [used] for extracting and tabulating data across
heterogeneous statistical data sources” (Data Ferrett). In this project, the time period for a specific CPS
survey was chosen followed by specific variables that each observation needed to have. The filtered out
data was downloaded into an Excel file.
- 17 -
person who is not married (including widowed, separated, divorced, single, and
Number of hours worked (week): The number of hours worked in a typical work
week at a full-time job (hours greater than or equal to 35).
Worker Class: Binomial variable of 0 given for a government worker (state,
local, federal), and a variable of 1 given for a private worker (profit and non-
Geography region: The region was split into separate variables so that analysis
based on binomial variables could be done.
Midwest: Those residing in the Midwest were given a binomial variable
value of 1, all else given a 0.
North: Those residing in the North were given a binomial variable value
of 1, all else given a 0.
West: Those residing in the West were given a binomial variable value of
1, all else given a 0.
South (left out of regression equation): Those residing in the South were
given a binomial variable value of 1, all else given a 0.
Union member: Binomial variable of 1 given if the ith person was a member of a
union and a 0 given otherwise.
Size of urban area: The Data Ferrett extraction tool does not give specific city
size values but rather ranges of the city size (ex. 100,000-249,999 people).
Therefore the average of the range was taken for each observation so that the
model may be less susceptible to error. The values used for city size include:
0: Not in a metro area
174,999.5: 100,000-249,999 people
374,999.5: 250,000-499,999 people
749,999.5: 500,000-999,999 people
1,749,999.5: 1,000,000-2,499,999 people
3,749,999.5: 2,500,000-4,999,999 people
5,000,000: 5,000,000 or greater
Children: The number of children under the age of 18, for the ith individual.
Industry and occupation: Binomial variables are split into separate variables and
then given a value of 1 if the ith person works in that industry and a value of 0
otherwise (Set up like the Geography region variable). Variables consist of:
agriculture, mining, construction, manufacturing-durable and non-durable goods,
transportation, communication, utilities and sanitary services, trade (wholesale
and retail), finance-insurance-real estate, private households and personal
services, business-auto-repair services, entertainment-recreation services, medical
services including hospitals, educational services, other professional services,
- 18 -
forestry and fisheries, public administration, and armed forces. The variable
Trade is left out of the regression model to avoid multicollinearity and to use as a
Years of Schooling: Values obtained from the Data Ferrett were given in terms of
highest level of educational attainment and not in terms of years of schooling.
Therefore, based upon the average student that starts schooling at the age of 5,
values were calculated as an equivalent to the degree given.
Years of schooling Education level attained
6 Less than Elementary
9 Less than H.S (7-11)
12 H.S diploma or GED
13 Some College, no diploma
14 Associate Degree
16 Bachelor’s Degree
18 Master’s Degree
21 Doctorate Degree
Data obtained from the Data Ferrett program did not always provide the exact
variable values needed for the study; therefore, variables were re-coded and/or estimated.
The variable “years of schooling completed” was not given in the form of a number but
instead was given based upon level of diploma achieved. Therefore, the “years of
schooling” variable was re-coded based upon averages (see “years of schooling”
variable). Potential experience was not available through the Data Ferrett, and so values
were calculated via Microsoft Excel. The equation used for potential experience was
age minus “years of schooling” minus six. The “city size” variable was also given in
ranges, and these ranges were re-coded to the mean of each range (see “city size”
After the Data Ferrett had filtered out individuals based upon the variables listed
above, there were two observations without an hourly wage listed. These observations
were removed from the sample (sample size reduced from 4396 to 4394) .
- 19 -
The data included a sample size of 4,394 individuals. There were 1,937 women
and 2,457 men in the sample. The average age of the individuals in the sample was 36.2
years old. The Northeast had 883 people in the sample, the South had 1,165 people
represented, the West had 1,200 people represented, and the Midwest had 1,146 people
represented in the sample. The average number of children for the ith person was 1.
There were 714 people in unions and 3,680 people not in unions. There were 1,924
people married while 2,470 people were not married. Government workers consisted of
579 people while 3,815 people were employed by private organizations. The average
years of schooling completed was 12.5 years, and the average years of potential
experience was 17.7 years. The average city size was 1,718,861 people. In depth
descriptive statistics of the data can be found in Table 1: Descriptive Statistics.
The hypothesized sign of “hours worked” was positive, because the more hours a
person works, the more he/she earns. All the geographical regions were expected to be
positive as compared to the South; the South is generally characterized by farming and
industries that typically are classified as low-waged and low-skilled. The “children”
variable was expected to be positive for males, because if a male has children, it
generally sends the signal to employers that he is responsible. Likewise, if a male is
married, it sends the signal to employers that he is responsible. However for women, the
expected sign of “children” is expected to be negative. Women who have children
typically lose time in the labor force due to pregnancy absences, decreasing a person’s
wage (Oaxaca). Unions push for higher wages and better employment benefits, making
the hypothesized “union” variable positive. “Years of schooling” is hypothesized to be
- 20 -
positive, because if you have more years of schooling, you are more skilled, more
knowledgeable, and it sends a positive signal to employers. “Potential experience”
should be positive, because as a person gains experience, he/she is more knowledgeable
and capable in that field of work.
Originally, the model was set up with the variable “age” included as an
independent variable. The variable “industry” was set up as one variable and was not re-
coded into independent sub-variables, and the “city size” variable was in terms of dummy
variables (ex. 1 given for city size range of 100,000-249,999 people). OLS regression
output can be seen in Appendix 1: Preliminary Model. Surprisingly, “years of
schooling” affected the ith person’s wage negatively at -5.27%, and “potential
experience” affected the ith person’s wage negatively by -11%. All the variables except
“years of schooling” were significant at the 10% level, which was surprising because
years of schooling should be an important factor in a person’s wage. The variance
inflation factors for “age,” “years of schooling,” and “potential experience” were
extremely large, hinting that multicollinearity was an issue.
The model was re-structured in an attempt to fix the problems stated above.
“Age” was removed as an independent variable since it was included in the “potential
experience” equation. Moreover, theoretically, as age increases, potential experience will
also increase. The goal of removing “age” from the model was to reduce or eliminate
multicollinearity from the model. “Industry” was re-coded into separate sub-variables so
that each type of industry could be analyzed based on a dummy variable value of either 0
or 1. “City size” was re-coded into actual values, based upon the mean of the city size
range, instead of using multiple dummy variables. By recoding these variables, more
- 21 -
specific and accurate analysis could be achieved. The results of the OLS regression are
shown in Appendix 2.
A few of the significant variables that affect a person’s wage are shown below:
Variables β T V.I.F
Gender .107 8.492 1.344
Male = 1
Northeast .109 6.609 1.497
Midwest .0786 5.215 1.498
West .0826 5.527 1.514
Children -.00343 -.552 1.736
Union .151 9.413 1.204
Marital Status .0766 5.313 1.750
Years of Schooling .057 23.54 1.071
Potential Experience .00723 13.828 1.040
City Size 1.844E-08 6.621 1.106
Worker Class -.0552 -2.194 2.477
The wage differential between men and women, holding all other variables constant, is
around 11%. That is, males are likely to make 11% more than women, holding all other
variables constant. Those who are in a union are likely to make around 15% more than
those who are not in unions. With one more year of schooling, a person’s wage is likely
to increase by 5.7%. Those who live in the Northeast are likely to make around 11%
- 22 -
more than those in the South; those who live in the Midwest are likely to make around
7.9% more than those in the South; those who live in the West are likely to earn around
8.3% more than those in the South. Those who are married are expected to have a 7.7%
higher wage than those who are not married, and those with one more year of potential
experience are expected to have a wage increase of 7.2%. These results are all significant
at the 1% level and are consistent with theoretical belief.
Most variables are highly significant in regards to the t-statistic at the 1%
significance level. However, whether a person’s wage is affected if they work for a
government office or a private office is insignificant as well as the agricultural, private
household, public administration, entertainment/recreational services, and other
professions sector of the “industry” variable. The variable “children” is also insignificant
at the 1% significance level but has a negative co-efficient as predicted. Each variable
passes the variance inflation factor test of less than 5, fixing the problem from the
preliminary model. Furthermore, the “years of schooling” and “potential experience”
variables are now positive and significant at the 1% level.
The model has an adjusted R2 of .34, meaning that the independent variables
predict 34% of the fluctuations of the log wage variable. The F-statistic in this model is
77 and the critical value is 1. Therefore, the null is rejected and the overall fit of the
model is significant. The model has a Durbin-Watson of .645 with a lower critical value
is 1.53 and an upper critical value of 1.83, assuming positive auto-correlation. Because
this is not a time series model, auto-correlation generally should not be a problem. The
traces of positive auto-correlation are due to the arrangement of the hourly wages listed in
- 23 -
Heteroskedasticity with the variable “potential experience” was initially thought
to be a potential problem. Theoretically, those who have a low potential experience
could be making a wide range of hourly wages. Those who are young and begin working
right after high school may make a low hourly wage. While those who are young and
have completed some level of college or higher education will be making a high wage.
As age increases so does potential experience and wage. The variance of wages is likely
to decrease in size. This is graphically seen in Appendix 3: Heteroskedasticity Graph 1.
The Park test was conducted to test whether the presence of heteroskedasticity existed.
The test resulted in a t-stat of 4.441, leading to the rejection of the null at all levels of
significance and the conclusion that there is heteroskedasticity. This can be seen in
Appendix 3: Heteroskedasticity Park Test.
One fix to heteroskedasticity is weighted least squares, in which case the potential
variable causing the problem, in this case “age” since it is essentially quite similar to
potential experience, is divided throughout the entire OLS equation. Appendix 3:
Weighted Least Squares shows the results of this test. Potential experience increases in
significance to 14.151, and the coefficient states that if potential experience increases by
one year, the wage will go up by .75%, which is not that great of a change. The rest of
the variables are quite similar and no variables lost significance.
- 24 -
The hypothesis that an unexplained gender wage differential still exists was
verified by the results of the tests based on the February 2002 Current Population Survey.
Overall, when the independent variables were held constant, men’s wages were still
around 11% higher than women’s wages. Traditionally, when women decide to get
married and start a family, they take a break from the labor force, therefore lowering their
on the job training, becoming less valuable to employers. Because most women take this
path, a woman entering into the labor force automatically sends a negative signal to
employers based upon stereotype. These negative signals as well as various other social
stereotypes could be the cause of this unexplained wage gap. The results also indicate
that various factors such as union membership and increased years of schooling
dramatically influence a person’s wage. Though many factors were listed as
determinants of a person’s wage, there are still many factors that are not included, either
due to inaccessibility or ignorance.
Because the study was based on a sample of only one month, future studies could
incorporate a sample of multiple months in order to gauge a more accurate wage
differential. Breaking down the industries into inter-industries and occupations and then
analyzing the gender wage gap would also be an interesting future study, especially since
literary studies have shown that inter-industry wage gaps are quite large. The true value
of the gender wage differential will never be completely accurate. There are too many
factors, specific personal and job traits, and social issues that influence a person’s wage.
However, with further study on this issue, people will at least gain a better understanding
of the causes that most influence a person’s wage.